diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU')
55 files changed, 5911 insertions, 2607 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll b/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll index 2d7ef2c..98fbbe1 100644 --- a/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll +++ b/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll @@ -169,6 +169,6 @@ attributes #1 = { nounwind } ;. ; HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) } -; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll index 9e24023..ebbeab9 100644 --- a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll +++ b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll @@ -146,9 +146,9 @@ define void @no_free_vgprs_at_agpr_to_agpr_copy(float %v0, float %v1) #0 { ; GFX908-NEXT: ;;#ASMSTART ; GFX908-NEXT: ; copy ; GFX908-NEXT: ;;#ASMEND -; GFX908-NEXT: v_accvgpr_read_b32 v32, a2 +; GFX908-NEXT: v_accvgpr_read_b32 v39, a2 ; GFX908-NEXT: s_nop 1 -; GFX908-NEXT: v_accvgpr_write_b32 a3, v32 +; GFX908-NEXT: v_accvgpr_write_b32 a3, v39 ; GFX908-NEXT: ;;#ASMSTART ; GFX908-NEXT: ; use a3 v[0:31] ; GFX908-NEXT: ;;#ASMEND @@ -437,9 +437,9 @@ define void @v32_asm_def_use(float %v0, float %v1) #4 { ; GFX908-NEXT: ; copy ; GFX908-NEXT: ;;#ASMEND ; GFX908-NEXT: s_nop 7 -; GFX908-NEXT: v_accvgpr_read_b32 v33, a2 +; GFX908-NEXT: v_accvgpr_read_b32 v35, a2 ; GFX908-NEXT: s_nop 1 -; GFX908-NEXT: v_accvgpr_write_b32 a3, v33 +; GFX908-NEXT: v_accvgpr_write_b32 a3, v35 ; GFX908-NEXT: ;;#ASMSTART ; GFX908-NEXT: ; use a3 v[0:31] ; GFX908-NEXT: ;;#ASMEND @@ -1045,9 +1045,9 @@ define void @no_free_vgprs_at_sgpr_to_agpr_copy(float %v0, float %v1) #0 { ; GFX908-NEXT: ;;#ASMSTART ; GFX908-NEXT: ; copy ; GFX908-NEXT: ;;#ASMEND -; GFX908-NEXT: v_accvgpr_read_b32 v32, a2 +; GFX908-NEXT: v_accvgpr_read_b32 v39, a2 ; GFX908-NEXT: s_nop 1 -; GFX908-NEXT: v_accvgpr_write_b32 a3, v32 +; GFX908-NEXT: v_accvgpr_write_b32 a3, v39 ; GFX908-NEXT: ;;#ASMSTART ; GFX908-NEXT: ; use a3 v[0:31] ; GFX908-NEXT: ;;#ASMEND diff --git a/llvm/test/CodeGen/AMDGPU/agpr-copy-propagation.mir b/llvm/test/CodeGen/AMDGPU/agpr-copy-propagation.mir index a42cf43..7e82382d 100644 --- a/llvm/test/CodeGen/AMDGPU/agpr-copy-propagation.mir +++ b/llvm/test/CodeGen/AMDGPU/agpr-copy-propagation.mir @@ -40,8 +40,8 @@ body: | ; GFX908: liveins: $agpr0 ; GFX908-NEXT: {{ $}} ; GFX908-NEXT: renamable $vgpr0 = COPY renamable $agpr0, implicit $exec - ; GFX908-NEXT: renamable $agpr1 = COPY renamable $vgpr0, implicit $exec - ; GFX908-NEXT: renamable $agpr2 = COPY renamable $vgpr0, implicit $exec + ; GFX908-NEXT: renamable $agpr1 = COPY $agpr0, implicit $exec + ; GFX908-NEXT: renamable $agpr2 = COPY $agpr0, implicit $exec ; GFX908-NEXT: S_ENDPGM 0, implicit $vgpr0, implicit $agpr1, implicit $agpr2 ; ; GFX90A-LABEL: name: do_not_propagate_agpr_to_agpr diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll index 664dfa2..2ad6e68 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll @@ -1,103 +1,166 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --check-globals all --version 4 ; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx90a -passes=amdgpu-attributor %s | FileCheck %s +; Shrink result attribute list by preventing use of most attributes. +define internal void @use_most() { +; CHECK-LABEL: define internal void @use_most( +; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { +; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [256 x i8], align 1, addrspace(5) +; CHECK-NEXT: [[ALLOCA_CAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA]] to ptr +; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.amdgcn.workitem.id.z() +; CHECK-NEXT: [[TMP4:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x() +; CHECK-NEXT: [[TMP5:%.*]] = call i32 @llvm.amdgcn.workgroup.id.y() +; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.amdgcn.workgroup.id.z() +; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.amdgcn.cluster.id.x() +; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.amdgcn.cluster.id.y() +; CHECK-NEXT: [[TMP12:%.*]] = call i32 @llvm.amdgcn.cluster.id.z() +; CHECK-NEXT: [[TMP7:%.*]] = call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() +; CHECK-NEXT: [[TMP8:%.*]] = call ptr addrspace(4) @llvm.amdgcn.queue.ptr() +; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.amdgcn.dispatch.id() +; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.amdgcn.lds.kernel.id() +; CHECK-NEXT: [[IMPLICIT_ARG_PTR:%.*]] = call ptr addrspace(4) @llvm.amdgcn.implicitarg.ptr() +; CHECK-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr [[ALLOCA_CAST]], ptr addrspace(4) [[IMPLICIT_ARG_PTR]], i64 256, i1 false) +; CHECK-NEXT: ret void +; + %alloca = alloca [256 x i8], addrspace(5) + %alloca.cast = addrspacecast ptr addrspace(5) %alloca to ptr + call i32 @llvm.amdgcn.workitem.id.x() + call i32 @llvm.amdgcn.workitem.id.y() + call i32 @llvm.amdgcn.workitem.id.z() + call i32 @llvm.amdgcn.workgroup.id.x() + call i32 @llvm.amdgcn.workgroup.id.y() + call i32 @llvm.amdgcn.workgroup.id.z() + call i32 @llvm.amdgcn.cluster.id.x() + call i32 @llvm.amdgcn.cluster.id.y() + call i32 @llvm.amdgcn.cluster.id.z() + call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() + call ptr addrspace(4) @llvm.amdgcn.queue.ptr() + call i64 @llvm.amdgcn.dispatch.id() + call i32 @llvm.amdgcn.lds.kernel.id() + %implicit.arg.ptr = call ptr addrspace(4) @llvm.amdgcn.implicitarg.ptr() + call void @llvm.memcpy.p0.p4(ptr %alloca.cast, ptr addrspace(4) %implicit.arg.ptr, i64 256, i1 false) + ret void +} + define amdgpu_kernel void @kernel_uses_asm_virtreg() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg( -; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { +; CHECK-SAME: ) #[[ATTR1:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "a"(i32 poison) + call void @use_most() ret void } define amdgpu_kernel void @kernel_uses_asm_virtreg_def() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: [[DEF:%.*]] = call i32 asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; %def = call i32 asm sideeffect "; def $0", "=a"() + call void @use_most() ret void } define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: [[DEF:%.*]] = call i64 asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; %def = call i64 asm sideeffect "; def $0", "={a[0:1]}"() + call void @use_most() ret void } define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "v,a"(i32 poison, i32 poison) + call void @use_most() ret void } define amdgpu_kernel void @kernel_uses_non_agpr_asm() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_non_agpr_asm( -; CHECK-SAME: ) #[[ATTR1:[0-9]+]] { +; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "v"(i32 poison) + call void @use_most() ret void } define amdgpu_kernel void @kernel_uses_asm_physreg() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "{a0}"(i32 poison) + call void @use_most() ret void } define amdgpu_kernel void @kernel_uses_asm_physreg_tuple() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_tuple( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) + call void @use_most() ret void } define void @func_uses_asm_virtreg_agpr() { ; CHECK-LABEL: define void @func_uses_asm_virtreg_agpr( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "a"(i32 poison) + call void @use_most() ret void } define void @func_uses_asm_physreg_agpr() { ; CHECK-LABEL: define void @func_uses_asm_physreg_agpr( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "{a0}"(i32 poison) + call void @use_most() ret void } define void @func_uses_asm_physreg_agpr_tuple() { ; CHECK-LABEL: define void @func_uses_asm_physreg_agpr_tuple( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) + call void @use_most() ret void } @@ -105,99 +168,119 @@ declare void @unknown() define amdgpu_kernel void @kernel_calls_extern() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern( -; CHECK-SAME: ) #[[ATTR2:[0-9]+]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void @unknown() +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void @unknown() + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern_marked_callsite( -; CHECK-SAME: ) #[[ATTR2]] { -; CHECK-NEXT: call void @unknown() #[[ATTR6:[0-9]+]] +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void @unknown() #[[ATTR10:[0-9]+]] +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void @unknown() #0 + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_indirect(ptr %indirect) { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect( -; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR2]] { +; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: call void [[INDIRECT]]() +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void %indirect() + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_indirect_marked_callsite(ptr %indirect) { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect_marked_callsite( -; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR2]] { -; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR6]] +; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR1]] { +; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR10]] +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void %indirect() #0 + call void @use_most() ret void } define amdgpu_kernel void @kernel_transitively_uses_agpr_asm() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_transitively_uses_agpr_asm( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void @func_uses_asm_physreg_agpr() + call void @use_most() ret void } define void @empty() { ; CHECK-LABEL: define void @empty( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; + call void @use_most() ret void } define void @also_empty() { ; CHECK-LABEL: define void @also_empty( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_empty() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_empty( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: call void @empty() +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void @empty() + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr( -; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-SAME: ) #[[ATTR1]] { ; CHECK-NEXT: call void @empty() ; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void @empty() call void @func_uses_asm_physreg_agpr() + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_generic_intrinsic(ptr %ptr0, ptr %ptr1, i64 %size) { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_generic_intrinsic( -; CHECK-SAME: ptr [[PTR0:%.*]], ptr [[PTR1:%.*]], i64 [[SIZE:%.*]]) #[[ATTR1]] { +; CHECK-SAME: ptr [[PTR0:%.*]], ptr [[PTR1:%.*]], i64 [[SIZE:%.*]]) #[[ATTR0]] { ; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[PTR0]], ptr [[PTR1]], i64 [[SIZE]], i1 false) +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; call void @llvm.memcpy.p0.p0.i64(ptr %ptr0, ptr %ptr1, i64 %size, i1 false) + call void @use_most() ret void } @@ -205,31 +288,35 @@ declare <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float, float, <32 x float> define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32(ptr addrspace(1) %out, float %a, float %b, <32 x float> %c) { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32( -; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]], <32 x float> [[C:%.*]]) #[[ATTR1]] { +; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]], <32 x float> [[C:%.*]]) #[[ATTR0]] { ; CHECK-NEXT: [[RESULT:%.*]] = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float [[A]], float [[B]], <32 x float> [[C]], i32 0, i32 0, i32 0) ; CHECK-NEXT: store <32 x float> [[RESULT]], ptr addrspace(1) [[OUT]], align 128 +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; %result = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float %a, float %b, <32 x float> %c, i32 0, i32 0, i32 0) store <32 x float> %result, ptr addrspace(1) %out + call void @use_most() ret void } define amdgpu_kernel void @kernel_calls_workitem_id_x(ptr addrspace(1) %out) { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_workitem_id_x( -; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR1]] { +; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { ; CHECK-NEXT: [[RESULT:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() ; CHECK-NEXT: store i32 [[RESULT]], ptr addrspace(1) [[OUT]], align 4 +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; %result = call i32 @llvm.amdgcn.workitem.id.x() store i32 %result, ptr addrspace(1) %out + call void @use_most() ret void } define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { ; CHECK-LABEL: define amdgpu_kernel void @indirect_calls_none_agpr( -; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty ; CHECK-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]] @@ -244,21 +331,476 @@ define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { ; CHECK: 5: ; CHECK-NEXT: unreachable ; CHECK: 6: +; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; %fptr = select i1 %cond, ptr @empty, ptr @also_empty call void %fptr() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, i32} asm sideeffect "; def $0", "=a,=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_1() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_1( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, <2 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, <2 x i32>} asm sideeffect "; def $0", "=a,=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, <2 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, <2 x i32>} asm sideeffect "; def $0", "=a,=v"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(ptr poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call ptr asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call ptr asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call <2 x ptr> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <2 x ptr> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, i32} asm sideeffect "; def $0", "={a0},={a[4:5]}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a4}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber_tuple() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_tuple( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a[10:13]}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber_oob() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_oob( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a256}"() + call void @use_most() ret void } +define amdgpu_kernel void @kernel_uses_asm_clobber_max() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_max( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a255}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg_oob() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_oob( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "{a256}"(i32 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <32 x i32> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(<32 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <32 x i32> asm sideeffect "; use $0", "=a,a"(<32 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @vreg_use_exceeds_register_file() { +; CHECK-LABEL: define amdgpu_kernel void @vreg_use_exceeds_register_file( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(<257 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @vreg_def_exceeds_register_file() { +; CHECK-LABEL: define amdgpu_kernel void @vreg_def_exceeds_register_file( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call <257 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <257 x i32> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @multiple() { +; CHECK-LABEL: define amdgpu_kernel void @multiple( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { <16 x i32>, <8 x i32>, <8 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {<16 x i32>, <8 x i32>, <8 x i32>} asm sideeffect "; def $0", "=a,=a,=a,a,a,a"(<4 x i32> splat (i32 0), <8 x i32> splat (i32 1), i64 999) + call void @use_most() + ret void +} + +define amdgpu_kernel void @earlyclobber_0() { +; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_0( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call <8 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <8 x i32> asm sideeffect "; def $0", "=&a,a"(i32 0) + call void @use_most() + ret void +} + +define amdgpu_kernel void @earlyclobber_1() { +; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_1( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { <8 x i32>, <16 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call { <8 x i32>, <16 x i32 > } asm sideeffect "; def $0, $1", "=&a,=&a,a,a"(i32 0, <16 x i32> splat (i32 1)) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1, $2", "{a16},a,a"(i32 poison, <8 x i32> poison, <16 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[TMP1:%.*]] = call { i32, <8 x i32>, <16 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call {i32, <8 x i32>, <16 x i32>} asm sideeffect "; def $0, $1, $2", "={a16},=a,=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[TMP1:%.*]] = call { i32, <16 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call {i32, <16 x i32>} asm sideeffect "; def $0, $1, $2", "={a16},=a,a"(<8 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @mixed_physreg_vreg_tuples_0() { +; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_0( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "{a[1:4]},a"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @mixed_physreg_vreg_tuples_1() { +; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_1( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,{a[0:3]}"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_raises_limit() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_raises_limit( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,{a[5:8]}"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +; FIXME: This should require 9. We cannot allocate an a128 at a0. +define amdgpu_kernel void @physreg_tuple_alignment_raises_limit() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_tuple_alignment_raises_limit( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,{a[1:4]}"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @align3_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @align3_virtreg( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,a"(<3 x i32> poison, <3 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @align3_align4_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @align3_align4_virtreg( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,a"(<3 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @align2_align4_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @align2_align4_virtreg( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,a"(<2 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_a55() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55( +; CHECK-SAME: ) #[[ATTR3:[0-9]+]] { +; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META0:![0-9]+]], i32 0) +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !0, i32 0) + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_v55() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_v55( +; CHECK-SAME: ) #[[ATTR4:[0-9]+]] { +; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META1:![0-9]+]], i32 0) +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !1, i32 0) + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_a55_57() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55_57( +; CHECK-SAME: ) #[[ATTR3]] { +; CHECK-NEXT: call void @llvm.write_register.i96(metadata [[META2:![0-9]+]], i96 0) +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !2, i96 0) + ret void +} + +define amdgpu_kernel void @kernel_uses_read_register_a55(ptr addrspace(1) %ptr) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_register_a55( +; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: [[REG:%.*]] = call i32 @llvm.read_register.i32(metadata [[META0]]) +; CHECK-NEXT: store i32 [[REG]], ptr addrspace(1) [[PTR]], align 4 +; CHECK-NEXT: ret void +; + %reg = call i32 @llvm.read_register.i64(metadata !0) + store i32 %reg, ptr addrspace(1) %ptr + ret void +} + +define amdgpu_kernel void @kernel_uses_read_volatile_register_a55(ptr addrspace(1) %ptr) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_volatile_register_a55( +; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: [[REG:%.*]] = call i32 @llvm.read_volatile_register.i32(metadata [[META0]]) +; CHECK-NEXT: store i32 [[REG]], ptr addrspace(1) [[PTR]], align 4 +; CHECK-NEXT: ret void +; + %reg = call i32 @llvm.read_volatile_register.i64(metadata !0) + store i32 %reg, ptr addrspace(1) %ptr + ret void +} + +define amdgpu_kernel void @kernel_uses_read_register_a56_59(ptr addrspace(1) %ptr) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_register_a56_59( +; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: [[REG:%.*]] = call i128 @llvm.read_register.i128(metadata [[META3:![0-9]+]]) +; CHECK-NEXT: store i128 [[REG]], ptr addrspace(1) [[PTR]], align 8 +; CHECK-NEXT: ret void +; + %reg = call i128 @llvm.read_register.i64(metadata !3) + store i128 %reg, ptr addrspace(1) %ptr + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_out_of_bounds_a256() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_out_of_bounds_a256( +; CHECK-SAME: ) #[[ATTR3]] { +; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META4:![0-9]+]], i32 0) +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !4, i32 0) + ret void +} attributes #0 = { "amdgpu-agpr-alloc"="0" } + +!0 = !{!"a55"} +!1 = !{!"v55"} +!2 = !{!"a[55:57]"} +!3 = !{!"a[56:59]"} +!4 = !{!"a256"} + +;. +; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2:[0-9]+]] = { convergent nocallback nofree nosync nounwind willreturn memory(none) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR6:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR7:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(read) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR8:[0-9]+]] = { nounwind "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR9:[0-9]+]] = { nocallback nounwind "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR10]] = { "amdgpu-agpr-alloc"="0" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2:[0-9]+]] = { "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3:[0-9]+]] = { convergent nocallback nofree nosync nounwind willreturn memory(none) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR5:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="0" } +; CHECK: [[META0]] = !{!"a55"} +; CHECK: [[META1]] = !{!"v55"} +; CHECK: [[META2]] = !{!"a[55:57]"} +; CHECK: [[META3]] = !{!"a[56:59]"} +; CHECK: [[META4]] = !{!"a256"} ;. diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll index fb566e5..9283bd5 100644 --- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll +++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll @@ -691,29 +691,29 @@ attributes #6 = { "enqueued-block" } ;. ; ATTRIBUTOR_HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; ATTRIBUTOR_HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR14]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR15:[0-9]+]] = { nounwind "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR16]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR17]] = { nounwind sanitize_address "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR18]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR14]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR15]] = { nounwind "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR16]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR17]] = { nounwind sanitize_address "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR18]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR19:[0-9]+]] = { nounwind sanitize_address "amdgpu-no-implicitarg-ptr" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR20:[0-9]+]] = { "enqueued-block" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR21]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "enqueued-block" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR21]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "enqueued-block" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR22]] = { "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR23]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR23]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR24]] = { nounwind } ; ATTRIBUTOR_HSA: attributes #[[ATTR25]] = { "enqueued-block" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll index 484ff77..8554485 100644 --- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll +++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll @@ -474,19 +474,19 @@ attributes #1 = { nounwind } ; AKF_HSA: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 500} ;. ; HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. ; HSA: [[META0]] = !{i32 1, i32 3, i32 4, i32 10} ; HSA: [[META1]] = !{i32 1, i32 5, i32 6, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll index 2efe024..e2a2deb 100644 --- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll +++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll @@ -294,13 +294,13 @@ attributes #1 = { nounwind } ;. ; CHECK: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; CHECK: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR6]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR7]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR8]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR9]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR9]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll b/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll index aaedb85..e67d7fdb 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll @@ -3,6 +3,8 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-TRUE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-FAKE16 %s define i8 @atomic_load_monotonic_i8(ptr addrspace(3) %ptr) { ; CI-LABEL: atomic_load_monotonic_i8: @@ -33,6 +35,14 @@ define i8 @atomic_load_monotonic_i8(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u8 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i8: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u8 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i8, ptr addrspace(3) %ptr monotonic, align 1 ret i8 %load } @@ -66,6 +76,14 @@ define i8 @atomic_load_monotonic_i8_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u8 v0, v0 offset:16 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i8_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u8 v0, v0 offset:16 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 16 %load = load atomic i8, ptr addrspace(3) %gep monotonic, align 1 ret i8 %load @@ -100,6 +118,14 @@ define i16 @atomic_load_monotonic_i16(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i16: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i16, ptr addrspace(3) %ptr monotonic, align 2 ret i16 %load } @@ -133,6 +159,14 @@ define i16 @atomic_load_monotonic_i16_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i16_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 offset:32 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i16, ptr addrspace(3) %ptr, i16 16 %load = load atomic i16, ptr addrspace(3) %gep monotonic, align 2 ret i16 %load @@ -160,6 +194,14 @@ define i32 @atomic_load_monotonic_i32(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b32 v0, v0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i32, ptr addrspace(3) %ptr monotonic, align 4 ret i32 %load } @@ -186,6 +228,14 @@ define i32 @atomic_load_monotonic_i32_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b32 v0, v0 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i32_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i32, ptr addrspace(3) %ptr, i32 16 %load = load atomic i32, ptr addrspace(3) %gep monotonic, align 4 ret i32 %load @@ -213,6 +263,14 @@ define i64 @atomic_load_monotonic_i64(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i64: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i64, ptr addrspace(3) %ptr monotonic, align 8 ret i64 %load } @@ -239,6 +297,14 @@ define i64 @atomic_load_monotonic_i64_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i64_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i64, ptr addrspace(3) %ptr, i32 16 %load = load atomic i64, ptr addrspace(3) %gep monotonic, align 8 ret i64 %load @@ -266,6 +332,14 @@ define float @atomic_load_monotonic_f32_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b32 v0, v0 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f32_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds float, ptr addrspace(3) %ptr, i32 16 %load = load atomic float, ptr addrspace(3) %gep monotonic, align 4 ret float %load @@ -293,6 +367,14 @@ define double @atomic_load_monotonic_f64_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f64_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds double, ptr addrspace(3) %ptr, i32 16 %load = load atomic double, ptr addrspace(3) %gep monotonic, align 8 ret double %load @@ -320,6 +402,14 @@ define ptr @atomic_load_monotonic_p0i8_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_p0i8_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds ptr, ptr addrspace(3) %ptr, i32 16 %load = load atomic ptr, ptr addrspace(3) %gep monotonic, align 8 ret ptr %load @@ -347,6 +437,14 @@ define ptr addrspace(3) @atomic_load_monotonic_p3i8_offset(ptr addrspace(3) %ptr ; GFX11-NEXT: ds_load_b32 v0, v0 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_p3i8_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds ptr addrspace(3), ptr addrspace(3) %ptr, i32 16 %load = load atomic ptr addrspace(3), ptr addrspace(3) %gep monotonic, align 4 ret ptr addrspace(3) %load @@ -381,6 +479,14 @@ define i16 @atomic_load_monotonic_f16(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f16: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic half, ptr addrspace(3) %ptr monotonic, align 2 %ret = bitcast half %load to i16 ret i16 %ret @@ -415,6 +521,14 @@ define i16 @atomic_load_monotonic_f16_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f16_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 offset:32 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds half, ptr addrspace(3) %ptr, i32 16 %load = load atomic half, ptr addrspace(3) %gep monotonic, align 2 %ret = bitcast half %load to i16 @@ -450,6 +564,14 @@ define i16 @atomic_load_monotonic_bf16(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_bf16: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic bfloat, ptr addrspace(3) %ptr monotonic, align 2 %ret = bitcast bfloat %load to i16 ret i16 %ret @@ -484,6 +606,14 @@ define i16 @atomic_load_monotonic_bf16_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_bf16_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 offset:32 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds bfloat, ptr addrspace(3) %ptr, i32 16 %load = load atomic bfloat, ptr addrspace(3) %gep monotonic, align 2 %ret = bitcast bfloat %load to i16 @@ -491,3 +621,5 @@ define i16 @atomic_load_monotonic_bf16_offset(ptr addrspace(3) %ptr) { } ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: ; GCN: {{.*}} +; GFX1250-FAKE16: {{.*}} +; GFX1250-TRUE16: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll b/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll index c2bb4f00..31065f2 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll @@ -3,6 +3,8 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-TRUE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-FAKE16 %s define void @atomic_store_monotonic_i8(ptr addrspace(3) %ptr, i8 %val) { ; CI-LABEL: atomic_store_monotonic_i8: @@ -41,6 +43,26 @@ define void @atomic_store_monotonic_i8(ptr addrspace(3) %ptr, i8 %val) { ; GFX11-FAKE16-NEXT: ds_store_b8 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_i8: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b8 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b8_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_i8: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i8 %val, 2 store atomic i8 %val, ptr addrspace(3) %ptr monotonic, align 1 store atomic i8 %val1, ptr addrspace(3) %ptr monotonic, align 1 @@ -84,6 +106,26 @@ define void @atomic_store_monotonic_offset_i8(ptr addrspace(3) %ptr, i8 %val) { ; GFX11-FAKE16-NEXT: ds_store_b8 v0, v2 offset:16 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_i8: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b8 v0, v1 offset:8 +; GFX1250-TRUE16-NEXT: ds_store_b8_d16_hi v0, v1 offset:16 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_i8: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v1 offset:8 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v2 offset:16 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i8 %val, 2 %gep_1 = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 8 %gep_2 = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 16 @@ -129,6 +171,26 @@ define void @atomic_store_monotonic_i16(ptr addrspace(3) %ptr, i16 %val) { ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_i16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_i16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i16 %val, 2 store atomic i16 %val, ptr addrspace(3) %ptr monotonic, align 2 store atomic i16 %val1, ptr addrspace(3) %ptr monotonic, align 2 @@ -172,6 +234,26 @@ define void @atomic_store_monotonic_offset_i16(ptr addrspace(3) %ptr, i16 %val) ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_i16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_i16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i16 %val, 2 %gep = getelementptr inbounds i16, ptr addrspace(3) %ptr, i16 16 store atomic i16 %val, ptr addrspace(3) %gep monotonic, align 2 @@ -201,6 +283,14 @@ define void @atomic_store_monotonic_i32(ptr addrspace(3) %ptr, i32 %val) { ; GFX11-NEXT: ds_store_b32 v0, v1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_i32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_store_b32 v0, v1 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] store atomic i32 %val, ptr addrspace(3) %ptr monotonic, align 4 ret void } @@ -227,6 +317,14 @@ define void @atomic_store_monotonic_offset_i32(ptr addrspace(3) %ptr, i32 %val) ; GFX11-NEXT: ds_store_b32 v0, v1 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_offset_i32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_store_b32 v0, v1 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i32, ptr addrspace(3) %ptr, i32 16 store atomic i32 %val, ptr addrspace(3) %gep monotonic, align 4 ret void @@ -254,6 +352,15 @@ define void @atomic_store_monotonic_i64(ptr addrspace(3) %ptr, i64 %val) { ; GFX11-NEXT: ds_store_b64 v0, v[1:2] ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_i64: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 +; GFX1250-NEXT: ds_store_b64 v0, v[2:3] +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] store atomic i64 %val, ptr addrspace(3) %ptr monotonic, align 8 ret void } @@ -280,6 +387,15 @@ define void @atomic_store_monotonic_offset_i64(ptr addrspace(3) %ptr, i64 %val) ; GFX11-NEXT: ds_store_b64 v0, v[1:2] offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_offset_i64: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 +; GFX1250-NEXT: ds_store_b64 v0, v[2:3] offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i64, ptr addrspace(3) %ptr, i64 16 store atomic i64 %val, ptr addrspace(3) %gep monotonic, align 8 ret void @@ -322,6 +438,26 @@ define void @atomic_store_monotonic_f16(ptr addrspace(3) %ptr, i16 %arg.val) { ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_f16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_f16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val = bitcast i16 %arg.val to half %val1 = bitcast i16 %arg.val1 to half @@ -367,6 +503,26 @@ define void @atomic_store_monotonic_offset_f16(ptr addrspace(3) %ptr, i16 %arg.v ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_f16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_f16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val1 = bitcast i16 %arg.val1 to half %val = bitcast i16 %arg.val to half @@ -413,6 +569,26 @@ define void @atomic_store_monotonic_bf16(ptr addrspace(3) %ptr, i16 %arg.val) { ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_bf16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_bf16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val1 = bitcast i16 %arg.val1 to bfloat %val = bitcast i16 %arg.val to bfloat @@ -458,6 +634,26 @@ define void @atomic_store_monotonic_offset_bf16(ptr addrspace(3) %ptr, i16 %arg. ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_bf16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_bf16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val1 = bitcast i16 %arg.val1 to bfloat %val = bitcast i16 %arg.val to bfloat diff --git a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll index f63dd6e..c90611f 100644 --- a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll +++ b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll @@ -147,10 +147,10 @@ define amdgpu_kernel void @call_calls_intrin_ascast_cc_kernel(ptr addrspace(3) % attributes #0 = { "amdgpu-no-flat-scratch-init" } ;. -; GFX9: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } ; GFX9: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx900" } ;. -; GFX10: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } ; GFX10: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1010" } ;. ; GFX9: [[META0]] = !{i32 1, i32 5, i32 6, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll index 60cd252..c005695a 100644 --- a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll +++ b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll @@ -723,7 +723,7 @@ define void @also_empty() { define amdgpu_kernel void @indirect_call_known_callees(i1 %cond) { ; GFX9-LABEL: define amdgpu_kernel void @indirect_call_known_callees( -; GFX9-SAME: i1 [[COND:%.*]]) #[[ATTR3:[0-9]+]] { +; GFX9-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { ; GFX9-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty ; GFX9-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty ; GFX9-NEXT: br i1 [[TMP1]], label %[[BB2:.*]], label %[[BB3:.*]] @@ -741,7 +741,7 @@ define amdgpu_kernel void @indirect_call_known_callees(i1 %cond) { ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define amdgpu_kernel void @indirect_call_known_callees( -; GFX10-SAME: i1 [[COND:%.*]]) #[[ATTR3:[0-9]+]] { +; GFX10-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { ; GFX10-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty ; GFX10-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty ; GFX10-NEXT: br i1 [[TMP1]], label %[[BB2:.*]], label %[[BB3:.*]] @@ -767,13 +767,13 @@ declare i32 @llvm.amdgcn.workgroup.id.x() define void @use_intrinsic_workitem_id_x() { ; GFX9-LABEL: define void @use_intrinsic_workitem_id_x( -; GFX9-SAME: ) #[[ATTR5:[0-9]+]] { +; GFX9-SAME: ) #[[ATTR4:[0-9]+]] { ; GFX9-NEXT: [[VAL:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() ; GFX9-NEXT: store volatile i32 [[VAL]], ptr addrspace(1) null, align 4 ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define void @use_intrinsic_workitem_id_x( -; GFX10-SAME: ) #[[ATTR5:[0-9]+]] { +; GFX10-SAME: ) #[[ATTR4:[0-9]+]] { ; GFX10-NEXT: [[VAL:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() ; GFX10-NEXT: store volatile i32 [[VAL]], ptr addrspace(1) null, align 4 ; GFX10-NEXT: ret void @@ -803,12 +803,12 @@ define amdgpu_kernel void @use_intrinsic_workitem_id_x_cc_kernel() { define void @call_use_intrinsic_workitem_id_x() { ; GFX9-LABEL: define void @call_use_intrinsic_workitem_id_x( -; GFX9-SAME: ) #[[ATTR5]] { +; GFX9-SAME: ) #[[ATTR4]] { ; GFX9-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define void @call_use_intrinsic_workitem_id_x( -; GFX10-SAME: ) #[[ATTR5]] { +; GFX10-SAME: ) #[[ATTR4]] { ; GFX10-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX10-NEXT: ret void ; @@ -818,12 +818,12 @@ define void @call_use_intrinsic_workitem_id_x() { define amdgpu_kernel void @call_use_intrinsic_workitem_id_x_cc_kernel() { ; GFX9-LABEL: define amdgpu_kernel void @call_use_intrinsic_workitem_id_x_cc_kernel( -; GFX9-SAME: ) #[[ATTR5]] { +; GFX9-SAME: ) #[[ATTR4]] { ; GFX9-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define amdgpu_kernel void @call_use_intrinsic_workitem_id_x_cc_kernel( -; GFX10-SAME: ) #[[ATTR5]] { +; GFX10-SAME: ) #[[ATTR4]] { ; GFX10-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX10-NEXT: ret void ; @@ -851,12 +851,12 @@ define amdgpu_kernel void @calls_intrin_ascast_cc_kernel(ptr addrspace(3) %ptr) define amdgpu_kernel void @with_inline_asm() { ; GFX9-LABEL: define amdgpu_kernel void @with_inline_asm( -; GFX9-SAME: ) #[[ATTR3]] { +; GFX9-SAME: ) #[[ATTR0]] { ; GFX9-NEXT: call void asm sideeffect " ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define amdgpu_kernel void @with_inline_asm( -; GFX10-SAME: ) #[[ATTR3]] { +; GFX10-SAME: ) #[[ATTR0]] { ; GFX10-NEXT: call void asm sideeffect " ; GFX10-NEXT: ret void ; @@ -865,19 +865,17 @@ define amdgpu_kernel void @with_inline_asm() { } ;. -; GFX9: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; GFX9: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } ; GFX9: attributes #[[ATTR2]] = { "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; GFX9: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; GFX9: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx900" } -; GFX9: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx900" } +; GFX9: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } ;. -; GFX10: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } -; GFX10: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } ; GFX10: attributes #[[ATTR2]] = { "target-cpu"="gfx1010" "uniform-work-group-size"="false" } -; GFX10: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } -; GFX10: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1010" } -; GFX10: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1010" } +; GFX10: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } ;. ; GFX9: [[META0]] = !{i32 2, i32 10} ; GFX9: [[META1]] = !{i32 1, i32 2, i32 3, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/bf16.ll b/llvm/test/CodeGen/AMDGPU/bf16.ll index 6b5647e..4b14dc6 100644 --- a/llvm/test/CodeGen/AMDGPU/bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/bf16.ll @@ -7,11 +7,9 @@ ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1010 | FileCheck %s -check-prefixes=GFX10 ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 | FileCheck %s -check-prefixes=GFX11,GFX11TRUE16 ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 | FileCheck %s -check-prefixes=GFX11,GFX11FAKE16 -; xUN: llc < %s -mtriple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 | FileCheck %s -check-prefixes=GFX1250,GFX1250TRUE16 +; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 | FileCheck %s -check-prefixes=GFX1250,GFX1250TRUE16 ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 | FileCheck %s -check-prefixes=GFX1250,GFX1250FAKE16 -; FIXME: real-true16 version of gfx1250 test fails - define void @test_load_store(ptr addrspace(1) %in, ptr addrspace(1) %out) { ; GCN-LABEL: test_load_store: ; GCN: ; %bb.0: @@ -2393,15 +2391,25 @@ define void @test_store_fpimm(ptr addrspace(1) %ptr0, ptr addrspace(1) %ptr1) { ; GFX11FAKE16-NEXT: global_store_b16 v[2:3], v5, off ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_store_fpimm: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_mov_b32_e32 v4, 0x3f80 -; GFX1250-NEXT: v_mov_b32_e32 v5, 0x4228 -; GFX1250-NEXT: global_store_b16 v[0:1], v4, off -; GFX1250-NEXT: global_store_b16 v[2:3], v5, off -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_store_fpimm: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v4.l, 0x3f80 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v4.h, 0x4228 +; GFX1250TRUE16-NEXT: global_store_b16 v[0:1], v4, off +; GFX1250TRUE16-NEXT: global_store_d16_hi_b16 v[2:3], v4, off +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_store_fpimm: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v4, 0x3f80 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v5, 0x4228 +; GFX1250FAKE16-NEXT: global_store_b16 v[0:1], v4, off +; GFX1250FAKE16-NEXT: global_store_b16 v[2:3], v5, off +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] store bfloat 1.0, ptr addrspace(1) %ptr0 store bfloat 42.0, ptr addrspace(1) %ptr1 ret void @@ -3796,13 +3804,21 @@ define amdgpu_gfx void @test_inreg_arg_store(bfloat inreg %in, ptr addrspace(1) ; GFX11FAKE16-NEXT: global_store_b16 v[0:1], v2, off ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_inreg_arg_store: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_mov_b32_e32 v2, s4 -; GFX1250-NEXT: global_store_b16 v[0:1], v2, off -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_inreg_arg_store: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, s4 +; GFX1250TRUE16-NEXT: global_store_b16 v[0:1], v2, off +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_inreg_arg_store: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v2, s4 +; GFX1250FAKE16-NEXT: global_store_b16 v[0:1], v2, off +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] store bfloat %in, ptr addrspace(1) %out ret void } @@ -3866,12 +3882,20 @@ define bfloat @test_byval(ptr addrspace(5) byval(bfloat) %bv, bfloat %val) { ; GFX11FAKE16-NEXT: scratch_store_b16 off, v0, s32 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_byval: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: scratch_store_b16 off, v0, s32 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_byval: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l +; GFX1250TRUE16-NEXT: scratch_store_b16 off, v1, s32 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_byval: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: scratch_store_b16 off, v0, s32 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] store bfloat %val, ptr addrspace(5) %bv %retval = load bfloat, ptr addrspace(5) %bv ret bfloat %retval @@ -6708,27 +6732,50 @@ define { <32 x i32>, bfloat } @test_overflow_stack(bfloat %a, <32 x i32> %b) { ; GFX11FAKE16-NEXT: scratch_store_b16 v0, v1, off offset:128 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_overflow_stack: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x2 -; GFX1250-NEXT: scratch_load_b32 v33, off, s32 offset:8 -; GFX1250-NEXT: scratch_load_b32 v32, off, s32 offset:4 -; GFX1250-NEXT: scratch_load_b32 v31, off, s32 -; GFX1250-NEXT: s_clause 0x5 -; GFX1250-NEXT: scratch_store_b128 v0, v[22:25], off offset:80 -; GFX1250-NEXT: scratch_store_b128 v0, v[18:21], off offset:64 -; GFX1250-NEXT: scratch_store_b128 v0, v[14:17], off offset:48 -; GFX1250-NEXT: scratch_store_b128 v0, v[10:13], off offset:32 -; GFX1250-NEXT: scratch_store_b128 v0, v[6:9], off offset:16 -; GFX1250-NEXT: scratch_store_b128 v0, v[2:5], off -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: s_clause 0x2 -; GFX1250-NEXT: scratch_store_b128 v0, v[30:33], off offset:112 -; GFX1250-NEXT: scratch_store_b128 v0, v[26:29], off offset:96 -; GFX1250-NEXT: scratch_store_b16 v0, v1, off offset:128 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_overflow_stack: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: s_clause 0x2 +; GFX1250TRUE16-NEXT: scratch_load_b32 v33, off, s32 offset:8 +; GFX1250TRUE16-NEXT: scratch_load_b32 v32, off, s32 offset:4 +; GFX1250TRUE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250TRUE16-NEXT: s_clause 0x3 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[22:25], off offset:80 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[18:21], off offset:64 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[14:17], off offset:48 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[10:13], off offset:32 +; GFX1250TRUE16-NEXT: s_clause 0x1 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[6:9], off offset:16 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[2:5], off +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250TRUE16-NEXT: s_clause 0x2 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[30:33], off offset:112 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[26:29], off offset:96 +; GFX1250TRUE16-NEXT: scratch_store_b16 v0, v1, off offset:128 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_overflow_stack: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: s_clause 0x2 +; GFX1250FAKE16-NEXT: scratch_load_b32 v33, off, s32 offset:8 +; GFX1250FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:4 +; GFX1250FAKE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250FAKE16-NEXT: s_clause 0x5 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[22:25], off offset:80 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[18:21], off offset:64 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[14:17], off offset:48 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[10:13], off offset:32 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[6:9], off offset:16 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[2:5], off +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250FAKE16-NEXT: s_clause 0x2 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[30:33], off offset:112 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[26:29], off offset:96 +; GFX1250FAKE16-NEXT: scratch_store_b16 v0, v1, off offset:128 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %ins.0 = insertvalue { <32 x i32>, bfloat } poison, <32 x i32> %b, 0 %ins.1 = insertvalue { <32 x i32>, bfloat } %ins.0 ,bfloat %a, 1 ret { <32 x i32>, bfloat } %ins.1 @@ -10726,15 +10773,29 @@ define bfloat @v_fadd_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fadd_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fadd_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fadd_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fadd bfloat %a, %b ret bfloat %op } @@ -15268,15 +15329,26 @@ define bfloat @v_fadd_bf16_fpimm_0(bfloat %arg0) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fadd_bf16_fpimm_0: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v0, 1.0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fadd_bf16_fpimm_0: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, 1.0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fadd_bf16_fpimm_0: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, 1.0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %add = fadd bfloat %arg0, 1.0 ret bfloat %add } @@ -15382,15 +15454,26 @@ define bfloat @v_fadd_bf16_fpimm_1(bfloat %arg0) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fadd_bf16_fpimm_1: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v0, 0x42280000, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fadd_bf16_fpimm_1: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, 0x42280000, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fadd_bf16_fpimm_1: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, 0x42280000, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %add = fadd bfloat %arg0, 42.0 ret bfloat %add } @@ -15507,15 +15590,29 @@ define bfloat @v_fsub_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fsub_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fsub_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fsub_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fsub bfloat %a, %b ret bfloat %op } @@ -15931,21 +16028,37 @@ define <3 x bfloat> @v_fsub_v3bf16(<3 x bfloat> %a, <3 x bfloat> %b) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fsub_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v3, 16, v3 -; GFX1250-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 -; GFX1250-NEXT: v_and_b32_e32 v5, 0xffff0000, v0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_dual_sub_f32 v4, v5, v4 :: v_dual_lshlrev_b32 v1, 16, v1 -; GFX1250-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v4 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fsub_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v3, 16, v3 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v1, v1, v3 +; GFX1250TRUE16-NEXT: v_dual_sub_f32 v3, v5, v4 :: v_dual_sub_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v3 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fsub_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_dual_sub_f32 v4, v5, v4 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250FAKE16-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v4 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fsub <3 x bfloat> %a, %b ret <3 x bfloat> %op } @@ -16371,12 +16484,26 @@ define bfloat @v_fmul_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fmul_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_fma_mixlo_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fmul_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fmul_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_fma_mixlo_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fmul bfloat %a, %b ret bfloat %op } @@ -21012,31 +21139,60 @@ define bfloat @v_fdiv_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fdiv_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_div_scale_f32 v2, null, v1, v1, v0 -; GFX1250-NEXT: v_rcp_f32_e32 v3, v2 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fma_f32 v4, -v2, v3, 1.0 -; GFX1250-NEXT: v_fmac_f32_e32 v3, v4, v3 -; GFX1250-NEXT: v_div_scale_f32 v4, vcc_lo, v0, v1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX1250-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fmac_f32_e32 v5, v6, v3 -; GFX1250-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX1250-NEXT: v_div_fixup_f32 v0, v2, v1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fdiv_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v0.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, v2.l +; GFX1250TRUE16-NEXT: v_div_scale_f32 v1, null, v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_rcp_f32_e32 v3, v1 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_fma_f32 v4, -v1, v3, 1.0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fmac_f32_e32 v3, v4, v3 +; GFX1250TRUE16-NEXT: v_div_scale_f32 v4, vcc_lo, v2, v0, v2 +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v6, -v1, v5, v4 +; GFX1250TRUE16-NEXT: v_fmac_f32_e32 v5, v6, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v1, -v1, v5, v4 +; GFX1250TRUE16-NEXT: v_div_fmas_f32 v1, v1, v3, v5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_div_fixup_f32 v0, v1, v0, v2 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fdiv_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_div_scale_f32 v2, null, v1, v1, v0 +; GFX1250FAKE16-NEXT: v_rcp_f32_e32 v3, v2 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fma_f32 v4, -v2, v3, 1.0 +; GFX1250FAKE16-NEXT: v_fmac_f32_e32 v3, v4, v3 +; GFX1250FAKE16-NEXT: v_div_scale_f32 v4, vcc_lo, v0, v1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX1250FAKE16-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fmac_f32_e32 v5, v6, v3 +; GFX1250FAKE16-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX1250FAKE16-NEXT: v_div_fixup_f32 v0, v2, v1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fdiv bfloat %a, %b ret bfloat %op } @@ -21092,12 +21248,19 @@ define bfloat @v_fabs_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fabs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0x7fff, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fabs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 0x7fff, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fabs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.fabs.bf16(bfloat %a) ret bfloat %op } @@ -21198,12 +21361,19 @@ define bfloat @v_fneg_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_xor_b32_e32 v0, 0x8000, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fneg_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_xor_b32_e32 v0, 0x8000, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fneg_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fneg_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v0, 0x8000, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fneg bfloat %a ret bfloat %op } @@ -21317,12 +21487,19 @@ define bfloat @v_fneg_fabs_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_or_b32_e32 v0, 0x8000, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fneg_fabs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_or_b32_e32 v0, 0x8000, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fneg_fabs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_or_b16 v0.l, 0x8000, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fneg_fabs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_or_b32_e32 v0, 0x8000, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %fabs = call bfloat @llvm.fabs.bf16(bfloat %a) %op = fneg bfloat %fabs ret bfloat %op @@ -21511,15 +21688,29 @@ define bfloat @v_minnum_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_minnum_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_min_num_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_minnum_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_min_num_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_minnum_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_min_num_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.minnum.bf16(bfloat %a, bfloat %b) ret bfloat %op } @@ -26073,15 +26264,29 @@ define bfloat @v_maxnum_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_maxnum_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_max_num_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_maxnum_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_max_num_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_maxnum_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_max_num_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.maxnum.bf16(bfloat %a, bfloat %b) ret bfloat %op } @@ -30764,12 +30969,19 @@ define bfloat @v_sqrt_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sqrt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_sqrt_bf16_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sqrt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_sqrt_bf16_e32 v0.l, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sqrt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_sqrt_bf16_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.sqrt.bf16(bfloat %a) ret bfloat %op } @@ -30877,15 +31089,26 @@ define bfloat @v_ldexp_bf16_i32(bfloat %a, i32 %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_ldexp_bf16_i32: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_ldexp_bf16_i32: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v2, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_ldexp_bf16_i32: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.ldexp.bf16.i32(bfloat %a, i32 %b) ret bfloat %op } @@ -31005,16 +31228,28 @@ define { bfloat, i16 } @v_frexp_bf16_i16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_frexp_bf16_i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_frexp_mant_f32_e32 v0, v1 -; GFX1250-NEXT: v_frexp_exp_i32_f32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_frexp_bf16_i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_frexp_mant_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_frexp_exp_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_frexp_bf16_i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_frexp_mant_f32_e32 v0, v1 +; GFX1250FAKE16-NEXT: v_frexp_exp_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call { bfloat, i16 } @llvm.frexp.bf16.i16(bfloat %a) ret { bfloat, i16 } %op } @@ -31254,31 +31489,58 @@ define bfloat @v_log_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_log_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) -; GFX1250-NEXT: v_log_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: v_mul_f32_e32 v1, 0x3f317217, v0 -; GFX1250-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fma_f32 v2, 0x3f317217, v0, -v1 -; GFX1250-NEXT: v_fmamk_f32 v2, v0, 0x3377d1cf, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v1, v1, v2 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 0x41b17218, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_log_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 32, vcc_lo +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v1, 0x3f317217, v0 +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v2, 0x3f317217, v0, -v1 +; GFX1250TRUE16-NEXT: v_fmamk_f32 v2, v0, 0x3377d1cf, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x41b17218, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_log_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250FAKE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v1, 0x3f317217, v0 +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fma_f32 v2, 0x3f317217, v0, -v1 +; GFX1250FAKE16-NEXT: v_fmamk_f32 v2, v0, 0x3377d1cf, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x41b17218, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.log.bf16(bfloat %a) ret bfloat %op } @@ -31439,12 +31701,19 @@ define bfloat @v_log2_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_log2_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_log_bf16_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_log2_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_log_bf16_e32 v0.l, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_log2_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_log_bf16_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.log2.bf16(bfloat %a) ret bfloat %op } @@ -31679,31 +31948,58 @@ define bfloat @v_log10_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_log10_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) -; GFX1250-NEXT: v_log_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: v_mul_f32_e32 v1, 0x3e9a209a, v0 -; GFX1250-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fma_f32 v2, 0x3e9a209a, v0, -v1 -; GFX1250-NEXT: v_fmamk_f32 v2, v0, 0x3284fbcf, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v1, v1, v2 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 0x411a209b, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_log10_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 32, vcc_lo +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v1, 0x3e9a209a, v0 +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v2, 0x3e9a209a, v0, -v1 +; GFX1250TRUE16-NEXT: v_fmamk_f32 v2, v0, 0x3284fbcf, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x411a209b, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_log10_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250FAKE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v1, 0x3e9a209a, v0 +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fma_f32 v2, 0x3e9a209a, v0, -v1 +; GFX1250FAKE16-NEXT: v_fmamk_f32 v2, v0, 0x3284fbcf, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x411a209b, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.log10.bf16(bfloat %a) ret bfloat %op } @@ -31946,34 +32242,65 @@ define bfloat @v_exp_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_exp_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: s_mov_b32 s0, 0x3fb8aa3b -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v1 -; GFX1250-NEXT: v_rndne_f32_e32 v3, v2 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] -; GFX1250-NEXT: s_mov_b32 s0, 0x32a5705f -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_sub_f32_e32 v2, v2, v3 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] -; GFX1250-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc2ce8ed0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_add_f32_e32 v0, v2, v0 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v3 -; GFX1250-NEXT: v_exp_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo -; GFX1250-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x42b17218, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_exp_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x3fb8aa3b +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v1 +; GFX1250TRUE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc2ce8ed0, v1 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v3, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v4, v2 +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x32a5705f +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v3 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v4 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250TRUE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x42b17218, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_exp_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x3fb8aa3b +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v1 +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v3, v2 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x32a5705f +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v2, v2, v3 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc2ce8ed0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v3 +; GFX1250FAKE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x42b17218, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.exp.bf16(bfloat %a) ret bfloat %op } @@ -32138,12 +32465,19 @@ define bfloat @v_exp2_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_exp2_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_exp_bf16_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_exp2_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_exp_bf16_e32 v0.l, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_exp2_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_exp_bf16_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.exp2.bf16(bfloat %a) ret bfloat %op } @@ -32382,34 +32716,65 @@ define bfloat @v_exp10_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_exp10_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: s_mov_b32 s0, 0x40549a78 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mul_f32_e32 v2, 0x40549a78, v1 -; GFX1250-NEXT: v_rndne_f32_e32 v3, v2 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] -; GFX1250-NEXT: s_mov_b32 s0, 0x33979a37 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_sub_f32_e32 v2, v2, v3 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] -; GFX1250-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc23369f4, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_add_f32_e32 v0, v2, v0 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v3 -; GFX1250-NEXT: v_exp_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo -; GFX1250-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x421a209b, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_exp10_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x40549a78 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v2, 0x40549a78, v1 +; GFX1250TRUE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc23369f4, v1 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v3, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v4, v2 +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x33979a37 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v3 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v4 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250TRUE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x421a209b, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_exp10_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x40549a78 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v2, 0x40549a78, v1 +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v3, v2 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x33979a37 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v2, v2, v3 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc23369f4, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v3 +; GFX1250FAKE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x421a209b, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.exp10.bf16(bfloat %a) ret bfloat %op } @@ -32517,15 +32882,26 @@ define bfloat @v_ceil_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_ceil_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ceil_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_ceil_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_ceil_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_ceil_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ceil_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.ceil.bf16(bfloat %a) ret bfloat %op } @@ -32633,15 +33009,26 @@ define bfloat @v_trunc_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_trunc_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_trunc_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_trunc_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_trunc_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_trunc_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_trunc_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.trunc.bf16(bfloat %a) ret bfloat %op } @@ -32749,15 +33136,26 @@ define bfloat @v_rint_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_rint_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_rndne_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_rint_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_rint_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.rint.bf16(bfloat %a) ret bfloat %op } @@ -32865,15 +33263,26 @@ define bfloat @v_nearbyint_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_nearbyint_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_rndne_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_nearbyint_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_nearbyint_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.nearbyint.bf16(bfloat %a) ret bfloat %op } @@ -33031,23 +33440,42 @@ define bfloat @v_round_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_round_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_trunc_f32_e32 v1, v0 -; GFX1250-NEXT: v_sub_f32_e32 v2, v0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_ge_f32_e64 s0, |v2|, 0.5 -; GFX1250-NEXT: v_cndmask_b32_e64 v2, 0, 1.0, s0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_bfi_b32 v0, 0x7fffffff, v2, v0 -; GFX1250-NEXT: v_add_f32_e32 v0, v1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_round_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_trunc_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v2, v1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_ge_f32_e64 s0, |v2|, 0.5 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v2, 0, 1.0, s0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_bfi_b32 v1, 0x7fffffff, v2, v1 +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_round_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_trunc_f32_e32 v1, v0 +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v2, v0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_ge_f32_e64 s0, |v2|, 0.5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v2, 0, 1.0, s0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_bfi_b32 v0, 0x7fffffff, v2, v0 +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.round.bf16(bfloat %a) ret bfloat %op } @@ -33155,15 +33583,26 @@ define bfloat @v_roundeven_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_roundeven_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_rndne_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_roundeven_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_roundeven_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.roundeven.bf16(bfloat %a) ret bfloat %op } @@ -33271,15 +33710,26 @@ define bfloat @v_floor_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_floor_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_floor_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_floor_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_floor_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_floor_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_floor_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.floor.bf16(bfloat %a) ret bfloat %op } @@ -33385,15 +33835,26 @@ define bfloat @v_canonicalize_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_canonicalize_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_max_num_f32_e32 v0, v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_canonicalize_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_max_num_f32_e32 v0, v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_canonicalize_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_max_num_f32_e32 v0, v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.canonicalize.bf16(bfloat %a) ret bfloat %op } @@ -33535,15 +33996,28 @@ define i1 @v_fcmp_oeq_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_oeq_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_oeq_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_oeq_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp oeq bfloat %a, %b ret i1 %op } @@ -33630,15 +34104,28 @@ define i1 @v_fcmp_ogt_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ogt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_gt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ogt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ogt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ogt bfloat %a, %b ret i1 %op } @@ -33725,15 +34212,28 @@ define i1 @v_fcmp_oge_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_oge_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_oge_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_ge_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_oge_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp oge bfloat %a, %b ret i1 %op } @@ -33820,15 +34320,28 @@ define i1 @v_fcmp_olt_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_olt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_olt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_lt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_olt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp olt bfloat %a, %b ret i1 %op } @@ -33915,15 +34428,28 @@ define i1 @v_fcmp_ole_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ole_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ole_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_le_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ole_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ole bfloat %a, %b ret i1 %op } @@ -34010,15 +34536,28 @@ define i1 @v_fcmp_one_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_one_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_lg_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_one_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_lg_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_one_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_lg_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp one bfloat %a, %b ret i1 %op } @@ -34105,15 +34644,28 @@ define i1 @v_fcmp_uno_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_uno_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_uno_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_uno_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp uno bfloat %a, %b ret i1 %op } @@ -34200,15 +34752,28 @@ define i1 @v_fcmp_ueq_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ueq_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nlg_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ueq_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nlg_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ueq_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nlg_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ueq bfloat %a, %b ret i1 %op } @@ -34295,15 +34860,28 @@ define i1 @v_fcmp_ugt_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ugt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nle_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ugt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nle_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ugt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nle_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ugt bfloat %a, %b ret i1 %op } @@ -34390,15 +34968,28 @@ define i1 @v_fcmp_uge_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_uge_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_uge_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_uge_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp uge bfloat %a, %b ret i1 %op } @@ -34485,15 +35076,28 @@ define i1 @v_fcmp_ult_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ult_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ult_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nge_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ult_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ult bfloat %a, %b ret i1 %op } @@ -34580,15 +35184,28 @@ define i1 @v_fcmp_ule_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ule_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ule_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ule_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ule bfloat %a, %b ret i1 %op } @@ -34675,15 +35292,28 @@ define i1 @v_fcmp_une_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_une_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_neq_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_une_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_neq_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_une_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_neq_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp une bfloat %a, %b ret i1 %op } @@ -34790,14 +35420,24 @@ define i16 @v_fptosi_bf16_to_i16(bfloat %x) { ; GFX11FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_bf16_to_i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_bf16_to_i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_bf16_to_i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi bfloat %x to i16 ret i16 %op } @@ -34899,18 +35539,31 @@ define <2 x i16> @v_fptosi_v2bf16_to_v2i16(<2 x bfloat> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_v2bf16_to_v2i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_v2bf16_to_v2i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v1, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_v2bf16_to_v2i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi <2 x bfloat> %x to <2 x i16> ret <2 x i16> %op } @@ -35032,19 +35685,33 @@ define <3 x i16> @v_fptosi_v3bf16_to_v3i16(<3 x bfloat> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v0, v0, v2, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_v3bf16_to_v3i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: v_perm_b32 v0, v0, v2, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_v3bf16_to_v3i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v2.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_v3bf16_to_v3i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v2, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi <3 x bfloat> %x to <3 x i16> ret <3 x i16> %op } @@ -35198,23 +35865,41 @@ define <4 x i16> @v_fptosi_v4bf16_to_v4i16(<4 x bfloat> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v1, v1, v2, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_v4bf16_to_v4i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v1 :: v_dual_lshlrev_b32 v3, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX1250-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v3, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_perm_b32 v0, v0, v3, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v1, v2, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_v4bf16_to_v4i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v2.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v3.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_v4bf16_to_v4i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v1 :: v_dual_lshlrev_b32 v3, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v3, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v1, v2, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi <4 x bfloat> %x to <4 x i16> ret <4 x i16> %op } @@ -35274,14 +35959,24 @@ define i32 @v_fptosi_bf16_to_i32(bfloat %x) { ; GFX11FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_bf16_to_i32: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_bf16_to_i32: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_bf16_to_i32: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi bfloat %x to i32 ret i32 %op } @@ -35729,26 +36424,48 @@ define i64 @v_fptosi_bf16_to_i64(bfloat %x) { ; GFX11FAKE16-NEXT: v_sub_co_ci_u32_e64 v1, null, v1, v3, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_bf16_to_i64: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_trunc_f32_e32 v0, v0 -; GFX1250-NEXT: v_mul_f32_e64 v1, 0x2f800000, |v0| -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_floor_f32_e32 v1, v1 -; GFX1250-NEXT: v_fma_f32 v2, 0xcf800000, v1, |v0| -; GFX1250-NEXT: v_ashrrev_i32_e32 v0, 31, v0 -; GFX1250-NEXT: v_cvt_u32_f32_e32 v3, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_u32_f32_e32 v2, v2 -; GFX1250-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_bitop2_b32 v3, v3, v0 bitop3:0x14 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_xor_b32_e32 v2, v2, v0 -; GFX1250-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_bf16_to_i64: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_trunc_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_mul_f32_e64 v1, 0x2f800000, |v0| +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_floor_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_fma_f32 v2, 0xcf800000, v1, |v0| +; GFX1250TRUE16-NEXT: v_ashrrev_i32_e32 v0, 31, v0 +; GFX1250TRUE16-NEXT: v_cvt_u32_f32_e32 v3, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_bitop2_b32 v3, v3, v0 bitop3:0x14 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v2, v2, v0 +; GFX1250TRUE16-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1] +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_bf16_to_i64: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_trunc_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_mul_f32_e64 v1, 0x2f800000, |v0| +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_floor_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_fma_f32 v2, 0xcf800000, v1, |v0| +; GFX1250FAKE16-NEXT: v_ashrrev_i32_e32 v0, 31, v0 +; GFX1250FAKE16-NEXT: v_cvt_u32_f32_e32 v3, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_bitop2_b32 v3, v3, v0 bitop3:0x14 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v2, v2, v0 +; GFX1250FAKE16-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi bfloat %x to i64 ret i64 %op } @@ -37293,22 +38010,39 @@ define <3 x bfloat> @v_sitofp_v3i16_to_v3bf16(<3 x i16> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sitofp_v3i16_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_ashrrev_i32_e32 v2, 16, v0 -; GFX1250-NEXT: v_bfe_i32 v0, v0, 0, 16 -; GFX1250-NEXT: v_bfe_i32 v1, v1, 0, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_i32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_i32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sitofp_v3i16_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_bfe_i32 v1, v1, 0, 16 +; GFX1250TRUE16-NEXT: v_ashrrev_i32_e32 v2, 16, v0 +; GFX1250TRUE16-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sitofp_v3i16_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_ashrrev_i32_e32 v2, 16, v0 +; GFX1250FAKE16-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GFX1250FAKE16-NEXT: v_bfe_i32 v1, v1, 0, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = sitofp <3 x i16> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -37972,17 +38706,31 @@ define <3 x bfloat> @v_sitofp_v3i32_to_v3bf16(<3 x i32> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v2, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sitofp_v3i32_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v2, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sitofp_v3i32_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sitofp_v3i32_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = sitofp <3 x i32> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -39232,52 +39980,101 @@ define <3 x bfloat> @v_sitofp_v3i64_to_v3bf16(<3 x i64> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sitofp_v3i64_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_xor_b32_e32 v8, v4, v5 -; GFX1250-NEXT: v_xor_b32_e32 v6, v2, v3 -; GFX1250-NEXT: v_cls_i32_e32 v10, v3 -; GFX1250-NEXT: v_cls_i32_e32 v9, v5 -; GFX1250-NEXT: v_cls_i32_e32 v11, v1 -; GFX1250-NEXT: v_dual_ashrrev_i32 v8, 31, v8 :: v_dual_bitop2_b32 v7, v0, v1 bitop3:0x14 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_dual_ashrrev_i32 v6, 31, v6 :: v_dual_ashrrev_i32 v7, 31, v7 -; GFX1250-NEXT: v_dual_add_nc_u32 v6, 32, v6 :: v_dual_add_nc_u32 v7, 32, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_add_min_u32_e64 v6, v10, -1, v6 -; GFX1250-NEXT: v_add_min_u32_e64 v7, v11, -1, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] -; GFX1250-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v2, 1, v2 -; GFX1250-NEXT: v_add_nc_u32_e32 v8, 32, v8 -; GFX1250-NEXT: v_min_u32_e32 v0, 1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX1250-NEXT: v_add_min_u32_e64 v8, v9, -1, v8 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v2, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] -; GFX1250-NEXT: v_sub_nc_u32_e32 v8, 32, v8 -; GFX1250-NEXT: v_ldexp_f32 v2, v2, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_min_u32_e32 v4, 1, v4 -; GFX1250-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_i32_e32 v1, v1 -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v4 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_ldexp_f32 v1, v1, v8 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sitofp_v3i64_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v7, v2, v3 +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v6, v4, v5 +; GFX1250TRUE16-NEXT: v_cls_i32_e32 v10, v3 +; GFX1250TRUE16-NEXT: v_cls_i32_e32 v9, v5 +; GFX1250TRUE16-NEXT: v_cls_i32_e32 v11, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_dual_ashrrev_i32 v7, 31, v7 :: v_dual_ashrrev_i32 v6, 31, v6 +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v8, v0, v1 +; GFX1250TRUE16-NEXT: v_dual_add_nc_u32 v7, 32, v7 :: v_dual_add_nc_u32 v6, 32, v6 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_ashrrev_i32_e32 v8, 31, v8 +; GFX1250TRUE16-NEXT: v_add_min_u32_e64 v7, v10, -1, v7 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_add_min_u32_e64 v6, v9, -1, v6 +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[2:3], v7, v[2:3] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[4:5], v6, v[4:5] +; GFX1250TRUE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250TRUE16-NEXT: v_add_nc_u32_e32 v8, 32, v8 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250TRUE16-NEXT: v_or_b32_e32 v2, v3, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_add_min_u32_e64 v8, v11, -1, v8 +; GFX1250TRUE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v4, v5, v4 bitop3:0x54 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[0:1], v8, v[0:1] +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v5, 32, v8 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: v_or_b32_e32 v0, v1, v0 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v4 +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v4, 32, v7 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v1, v1, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_ldexp_f32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sitofp_v3i64_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v8, v4, v5 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v6, v2, v3 +; GFX1250FAKE16-NEXT: v_cls_i32_e32 v10, v3 +; GFX1250FAKE16-NEXT: v_cls_i32_e32 v9, v5 +; GFX1250FAKE16-NEXT: v_cls_i32_e32 v11, v1 +; GFX1250FAKE16-NEXT: v_dual_ashrrev_i32 v8, 31, v8 :: v_dual_bitop2_b32 v7, v0, v1 bitop3:0x14 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_dual_ashrrev_i32 v6, 31, v6 :: v_dual_ashrrev_i32 v7, 31, v7 +; GFX1250FAKE16-NEXT: v_dual_add_nc_u32 v6, 32, v6 :: v_dual_add_nc_u32 v7, 32, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_add_min_u32_e64 v6, v10, -1, v6 +; GFX1250FAKE16-NEXT: v_add_min_u32_e64 v7, v11, -1, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250FAKE16-NEXT: v_add_nc_u32_e32 v8, 32, v8 +; GFX1250FAKE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_or_b32_e32 v2, v3, v2 +; GFX1250FAKE16-NEXT: v_add_min_u32_e64 v8, v9, -1, v8 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] +; GFX1250FAKE16-NEXT: v_sub_nc_u32_e32 v8, 32, v8 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v2, v2, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v4 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v1, v1, v8 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = sitofp <3 x i64> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -40015,15 +40812,26 @@ define bfloat @v_uitofp_i16_to_bf16(i16 %x) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_i16_to_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_i16_to_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_i16_to_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp i16 %x to bfloat ret bfloat %op } @@ -40167,18 +40975,32 @@ define <2 x bfloat> @v_uitofp_v2i16_to_v2bf16(<2 x i16> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x7060302 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v2i16_to_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v2i16_to_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.h +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v2i16_to_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <2 x i16> %x to <2 x bfloat> ret <2 x bfloat> %op } @@ -40373,22 +41195,41 @@ define <3 x bfloat> @v_uitofp_v3i16_to_v3bf16(<3 x i16> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v3i16_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: v_and_b32_e32 v1, 0xffff, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v3i16_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v3.l, v0.h +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v3 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v3, v0, s0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v1, v2 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v3.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v3i16_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <3 x i16> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -40626,23 +41467,43 @@ define <4 x bfloat> @v_uitofp_v4i16_to_v4bf16(<4 x i16> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v1, v1, v2, 0x7060302 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v4i16_to_v4bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v2, 16, v1 :: v_dual_lshrrev_b32 v3, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: v_and_b32_e32 v1, 0xffff, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v3, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v3 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, v2 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v4i16_to_v4bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, v1.h +; GFX1250TRUE16-NEXT: v_and_b32_e32 v3, 0xffff, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v4, v2 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.h +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 0xffff, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v3, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v1, v2 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v3, v4 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v4i16_to_v4bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v2, 16, v1 :: v_dual_lshrrev_b32 v3, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v3, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v3 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, v2 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <4 x i16> %x to <4 x bfloat> ret <4 x bfloat> %op } @@ -41058,17 +41919,31 @@ define <3 x bfloat> @v_uitofp_v3i32_to_v3bf16(<3 x i32> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v2, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v3i32_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v3i32_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v3i32_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <3 x i32> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -42105,44 +42980,84 @@ define <3 x bfloat> @v_uitofp_v3i64_to_v3bf16(<3 x i64> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v3i64_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_clz_i32_u32_e32 v6, v3 -; GFX1250-NEXT: v_clz_i32_u32_e32 v7, v1 -; GFX1250-NEXT: v_clz_i32_u32_e32 v8, v5 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v6, 32, v6 -; GFX1250-NEXT: v_min_u32_e32 v7, 32, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v8, 32, v8 -; GFX1250-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] -; GFX1250-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v2, 1, v2 -; GFX1250-NEXT: v_min_u32_e32 v0, 1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v4, 1, v4 -; GFX1250-NEXT: v_dual_sub_nc_u32 v8, 32, v8 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 -; GFX1250-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_ldexp_f32 v2, v2, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v4 -; GFX1250-NEXT: v_ldexp_f32 v1, v1, v8 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v3i64_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_clz_i32_u32_e32 v6, v5 +; GFX1250TRUE16-NEXT: v_clz_i32_u32_e32 v7, v3 +; GFX1250TRUE16-NEXT: v_clz_i32_u32_e32 v8, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v6, 32, v6 +; GFX1250TRUE16-NEXT: v_min_u32_e32 v7, 32, v7 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v8, 32, v8 +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[4:5], v6, v[4:5] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[2:3], v7, v[2:3] +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[0:1], v8, v[0:1] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250TRUE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: v_or_b32_e32 v4, v5, v4 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_or_b32_e32 v2, v3, v2 +; GFX1250TRUE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v5, 32, v8 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_3) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v4 +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v4, 32, v7 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v1, v1, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_ldexp_f32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v3i64_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_clz_i32_u32_e32 v6, v3 +; GFX1250FAKE16-NEXT: v_clz_i32_u32_e32 v7, v1 +; GFX1250FAKE16-NEXT: v_clz_i32_u32_e32 v8, v5 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v6, 32, v6 +; GFX1250FAKE16-NEXT: v_min_u32_e32 v7, 32, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v8, 32, v8 +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250FAKE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v8, 32, v8 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v2, v2, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v4 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v1, v1, v8 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <3 x i64> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -42717,15 +43632,25 @@ define bfloat @v_select_bf16(i1 %cond, bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select i1 %cond, bfloat %a, bfloat %b ret bfloat %op } @@ -42810,16 +43735,27 @@ define bfloat @v_select_fneg_lhs_bf16(i1 %cond, bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_fneg_lhs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0 -; GFX1250-NEXT: v_xor_b32_e32 v1, 0x8000, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_fneg_lhs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v1.l +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_fneg_lhs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v1, 0x8000, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %neg.a = fneg bfloat %a %op = select i1 %cond, bfloat %neg.a, bfloat %b ret bfloat %op @@ -42905,16 +43841,27 @@ define bfloat @v_select_fneg_rhs_bf16(i1 %cond, bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_fneg_rhs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0 -; GFX1250-NEXT: v_xor_b32_e32 v2, 0x8000, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_fneg_rhs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v2.l +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_fneg_rhs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v2, 0x8000, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %neg.b = fneg bfloat %b %op = select i1 %cond, bfloat %a, bfloat %neg.b ret bfloat %op @@ -43025,18 +43972,29 @@ define <2 x bfloat> @v_select_v2bf16(i1 %cond, <2 x bfloat> %a, <2 x bfloat> %b) ; GFX11FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v3, 16, v1 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_cndmask_b32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v1.h, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v3, 16, v1 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_cndmask_b32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select i1 %cond, <2 x bfloat> %a, <2 x bfloat> %b ret <2 x bfloat> %op } @@ -43155,20 +44113,34 @@ define <2 x bfloat> @v_vselect_v2bf16(<2 x i1> %cond, <2 x bfloat> %a, <2 x bflo ; GFX11FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v5, 16, v3 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v5, 16, v3 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <2 x i1> %cond, <2 x bfloat> %a, <2 x bfloat> %b ret <2 x bfloat> %op } @@ -43256,16 +44228,26 @@ define amdgpu_ps i32 @s_select_bf16(bfloat inreg %a, bfloat inreg %b, i32 %c) { ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_select_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: v_mov_b32_e32 v1, s0 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s1, v1, vcc_lo -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_select_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, s0 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, 0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, s1, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_select_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v1, s0 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s1, v1, vcc_lo +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq i32 %c, 0 %op = select i1 %cond, bfloat %a, bfloat %b %cast = bitcast bfloat %op to i16 @@ -43402,20 +44384,34 @@ define amdgpu_ps i32 @s_select_v2bf16(<2 x bfloat> inreg %a, <2 x bfloat> inreg ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_select_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_lshr_b32 s2, s0, 16 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: v_dual_mov_b32 v1, s2 :: v_dual_mov_b32 v2, s0 -; GFX1250-NEXT: s_lshr_b32 s3, s1, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s3, v1, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, s1, v2, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_select_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_lshr_b32 s2, s0, 16 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, s2 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, s0 +; GFX1250TRUE16-NEXT: s_lshr_b32 s0, s1, 16 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, s0, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, s1, v0.l, vcc_lo +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_select_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_lshr_b32 s2, s0, 16 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v1, s2 :: v_dual_mov_b32 v2, s0 +; GFX1250FAKE16-NEXT: s_lshr_b32 s3, s1, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s3, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, s1, v2, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq i32 %c, 0 %op = select i1 %cond, <2 x bfloat> %a, <2 x bfloat> %b %cast = bitcast <2 x bfloat> %op to i32 @@ -43554,21 +44550,36 @@ define amdgpu_ps i32 @s_vselect_v2bf16(<2 x bfloat> inreg %a, <2 x bfloat> inreg ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_vselect_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_lshr_b32 s2, s0, 16 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX1250-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s0 -; GFX1250-NEXT: s_lshr_b32 s0, s1, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, s0, v2, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s1, v3, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_vselect_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_lshr_b32 s3, s0, 16 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s2, 0, v1 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, s3 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, s0 +; GFX1250TRUE16-NEXT: s_lshr_b32 s0, s1, 16 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, s0, v0.l, s2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, s1, v0.h, vcc_lo +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v1 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_vselect_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_lshr_b32 s2, s0, 16 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s0 +; GFX1250FAKE16-NEXT: s_lshr_b32 s0, s1, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, s0, v2, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s1, v3, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq <2 x i32> %c, zeroinitializer %op = select <2 x i1> %cond, <2 x bfloat> %a, <2 x bfloat> %b %cast = bitcast <2 x bfloat> %op to i32 @@ -45557,32 +46568,55 @@ define amdgpu_ps <2 x i32> @s_vselect_v4bf16(<4 x bfloat> inreg %a, <4 x bfloat> ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s1, v1 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_vselect_v4bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_lshr_b32 s4, s1, 16 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v3 -; GFX1250-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s1 -; GFX1250-NEXT: s_lshr_b32 s4, s3, 16 -; GFX1250-NEXT: s_lshr_b32 s5, s0, 16 -; GFX1250-NEXT: v_mov_b32_e32 v6, s0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cndmask_b32_e32 v3, s4, v4, vcc_lo -; GFX1250-NEXT: v_mov_b32_e32 v4, s5 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX1250-NEXT: s_lshr_b32 s0, s2, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, s0, v4, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s2, v6, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_cndmask_b32_e32 v2, s3, v5, vcc_lo -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_readfirstlane_b32 s1, v1 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_vselect_v4bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_lshr_b32 s7, s1, 16 +; GFX1250TRUE16-NEXT: s_lshr_b32 s9, s0, 16 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s4, 0, v1 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s5, 0, v2 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s6, 0, v3 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, s7 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, s9 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, s0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, s1 +; GFX1250TRUE16-NEXT: s_lshr_b32 s8, s3, 16 +; GFX1250TRUE16-NEXT: s_lshr_b32 s0, s2, 16 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, s8, v0.l, s6 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, s0, v0.h, s4 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, s2, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, s3, v1.h, s5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s1, v2 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_vselect_v4bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_lshr_b32 s4, s1, 16 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v3 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s1 +; GFX1250FAKE16-NEXT: s_lshr_b32 s4, s3, 16 +; GFX1250FAKE16-NEXT: s_lshr_b32 s5, s0, 16 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v6, s0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, s4, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v4, s5 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX1250FAKE16-NEXT: s_lshr_b32 s0, s2, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, s0, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s2, v6, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v2, s3, v5, vcc_lo +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s1, v1 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq <4 x i32> %c, zeroinitializer %op = select <4 x i1> %cond, <4 x bfloat> %a, <4 x bfloat> %b %cast = bitcast <4 x bfloat> %op to <2 x i32> @@ -45787,27 +46821,49 @@ define <4 x bfloat> @v_vselect_v4bf16(<4 x i1> %cond, <4 x bfloat> %a, <4 x bflo ; GFX11FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v4bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v2, 1, v2 -; GFX1250-NEXT: v_dual_lshrrev_b32 v8, 16, v4 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v9, 16, v6 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_dual_cndmask_b32 v2, v7, v5, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v7, 16, v7 :: v_dual_lshrrev_b32 v5, 16, v5 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v6, v4, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v9, v8, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v4bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v3.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v1.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v7.l, v5.l, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v6.l, v4.l, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v6.h, v4.h, s1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v7.h, v5.h, s2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v4bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v2, 1, v2 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v8, 16, v4 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v9, 16, v6 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v2, v7, v5, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v7, 16, v7 :: v_dual_lshrrev_b32 v5, 16, v5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v6, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v9, v8, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <4 x i1> %cond, <4 x bfloat> %a, <4 x bfloat> %b ret <4 x bfloat> %op } @@ -46161,45 +47217,77 @@ define <8 x bfloat> @v_vselect_v8bf16(<8 x i1> %cond, <8 x bfloat> %a, <8 x bflo ; GFX11FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v8bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v6, 1, v6 -; GFX1250-NEXT: v_and_b32_e32 v4, 1, v4 -; GFX1250-NEXT: v_dual_lshrrev_b32 v17, 16, v14 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v16, 16, v10 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 -; GFX1250-NEXT: v_and_b32_e32 v1, 1, v1 -; GFX1250-NEXT: v_dual_cndmask_b32 v6, v15, v11, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 -; GFX1250-NEXT: v_and_b32_e32 v7, 1, v7 -; GFX1250-NEXT: v_lshrrev_b32_e32 v11, 16, v11 -; GFX1250-NEXT: v_dual_cndmask_b32 v4, v14, v10 :: v_dual_lshrrev_b32 v15, 16, v15 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 -; GFX1250-NEXT: v_dual_lshrrev_b32 v14, 16, v12 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 -; GFX1250-NEXT: v_lshrrev_b32_e32 v10, 16, v8 -; GFX1250-NEXT: v_cndmask_b32_e32 v5, v17, v16, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v2, v13, v9, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v9, 16, v9 -; GFX1250-NEXT: v_dual_cndmask_b32 v0, v12, v8 :: v_dual_lshrrev_b32 v13, 16, v13 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v14, v10, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v13, v9, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 -; GFX1250-NEXT: v_cndmask_b32_e32 v7, v15, v11, vcc_lo -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v8bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v3.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v5.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v6.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v4.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.l, 1, v7.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s5, 1, v1.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s3, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s4, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s6, 1, v2.l +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v12.l, v8.l, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.l, v15.l, v11.l, s2 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, v14.l, v10.l, s3 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v13.l, v9.l, s4 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v12.h, v8.h, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v13.h, v9.h, s1 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, v14.h, v10.h, s5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.h, v15.h, v11.h, s6 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v8bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v6, 1, v6 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v17, 16, v14 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v16, 16, v10 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 1, v1 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v6, v15, v11, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v7, 1, v7 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v11, 16, v11 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v4, v14, v10 :: v_dual_lshrrev_b32 v15, 16, v15 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v14, 16, v12 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v10, 16, v8 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v5, v17, v16, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v2, v13, v9, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v9, 16, v9 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v0, v12, v8 :: v_dual_lshrrev_b32 v13, 16, v13 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v14, v10, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v13, v9, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v7, v15, v11, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <8 x i1> %cond, <8 x bfloat> %a, <8 x bfloat> %b ret <8 x bfloat> %op } @@ -46939,73 +48027,129 @@ define <16 x bfloat> @v_vselect_v16bf16(<16 x i1> %cond, <16 x bfloat> %a, <16 x ; GFX11FAKE16-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v16bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: scratch_load_b32 v31, off, s32 -; GFX1250-NEXT: v_dual_lshrrev_b32 v52, 16, v25 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v53, 16, v16 :: v_dual_bitop2_b32 v13, 1, v13 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v33, 16, v22 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 -; GFX1250-NEXT: v_dual_lshrrev_b32 v34, 16, v30 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v51, 16, v17 :: v_dual_bitop2_b32 v10, 1, v10 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v12, v30, v22, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 -; GFX1250-NEXT: v_dual_lshrrev_b32 v50, 16, v26 :: v_dual_bitop2_b32 v11, 1, v11 bitop3:0x40 -; GFX1250-NEXT: v_and_b32_e32 v14, 1, v14 -; GFX1250-NEXT: v_dual_lshrrev_b32 v35, 16, v21 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v13, v34, v33, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 -; GFX1250-NEXT: v_dual_lshrrev_b32 v36, 16, v29 :: v_dual_bitop2_b32 v4, 1, v4 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v49, 16, v18 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v10, v29, v21, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 -; GFX1250-NEXT: v_dual_lshrrev_b32 v37, 16, v20 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v38, 16, v28 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v48, 16, v27 :: v_dual_bitop2_b32 v9, 1, v9 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v11, v36, v35, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 -; GFX1250-NEXT: v_dual_lshrrev_b32 v39, 16, v19 :: v_dual_bitop2_b32 v6, 1, v6 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v32, 16, v23 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v8, v28, v20, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 -; GFX1250-NEXT: v_dual_lshrrev_b32 v54, 16, v24 :: v_dual_bitop2_b32 v15, 1, v15 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v9, v38, v37, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 -; GFX1250-NEXT: v_cndmask_b32_e32 v6, v27, v19, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 -; GFX1250-NEXT: v_cndmask_b32_e32 v4, v26, v18, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v2, v25, v17, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v52, v51, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v24, v16, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v54, v53, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 -; GFX1250-NEXT: v_cndmask_b32_e32 v5, v50, v49, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v3, 16, v31 -; GFX1250-NEXT: v_cndmask_b32_e32 v7, v48, v39, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 -; GFX1250-NEXT: v_cndmask_b32_e32 v14, v31, v23, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 -; GFX1250-NEXT: v_cndmask_b32_e32 v15, v3, v32, vcc_lo -; GFX1250-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v16bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v3.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.l, 1, v5.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.h, 1, v4.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.l, 1, v7.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.h, 1, v6.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.l, 1, v9.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.h, 1, v8.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.l, 1, v11.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.h, 1, v10.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.l, 1, v13.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.h, 1, v12.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.l, 1, v15.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.h, 1, v14.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v1.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s3, 1, v2.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s4, 1, v2.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s5, 1, v3.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s6, 1, v3.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s7, 1, v4.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s8, 1, v4.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s9, 1, v5.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s10, 1, v6.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s11, 1, v6.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s12, 1, v5.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s13, 1, v7.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s14, 1, v7.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.l, v30.l, v22.l, s10 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.h, v30.h, v22.h, s11 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.l, v29.l, v21.l, s12 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.h, v29.h, v21.h, s9 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.l, v28.l, v20.l, s8 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.h, v28.h, v20.h, s7 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.l, v27.l, v19.l, s6 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.h, v27.h, v19.h, s5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, v26.l, v18.l, s4 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v25.l, v17.l, s2 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v24.l, v16.l, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v24.h, v16.h, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v25.h, v17.h, s1 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, v26.h, v18.h, s3 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.l, v31.l, v23.l, s14 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.h, v31.h, v23.h, s13 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v16bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v52, 16, v25 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v53, 16, v16 :: v_dual_bitop2_b32 v13, 1, v13 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v33, 16, v22 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v34, 16, v30 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v51, 16, v17 :: v_dual_bitop2_b32 v10, 1, v10 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v12, v30, v22, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v50, 16, v26 :: v_dual_bitop2_b32 v11, 1, v11 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v14, 1, v14 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v35, 16, v21 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v13, v34, v33, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v36, 16, v29 :: v_dual_bitop2_b32 v4, 1, v4 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v49, 16, v18 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v10, v29, v21, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v37, 16, v20 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v38, 16, v28 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v48, 16, v27 :: v_dual_bitop2_b32 v9, 1, v9 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v11, v36, v35, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v39, 16, v19 :: v_dual_bitop2_b32 v6, 1, v6 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v32, 16, v23 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v8, v28, v20, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v54, 16, v24 :: v_dual_bitop2_b32 v15, 1, v15 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v9, v38, v37, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v6, v27, v19, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v4, v26, v18, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v2, v25, v17, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v52, v51, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v24, v16, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v54, v53, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v5, v50, v49, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v31 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v7, v48, v39, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v14, v31, v23, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v15, v3, v32, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <16 x i1> %cond, <16 x bfloat> %a, <16 x bfloat> %b ret <16 x bfloat> %op } @@ -48861,177 +50005,330 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x ; GFX11FAKE16-NEXT: v_perm_b32 v15, v31, v30, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v32bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x1b -; GFX1250-NEXT: scratch_load_b32 v31, off, s32 offset:60 -; GFX1250-NEXT: scratch_load_b32 v32, off, s32 offset:124 -; GFX1250-NEXT: scratch_load_u16 v33, off, s32 -; GFX1250-NEXT: scratch_load_b32 v34, off, s32 offset:128 -; GFX1250-NEXT: scratch_load_b32 v35, off, s32 offset:64 -; GFX1250-NEXT: scratch_load_b32 v36, off, s32 offset:120 -; GFX1250-NEXT: scratch_load_b32 v37, off, s32 offset:56 -; GFX1250-NEXT: scratch_load_b32 v38, off, s32 offset:116 -; GFX1250-NEXT: scratch_load_b32 v39, off, s32 offset:52 -; GFX1250-NEXT: scratch_load_b32 v48, off, s32 offset:112 -; GFX1250-NEXT: scratch_load_b32 v49, off, s32 offset:48 -; GFX1250-NEXT: scratch_load_b32 v50, off, s32 offset:108 -; GFX1250-NEXT: scratch_load_b32 v51, off, s32 offset:44 -; GFX1250-NEXT: scratch_load_b32 v52, off, s32 offset:104 -; GFX1250-NEXT: scratch_load_b32 v53, off, s32 offset:40 -; GFX1250-NEXT: scratch_load_b32 v54, off, s32 offset:100 -; GFX1250-NEXT: scratch_load_b32 v55, off, s32 offset:36 -; GFX1250-NEXT: scratch_load_b32 v64, off, s32 offset:76 -; GFX1250-NEXT: scratch_load_b32 v65, off, s32 offset:12 -; GFX1250-NEXT: scratch_load_b32 v66, off, s32 offset:96 -; GFX1250-NEXT: scratch_load_b32 v67, off, s32 offset:32 -; GFX1250-NEXT: scratch_load_b32 v68, off, s32 offset:80 -; GFX1250-NEXT: scratch_load_b32 v69, off, s32 offset:84 -; GFX1250-NEXT: scratch_load_b32 v70, off, s32 offset:92 -; GFX1250-NEXT: scratch_load_b32 v71, off, s32 offset:28 -; GFX1250-NEXT: scratch_load_b32 v80, off, s32 offset:20 -; GFX1250-NEXT: scratch_load_b32 v81, off, s32 offset:88 -; GFX1250-NEXT: scratch_load_b32 v82, off, s32 offset:24 -; GFX1250-NEXT: v_and_b32_e32 v30, 1, v30 -; GFX1250-NEXT: v_and_b32_e32 v29, 1, v29 -; GFX1250-NEXT: v_and_b32_e32 v26, 1, v26 -; GFX1250-NEXT: v_and_b32_e32 v24, 1, v24 -; GFX1250-NEXT: v_and_b32_e32 v22, 1, v22 -; GFX1250-NEXT: v_and_b32_e32 v20, 1, v20 -; GFX1250-NEXT: v_and_b32_e32 v18, 1, v18 -; GFX1250-NEXT: v_and_b32_e32 v16, 1, v16 -; GFX1250-NEXT: v_and_b32_e32 v10, 1, v10 -; GFX1250-NEXT: v_and_b32_e32 v6, 1, v6 -; GFX1250-NEXT: v_and_b32_e32 v4, 1, v4 -; GFX1250-NEXT: v_and_b32_e32 v1, 1, v1 -; GFX1250-NEXT: v_and_b32_e32 v3, 1, v3 -; GFX1250-NEXT: v_and_b32_e32 v5, 1, v5 -; GFX1250-NEXT: v_and_b32_e32 v23, 1, v23 -; GFX1250-NEXT: v_and_b32_e32 v9, 1, v9 -; GFX1250-NEXT: v_and_b32_e32 v13, 1, v13 -; GFX1250-NEXT: v_and_b32_e32 v15, 1, v15 -; GFX1250-NEXT: v_and_b32_e32 v21, 1, v21 -; GFX1250-NEXT: v_and_b32_e32 v11, 1, v11 -; GFX1250-NEXT: v_and_b32_e32 v19, 1, v19 -; GFX1250-NEXT: s_wait_loadcnt 0x1a -; GFX1250-NEXT: v_dual_lshrrev_b32 v83, 16, v32 :: v_dual_bitop2_b32 v17, 1, v17 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e64 s1, 1, v30 -; GFX1250-NEXT: v_and_b32_e32 v28, 1, v28 -; GFX1250-NEXT: s_wait_loadcnt 0x17 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_dual_cndmask_b32 v30, v34, v35, s1 :: v_dual_bitop2_b32 v33, 1, v33 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v28 -; GFX1250-NEXT: v_lshrrev_b32_e32 v28, 16, v31 -; GFX1250-NEXT: v_cmp_eq_u32_e64 s0, 1, v29 -; GFX1250-NEXT: scratch_load_b32 v29, off, s32 offset:16 -; GFX1250-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_lshrrev_b32 v34, 16, v34 -; GFX1250-NEXT: v_cndmask_b32_e32 v31, v32, v31, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v33 -; GFX1250-NEXT: scratch_load_b32 v32, off, s32 offset:72 -; GFX1250-NEXT: v_cndmask_b32_e64 v28, v83, v28, s0 -; GFX1250-NEXT: scratch_load_b32 v83, off, s32 offset:4 -; GFX1250-NEXT: v_cndmask_b32_e32 v34, v34, v35, vcc_lo -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: scratch_load_b32 v35, off, s32 offset:68 -; GFX1250-NEXT: scratch_load_b32 v33, off, s32 offset:8 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v26 -; GFX1250-NEXT: s_wait_loadcnt 0x1a -; GFX1250-NEXT: v_dual_cndmask_b32 v26, v36, v37, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v24 -; GFX1250-NEXT: v_dual_lshrrev_b32 v37, 16, v37 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x18 -; GFX1250-NEXT: v_dual_lshrrev_b32 v36, 16, v36 :: v_dual_cndmask_b32 v24, v38, v39, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v22 -; GFX1250-NEXT: v_dual_lshrrev_b32 v38, 16, v38 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x16 -; GFX1250-NEXT: v_dual_cndmask_b32 v22, v48, v49 :: v_dual_lshrrev_b32 v39, 16, v39 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v20 -; GFX1250-NEXT: v_dual_lshrrev_b32 v49, 16, v49 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x14 -; GFX1250-NEXT: v_dual_lshrrev_b32 v48, 16, v48 :: v_dual_cndmask_b32 v20, v50, v51, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v18 -; GFX1250-NEXT: v_dual_lshrrev_b32 v51, 16, v51 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x12 -; GFX1250-NEXT: v_dual_lshrrev_b32 v50, 16, v50 :: v_dual_cndmask_b32 v18, v52, v53, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v16 -; GFX1250-NEXT: v_dual_lshrrev_b32 v53, 16, v53 :: v_dual_bitop2_b32 v14, 1, v14 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x10 -; GFX1250-NEXT: v_dual_lshrrev_b32 v52, 16, v52 :: v_dual_cndmask_b32 v16, v54, v55, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 -; GFX1250-NEXT: v_dual_lshrrev_b32 v55, 16, v55 :: v_dual_lshrrev_b32 v54, 16, v54 -; GFX1250-NEXT: s_wait_loadcnt 0xc -; GFX1250-NEXT: v_cndmask_b32_e32 v14, v66, v67, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 -; GFX1250-NEXT: v_dual_lshrrev_b32 v67, 16, v67 :: v_dual_lshrrev_b32 v66, 16, v66 -; GFX1250-NEXT: s_wait_loadcnt 0x8 -; GFX1250-NEXT: v_cndmask_b32_e32 v12, v70, v71, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 -; GFX1250-NEXT: v_dual_lshrrev_b32 v70, 16, v70 :: v_dual_bitop2_b32 v25, 1, v25 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x5 -; GFX1250-NEXT: v_dual_cndmask_b32 v10, v81, v82 :: v_dual_lshrrev_b32 v71, 16, v71 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 -; GFX1250-NEXT: v_dual_lshrrev_b32 v82, 16, v82 :: v_dual_bitop2_b32 v27, 1, v27 bitop3:0x40 -; GFX1250-NEXT: v_dual_cndmask_b32 v8, v69, v80 :: v_dual_lshrrev_b32 v81, 16, v81 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 -; GFX1250-NEXT: v_dual_lshrrev_b32 v80, 16, v80 :: v_dual_lshrrev_b32 v69, 16, v69 -; GFX1250-NEXT: s_wait_loadcnt 0x4 -; GFX1250-NEXT: v_dual_cndmask_b32 v6, v68, v29 :: v_dual_lshrrev_b32 v29, 16, v29 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 -; GFX1250-NEXT: v_dual_lshrrev_b32 v68, 16, v68 :: v_dual_cndmask_b32 v4, v64, v65, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_dual_lshrrev_b32 v65, 16, v65 :: v_dual_lshrrev_b32 v64, 16, v64 -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_dual_cndmask_b32 v2, v32, v33 :: v_dual_lshrrev_b32 v33, 16, v33 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v32, 16, v32 :: v_dual_cndmask_b32 v0, v35, v83, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v27 -; GFX1250-NEXT: v_dual_lshrrev_b32 v83, 16, v83 :: v_dual_cndmask_b32 v27, v36, v37, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v25 -; GFX1250-NEXT: v_cndmask_b32_e32 v25, v38, v39, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v23 -; GFX1250-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_cndmask_b32 v23, v48, v49, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v21 -; GFX1250-NEXT: v_cndmask_b32_e32 v21, v50, v51, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v19 -; GFX1250-NEXT: v_cndmask_b32_e32 v19, v52, v53, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v17 -; GFX1250-NEXT: v_cndmask_b32_e32 v17, v54, v55, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 -; GFX1250-NEXT: v_cndmask_b32_e32 v15, v66, v67, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 -; GFX1250-NEXT: v_cndmask_b32_e32 v13, v70, v71, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 -; GFX1250-NEXT: v_cndmask_b32_e32 v11, v81, v82, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 -; GFX1250-NEXT: v_cndmask_b32_e32 v7, v68, v29, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v32, v33, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v35, v83, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 -; GFX1250-NEXT: v_cndmask_b32_e32 v5, v64, v65, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 -; GFX1250-NEXT: v_cndmask_b32_e32 v9, v69, v80, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v8, v17, v16, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v9, v19, v18, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v10, v21, v20, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v11, v23, v22, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v12, v25, v24, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v13, v27, v26, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v14, v28, v31, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v15, v34, v30, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v32bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: s_clause 0x20 +; GFX1250TRUE16-NEXT: scratch_load_u16 v31, off, s32 +; GFX1250TRUE16-NEXT: scratch_load_b32 v32, off, s32 offset:68 +; GFX1250TRUE16-NEXT: scratch_load_b32 v33, off, s32 offset:72 +; GFX1250TRUE16-NEXT: scratch_load_b32 v34, off, s32 offset:76 +; GFX1250TRUE16-NEXT: scratch_load_b32 v35, off, s32 offset:124 +; GFX1250TRUE16-NEXT: scratch_load_b32 v36, off, s32 offset:128 +; GFX1250TRUE16-NEXT: scratch_load_b32 v37, off, s32 offset:64 +; GFX1250TRUE16-NEXT: scratch_load_b32 v38, off, s32 offset:60 +; GFX1250TRUE16-NEXT: scratch_load_b32 v39, off, s32 offset:120 +; GFX1250TRUE16-NEXT: scratch_load_b32 v48, off, s32 offset:56 +; GFX1250TRUE16-NEXT: scratch_load_b32 v49, off, s32 offset:116 +; GFX1250TRUE16-NEXT: scratch_load_b32 v50, off, s32 offset:52 +; GFX1250TRUE16-NEXT: scratch_load_b32 v51, off, s32 offset:112 +; GFX1250TRUE16-NEXT: scratch_load_b32 v52, off, s32 offset:48 +; GFX1250TRUE16-NEXT: scratch_load_b32 v53, off, s32 offset:108 +; GFX1250TRUE16-NEXT: scratch_load_b32 v54, off, s32 offset:44 +; GFX1250TRUE16-NEXT: scratch_load_b32 v55, off, s32 offset:104 +; GFX1250TRUE16-NEXT: scratch_load_b32 v64, off, s32 offset:40 +; GFX1250TRUE16-NEXT: scratch_load_b32 v65, off, s32 offset:100 +; GFX1250TRUE16-NEXT: scratch_load_b32 v66, off, s32 offset:36 +; GFX1250TRUE16-NEXT: scratch_load_b32 v67, off, s32 offset:96 +; GFX1250TRUE16-NEXT: scratch_load_b32 v68, off, s32 offset:32 +; GFX1250TRUE16-NEXT: scratch_load_b32 v69, off, s32 offset:92 +; GFX1250TRUE16-NEXT: scratch_load_b32 v70, off, s32 offset:28 +; GFX1250TRUE16-NEXT: scratch_load_b32 v71, off, s32 offset:88 +; GFX1250TRUE16-NEXT: scratch_load_b32 v80, off, s32 offset:24 +; GFX1250TRUE16-NEXT: scratch_load_b32 v81, off, s32 offset:84 +; GFX1250TRUE16-NEXT: scratch_load_b32 v82, off, s32 offset:20 +; GFX1250TRUE16-NEXT: scratch_load_b32 v83, off, s32 offset:80 +; GFX1250TRUE16-NEXT: scratch_load_b32 v84, off, s32 offset:16 +; GFX1250TRUE16-NEXT: scratch_load_b32 v85, off, s32 offset:12 +; GFX1250TRUE16-NEXT: scratch_load_b32 v86, off, s32 offset:8 +; GFX1250TRUE16-NEXT: scratch_load_b32 v87, off, s32 offset:4 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v3.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.l, 1, v9.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v4.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v1.h +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v5.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v7.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v6.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.h, 1, v8.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.l, 1, v11.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.h, 1, v10.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.l, 1, v13.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.h, 1, v12.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.l, 1, v15.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.h, 1, v14.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.l, 1, v17.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.h, 1, v16.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.l, 1, v19.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.h, 1, v18.l +; GFX1250TRUE16-NEXT: v_and_b16 v8.l, 1, v21.l +; GFX1250TRUE16-NEXT: v_and_b16 v8.h, 1, v20.l +; GFX1250TRUE16-NEXT: v_and_b16 v9.l, 1, v23.l +; GFX1250TRUE16-NEXT: v_and_b16 v9.h, 1, v22.l +; GFX1250TRUE16-NEXT: v_and_b16 v10.l, 1, v25.l +; GFX1250TRUE16-NEXT: v_and_b16 v10.h, 1, v24.l +; GFX1250TRUE16-NEXT: v_and_b16 v11.l, 1, v27.l +; GFX1250TRUE16-NEXT: v_and_b16 v11.h, 1, v26.l +; GFX1250TRUE16-NEXT: v_and_b16 v12.l, 1, v29.l +; GFX1250TRUE16-NEXT: v_and_b16 v12.h, 1, v28.l +; GFX1250TRUE16-NEXT: v_and_b16 v13.l, 1, v30.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s4, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s3, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s5, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s6, 1, v1.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s7, 1, v2.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s8, 1, v2.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s9, 1, v3.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s10, 1, v3.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s11, 1, v4.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s12, 1, v4.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s13, 1, v5.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s14, 1, v5.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s15, 1, v6.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s16, 1, v6.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s17, 1, v7.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s18, 1, v7.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s19, 1, v8.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s20, 1, v8.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s21, 1, v9.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s22, 1, v9.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s23, 1, v10.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s24, 1, v10.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s25, 1, v11.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s26, 1, v13.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s27, 1, v12.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s28, 1, v12.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s29, 1, v11.h +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x20 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v31.l +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x1a +; GFX1250TRUE16-NEXT: v_cndmask_b16 v15.l, v36.l, v37.l, s26 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x19 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v14.l, v35.l, v38.l, s27 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v14.h, v35.h, v38.h, s28 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x17 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v13.l, v39.l, v48.l, s29 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v13.h, v39.h, v48.h, s25 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x15 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v12.l, v49.l, v50.l, s24 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v12.h, v49.h, v50.h, s23 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x13 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v11.l, v51.l, v52.l, s22 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v11.h, v51.h, v52.h, s21 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x11 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v10.l, v53.l, v54.l, s20 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v10.h, v53.h, v54.h, s19 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0xf +; GFX1250TRUE16-NEXT: v_cndmask_b16 v9.l, v55.l, v64.l, s18 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v9.h, v55.h, v64.h, s17 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0xd +; GFX1250TRUE16-NEXT: v_cndmask_b16 v8.l, v65.l, v66.l, s16 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v8.h, v65.h, v66.h, s15 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0xb +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.l, v67.l, v68.l, s14 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.h, v67.h, v68.h, s13 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x9 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.l, v69.l, v70.l, s12 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.h, v69.h, v70.h, s11 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x7 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.l, v71.l, v80.l, s10 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.h, v71.h, v80.h, s9 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.l, v81.l, v82.l, s8 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.h, v81.h, v82.h, s7 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x3 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.l, v83.l, v84.l, s6 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x2 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, v34.l, v85.l, s4 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x1 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v33.l, v86.l, s2 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v32.l, v87.l, s1 +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v0.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v32.h, v87.h, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v33.h, v86.h, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, v34.h, v85.h, s3 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.h, v83.h, v84.h, s5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v15.h, v36.h, v37.h, s1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v32bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: s_clause 0x1b +; GFX1250FAKE16-NEXT: scratch_load_b32 v31, off, s32 offset:60 +; GFX1250FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:124 +; GFX1250FAKE16-NEXT: scratch_load_u16 v33, off, s32 +; GFX1250FAKE16-NEXT: scratch_load_b32 v34, off, s32 offset:128 +; GFX1250FAKE16-NEXT: scratch_load_b32 v35, off, s32 offset:64 +; GFX1250FAKE16-NEXT: scratch_load_b32 v36, off, s32 offset:120 +; GFX1250FAKE16-NEXT: scratch_load_b32 v37, off, s32 offset:56 +; GFX1250FAKE16-NEXT: scratch_load_b32 v38, off, s32 offset:116 +; GFX1250FAKE16-NEXT: scratch_load_b32 v39, off, s32 offset:52 +; GFX1250FAKE16-NEXT: scratch_load_b32 v48, off, s32 offset:112 +; GFX1250FAKE16-NEXT: scratch_load_b32 v49, off, s32 offset:48 +; GFX1250FAKE16-NEXT: scratch_load_b32 v50, off, s32 offset:108 +; GFX1250FAKE16-NEXT: scratch_load_b32 v51, off, s32 offset:44 +; GFX1250FAKE16-NEXT: scratch_load_b32 v52, off, s32 offset:104 +; GFX1250FAKE16-NEXT: scratch_load_b32 v53, off, s32 offset:40 +; GFX1250FAKE16-NEXT: scratch_load_b32 v54, off, s32 offset:100 +; GFX1250FAKE16-NEXT: scratch_load_b32 v55, off, s32 offset:36 +; GFX1250FAKE16-NEXT: scratch_load_b32 v64, off, s32 offset:76 +; GFX1250FAKE16-NEXT: scratch_load_b32 v65, off, s32 offset:12 +; GFX1250FAKE16-NEXT: scratch_load_b32 v66, off, s32 offset:96 +; GFX1250FAKE16-NEXT: scratch_load_b32 v67, off, s32 offset:32 +; GFX1250FAKE16-NEXT: scratch_load_b32 v68, off, s32 offset:80 +; GFX1250FAKE16-NEXT: scratch_load_b32 v69, off, s32 offset:84 +; GFX1250FAKE16-NEXT: scratch_load_b32 v70, off, s32 offset:92 +; GFX1250FAKE16-NEXT: scratch_load_b32 v71, off, s32 offset:28 +; GFX1250FAKE16-NEXT: scratch_load_b32 v80, off, s32 offset:20 +; GFX1250FAKE16-NEXT: scratch_load_b32 v81, off, s32 offset:88 +; GFX1250FAKE16-NEXT: scratch_load_b32 v82, off, s32 offset:24 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v30, 1, v30 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v29, 1, v29 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v26, 1, v26 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v24, 1, v24 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v22, 1, v22 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v20, 1, v20 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v18, 1, v18 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v16, 1, v16 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v10, 1, v10 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v6, 1, v6 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 1, v1 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v3, 1, v3 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v5, 1, v5 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v23, 1, v23 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v9, 1, v9 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v13, 1, v13 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v15, 1, v15 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v21, 1, v21 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v11, 1, v11 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v19, 1, v19 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x1a +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v83, 16, v32 :: v_dual_bitop2_b32 v17, 1, v17 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e64 s1, 1, v30 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v28, 1, v28 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x17 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v30, v34, v35, s1 :: v_dual_bitop2_b32 v33, 1, v33 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v28 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v28, 16, v31 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e64 s0, 1, v29 +; GFX1250FAKE16-NEXT: scratch_load_b32 v29, off, s32 offset:16 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_lshrrev_b32 v34, 16, v34 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v31, v32, v31, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v33 +; GFX1250FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:72 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v28, v83, v28, s0 +; GFX1250FAKE16-NEXT: scratch_load_b32 v83, off, s32 offset:4 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v34, v34, v35, vcc_lo +; GFX1250FAKE16-NEXT: s_clause 0x1 +; GFX1250FAKE16-NEXT: scratch_load_b32 v35, off, s32 offset:68 +; GFX1250FAKE16-NEXT: scratch_load_b32 v33, off, s32 offset:8 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v26 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x1a +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v26, v36, v37, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v24 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v37, 16, v37 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x18 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v36, 16, v36 :: v_dual_cndmask_b32 v24, v38, v39, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v22 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v38, 16, v38 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x16 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v22, v48, v49 :: v_dual_lshrrev_b32 v39, 16, v39 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v20 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v49, 16, v49 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x14 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v48, 16, v48 :: v_dual_cndmask_b32 v20, v50, v51, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v18 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v51, 16, v51 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x12 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v50, 16, v50 :: v_dual_cndmask_b32 v18, v52, v53, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v16 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v53, 16, v53 :: v_dual_bitop2_b32 v14, 1, v14 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x10 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v52, 16, v52 :: v_dual_cndmask_b32 v16, v54, v55, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v55, 16, v55 :: v_dual_lshrrev_b32 v54, 16, v54 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0xc +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v14, v66, v67, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v67, 16, v67 :: v_dual_lshrrev_b32 v66, 16, v66 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x8 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v12, v70, v71, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v70, 16, v70 :: v_dual_bitop2_b32 v25, 1, v25 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x5 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v10, v81, v82 :: v_dual_lshrrev_b32 v71, 16, v71 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v82, 16, v82 :: v_dual_bitop2_b32 v27, 1, v27 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v8, v69, v80 :: v_dual_lshrrev_b32 v81, 16, v81 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v80, 16, v80 :: v_dual_lshrrev_b32 v69, 16, v69 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x4 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v6, v68, v29 :: v_dual_lshrrev_b32 v29, 16, v29 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v68, 16, v68 :: v_dual_cndmask_b32 v4, v64, v65, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v65, 16, v65 :: v_dual_lshrrev_b32 v64, 16, v64 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v2, v32, v33 :: v_dual_lshrrev_b32 v33, 16, v33 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v32, 16, v32 :: v_dual_cndmask_b32 v0, v35, v83, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v27 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v83, 16, v83 :: v_dual_cndmask_b32 v27, v36, v37, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v25 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v25, v38, v39, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v23 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_cndmask_b32 v23, v48, v49, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v21 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v21, v50, v51, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v19 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v19, v52, v53, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v17 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v17, v54, v55, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v15, v66, v67, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v13, v70, v71, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v11, v81, v82, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v7, v68, v29, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v32, v33, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v35, v83, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v5, v64, v65, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v9, v69, v80, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v8, v17, v16, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v9, v19, v18, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v10, v21, v20, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v11, v23, v22, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v12, v25, v24, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v13, v27, v26, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v14, v28, v31, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v15, v34, v30, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <32 x i1> %cond, <32 x bfloat> %a, <32 x bfloat> %b ret <32 x bfloat> %op } @@ -49167,12 +50464,21 @@ define bfloat @v_fma_bf16(bfloat %a, bfloat %b, bfloat %c) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fma_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fma_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fma_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %c) ret bfloat %op } @@ -54791,12 +56097,21 @@ define bfloat @v_fmuladd_bf16(bfloat %a, bfloat %b, bfloat %c) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fmuladd_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fmuladd_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fmuladd_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.fmuladd.bf16(bfloat %a, bfloat %b, bfloat %c) ret bfloat %op } @@ -55652,5 +56967,3 @@ define <4 x bfloat> @v_fmuladd_v4bf16(<4 x bfloat> %a, <4 x bfloat> %b, <4 x bfl %op = call <4 x bfloat> @llvm.fmuladd.v4bf16(<4 x bfloat> %a, <4 x bfloat> %b, <4 x bfloat> %c) ret <4 x bfloat> %op } -;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: -; GFX1250FAKE16: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll index 363a248..cbf6b66 100644 --- a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll +++ b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll @@ -1262,7 +1262,7 @@ define amdgpu_ps void @ps_mesa_i16(i16 %arg0) { ; GFX1250-TRUE16-LABEL: ps_mesa_i16: ; GFX1250-TRUE16: ; %bb.0: ; GFX1250-TRUE16-NEXT: v_add_nc_u16 v0.l, v0.l, v0.l -; GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v0 +; GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v0, off ; GFX1250-TRUE16-NEXT: s_endpgm ; ; GFX1250-FAKE16-LABEL: ps_mesa_i16: @@ -3013,7 +3013,7 @@ define amdgpu_cs void @amdgpu_cs_v8i1(<8 x i1> %arg0) { ; GFX1250-TRUE16-NEXT: v_lshlrev_b16 v0.h, 4, v0.h ; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX1250-TRUE16-NEXT: v_bitop3_b16 v0.l, v0.l, v0.h, 15 bitop3:0xec -; GFX1250-TRUE16-NEXT: flat_store_b8 v[0:1], v0 +; GFX1250-TRUE16-NEXT: global_store_b8 v[0:1], v0, off ; GFX1250-TRUE16-NEXT: s_endpgm ; ; GFX1250-FAKE16-LABEL: amdgpu_cs_v8i1: @@ -3297,7 +3297,7 @@ define amdgpu_cs void @amdgpu_cs_v16i1(<16 x i1> %arg0) { ; GFX1250-TRUE16-NEXT: v_or_b16 v0.h, v1.h, v1.l ; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX1250-TRUE16-NEXT: v_bitop3_b16 v0.l, v0.l, v0.h, 0xff bitop3:0xec -; GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v0 +; GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v0, off ; GFX1250-TRUE16-NEXT: s_endpgm ; ; GFX1250-FAKE16-LABEL: amdgpu_cs_v16i1: diff --git a/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll b/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll index f706f53..eb40e5c 100644 --- a/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll +++ b/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll @@ -35,6 +35,6 @@ define amdgpu_kernel void @test_direct_indirect_call() { ret void } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll b/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll index 8da204b..c02ff28 100644 --- a/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll +++ b/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll @@ -28,6 +28,6 @@ define amdgpu_kernel void @test_simple_indirect_call() #0 { attributes #0 = { "amdgpu-no-dispatch-id" } ;. -; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_GCN: attributes #[[ATTR1]] = { "amdgpu-no-dispatch-id" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/elf-header-flags-sramecc.ll b/llvm/test/CodeGen/AMDGPU/elf-header-flags-sramecc.ll index c4479b3..e3bc516 100644 --- a/llvm/test/CodeGen/AMDGPU/elf-header-flags-sramecc.ll +++ b/llvm/test/CodeGen/AMDGPU/elf-header-flags-sramecc.ll @@ -15,6 +15,9 @@ ; RUN: llc -filetype=obj -mtriple=amdgcn -mcpu=gfx950 < %s | llvm-readobj --file-header - | FileCheck --check-prefix=SRAM-ECC-GFX950 %s ; RUN: llc -filetype=obj -mtriple=amdgcn -mcpu=gfx950 -mattr=+sramecc < %s | llvm-readobj --file-header - | FileCheck --check-prefix=SRAM-ECC-GFX950 %s +; RUN: llc -filetype=obj -mtriple=amdgcn -mcpu=gfx1250 < %s | llvm-readobj --file-header - | FileCheck --check-prefix=SRAM-ECC-GFX1250 %s +; RUN: llc -filetype=obj -mtriple=amdgcn -mcpu=gfx1250 -mattr=+sramecc < %s | llvm-readobj --file-header - | FileCheck --check-prefix=SRAM-ECC-GFX1250 %s + ; NO-SRAM-ECC-GFX906: Flags [ ; NO-SRAM-ECC-GFX906-NEXT: EF_AMDGPU_FEATURE_XNACK_V3 (0x100) ; NO-SRAM-ECC-GFX906-NEXT: EF_AMDGPU_MACH_AMDGCN_GFX906 (0x2F) @@ -52,6 +55,11 @@ ; SRAM-ECC-GFX950: EF_AMDGPU_MACH_AMDGCN_GFX950 (0x4F) ; SRAM-ECC-GFX950: ] +; SRAM-ECC-GFX1250: Flags [ +; SRAM-ECC-GFX1250: EF_AMDGPU_FEATURE_SRAMECC_V3 (0x200) +; SRAM-ECC-GFX1250: EF_AMDGPU_MACH_AMDGCN_GFX1250 (0x49) +; SRAM-ECC-GFX1250: ] + define amdgpu_kernel void @elf_header() { ret void } diff --git a/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll b/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll index ab51693..05d3e9c3 100644 --- a/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll +++ b/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll @@ -497,12 +497,10 @@ define amdgpu_kernel void @test_fold_canonicalize_minnum_value_f32(ptr addrspace ret void } -; FIXME: Should there be more checks here? minnum with NaN operand is simplified away. +; FIXME: Should there be more checks here? minnum with sNaN operand is simplified to qNaN. ; GCN-LABEL: test_fold_canonicalize_sNaN_value_f32: -; GCN: {{flat|global}}_load_dword [[LOAD:v[0-9]+]] -; VI: v_mul_f32_e32 v{{[0-9]+}}, 1.0, [[LOAD]] -; GFX9: v_max_f32_e32 v{{[0-9]+}}, [[LOAD]], [[LOAD]] +; GCN: v_mov_b32_e32 v{{.+}}, 0x7fc00000 define amdgpu_kernel void @test_fold_canonicalize_sNaN_value_f32(ptr addrspace(1) %arg) { %id = tail call i32 @llvm.amdgcn.workitem.id.x() %gep = getelementptr inbounds float, ptr addrspace(1) %arg, i32 %id diff --git a/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll b/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll index 3de6df2..833be20 100644 --- a/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll +++ b/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll @@ -1949,8 +1949,7 @@ define float @v_fneg_self_minimumnum_f32_ieee(float %a) #0 { ; GCN-LABEL: v_fneg_self_minimumnum_f32_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_mul_f32_e32 v0, -1.0, v0 -; GCN-NEXT: v_max_f32_e32 v0, v0, v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %min = call float @llvm.minimumnum.f32(float %a, float %a) %min.fneg = fneg float %min @@ -1961,7 +1960,7 @@ define float @v_fneg_self_minimumnum_f32_no_ieee(float %a) #4 { ; GCN-LABEL: v_fneg_self_minimumnum_f32_no_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_max_f32_e64 v0, -v0, -v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %min = call float @llvm.minimumnum.f32(float %a, float %a) %min.fneg = fneg float %min @@ -2285,8 +2284,7 @@ define float @v_fneg_self_maximumnum_f32_ieee(float %a) #0 { ; GCN-LABEL: v_fneg_self_maximumnum_f32_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_mul_f32_e32 v0, -1.0, v0 -; GCN-NEXT: v_min_f32_e32 v0, v0, v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %max = call float @llvm.maximumnum.f32(float %a, float %a) %max.fneg = fneg float %max @@ -2297,7 +2295,7 @@ define float @v_fneg_self_maximumnum_f32_no_ieee(float %a) #4 { ; GCN-LABEL: v_fneg_self_maximumnum_f32_no_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_min_f32_e64 v0, -v0, -v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %max = call float @llvm.maximumnum.f32(float %a, float %a) %max.fneg = fneg float %max diff --git a/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll b/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll index 40d2765..b0dd187 100644 --- a/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll @@ -11,9 +11,9 @@ ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -global-isel=0 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-SDAG-FAKE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GISEL-TRUE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GISEL-FAKE16 %s -; TODO: FIXME-TRUE16 llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=0 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-SDAG-TRUE16 %s +; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=0 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-SDAG-TRUE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=0 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-SDAG-FAKE16 %s -; TODO: FIXME-TRUE16 llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=1 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-GISEL-TRUE16 %s +; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=1 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-GISEL-TRUE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=1 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-GISEL-FAKE16 %s define amdgpu_kernel void @fptrunc_f32_to_f16( @@ -197,6 +197,24 @@ define amdgpu_kernel void @fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -215,6 +233,21 @@ define amdgpu_kernel void @fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -419,6 +452,24 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_afn(ptr addrspace(1) %r, ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16_afn: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16_afn: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -437,6 +488,21 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_afn(ptr addrspace(1) %r, ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16_afn: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16_afn: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1160,6 +1226,73 @@ define amdgpu_kernel void @fptrunc_f64_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f64_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s2, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s3, s2, 0x1ff +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s2, 8 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v0, s3, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_bfe_u32 s3, s2, 0xb0014 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-SDAG-TRUE16-NEXT: s_sub_co_i32 s4, 0x3f1, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2) +; GFX1250-SDAG-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX1250-SDAG-TRUE16-NEXT: v_med3_i32 v1, s4, 0, 13 +; GFX1250-SDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s8, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s4, s5, s4 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s4, 0x1000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s9, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s9, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s8, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_addk_co_i32 s3, 0xfc10 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s9, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s3, 12 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s4, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s8, s5, 7 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_gt_i32 s8, 5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s9, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s8, 3 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s5, 2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s8, s9 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_add_co_i32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 31 +; GFX1250-SDAG-TRUE16-NEXT: s_movk_i32 s8, 0x7e00 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s4, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s4, s8, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s3, 0x40f +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s4, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s2, s2, 16 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s2, s2, 0x8000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s2, s2, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f64_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1227,6 +1360,63 @@ define amdgpu_kernel void @fptrunc_f64_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f64_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b64 s[2:3], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s6, s3, 0x1ff +; GFX1250-GISEL-TRUE16-NEXT: s_bfe_u32 s4, s3, 0xb0014 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s5, s3, 8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s6, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_addk_co_i32 s4, 0xfc10 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s2, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s5, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s2, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_sub_co_i32 s6, 1, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s8, s2, 0x1000 +; GFX1250-GISEL-TRUE16-NEXT: s_max_i32 s6, s6, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s7, s4, 12 +; GFX1250-GISEL-TRUE16-NEXT: s_min_i32 s6, s6, 13 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s5, s5, 9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s9, s8, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s2, s7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s6, s9, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s5, s5, 0x7c00 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s6, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s9, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lt_i32 s4, 1 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, s6, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s6, s2, 7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s2, s2, 2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s6, 3 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s7, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s6, 5 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s7, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_add_co_i32 s2, s2, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s4, 30 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, 0x7c00, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s4, 0x40f +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, s5, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s3, 16 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s3, s3, 0x8000 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s3, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f64_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1489,6 +1679,26 @@ define amdgpu_kernel void @fptrunc_f64_to_f16_afn( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f64_to_f16_afn: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, v[0:1] +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f64_to_f16_afn: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1509,6 +1719,20 @@ define amdgpu_kernel void @fptrunc_f64_to_f16_afn( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f64_to_f16_afn: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b64 s[2:3], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, s[2:3] +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f64_to_f16_afn: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1740,6 +1964,24 @@ define amdgpu_kernel void @fptrunc_v2f32_to_v2f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_v2f32_to_v2f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v1 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_v2f32_to_v2f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1758,6 +2000,20 @@ define amdgpu_kernel void @fptrunc_v2f32_to_v2f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_v2f32_to_v2f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b64 s[2:3], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b64_e32 v[0:1], s[2:3] +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_v2f32_to_v2f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3017,6 +3273,122 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_v2f64_to_v2f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b128 v[0:3], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s2, v3 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s3, s2, 0x1ff +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s2, 8 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v2, s3, v2 +; GFX1250-SDAG-TRUE16-NEXT: s_bfe_u32 s3, s2, 0xb0014 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-SDAG-TRUE16-NEXT: s_sub_co_i32 s4, 0x3f1, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2) +; GFX1250-SDAG-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v2 +; GFX1250-SDAG-TRUE16-NEXT: v_med3_i32 v3, s4, 0, 13 +; GFX1250-SDAG-TRUE16-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s8, v3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s4, v2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s4, s5, s4 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s4, 0x1000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s9, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s9, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s8, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_addk_co_i32 s3, 0xfc10 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s9, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s3, 12 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s4, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s8, s5, 7 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_gt_i32 s8, 5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s9, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s8, 3 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s5, 2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s8, s9 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_add_co_i32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 31 +; GFX1250-SDAG-TRUE16-NEXT: s_movk_i32 s8, 0x7e00 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s4, 0 +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s4, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s9, s8, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s3, 0x40f +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s9, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s5, s4, 0x1ff +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s10, s4, 8 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v0, s5, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_bfe_u32 s5, s4, 0xb0014 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s10, s10, 0xffe +; GFX1250-SDAG-TRUE16-NEXT: s_sub_co_i32 s9, 0x3f1, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s2, s2, 16 +; GFX1250-SDAG-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX1250-SDAG-TRUE16-NEXT: v_med3_i32 v1, s9, 0, 13 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s2, s2, 0x8000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s2, s2, s3 +; GFX1250-SDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s11, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s9, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s9, s10, s9 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s10, s9, 0x1000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s12, s10, s11 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s11, s12, s11 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s11, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_addk_co_i32 s5, 0xfc10 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s3, s12, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s10, s5, 12 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s10, s9, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s5, 1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s3, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s10, s3, 7 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_gt_i32 s10, 5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s11, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s10, 3 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s10, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s3, s3, 2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s10, s10, s11 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_add_co_i32 s3, s3, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s5, 31 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s3, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s9, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s8, s8, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s5, 0x40f +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s8, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s4, s4, 16 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s4, s4, 0x8000 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s3, s4, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: s_pack_ll_b32_b16 s2, s3, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_v2f64_to_v2f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3133,6 +3505,109 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_v2f64_to_v2f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[4:7], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s8, s5, 0x1ff +; GFX1250-GISEL-TRUE16-NEXT: s_bfe_u32 s2, s5, 0xb0014 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s5, 8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s4, s8, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_addk_co_i32 s2, 0xfc10 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s3, s3, 0xffe +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s4, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s4, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s3, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s3, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s4, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_sub_co_i32 s8, 1, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s10, s3, 0x1000 +; GFX1250-GISEL-TRUE16-NEXT: s_max_i32 s8, s8, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s9, s2, 12 +; GFX1250-GISEL-TRUE16-NEXT: s_min_i32 s8, s8, 13 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s4, s4, 9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s11, s10, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s3, s9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s8, s11, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s4, s4, 0x7c00 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s8, s10 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s8, s11, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lt_i32 s2, 1 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, s8, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s8, s3, 7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s3, 2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s8, 3 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s9, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s8, 5 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s8, s9, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_add_co_i32 s3, s3, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s2, 30 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, 0x7c00, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s2, 0x40f +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, s4, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s5, 16 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s8, s7, 0x1ff +; GFX1250-GISEL-TRUE16-NEXT: s_bfe_u32 s4, s7, 0xb0014 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s5, s7, 8 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s3, s3, 0x8000 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s8, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_addk_co_i32 s4, 0xfc10 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s3, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s6, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s5, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s3, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_sub_co_i32 s6, 1, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s9, s3, 0x1000 +; GFX1250-GISEL-TRUE16-NEXT: s_max_i32 s6, s6, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s8, s4, 12 +; GFX1250-GISEL-TRUE16-NEXT: s_min_i32 s6, s6, 13 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s5, s5, 9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s10, s9, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s3, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s6, s10, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s5, s5, 0x7c00 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s6, s9 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s10, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lt_i32 s4, 1 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, s6, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s6, s3, 7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s3, 2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s6, 3 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s6, 5 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s8, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_add_co_i32 s3, s3, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s4, 30 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, 0x7c00, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s4, 0x40f +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, s5, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s4, s7, 16 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s4, s4, 0x8000 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s4, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_pack_ll_b32_b16 s2, s2, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_v2f64_to_v2f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3481,6 +3956,27 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16_afn( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_v2f64_to_v2f16_afn: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b128 v[0:3], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f32_f64_e32 v2, v[2:3] +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, v[0:1] +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v2 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_v2f64_to_v2f16_afn: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3502,6 +3998,25 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16_afn( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_v2f64_to_v2f16_afn: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[4:7], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, s[4:5] +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f32_f64_e32 v1, s[6:7] +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v1 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-GISEL-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_v2f64_to_v2f16_afn: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3710,6 +4225,26 @@ define amdgpu_kernel void @fneg_fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fneg_fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fneg_fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3730,6 +4265,22 @@ define amdgpu_kernel void @fneg_fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fneg_fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_xor_b32 s2, s2, 0x80000000 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fneg_fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3936,6 +4487,26 @@ define amdgpu_kernel void @fabs_fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fabs_fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fabs_fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3956,6 +4527,22 @@ define amdgpu_kernel void @fabs_fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fabs_fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_bitset0_b32 s2, 31 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fabs_fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4162,6 +4749,26 @@ define amdgpu_kernel void @fneg_fabs_fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v0, 0x80000000, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4182,6 +4789,22 @@ define amdgpu_kernel void @fneg_fabs_fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_bitset1_b32 s2, 31 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4396,6 +5019,26 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_zext_i32( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16_zext_i32: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16_zext_i32: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4416,6 +5059,22 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_zext_i32( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16_zext_i32: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s2, 0xffff, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16_zext_i32: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4630,6 +5289,27 @@ define amdgpu_kernel void @fptrunc_fabs_f32_to_f16_zext_i32( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4651,6 +5331,24 @@ define amdgpu_kernel void @fptrunc_fabs_f32_to_f16_zext_i32( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_bitset0_b32 s2, 31 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s2, 0xffff, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4877,6 +5575,26 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_sext_i32( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16_sext_i32: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16_sext_i32: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4897,6 +5615,22 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_sext_i32( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16_sext_i32: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_sext_i32_i16 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16_sext_i32: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 diff --git a/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll b/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll index 3089054..32f7d6b 100644 --- a/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll +++ b/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll @@ -276,23 +276,23 @@ attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memo ;. ; V4: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; V4: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR5]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } ;. ; V5: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; V5: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V5: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V5: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V5: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } ;. ; V6: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; V6: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V6: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V6: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V6: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } ;. ; V4: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. diff --git a/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll b/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll index d3ef1b7..a0f5d2f 100644 --- a/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll +++ b/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll @@ -68,6 +68,6 @@ if.end: ret void } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll b/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll index 71a330e..4e952b6 100644 --- a/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll +++ b/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll @@ -55,8 +55,8 @@ define amdgpu_kernel void @issue120256_private(ptr addrspace(1) %out) { ; FIXME: Inference of amdgpu-no-queue-ptr should not depend on code object version. !0 = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } ;. ; CHECK: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll index 6ccfad7..ff47563 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll @@ -14,7 +14,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_bf8_f16 v0.l, v0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[2:3], v0 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[2:3], v0, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_bf8_f16_v: @@ -28,7 +28,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_dual_mov_b32 v4, v1 :: v_dual_mov_b32 v5, v2 ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_bf8_f16 v0.l, v0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[4:5], v0 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[4:5], v0, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_bf8_f16_v: @@ -46,7 +46,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_bf8_f16_s: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, s0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_bf8_f16_s: @@ -58,7 +58,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_bf8_f16_s: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, s0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_bf8_f16_s: @@ -75,7 +75,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_l(ptr addrspace(1) %out) { ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_bf8_f16_l: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, 0x56400000 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_bf8_f16_l: @@ -87,7 +87,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_l(ptr addrspace(1) %out) { ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_bf8_f16_l: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, 0x56400000 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_bf8_f16_l: @@ -105,7 +105,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_fp8_f16 v0.l, v0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[2:3], v0 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[2:3], v0, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_fp8_f16_v: @@ -119,7 +119,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_dual_mov_b32 v4, v1 :: v_dual_mov_b32 v5, v2 ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_fp8_f16 v0.l, v0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[4:5], v0 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[4:5], v0, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_fp8_f16_v: @@ -137,7 +137,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_fp8_f16_s: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, s0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_fp8_f16_s: @@ -149,7 +149,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_fp8_f16_s: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, s0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_fp8_f16_s: @@ -166,7 +166,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_l(ptr addrspace(1) %out) { ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_fp8_f16_l: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, 0x56400000 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_fp8_f16_l: @@ -178,7 +178,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_l(ptr addrspace(1) %out) { ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_fp8_f16_l: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, 0x56400000 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_fp8_f16_l: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.form.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.form.ll index 87a7c2e..cc4cc8e 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.form.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.form.ll @@ -72,5 +72,206 @@ define <4 x float> @request_no_agpr(<8 x half> %arg0, <8 x half> %arg1, <4 x flo ret <4 x float> %result } +; Make sure this selects the VGPR form, if AGPRs available, but not +; enough. +define amdgpu_kernel void @not_enough_agprs(ptr addrspace(1) %arg) #2 { +; HEURRC-LABEL: not_enough_agprs: +; HEURRC: ; %bb.0: ; %bb +; HEURRC-NEXT: s_load_dwordx2 s[34:35], s[4:5], 0x24 +; HEURRC-NEXT: v_mov_b32_e32 v33, 1.0 +; HEURRC-NEXT: v_mov_b32_e32 v34, 2.0 +; HEURRC-NEXT: v_mov_b32_e32 v32, 0 +; HEURRC-NEXT: s_waitcnt lgkmcnt(0) +; HEURRC-NEXT: s_load_dwordx16 s[16:31], s[34:35], 0x0 +; HEURRC-NEXT: s_load_dwordx16 s[0:15], s[34:35], 0x40 +; HEURRC-NEXT: s_waitcnt lgkmcnt(0) +; HEURRC-NEXT: v_mov_b32_e32 v0, s16 +; HEURRC-NEXT: v_mov_b32_e32 v1, s17 +; HEURRC-NEXT: v_mov_b32_e32 v2, s18 +; HEURRC-NEXT: v_mov_b32_e32 v3, s19 +; HEURRC-NEXT: v_mov_b32_e32 v4, s20 +; HEURRC-NEXT: v_mov_b32_e32 v5, s21 +; HEURRC-NEXT: v_mov_b32_e32 v6, s22 +; HEURRC-NEXT: v_mov_b32_e32 v7, s23 +; HEURRC-NEXT: v_mov_b32_e32 v8, s24 +; HEURRC-NEXT: v_mov_b32_e32 v9, s25 +; HEURRC-NEXT: v_mov_b32_e32 v10, s26 +; HEURRC-NEXT: v_mov_b32_e32 v11, s27 +; HEURRC-NEXT: v_mov_b32_e32 v12, s28 +; HEURRC-NEXT: v_mov_b32_e32 v13, s29 +; HEURRC-NEXT: v_mov_b32_e32 v14, s30 +; HEURRC-NEXT: v_mov_b32_e32 v15, s31 +; HEURRC-NEXT: v_mov_b32_e32 v16, s0 +; HEURRC-NEXT: v_mov_b32_e32 v17, s1 +; HEURRC-NEXT: v_mov_b32_e32 v18, s2 +; HEURRC-NEXT: v_mov_b32_e32 v19, s3 +; HEURRC-NEXT: v_mov_b32_e32 v20, s4 +; HEURRC-NEXT: v_mov_b32_e32 v21, s5 +; HEURRC-NEXT: v_mov_b32_e32 v22, s6 +; HEURRC-NEXT: v_mov_b32_e32 v23, s7 +; HEURRC-NEXT: v_mov_b32_e32 v24, s8 +; HEURRC-NEXT: v_mov_b32_e32 v25, s9 +; HEURRC-NEXT: v_mov_b32_e32 v26, s10 +; HEURRC-NEXT: v_mov_b32_e32 v27, s11 +; HEURRC-NEXT: v_mov_b32_e32 v28, s12 +; HEURRC-NEXT: v_mov_b32_e32 v29, s13 +; HEURRC-NEXT: v_mov_b32_e32 v30, s14 +; HEURRC-NEXT: v_mov_b32_e32 v31, s15 +; HEURRC-NEXT: s_nop 1 +; HEURRC-NEXT: v_mfma_f32_32x32x1_2b_f32 v[0:31], v33, v34, v[0:31] cbsz:1 abid:2 blgp:3 +; HEURRC-NEXT: s_nop 15 +; HEURRC-NEXT: s_nop 1 +; HEURRC-NEXT: global_store_dwordx4 v32, v[24:27], s[34:35] offset:96 +; HEURRC-NEXT: global_store_dwordx4 v32, v[28:31], s[34:35] offset:112 +; HEURRC-NEXT: global_store_dwordx4 v32, v[16:19], s[34:35] offset:64 +; HEURRC-NEXT: global_store_dwordx4 v32, v[20:23], s[34:35] offset:80 +; HEURRC-NEXT: global_store_dwordx4 v32, v[8:11], s[34:35] offset:32 +; HEURRC-NEXT: global_store_dwordx4 v32, v[12:15], s[34:35] offset:48 +; HEURRC-NEXT: global_store_dwordx4 v32, v[0:3], s[34:35] +; HEURRC-NEXT: global_store_dwordx4 v32, v[4:7], s[34:35] offset:16 +; HEURRC-NEXT: s_endpgm +; +; VGPRRC-LABEL: not_enough_agprs: +; VGPRRC: ; %bb.0: ; %bb +; VGPRRC-NEXT: s_load_dwordx2 s[34:35], s[4:5], 0x24 +; VGPRRC-NEXT: v_mov_b32_e32 v33, 1.0 +; VGPRRC-NEXT: v_mov_b32_e32 v34, 2.0 +; VGPRRC-NEXT: v_mov_b32_e32 v32, 0 +; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) +; VGPRRC-NEXT: s_load_dwordx16 s[16:31], s[34:35], 0x0 +; VGPRRC-NEXT: s_load_dwordx16 s[0:15], s[34:35], 0x40 +; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) +; VGPRRC-NEXT: v_mov_b32_e32 v0, s16 +; VGPRRC-NEXT: v_mov_b32_e32 v1, s17 +; VGPRRC-NEXT: v_mov_b32_e32 v2, s18 +; VGPRRC-NEXT: v_mov_b32_e32 v3, s19 +; VGPRRC-NEXT: v_mov_b32_e32 v4, s20 +; VGPRRC-NEXT: v_mov_b32_e32 v5, s21 +; VGPRRC-NEXT: v_mov_b32_e32 v6, s22 +; VGPRRC-NEXT: v_mov_b32_e32 v7, s23 +; VGPRRC-NEXT: v_mov_b32_e32 v8, s24 +; VGPRRC-NEXT: v_mov_b32_e32 v9, s25 +; VGPRRC-NEXT: v_mov_b32_e32 v10, s26 +; VGPRRC-NEXT: v_mov_b32_e32 v11, s27 +; VGPRRC-NEXT: v_mov_b32_e32 v12, s28 +; VGPRRC-NEXT: v_mov_b32_e32 v13, s29 +; VGPRRC-NEXT: v_mov_b32_e32 v14, s30 +; VGPRRC-NEXT: v_mov_b32_e32 v15, s31 +; VGPRRC-NEXT: v_mov_b32_e32 v16, s0 +; VGPRRC-NEXT: v_mov_b32_e32 v17, s1 +; VGPRRC-NEXT: v_mov_b32_e32 v18, s2 +; VGPRRC-NEXT: v_mov_b32_e32 v19, s3 +; VGPRRC-NEXT: v_mov_b32_e32 v20, s4 +; VGPRRC-NEXT: v_mov_b32_e32 v21, s5 +; VGPRRC-NEXT: v_mov_b32_e32 v22, s6 +; VGPRRC-NEXT: v_mov_b32_e32 v23, s7 +; VGPRRC-NEXT: v_mov_b32_e32 v24, s8 +; VGPRRC-NEXT: v_mov_b32_e32 v25, s9 +; VGPRRC-NEXT: v_mov_b32_e32 v26, s10 +; VGPRRC-NEXT: v_mov_b32_e32 v27, s11 +; VGPRRC-NEXT: v_mov_b32_e32 v28, s12 +; VGPRRC-NEXT: v_mov_b32_e32 v29, s13 +; VGPRRC-NEXT: v_mov_b32_e32 v30, s14 +; VGPRRC-NEXT: v_mov_b32_e32 v31, s15 +; VGPRRC-NEXT: s_nop 1 +; VGPRRC-NEXT: v_mfma_f32_32x32x1_2b_f32 v[0:31], v33, v34, v[0:31] cbsz:1 abid:2 blgp:3 +; VGPRRC-NEXT: s_nop 15 +; VGPRRC-NEXT: s_nop 1 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[24:27], s[34:35] offset:96 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[28:31], s[34:35] offset:112 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[16:19], s[34:35] offset:64 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[20:23], s[34:35] offset:80 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[8:11], s[34:35] offset:32 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[12:15], s[34:35] offset:48 +; VGPRRC-NEXT: global_store_dwordx4 v32, v[0:3], s[34:35] +; VGPRRC-NEXT: global_store_dwordx4 v32, v[4:7], s[34:35] offset:16 +; VGPRRC-NEXT: s_endpgm +bb: + %in.1 = load <32 x float>, ptr addrspace(1) %arg, align 128 + %mai.1 = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 1.000000e+00, float 2.000000e+00, <32 x float> %in.1, i32 1, i32 2, i32 3) + store <32 x float> %mai.1, ptr addrspace(1) %arg, align 128 + ret void +} + +define <16 x float> @mfma_scale_respect_flag(<8 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, i32 %scale0, i32 %scale1) #2 { +; HEURRC-LABEL: mfma_scale_respect_flag: +; HEURRC: ; %bb.0: +; HEURRC-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; HEURRC-NEXT: scratch_load_dword a15, off, s32 +; HEURRC-NEXT: scratch_load_dword v31, off, s32 offset:8 +; HEURRC-NEXT: scratch_load_dword v32, off, s32 offset:4 +; HEURRC-NEXT: v_accvgpr_write_b32 a0, v16 +; HEURRC-NEXT: v_accvgpr_write_b32 a1, v17 +; HEURRC-NEXT: v_accvgpr_write_b32 a2, v18 +; HEURRC-NEXT: v_accvgpr_write_b32 a3, v19 +; HEURRC-NEXT: v_accvgpr_write_b32 a4, v20 +; HEURRC-NEXT: v_accvgpr_write_b32 a5, v21 +; HEURRC-NEXT: v_accvgpr_write_b32 a6, v22 +; HEURRC-NEXT: v_accvgpr_write_b32 a7, v23 +; HEURRC-NEXT: v_accvgpr_write_b32 a8, v24 +; HEURRC-NEXT: v_accvgpr_write_b32 a9, v25 +; HEURRC-NEXT: v_accvgpr_write_b32 a10, v26 +; HEURRC-NEXT: v_accvgpr_write_b32 a11, v27 +; HEURRC-NEXT: v_accvgpr_write_b32 a12, v28 +; HEURRC-NEXT: v_accvgpr_write_b32 a13, v29 +; HEURRC-NEXT: v_accvgpr_write_b32 a14, v30 +; HEURRC-NEXT: s_waitcnt vmcnt(0) +; HEURRC-NEXT: s_nop 0 +; HEURRC-NEXT: v_mfma_scale_f32_32x32x64_f8f6f4 a[0:15], v[0:7], v[8:15], a[0:15], v32, v31 op_sel_hi:[0,0,0] +; HEURRC-NEXT: s_nop 15 +; HEURRC-NEXT: s_nop 3 +; HEURRC-NEXT: v_accvgpr_read_b32 v0, a0 +; HEURRC-NEXT: v_accvgpr_read_b32 v1, a1 +; HEURRC-NEXT: v_accvgpr_read_b32 v2, a2 +; HEURRC-NEXT: v_accvgpr_read_b32 v3, a3 +; HEURRC-NEXT: v_accvgpr_read_b32 v4, a4 +; HEURRC-NEXT: v_accvgpr_read_b32 v5, a5 +; HEURRC-NEXT: v_accvgpr_read_b32 v6, a6 +; HEURRC-NEXT: v_accvgpr_read_b32 v7, a7 +; HEURRC-NEXT: v_accvgpr_read_b32 v8, a8 +; HEURRC-NEXT: v_accvgpr_read_b32 v9, a9 +; HEURRC-NEXT: v_accvgpr_read_b32 v10, a10 +; HEURRC-NEXT: v_accvgpr_read_b32 v11, a11 +; HEURRC-NEXT: v_accvgpr_read_b32 v12, a12 +; HEURRC-NEXT: v_accvgpr_read_b32 v13, a13 +; HEURRC-NEXT: v_accvgpr_read_b32 v14, a14 +; HEURRC-NEXT: v_accvgpr_read_b32 v15, a15 +; HEURRC-NEXT: s_setpc_b64 s[30:31] +; +; VGPRRC-LABEL: mfma_scale_respect_flag: +; VGPRRC: ; %bb.0: +; VGPRRC-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VGPRRC-NEXT: scratch_load_dword v31, off, s32 +; VGPRRC-NEXT: scratch_load_dword v32, off, s32 offset:8 +; VGPRRC-NEXT: scratch_load_dword v33, off, s32 offset:4 +; VGPRRC-NEXT: s_waitcnt vmcnt(0) +; VGPRRC-NEXT: v_mfma_scale_f32_32x32x64_f8f6f4 v[16:31], v[0:7], v[8:15], v[16:31], v33, v32 op_sel_hi:[0,0,0] +; VGPRRC-NEXT: s_nop 15 +; VGPRRC-NEXT: s_nop 3 +; VGPRRC-NEXT: v_mov_b32_e32 v0, v16 +; VGPRRC-NEXT: v_mov_b32_e32 v1, v17 +; VGPRRC-NEXT: v_mov_b32_e32 v2, v18 +; VGPRRC-NEXT: v_mov_b32_e32 v3, v19 +; VGPRRC-NEXT: v_mov_b32_e32 v4, v20 +; VGPRRC-NEXT: v_mov_b32_e32 v5, v21 +; VGPRRC-NEXT: v_mov_b32_e32 v6, v22 +; VGPRRC-NEXT: v_mov_b32_e32 v7, v23 +; VGPRRC-NEXT: v_mov_b32_e32 v8, v24 +; VGPRRC-NEXT: v_mov_b32_e32 v9, v25 +; VGPRRC-NEXT: v_mov_b32_e32 v10, v26 +; VGPRRC-NEXT: v_mov_b32_e32 v11, v27 +; VGPRRC-NEXT: v_mov_b32_e32 v12, v28 +; VGPRRC-NEXT: v_mov_b32_e32 v13, v29 +; VGPRRC-NEXT: v_mov_b32_e32 v14, v30 +; VGPRRC-NEXT: v_mov_b32_e32 v15, v31 +; VGPRRC-NEXT: s_setpc_b64 s[30:31] + %result = call <16 x float> @llvm.amdgcn.mfma.scale.f32.32x32x64.f8f6f4.v8i32.v8i32(<8 x i32> %arg0, <8 x i32> %arg1, <16 x float> %arg2, + i32 0, ; cbsz + i32 0, ; blgp + i32 0, i32 %scale0, i32 0, i32 %scale1) + ret <16 x float> %result +} + attributes #0 = { "amdgpu-agpr-alloc"="32,256" } attributes #1 = { "amdgpu-agpr-alloc"="0,0" } +attributes #2 = { nounwind "amdgpu-agpr-alloc"="20" } diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx90a.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx90a.ll index 5ab8706..22bc62a 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx90a.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx90a.ll @@ -726,12 +726,12 @@ define amdgpu_kernel void @test_mfma_f64_4x4x4f64(ptr addrspace(1) %arg, double ; GFX90A-VGPR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 ; GFX90A-VGPR-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 ; GFX90A-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[0:1], s[2:3], s[2:3] op_sel:[0,1] -; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[2:3], s[6:7], s[6:7] op_sel:[0,1] +; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[2:3], s[2:3], s[2:3] op_sel:[0,1] +; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[4:5], s[6:7], s[6:7] op_sel:[0,1] ; GFX90A-VGPR-NEXT: s_nop 1 -; GFX90A-VGPR-NEXT: v_mfma_f64_4x4x4f64 v[4:5], v[0:1], v[2:3], 0 +; GFX90A-VGPR-NEXT: v_mfma_f64_4x4x4f64 v[0:1], v[2:3], v[4:5], 0 ; GFX90A-VGPR-NEXT: s_nop 3 -; GFX90A-VGPR-NEXT: v_mfma_f64_4x4x4f64 v[0:1], v[0:1], v[2:3], v[4:5] cbsz:1 abid:2 blgp:3 +; GFX90A-VGPR-NEXT: v_mfma_f64_4x4x4f64 v[0:1], v[2:3], v[4:5], v[0:1] cbsz:1 abid:2 blgp:3 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v2, 0 ; GFX90A-VGPR-NEXT: s_nop 7 ; GFX90A-VGPR-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] @@ -742,12 +742,12 @@ define amdgpu_kernel void @test_mfma_f64_4x4x4f64(ptr addrspace(1) %arg, double ; GFX942-VGPR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 ; GFX942-VGPR-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 ; GFX942-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-VGPR-NEXT: v_mov_b64_e32 v[0:1], s[2:3] -; GFX942-VGPR-NEXT: v_mov_b64_e32 v[2:3], s[6:7] +; GFX942-VGPR-NEXT: v_mov_b64_e32 v[2:3], s[2:3] +; GFX942-VGPR-NEXT: v_mov_b64_e32 v[4:5], s[6:7] ; GFX942-VGPR-NEXT: s_nop 1 -; GFX942-VGPR-NEXT: v_mfma_f64_4x4x4_4b_f64 v[4:5], v[0:1], v[2:3], 0 +; GFX942-VGPR-NEXT: v_mfma_f64_4x4x4_4b_f64 v[0:1], v[2:3], v[4:5], 0 ; GFX942-VGPR-NEXT: s_nop 3 -; GFX942-VGPR-NEXT: v_mfma_f64_4x4x4_4b_f64 v[0:1], v[0:1], v[2:3], v[4:5] cbsz:1 abid:2 neg:[1,1,0] +; GFX942-VGPR-NEXT: v_mfma_f64_4x4x4_4b_f64 v[0:1], v[2:3], v[4:5], v[0:1] cbsz:1 abid:2 neg:[1,1,0] ; GFX942-VGPR-NEXT: v_mov_b32_e32 v2, 0 ; GFX942-VGPR-NEXT: s_nop 7 ; GFX942-VGPR-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] @@ -765,10 +765,10 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64(ptr addrspace(1) %arg, doubl ; GFX90A-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x24 ; GFX90A-NEXT: s_load_dwordx2 s[12:13], s[4:5], 0x34 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_mov_b32_e32 v2, s10 +; GFX90A-NEXT: v_mov_b32_e32 v0, s10 ; GFX90A-NEXT: s_load_dwordx8 s[0:7], s[8:9], 0x0 -; GFX90A-NEXT: v_mov_b32_e32 v3, s11 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], s[12:13], s[12:13] op_sel:[0,1] +; GFX90A-NEXT: v_mov_b32_e32 v1, s11 +; GFX90A-NEXT: v_pk_mov_b32 v[2:3], s[12:13], s[12:13] op_sel:[0,1] ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) ; GFX90A-NEXT: v_accvgpr_write_b32 a0, s0 ; GFX90A-NEXT: v_accvgpr_write_b32 a1, s1 @@ -779,7 +779,7 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64(ptr addrspace(1) %arg, doubl ; GFX90A-NEXT: v_accvgpr_write_b32 a6, s6 ; GFX90A-NEXT: v_accvgpr_write_b32 a7, s7 ; GFX90A-NEXT: s_nop 1 -; GFX90A-NEXT: v_mfma_f64_16x16x4f64 a[0:7], v[2:3], v[0:1], a[0:7] cbsz:1 abid:2 blgp:3 +; GFX90A-NEXT: v_mfma_f64_16x16x4f64 a[0:7], v[0:1], v[2:3], a[0:7] cbsz:1 abid:2 blgp:3 ; GFX90A-NEXT: v_mov_b32_e32 v0, 0 ; GFX90A-NEXT: s_nop 15 ; GFX90A-NEXT: s_nop 0 @@ -792,10 +792,10 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64(ptr addrspace(1) %arg, doubl ; GFX942-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x24 ; GFX942-NEXT: s_load_dwordx2 s[12:13], s[4:5], 0x34 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_mov_b32_e32 v2, s10 +; GFX942-NEXT: v_mov_b32_e32 v0, s10 ; GFX942-NEXT: s_load_dwordx8 s[0:7], s[8:9], 0x0 -; GFX942-NEXT: v_mov_b32_e32 v3, s11 -; GFX942-NEXT: v_mov_b64_e32 v[0:1], s[12:13] +; GFX942-NEXT: v_mov_b32_e32 v1, s11 +; GFX942-NEXT: v_mov_b64_e32 v[2:3], s[12:13] ; GFX942-NEXT: s_waitcnt lgkmcnt(0) ; GFX942-NEXT: v_accvgpr_write_b32 a0, s0 ; GFX942-NEXT: v_accvgpr_write_b32 a1, s1 @@ -806,7 +806,7 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64(ptr addrspace(1) %arg, doubl ; GFX942-NEXT: v_accvgpr_write_b32 a6, s6 ; GFX942-NEXT: v_accvgpr_write_b32 a7, s7 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_mfma_f64_16x16x4_f64 a[0:7], v[2:3], v[0:1], a[0:7] cbsz:1 abid:2 neg:[1,1,0] +; GFX942-NEXT: v_mfma_f64_16x16x4_f64 a[0:7], v[0:1], v[2:3], a[0:7] cbsz:1 abid:2 neg:[1,1,0] ; GFX942-NEXT: v_mov_b32_e32 v0, 0 ; GFX942-NEXT: s_nop 15 ; GFX942-NEXT: s_nop 0 @@ -819,17 +819,17 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64(ptr addrspace(1) %arg, doubl ; GFX90A-VGPR-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x24 ; GFX90A-VGPR-NEXT: s_load_dwordx2 s[12:13], s[4:5], 0x34 ; GFX90A-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-VGPR-NEXT: v_mov_b32_e32 v10, s10 +; GFX90A-VGPR-NEXT: v_mov_b32_e32 v8, s10 ; GFX90A-VGPR-NEXT: s_load_dwordx8 s[0:7], s[8:9], 0x0 -; GFX90A-VGPR-NEXT: v_mov_b32_e32 v11, s11 -; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[8:9], s[12:13], s[12:13] op_sel:[0,1] +; GFX90A-VGPR-NEXT: v_mov_b32_e32 v9, s11 +; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[10:11], s[12:13], s[12:13] op_sel:[0,1] ; GFX90A-VGPR-NEXT: s_waitcnt lgkmcnt(0) ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[0:1], s[0:1], s[0:1] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[2:3], s[2:3], s[2:3] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[4:5], s[4:5], s[4:5] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[6:7], s[6:7], s[6:7] op_sel:[0,1] ; GFX90A-VGPR-NEXT: s_nop 1 -; GFX90A-VGPR-NEXT: v_mfma_f64_16x16x4f64 v[0:7], v[10:11], v[8:9], v[0:7] cbsz:1 abid:2 blgp:3 +; GFX90A-VGPR-NEXT: v_mfma_f64_16x16x4f64 v[0:7], v[8:9], v[10:11], v[0:7] cbsz:1 abid:2 blgp:3 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v8, 0 ; GFX90A-VGPR-NEXT: s_nop 15 ; GFX90A-VGPR-NEXT: s_nop 0 @@ -842,17 +842,17 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64(ptr addrspace(1) %arg, doubl ; GFX942-VGPR-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x24 ; GFX942-VGPR-NEXT: s_load_dwordx2 s[12:13], s[4:5], 0x34 ; GFX942-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-VGPR-NEXT: v_mov_b32_e32 v10, s10 +; GFX942-VGPR-NEXT: v_mov_b32_e32 v8, s10 ; GFX942-VGPR-NEXT: s_load_dwordx8 s[0:7], s[8:9], 0x0 -; GFX942-VGPR-NEXT: v_mov_b32_e32 v11, s11 -; GFX942-VGPR-NEXT: v_mov_b64_e32 v[8:9], s[12:13] +; GFX942-VGPR-NEXT: v_mov_b32_e32 v9, s11 +; GFX942-VGPR-NEXT: v_mov_b64_e32 v[10:11], s[12:13] ; GFX942-VGPR-NEXT: s_waitcnt lgkmcnt(0) ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[0:1], s[0:1] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[4:5], s[4:5] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[6:7], s[6:7] ; GFX942-VGPR-NEXT: s_nop 1 -; GFX942-VGPR-NEXT: v_mfma_f64_16x16x4_f64 v[0:7], v[10:11], v[8:9], v[0:7] cbsz:1 abid:2 neg:[1,1,0] +; GFX942-VGPR-NEXT: v_mfma_f64_16x16x4_f64 v[0:7], v[8:9], v[10:11], v[0:7] cbsz:1 abid:2 neg:[1,1,0] ; GFX942-VGPR-NEXT: v_mov_b32_e32 v8, 0 ; GFX942-VGPR-NEXT: s_nop 15 ; GFX942-VGPR-NEXT: s_nop 0 @@ -1629,20 +1629,20 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64_imm(ptr addrspace(1) %arg, d ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v7, 0x3ff00000 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v2, v0 ; GFX90A-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-VGPR-NEXT: v_mov_b32_e32 v12, s2 -; GFX90A-VGPR-NEXT: v_mov_b32_e32 v13, s3 +; GFX90A-VGPR-NEXT: v_mov_b32_e32 v10, s2 +; GFX90A-VGPR-NEXT: v_mov_b32_e32 v11, s3 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v3, v0 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v4, v0 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v5, v0 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v6, v0 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v1, v0 ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[8:9], v[6:7], v[6:7] op_sel:[0,1] -; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[10:11], s[6:7], s[6:7] op_sel:[0,1] +; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[12:13], s[6:7], s[6:7] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[6:7], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1] ; GFX90A-VGPR-NEXT: s_nop 1 -; GFX90A-VGPR-NEXT: v_mfma_f64_16x16x4f64 v[2:9], v[12:13], v[10:11], v[2:9] +; GFX90A-VGPR-NEXT: v_mfma_f64_16x16x4f64 v[2:9], v[10:11], v[12:13], v[2:9] ; GFX90A-VGPR-NEXT: s_nop 15 ; GFX90A-VGPR-NEXT: s_nop 1 ; GFX90A-VGPR-NEXT: global_store_dwordx4 v0, v[6:9], s[0:1] offset:16 @@ -1657,20 +1657,20 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64_imm(ptr addrspace(1) %arg, d ; GFX942-VGPR-NEXT: v_mov_b32_e32 v7, 0x3ff00000 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v2, v0 ; GFX942-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-VGPR-NEXT: v_mov_b32_e32 v12, s2 -; GFX942-VGPR-NEXT: v_mov_b32_e32 v13, s3 +; GFX942-VGPR-NEXT: v_mov_b32_e32 v10, s2 +; GFX942-VGPR-NEXT: v_mov_b32_e32 v11, s3 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v3, v0 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v4, v0 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v5, v0 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v6, v0 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v1, v0 ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[8:9], v[6:7] -; GFX942-VGPR-NEXT: v_mov_b64_e32 v[10:11], s[6:7] +; GFX942-VGPR-NEXT: v_mov_b64_e32 v[12:13], s[6:7] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[6:7], v[4:5] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[4:5], v[2:3] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[2:3], v[0:1] ; GFX942-VGPR-NEXT: s_nop 1 -; GFX942-VGPR-NEXT: v_mfma_f64_16x16x4_f64 v[2:9], v[12:13], v[10:11], v[2:9] +; GFX942-VGPR-NEXT: v_mfma_f64_16x16x4_f64 v[2:9], v[10:11], v[12:13], v[2:9] ; GFX942-VGPR-NEXT: s_nop 15 ; GFX942-VGPR-NEXT: s_nop 1 ; GFX942-VGPR-NEXT: global_store_dwordx4 v0, v[6:9], s[0:1] offset:16 @@ -1743,20 +1743,20 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64_splat_lit(ptr addrspace(1) % ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v1, 0x405ec000 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v2, v0 ; GFX90A-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-VGPR-NEXT: v_mov_b32_e32 v12, s2 -; GFX90A-VGPR-NEXT: v_mov_b32_e32 v13, s3 +; GFX90A-VGPR-NEXT: v_mov_b32_e32 v10, s2 +; GFX90A-VGPR-NEXT: v_mov_b32_e32 v11, s3 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v3, v1 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v4, v0 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v5, v1 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v6, v0 ; GFX90A-VGPR-NEXT: v_mov_b32_e32 v7, v1 ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[8:9], v[6:7], v[6:7] op_sel:[0,1] -; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[10:11], s[6:7], s[6:7] op_sel:[0,1] +; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[12:13], s[6:7], s[6:7] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[6:7], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] ; GFX90A-VGPR-NEXT: v_pk_mov_b32 v[2:3], v[0:1], v[0:1] op_sel:[0,1] ; GFX90A-VGPR-NEXT: s_nop 1 -; GFX90A-VGPR-NEXT: v_mfma_f64_16x16x4f64 v[2:9], v[12:13], v[10:11], v[2:9] +; GFX90A-VGPR-NEXT: v_mfma_f64_16x16x4f64 v[2:9], v[10:11], v[12:13], v[2:9] ; GFX90A-VGPR-NEXT: s_nop 15 ; GFX90A-VGPR-NEXT: s_nop 1 ; GFX90A-VGPR-NEXT: global_store_dwordx4 v0, v[6:9], s[0:1] offset:16 @@ -1771,20 +1771,20 @@ define amdgpu_kernel void @test_mfma_f64_16x16x4f64_splat_lit(ptr addrspace(1) % ; GFX942-VGPR-NEXT: v_mov_b32_e32 v1, 0x405ec000 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v2, v0 ; GFX942-VGPR-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-VGPR-NEXT: v_mov_b32_e32 v12, s2 -; GFX942-VGPR-NEXT: v_mov_b32_e32 v13, s3 +; GFX942-VGPR-NEXT: v_mov_b32_e32 v10, s2 +; GFX942-VGPR-NEXT: v_mov_b32_e32 v11, s3 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v3, v1 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v4, v0 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v5, v1 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v6, v0 ; GFX942-VGPR-NEXT: v_mov_b32_e32 v7, v1 ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[8:9], v[6:7] -; GFX942-VGPR-NEXT: v_mov_b64_e32 v[10:11], s[6:7] +; GFX942-VGPR-NEXT: v_mov_b64_e32 v[12:13], s[6:7] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[6:7], v[4:5] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[4:5], v[2:3] ; GFX942-VGPR-NEXT: v_mov_b64_e32 v[2:3], v[0:1] ; GFX942-VGPR-NEXT: s_nop 1 -; GFX942-VGPR-NEXT: v_mfma_f64_16x16x4_f64 v[2:9], v[12:13], v[10:11], v[2:9] +; GFX942-VGPR-NEXT: v_mfma_f64_16x16x4_f64 v[2:9], v[10:11], v[12:13], v[2:9] ; GFX942-VGPR-NEXT: s_nop 15 ; GFX942-VGPR-NEXT: s_nop 1 ; GFX942-VGPR-NEXT: global_store_dwordx4 v0, v[6:9], s[0:1] offset:16 diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll index dc4c9291..2fb677e 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx942.ll @@ -1445,20 +1445,20 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x32_f16(ptr addrspace(1) %arg, < ; GFX942-SDAG: ; %bb.0: ; %bb ; GFX942-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 ; GFX942-SDAG-NEXT: s_load_dword s6, s[4:5], 0x44 -; GFX942-SDAG-NEXT: v_mov_b32_e32 v6, 0 +; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, 0 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX942-SDAG-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[10:11] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[12:13] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[14:15] -; GFX942-SDAG-NEXT: v_mov_b32_e32 v7, s6 +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[12:13] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[14:15] +; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s6 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[2:3] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[6:7], s[0:1] ; GFX942-SDAG-NEXT: s_nop 1 -; GFX942-SDAG-NEXT: v_smfmac_f32_16x16x32_f16 v[8:11], v[4:5], v[0:3], v7 cbsz:1 abid:2 +; GFX942-SDAG-NEXT: v_smfmac_f32_16x16x32_f16 v[6:9], v[10:11], v[2:5], v1 cbsz:1 abid:2 ; GFX942-SDAG-NEXT: s_nop 6 -; GFX942-SDAG-NEXT: global_store_dwordx4 v6, v[8:11], s[8:9] +; GFX942-SDAG-NEXT: global_store_dwordx4 v0, v[6:9], s[8:9] ; GFX942-SDAG-NEXT: s_endpgm ; ; GFX942-GISEL-LABEL: test_smfmac_f32_16x16x32_f16: @@ -1485,20 +1485,20 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x32_f16(ptr addrspace(1) %arg, < ; GFX950-SDAG: ; %bb.0: ; %bb ; GFX950-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 ; GFX950-SDAG-NEXT: s_load_dword s6, s[4:5], 0x44 -; GFX950-SDAG-NEXT: v_mov_b32_e32 v6, 0 +; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, 0 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX950-SDAG-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[10:11] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[12:13] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[14:15] -; GFX950-SDAG-NEXT: v_mov_b32_e32 v7, s6 +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[12:13] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[14:15] +; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s6 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[2:3] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[6:7], s[0:1] ; GFX950-SDAG-NEXT: s_nop 1 -; GFX950-SDAG-NEXT: v_smfmac_f32_16x16x32_f16 v[8:11], v[4:5], v[0:3], v7 cbsz:1 abid:2 +; GFX950-SDAG-NEXT: v_smfmac_f32_16x16x32_f16 v[6:9], v[10:11], v[2:5], v1 cbsz:1 abid:2 ; GFX950-SDAG-NEXT: s_nop 7 -; GFX950-SDAG-NEXT: global_store_dwordx4 v6, v[8:11], s[8:9] +; GFX950-SDAG-NEXT: global_store_dwordx4 v0, v[6:9], s[8:9] ; GFX950-SDAG-NEXT: s_endpgm ; ; GFX950-GISEL-LABEL: test_smfmac_f32_16x16x32_f16: @@ -1577,11 +1577,11 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_f16(ptr addrspace(1) %arg, < ; GFX942-SDAG-NEXT: s_load_dwordx8 s[16:23], s[4:5], 0x24 ; GFX942-SDAG-NEXT: s_load_dword s24, s[4:5], 0x44 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[18:19] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[22:23], s[18:19] ; GFX942-SDAG-NEXT: s_load_dwordx16 s[0:15], s[16:17], 0x0 -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[16:17], s[20:21] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[22:23] -; GFX942-SDAG-NEXT: v_mov_b32_e32 v22, s24 +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[20:21] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[22:23] +; GFX942-SDAG-NEXT: v_mov_b32_e32 v16, s24 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[0:1] ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[2:3] @@ -1592,7 +1592,7 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_f16(ptr addrspace(1) %arg, < ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[12:13], s[12:13] ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[14:15], s[14:15] ; GFX942-SDAG-NEXT: s_nop 1 -; GFX942-SDAG-NEXT: v_smfmac_f32_32x32x16_f16 v[0:15], v[20:21], v[16:19], v22 cbsz:1 abid:2 +; GFX942-SDAG-NEXT: v_smfmac_f32_32x32x16_f16 v[0:15], v[22:23], v[18:21], v16 cbsz:1 abid:2 ; GFX942-SDAG-NEXT: v_mov_b32_e32 v16, 0 ; GFX942-SDAG-NEXT: s_nop 9 ; GFX942-SDAG-NEXT: global_store_dwordx4 v16, v[12:15], s[16:17] offset:48 @@ -1635,11 +1635,11 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_f16(ptr addrspace(1) %arg, < ; GFX950-SDAG-NEXT: s_load_dwordx8 s[16:23], s[4:5], 0x24 ; GFX950-SDAG-NEXT: s_load_dword s24, s[4:5], 0x44 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[18:19] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[22:23], s[18:19] ; GFX950-SDAG-NEXT: s_load_dwordx16 s[0:15], s[16:17], 0x0 -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[16:17], s[20:21] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[22:23] -; GFX950-SDAG-NEXT: v_mov_b32_e32 v22, s24 +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[20:21] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[22:23] +; GFX950-SDAG-NEXT: v_mov_b32_e32 v16, s24 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[0:1] ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[2:3] @@ -1650,7 +1650,7 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_f16(ptr addrspace(1) %arg, < ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[12:13], s[12:13] ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[14:15], s[14:15] ; GFX950-SDAG-NEXT: s_nop 1 -; GFX950-SDAG-NEXT: v_smfmac_f32_32x32x16_f16 v[0:15], v[20:21], v[16:19], v22 cbsz:1 abid:2 +; GFX950-SDAG-NEXT: v_smfmac_f32_32x32x16_f16 v[0:15], v[22:23], v[18:21], v16 cbsz:1 abid:2 ; GFX950-SDAG-NEXT: v_mov_b32_e32 v16, 0 ; GFX950-SDAG-NEXT: s_nop 10 ; GFX950-SDAG-NEXT: global_store_dwordx4 v16, v[12:15], s[16:17] offset:48 @@ -1847,20 +1847,20 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x32_bf16(ptr addrspace(1) %arg, ; GFX942-SDAG: ; %bb.0: ; %bb ; GFX942-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 ; GFX942-SDAG-NEXT: s_load_dword s6, s[4:5], 0x44 -; GFX942-SDAG-NEXT: v_mov_b32_e32 v6, 0 +; GFX942-SDAG-NEXT: v_mov_b32_e32 v0, 0 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX942-SDAG-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[10:11] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[12:13] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[14:15] -; GFX942-SDAG-NEXT: v_mov_b32_e32 v7, s6 +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[12:13] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[14:15] +; GFX942-SDAG-NEXT: v_mov_b32_e32 v1, s6 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[2:3] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[6:7], s[0:1] ; GFX942-SDAG-NEXT: s_nop 1 -; GFX942-SDAG-NEXT: v_smfmac_f32_16x16x32_bf16 v[8:11], v[4:5], v[0:3], v7 cbsz:1 abid:2 +; GFX942-SDAG-NEXT: v_smfmac_f32_16x16x32_bf16 v[6:9], v[10:11], v[2:5], v1 cbsz:1 abid:2 ; GFX942-SDAG-NEXT: s_nop 6 -; GFX942-SDAG-NEXT: global_store_dwordx4 v6, v[8:11], s[8:9] +; GFX942-SDAG-NEXT: global_store_dwordx4 v0, v[6:9], s[8:9] ; GFX942-SDAG-NEXT: s_endpgm ; ; GFX942-GISEL-LABEL: test_smfmac_f32_16x16x32_bf16: @@ -1887,20 +1887,20 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x32_bf16(ptr addrspace(1) %arg, ; GFX950-SDAG: ; %bb.0: ; %bb ; GFX950-SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 ; GFX950-SDAG-NEXT: s_load_dword s6, s[4:5], 0x44 -; GFX950-SDAG-NEXT: v_mov_b32_e32 v6, 0 +; GFX950-SDAG-NEXT: v_mov_b32_e32 v0, 0 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX950-SDAG-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[10:11] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[12:13] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[14:15] -; GFX950-SDAG-NEXT: v_mov_b32_e32 v7, s6 +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[12:13] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[4:5], s[14:15] +; GFX950-SDAG-NEXT: v_mov_b32_e32 v1, s6 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[8:9], s[2:3] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[6:7], s[0:1] ; GFX950-SDAG-NEXT: s_nop 1 -; GFX950-SDAG-NEXT: v_smfmac_f32_16x16x32_bf16 v[8:11], v[4:5], v[0:3], v7 cbsz:1 abid:2 +; GFX950-SDAG-NEXT: v_smfmac_f32_16x16x32_bf16 v[6:9], v[10:11], v[2:5], v1 cbsz:1 abid:2 ; GFX950-SDAG-NEXT: s_nop 7 -; GFX950-SDAG-NEXT: global_store_dwordx4 v6, v[8:11], s[8:9] +; GFX950-SDAG-NEXT: global_store_dwordx4 v0, v[6:9], s[8:9] ; GFX950-SDAG-NEXT: s_endpgm ; ; GFX950-GISEL-LABEL: test_smfmac_f32_16x16x32_bf16: @@ -1979,11 +1979,11 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_bf16(ptr addrspace(1) %arg, ; GFX942-SDAG-NEXT: s_load_dwordx8 s[16:23], s[4:5], 0x24 ; GFX942-SDAG-NEXT: s_load_dword s24, s[4:5], 0x44 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[18:19] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[22:23], s[18:19] ; GFX942-SDAG-NEXT: s_load_dwordx16 s[0:15], s[16:17], 0x0 -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[16:17], s[20:21] -; GFX942-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[22:23] -; GFX942-SDAG-NEXT: v_mov_b32_e32 v22, s24 +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[20:21] +; GFX942-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[22:23] +; GFX942-SDAG-NEXT: v_mov_b32_e32 v16, s24 ; GFX942-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[0:1] ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[2:3] @@ -1994,7 +1994,7 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_bf16(ptr addrspace(1) %arg, ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[12:13], s[12:13] ; GFX942-SDAG-NEXT: v_mov_b64_e32 v[14:15], s[14:15] ; GFX942-SDAG-NEXT: s_nop 1 -; GFX942-SDAG-NEXT: v_smfmac_f32_32x32x16_bf16 v[0:15], v[20:21], v[16:19], v22 cbsz:1 abid:2 +; GFX942-SDAG-NEXT: v_smfmac_f32_32x32x16_bf16 v[0:15], v[22:23], v[18:21], v16 cbsz:1 abid:2 ; GFX942-SDAG-NEXT: v_mov_b32_e32 v16, 0 ; GFX942-SDAG-NEXT: s_nop 9 ; GFX942-SDAG-NEXT: global_store_dwordx4 v16, v[12:15], s[16:17] offset:48 @@ -2037,11 +2037,11 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_bf16(ptr addrspace(1) %arg, ; GFX950-SDAG-NEXT: s_load_dwordx8 s[16:23], s[4:5], 0x24 ; GFX950-SDAG-NEXT: s_load_dword s24, s[4:5], 0x44 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[18:19] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[22:23], s[18:19] ; GFX950-SDAG-NEXT: s_load_dwordx16 s[0:15], s[16:17], 0x0 -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[16:17], s[20:21] -; GFX950-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[22:23] -; GFX950-SDAG-NEXT: v_mov_b32_e32 v22, s24 +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[18:19], s[20:21] +; GFX950-SDAG-NEXT: v_mov_b64_e32 v[20:21], s[22:23] +; GFX950-SDAG-NEXT: v_mov_b32_e32 v16, s24 ; GFX950-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[0:1], s[0:1] ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[2:3], s[2:3] @@ -2052,7 +2052,7 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x16_bf16(ptr addrspace(1) %arg, ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[12:13], s[12:13] ; GFX950-SDAG-NEXT: v_mov_b64_e32 v[14:15], s[14:15] ; GFX950-SDAG-NEXT: s_nop 1 -; GFX950-SDAG-NEXT: v_smfmac_f32_32x32x16_bf16 v[0:15], v[20:21], v[16:19], v22 cbsz:1 abid:2 +; GFX950-SDAG-NEXT: v_smfmac_f32_32x32x16_bf16 v[0:15], v[22:23], v[18:21], v16 cbsz:1 abid:2 ; GFX950-SDAG-NEXT: v_mov_b32_e32 v16, 0 ; GFX950-SDAG-NEXT: s_nop 10 ; GFX950-SDAG-NEXT: global_store_dwordx4 v16, v[12:15], s[16:17] offset:48 diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll index 033a35f..13a96cf 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.bf16.ll @@ -15,15 +15,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16(<8 x bfloat> %arg0, <8 x ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; GCN-NEXT: v_mov_b64_e32 v[8:9], 48 -; GCN-NEXT: v_mov_b64_e32 v[10:11], 32 -; GCN-NEXT: v_mov_b64_e32 v[12:13], 16 +; GCN-NEXT: v_mov_b64_e32 v[0:1], 48 +; GCN-NEXT: v_mov_b64_e32 v[2:3], 32 +; GCN-NEXT: v_mov_b64_e32 v[4:5], 16 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[0:1], s[24:25] -; GCN-NEXT: v_mov_b64_e32 v[2:3], s[26:27] -; GCN-NEXT: v_mov_b64_e32 v[4:5], s[28:29] +; GCN-NEXT: v_mov_b64_e32 v[8:9], s[24:25] +; GCN-NEXT: v_mov_b64_e32 v[10:11], s[26:27] +; GCN-NEXT: v_mov_b64_e32 v[12:13], s[28:29] ; GCN-NEXT: v_accvgpr_write_b32 a0, s8 -; GCN-NEXT: v_mov_b64_e32 v[6:7], s[30:31] +; GCN-NEXT: v_mov_b64_e32 v[14:15], s[30:31] ; GCN-NEXT: v_accvgpr_write_b32 a1, s9 ; GCN-NEXT: v_accvgpr_write_b32 a2, s10 ; GCN-NEXT: v_accvgpr_write_b32 a3, s11 @@ -41,40 +41,39 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16(<8 x bfloat> %arg0, <8 x ; GCN-NEXT: v_accvgpr_write_b32 a15, s23 ; GCN-NEXT: v_mov_b32_e32 v16, s16 ; GCN-NEXT: v_mov_b32_e32 v17, s17 -; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[16:31], v[0:3], v[4:7], a[0:15] +; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[16:31], v[8:11], v[12:15], a[0:15] ; GCN-NEXT: v_mov_b32_e32 v18, s18 ; GCN-NEXT: v_mov_b32_e32 v19, s19 -; GCN-NEXT: v_mov_b32_e32 v0, s20 -; GCN-NEXT: v_mov_b32_e32 v1, s21 -; GCN-NEXT: v_mov_b32_e32 v2, s22 -; GCN-NEXT: v_mov_b32_e32 v3, s23 -; GCN-NEXT: v_mov_b64_e32 v[14:15], 0 +; GCN-NEXT: v_mov_b32_e32 v8, s20 +; GCN-NEXT: v_mov_b32_e32 v9, s21 +; GCN-NEXT: v_mov_b32_e32 v10, s22 +; GCN-NEXT: v_mov_b32_e32 v11, s23 +; GCN-NEXT: v_mov_b64_e32 v[6:7], 0 ; GCN-NEXT: s_nop 4 -; GCN-NEXT: global_store_dwordx4 v[8:9], a[28:31], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[0:1], a[28:31], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[10:11], a[24:27], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[2:3], a[24:27], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[12:13], a[20:23], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[4:5], a[20:23], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[14:15], a[16:19], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[6:7], a[16:19], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[10:11], v[16:19], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[2:3], v[16:19], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[0:1], v[8:11], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v0, s8 ; GCN-NEXT: v_mov_b32_e32 v1, s9 ; GCN-NEXT: v_mov_b32_e32 v2, s10 ; GCN-NEXT: v_mov_b32_e32 v3, s11 -; GCN-NEXT: global_store_dwordx4 v[14:15], v[0:3], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v0, s12 ; GCN-NEXT: v_mov_b32_e32 v1, s13 ; GCN-NEXT: v_mov_b32_e32 v2, s14 ; GCN-NEXT: v_mov_b32_e32 v3, s15 -; GCN-NEXT: global_store_dwordx4 v[12:13], v[0:3], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_endpgm %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0) @@ -88,15 +87,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__flags(<8 x bfloat> %arg0 ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; GCN-NEXT: v_mov_b64_e32 v[8:9], 48 -; GCN-NEXT: v_mov_b64_e32 v[10:11], 32 -; GCN-NEXT: v_mov_b64_e32 v[12:13], 16 +; GCN-NEXT: v_mov_b64_e32 v[0:1], 48 +; GCN-NEXT: v_mov_b64_e32 v[2:3], 32 +; GCN-NEXT: v_mov_b64_e32 v[4:5], 16 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[0:1], s[24:25] -; GCN-NEXT: v_mov_b64_e32 v[2:3], s[26:27] -; GCN-NEXT: v_mov_b64_e32 v[4:5], s[28:29] +; GCN-NEXT: v_mov_b64_e32 v[8:9], s[24:25] +; GCN-NEXT: v_mov_b64_e32 v[10:11], s[26:27] +; GCN-NEXT: v_mov_b64_e32 v[12:13], s[28:29] ; GCN-NEXT: v_accvgpr_write_b32 a0, s8 -; GCN-NEXT: v_mov_b64_e32 v[6:7], s[30:31] +; GCN-NEXT: v_mov_b64_e32 v[14:15], s[30:31] ; GCN-NEXT: v_accvgpr_write_b32 a1, s9 ; GCN-NEXT: v_accvgpr_write_b32 a2, s10 ; GCN-NEXT: v_accvgpr_write_b32 a3, s11 @@ -114,40 +113,39 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__flags(<8 x bfloat> %arg0 ; GCN-NEXT: v_accvgpr_write_b32 a15, s23 ; GCN-NEXT: v_mov_b32_e32 v16, s16 ; GCN-NEXT: v_mov_b32_e32 v17, s17 -; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[16:31], v[0:3], v[4:7], a[0:15] cbsz:2 abid:3 blgp:1 +; GCN-NEXT: v_mfma_f32_32x32x16_bf16 a[16:31], v[8:11], v[12:15], a[0:15] cbsz:2 abid:3 blgp:1 ; GCN-NEXT: v_mov_b32_e32 v18, s18 ; GCN-NEXT: v_mov_b32_e32 v19, s19 -; GCN-NEXT: v_mov_b32_e32 v0, s20 -; GCN-NEXT: v_mov_b32_e32 v1, s21 -; GCN-NEXT: v_mov_b32_e32 v2, s22 -; GCN-NEXT: v_mov_b32_e32 v3, s23 -; GCN-NEXT: v_mov_b64_e32 v[14:15], 0 +; GCN-NEXT: v_mov_b32_e32 v8, s20 +; GCN-NEXT: v_mov_b32_e32 v9, s21 +; GCN-NEXT: v_mov_b32_e32 v10, s22 +; GCN-NEXT: v_mov_b32_e32 v11, s23 +; GCN-NEXT: v_mov_b64_e32 v[6:7], 0 ; GCN-NEXT: s_nop 4 -; GCN-NEXT: global_store_dwordx4 v[8:9], a[28:31], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[0:1], a[28:31], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[10:11], a[24:27], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[2:3], a[24:27], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[12:13], a[20:23], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[4:5], a[20:23], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[14:15], a[16:19], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[6:7], a[16:19], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[10:11], v[16:19], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[2:3], v[16:19], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[0:1], v[8:11], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v0, s8 ; GCN-NEXT: v_mov_b32_e32 v1, s9 ; GCN-NEXT: v_mov_b32_e32 v2, s10 ; GCN-NEXT: v_mov_b32_e32 v3, s11 -; GCN-NEXT: global_store_dwordx4 v[14:15], v[0:3], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v0, s12 ; GCN-NEXT: v_mov_b32_e32 v1, s13 ; GCN-NEXT: v_mov_b32_e32 v2, s14 ; GCN-NEXT: v_mov_b32_e32 v3, s15 -; GCN-NEXT: global_store_dwordx4 v[12:13], v[0:3], off sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_endpgm %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 2, i32 3, i32 1) @@ -250,13 +248,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd(<8 x bfloat> %arg ; GCN-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; GCN-NEXT: v_mov_b32_e32 v44, 0 +; GCN-NEXT: v_mov_b32_e32 v36, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; GCN-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; GCN-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; GCN-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; GCN-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; GCN-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; GCN-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; GCN-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; GCN-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; GCN-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; GCN-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; GCN-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -264,41 +262,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd(<8 x bfloat> %arg ; GCN-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; GCN-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; GCN-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; GCN-NEXT: v_mov_b32_e32 v40, s20 -; GCN-NEXT: v_mov_b32_e32 v41, s21 -; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[32:35], v[36:39], v[16:31] -; GCN-NEXT: v_mov_b32_e32 v42, s22 -; GCN-NEXT: v_mov_b32_e32 v43, s23 -; GCN-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; GCN-NEXT: v_mov_b32_e32 v32, s20 +; GCN-NEXT: v_mov_b32_e32 v33, s21 +; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[38:41], v[42:45], v[16:31] +; GCN-NEXT: v_mov_b32_e32 v34, s22 +; GCN-NEXT: v_mov_b32_e32 v35, s23 +; GCN-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 2 ; GCN-NEXT: v_mov_b32_e32 v16, s16 ; GCN-NEXT: v_mov_b32_e32 v17, s17 ; GCN-NEXT: v_mov_b32_e32 v18, s18 ; GCN-NEXT: v_mov_b32_e32 v19, s19 -; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v16, s12 ; GCN-NEXT: v_mov_b32_e32 v17, s13 ; GCN-NEXT: v_mov_b32_e32 v18, s14 ; GCN-NEXT: v_mov_b32_e32 v19, s15 -; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v16, s8 ; GCN-NEXT: v_mov_b32_e32 v17, s9 ; GCN-NEXT: v_mov_b32_e32 v18, s10 ; GCN-NEXT: v_mov_b32_e32 v19, s11 -; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_endpgm %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 0, i32 0, i32 0) @@ -313,13 +311,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd__flags(<8 x bfloa ; GCN-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; GCN-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; GCN-NEXT: v_mov_b32_e32 v44, 0 +; GCN-NEXT: v_mov_b32_e32 v36, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; GCN-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; GCN-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; GCN-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; GCN-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; GCN-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; GCN-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; GCN-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; GCN-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; GCN-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; GCN-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; GCN-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -327,41 +325,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_bf16__vgprcd__flags(<8 x bfloa ; GCN-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; GCN-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; GCN-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; GCN-NEXT: v_mov_b32_e32 v40, s20 -; GCN-NEXT: v_mov_b32_e32 v41, s21 -; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3 -; GCN-NEXT: v_mov_b32_e32 v42, s22 -; GCN-NEXT: v_mov_b32_e32 v43, s23 -; GCN-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; GCN-NEXT: v_mov_b32_e32 v32, s20 +; GCN-NEXT: v_mov_b32_e32 v33, s21 +; GCN-NEXT: v_mfma_f32_32x32x16_bf16 v[0:15], v[38:41], v[42:45], v[16:31] cbsz:1 abid:2 blgp:3 +; GCN-NEXT: v_mov_b32_e32 v34, s22 +; GCN-NEXT: v_mov_b32_e32 v35, s23 +; GCN-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 2 ; GCN-NEXT: v_mov_b32_e32 v16, s16 ; GCN-NEXT: v_mov_b32_e32 v17, s17 ; GCN-NEXT: v_mov_b32_e32 v18, s18 ; GCN-NEXT: v_mov_b32_e32 v19, s19 -; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v16, s12 ; GCN-NEXT: v_mov_b32_e32 v17, s13 ; GCN-NEXT: v_mov_b32_e32 v18, s14 ; GCN-NEXT: v_mov_b32_e32 v19, s15 -; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 ; GCN-NEXT: v_mov_b32_e32 v16, s8 ; GCN-NEXT: v_mov_b32_e32 v17, s9 ; GCN-NEXT: v_mov_b32_e32 v18, s10 ; GCN-NEXT: v_mov_b32_e32 v19, s11 -; GCN-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) -; GCN-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; GCN-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_endpgm %result = call <16 x float> @llvm.amdgcn.mfma.f32.32x32x16.bf16(<8 x bfloat> %arg0, <8 x bfloat> %arg1, <16 x float> %arg2, i32 1, i32 2, i32 3) diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll index 7532062..ab0000f 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll @@ -141,18 +141,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp ; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; SDAG-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; SDAG-NEXT: v_mov_b32_e32 v12, 0 +; SDAG-NEXT: v_mov_b32_e32 v4, 0 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] +; SDAG-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[6:9], v[10:13], v[0:3] ; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; SDAG-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; SDAG-NEXT: s_endpgm ; ; GISEL-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd: @@ -179,18 +179,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp ; HEURRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; HEURRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; HEURRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; HEURRC-NEXT: v_mov_b32_e32 v12, 0 +; HEURRC-NEXT: v_mov_b32_e32 v4, 0 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; HEURRC-NEXT: s_nop 1 -; HEURRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] +; HEURRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[6:9], v[10:13], v[0:3] ; HEURRC-NEXT: s_nop 7 -; HEURRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; HEURRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; HEURRC-NEXT: s_endpgm ; ; VGPRRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd: @@ -198,18 +198,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd(ptr addrsp ; VGPRRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; VGPRRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; VGPRRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; VGPRRC-NEXT: v_mov_b32_e32 v12, 0 +; VGPRRC-NEXT: v_mov_b32_e32 v4, 0 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; VGPRRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; VGPRRC-NEXT: s_nop 1 -; VGPRRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] +; VGPRRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[6:9], v[10:13], v[0:3] ; VGPRRC-NEXT: s_nop 7 -; VGPRRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; VGPRRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd: ; AGPR: ; %bb.0: @@ -260,18 +260,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr ; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; SDAG-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; SDAG-NEXT: v_mov_b32_e32 v12, 0 +; SDAG-NEXT: v_mov_b32_e32 v4, 0 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1 +; SDAG-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1 ; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; SDAG-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; SDAG-NEXT: s_endpgm ; ; GISEL-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags: @@ -298,18 +298,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr ; HEURRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; HEURRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; HEURRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; HEURRC-NEXT: v_mov_b32_e32 v12, 0 +; HEURRC-NEXT: v_mov_b32_e32 v4, 0 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; HEURRC-NEXT: s_nop 1 -; HEURRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1 +; HEURRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1 ; HEURRC-NEXT: s_nop 7 -; HEURRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; HEURRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; HEURRC-NEXT: s_endpgm ; ; VGPRRC-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags: @@ -317,18 +317,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags(ptr ; VGPRRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; VGPRRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; VGPRRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; VGPRRC-NEXT: v_mov_b32_e32 v12, 0 +; VGPRRC-NEXT: v_mov_b32_e32 v4, 0 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; VGPRRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; VGPRRC-NEXT: s_nop 1 -; VGPRRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1 +; VGPRRC-NEXT: v_mfma_f32_16x16x32_f16 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1 ; VGPRRC-NEXT: s_nop 7 -; VGPRRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; VGPRRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_16x16x32_f16_no_agpr__vgprcd__flags: ; AGPR: ; %bb.0: @@ -382,15 +382,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16(<8 x half> %arg0, <8 x hal ; SDAG: ; %bb.0: ; SDAG-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; SDAG-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; SDAG-NEXT: v_mov_b64_e32 v[8:9], 48 -; SDAG-NEXT: v_mov_b64_e32 v[10:11], 32 -; SDAG-NEXT: v_mov_b64_e32 v[12:13], 16 +; SDAG-NEXT: v_mov_b64_e32 v[0:1], 48 +; SDAG-NEXT: v_mov_b64_e32 v[2:3], 32 +; SDAG-NEXT: v_mov_b64_e32 v[4:5], 16 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[24:25] -; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[26:27] -; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[28:29] +; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[24:25] +; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[26:27] +; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[28:29] ; SDAG-NEXT: v_accvgpr_write_b32 a0, s8 -; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[30:31] +; SDAG-NEXT: v_mov_b64_e32 v[14:15], s[30:31] ; SDAG-NEXT: v_accvgpr_write_b32 a1, s9 ; SDAG-NEXT: v_accvgpr_write_b32 a2, s10 ; SDAG-NEXT: v_accvgpr_write_b32 a3, s11 @@ -408,40 +408,39 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16(<8 x half> %arg0, <8 x hal ; SDAG-NEXT: v_accvgpr_write_b32 a15, s23 ; SDAG-NEXT: v_mov_b32_e32 v16, s16 ; SDAG-NEXT: v_mov_b32_e32 v17, s17 -; SDAG-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[0:3], v[4:7], a[0:15] +; SDAG-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[8:11], v[12:15], a[0:15] ; SDAG-NEXT: v_mov_b32_e32 v18, s18 ; SDAG-NEXT: v_mov_b32_e32 v19, s19 -; SDAG-NEXT: v_mov_b32_e32 v0, s20 -; SDAG-NEXT: v_mov_b32_e32 v1, s21 -; SDAG-NEXT: v_mov_b32_e32 v2, s22 -; SDAG-NEXT: v_mov_b32_e32 v3, s23 -; SDAG-NEXT: v_mov_b64_e32 v[14:15], 0 +; SDAG-NEXT: v_mov_b32_e32 v8, s20 +; SDAG-NEXT: v_mov_b32_e32 v9, s21 +; SDAG-NEXT: v_mov_b32_e32 v10, s22 +; SDAG-NEXT: v_mov_b32_e32 v11, s23 +; SDAG-NEXT: v_mov_b64_e32 v[6:7], 0 ; SDAG-NEXT: s_nop 4 -; SDAG-NEXT: global_store_dwordx4 v[8:9], a[28:31], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[0:1], a[28:31], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[10:11], a[24:27], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[2:3], a[24:27], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[12:13], a[20:23], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[4:5], a[20:23], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[14:15], a[16:19], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[6:7], a[16:19], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[10:11], v[16:19], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[2:3], v[16:19], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[0:1], v[8:11], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v0, s8 ; SDAG-NEXT: v_mov_b32_e32 v1, s9 ; SDAG-NEXT: v_mov_b32_e32 v2, s10 ; SDAG-NEXT: v_mov_b32_e32 v3, s11 -; SDAG-NEXT: global_store_dwordx4 v[14:15], v[0:3], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v0, s12 ; SDAG-NEXT: v_mov_b32_e32 v1, s13 ; SDAG-NEXT: v_mov_b32_e32 v2, s14 ; SDAG-NEXT: v_mov_b32_e32 v3, s15 -; SDAG-NEXT: global_store_dwordx4 v[12:13], v[0:3], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_endpgm ; @@ -508,15 +507,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16(<8 x half> %arg0, <8 x hal ; HEURRC: ; %bb.0: ; HEURRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; HEURRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; HEURRC-NEXT: v_mov_b64_e32 v[8:9], 48 -; HEURRC-NEXT: v_mov_b64_e32 v[10:11], 32 -; HEURRC-NEXT: v_mov_b64_e32 v[12:13], 16 +; HEURRC-NEXT: v_mov_b64_e32 v[0:1], 48 +; HEURRC-NEXT: v_mov_b64_e32 v[2:3], 32 +; HEURRC-NEXT: v_mov_b64_e32 v[4:5], 16 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[24:25] -; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[26:27] -; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[28:29] +; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[24:25] +; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[26:27] +; HEURRC-NEXT: v_mov_b64_e32 v[12:13], s[28:29] ; HEURRC-NEXT: v_accvgpr_write_b32 a0, s8 -; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[30:31] +; HEURRC-NEXT: v_mov_b64_e32 v[14:15], s[30:31] ; HEURRC-NEXT: v_accvgpr_write_b32 a1, s9 ; HEURRC-NEXT: v_accvgpr_write_b32 a2, s10 ; HEURRC-NEXT: v_accvgpr_write_b32 a3, s11 @@ -534,40 +533,39 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16(<8 x half> %arg0, <8 x hal ; HEURRC-NEXT: v_accvgpr_write_b32 a15, s23 ; HEURRC-NEXT: v_mov_b32_e32 v16, s16 ; HEURRC-NEXT: v_mov_b32_e32 v17, s17 -; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[0:3], v[4:7], a[0:15] +; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[8:11], v[12:15], a[0:15] ; HEURRC-NEXT: v_mov_b32_e32 v18, s18 ; HEURRC-NEXT: v_mov_b32_e32 v19, s19 -; HEURRC-NEXT: v_mov_b32_e32 v0, s20 -; HEURRC-NEXT: v_mov_b32_e32 v1, s21 -; HEURRC-NEXT: v_mov_b32_e32 v2, s22 -; HEURRC-NEXT: v_mov_b32_e32 v3, s23 -; HEURRC-NEXT: v_mov_b64_e32 v[14:15], 0 +; HEURRC-NEXT: v_mov_b32_e32 v8, s20 +; HEURRC-NEXT: v_mov_b32_e32 v9, s21 +; HEURRC-NEXT: v_mov_b32_e32 v10, s22 +; HEURRC-NEXT: v_mov_b32_e32 v11, s23 +; HEURRC-NEXT: v_mov_b64_e32 v[6:7], 0 ; HEURRC-NEXT: s_nop 4 -; HEURRC-NEXT: global_store_dwordx4 v[8:9], a[28:31], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[0:1], a[28:31], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[10:11], a[24:27], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[2:3], a[24:27], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[12:13], a[20:23], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[4:5], a[20:23], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[14:15], a[16:19], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[6:7], a[16:19], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[10:11], v[16:19], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[2:3], v[16:19], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[0:1], v[8:11], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v0, s8 ; HEURRC-NEXT: v_mov_b32_e32 v1, s9 ; HEURRC-NEXT: v_mov_b32_e32 v2, s10 ; HEURRC-NEXT: v_mov_b32_e32 v3, s11 -; HEURRC-NEXT: global_store_dwordx4 v[14:15], v[0:3], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v0, s12 ; HEURRC-NEXT: v_mov_b32_e32 v1, s13 ; HEURRC-NEXT: v_mov_b32_e32 v2, s14 ; HEURRC-NEXT: v_mov_b32_e32 v3, s15 -; HEURRC-NEXT: global_store_dwordx4 v[12:13], v[0:3], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_endpgm ; @@ -575,15 +573,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16(<8 x half> %arg0, <8 x hal ; VGPRRC: ; %bb.0: ; VGPRRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; VGPRRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; VGPRRC-NEXT: v_mov_b64_e32 v[40:41], 48 -; VGPRRC-NEXT: v_mov_b64_e32 v[42:43], 32 -; VGPRRC-NEXT: v_mov_b64_e32 v[44:45], 16 +; VGPRRC-NEXT: v_mov_b64_e32 v[32:33], 48 +; VGPRRC-NEXT: v_mov_b64_e32 v[34:35], 32 +; VGPRRC-NEXT: v_mov_b64_e32 v[36:37], 16 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; VGPRRC-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; VGPRRC-NEXT: v_mov_b64_e32 v[42:43], s[26:27] +; VGPRRC-NEXT: v_mov_b64_e32 v[40:41], s[24:25] +; VGPRRC-NEXT: v_mov_b64_e32 v[46:47], s[30:31] ; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; VGPRRC-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; VGPRRC-NEXT: v_mov_b64_e32 v[44:45], s[28:29] ; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] ; VGPRRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] ; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] @@ -593,40 +591,40 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16(<8 x half> %arg0, <8 x hal ; VGPRRC-NEXT: v_mov_b64_e32 v[14:15], s[22:23] ; VGPRRC-NEXT: v_mov_b32_e32 v48, s16 ; VGPRRC-NEXT: v_mov_b32_e32 v49, s17 -; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[16:31], v[32:35], v[36:39], v[0:15] +; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[16:31], v[40:43], v[44:47], v[0:15] ; VGPRRC-NEXT: v_mov_b32_e32 v50, s18 ; VGPRRC-NEXT: v_mov_b32_e32 v51, s19 -; VGPRRC-NEXT: v_mov_b64_e32 v[46:47], 0 +; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], 0 ; VGPRRC-NEXT: s_nop 8 -; VGPRRC-NEXT: global_store_dwordx4 v[40:41], v[28:31], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[32:33], v[28:31], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[42:43], v[24:27], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[34:35], v[24:27], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[44:45], v[20:23], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[36:37], v[20:23], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[46:47], v[16:19], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[38:39], v[16:19], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: v_mov_b32_e32 v0, s20 ; VGPRRC-NEXT: v_mov_b32_e32 v1, s21 ; VGPRRC-NEXT: v_mov_b32_e32 v2, s22 ; VGPRRC-NEXT: v_mov_b32_e32 v3, s23 -; VGPRRC-NEXT: global_store_dwordx4 v[42:43], v[48:51], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[34:35], v[48:51], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[40:41], v[0:3], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[32:33], v[0:3], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v0, s8 ; VGPRRC-NEXT: v_mov_b32_e32 v1, s9 ; VGPRRC-NEXT: v_mov_b32_e32 v2, s10 ; VGPRRC-NEXT: v_mov_b32_e32 v3, s11 -; VGPRRC-NEXT: global_store_dwordx4 v[46:47], v[0:3], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[38:39], v[0:3], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v0, s12 ; VGPRRC-NEXT: v_mov_b32_e32 v1, s13 ; VGPRRC-NEXT: v_mov_b32_e32 v2, s14 ; VGPRRC-NEXT: v_mov_b32_e32 v3, s15 -; VGPRRC-NEXT: global_store_dwordx4 v[44:45], v[0:3], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[36:37], v[0:3], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_32x32x16_f16: @@ -765,15 +763,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__flags(<8 x half> %arg0, < ; SDAG: ; %bb.0: ; SDAG-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; SDAG-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; SDAG-NEXT: v_mov_b64_e32 v[8:9], 48 -; SDAG-NEXT: v_mov_b64_e32 v[10:11], 32 -; SDAG-NEXT: v_mov_b64_e32 v[12:13], 16 +; SDAG-NEXT: v_mov_b64_e32 v[0:1], 48 +; SDAG-NEXT: v_mov_b64_e32 v[2:3], 32 +; SDAG-NEXT: v_mov_b64_e32 v[4:5], 16 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[24:25] -; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[26:27] -; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[28:29] +; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[24:25] +; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[26:27] +; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[28:29] ; SDAG-NEXT: v_accvgpr_write_b32 a0, s8 -; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[30:31] +; SDAG-NEXT: v_mov_b64_e32 v[14:15], s[30:31] ; SDAG-NEXT: v_accvgpr_write_b32 a1, s9 ; SDAG-NEXT: v_accvgpr_write_b32 a2, s10 ; SDAG-NEXT: v_accvgpr_write_b32 a3, s11 @@ -791,40 +789,39 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__flags(<8 x half> %arg0, < ; SDAG-NEXT: v_accvgpr_write_b32 a15, s23 ; SDAG-NEXT: v_mov_b32_e32 v16, s16 ; SDAG-NEXT: v_mov_b32_e32 v17, s17 -; SDAG-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[0:3], v[4:7], a[0:15] cbsz:2 abid:3 blgp:1 +; SDAG-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[8:11], v[12:15], a[0:15] cbsz:2 abid:3 blgp:1 ; SDAG-NEXT: v_mov_b32_e32 v18, s18 ; SDAG-NEXT: v_mov_b32_e32 v19, s19 -; SDAG-NEXT: v_mov_b32_e32 v0, s20 -; SDAG-NEXT: v_mov_b32_e32 v1, s21 -; SDAG-NEXT: v_mov_b32_e32 v2, s22 -; SDAG-NEXT: v_mov_b32_e32 v3, s23 -; SDAG-NEXT: v_mov_b64_e32 v[14:15], 0 +; SDAG-NEXT: v_mov_b32_e32 v8, s20 +; SDAG-NEXT: v_mov_b32_e32 v9, s21 +; SDAG-NEXT: v_mov_b32_e32 v10, s22 +; SDAG-NEXT: v_mov_b32_e32 v11, s23 +; SDAG-NEXT: v_mov_b64_e32 v[6:7], 0 ; SDAG-NEXT: s_nop 4 -; SDAG-NEXT: global_store_dwordx4 v[8:9], a[28:31], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[0:1], a[28:31], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[10:11], a[24:27], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[2:3], a[24:27], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[12:13], a[20:23], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[4:5], a[20:23], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[14:15], a[16:19], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[6:7], a[16:19], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[10:11], v[16:19], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[2:3], v[16:19], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[0:1], v[8:11], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v0, s8 ; SDAG-NEXT: v_mov_b32_e32 v1, s9 ; SDAG-NEXT: v_mov_b32_e32 v2, s10 ; SDAG-NEXT: v_mov_b32_e32 v3, s11 -; SDAG-NEXT: global_store_dwordx4 v[14:15], v[0:3], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v0, s12 ; SDAG-NEXT: v_mov_b32_e32 v1, s13 ; SDAG-NEXT: v_mov_b32_e32 v2, s14 ; SDAG-NEXT: v_mov_b32_e32 v3, s15 -; SDAG-NEXT: global_store_dwordx4 v[12:13], v[0:3], off sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_endpgm ; @@ -891,15 +888,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__flags(<8 x half> %arg0, < ; HEURRC: ; %bb.0: ; HEURRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; HEURRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; HEURRC-NEXT: v_mov_b64_e32 v[8:9], 48 -; HEURRC-NEXT: v_mov_b64_e32 v[10:11], 32 -; HEURRC-NEXT: v_mov_b64_e32 v[12:13], 16 +; HEURRC-NEXT: v_mov_b64_e32 v[0:1], 48 +; HEURRC-NEXT: v_mov_b64_e32 v[2:3], 32 +; HEURRC-NEXT: v_mov_b64_e32 v[4:5], 16 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[24:25] -; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[26:27] -; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[28:29] +; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[24:25] +; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[26:27] +; HEURRC-NEXT: v_mov_b64_e32 v[12:13], s[28:29] ; HEURRC-NEXT: v_accvgpr_write_b32 a0, s8 -; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[30:31] +; HEURRC-NEXT: v_mov_b64_e32 v[14:15], s[30:31] ; HEURRC-NEXT: v_accvgpr_write_b32 a1, s9 ; HEURRC-NEXT: v_accvgpr_write_b32 a2, s10 ; HEURRC-NEXT: v_accvgpr_write_b32 a3, s11 @@ -917,40 +914,39 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__flags(<8 x half> %arg0, < ; HEURRC-NEXT: v_accvgpr_write_b32 a15, s23 ; HEURRC-NEXT: v_mov_b32_e32 v16, s16 ; HEURRC-NEXT: v_mov_b32_e32 v17, s17 -; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[0:3], v[4:7], a[0:15] cbsz:2 abid:3 blgp:1 +; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 a[16:31], v[8:11], v[12:15], a[0:15] cbsz:2 abid:3 blgp:1 ; HEURRC-NEXT: v_mov_b32_e32 v18, s18 ; HEURRC-NEXT: v_mov_b32_e32 v19, s19 -; HEURRC-NEXT: v_mov_b32_e32 v0, s20 -; HEURRC-NEXT: v_mov_b32_e32 v1, s21 -; HEURRC-NEXT: v_mov_b32_e32 v2, s22 -; HEURRC-NEXT: v_mov_b32_e32 v3, s23 -; HEURRC-NEXT: v_mov_b64_e32 v[14:15], 0 +; HEURRC-NEXT: v_mov_b32_e32 v8, s20 +; HEURRC-NEXT: v_mov_b32_e32 v9, s21 +; HEURRC-NEXT: v_mov_b32_e32 v10, s22 +; HEURRC-NEXT: v_mov_b32_e32 v11, s23 +; HEURRC-NEXT: v_mov_b64_e32 v[6:7], 0 ; HEURRC-NEXT: s_nop 4 -; HEURRC-NEXT: global_store_dwordx4 v[8:9], a[28:31], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[0:1], a[28:31], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[10:11], a[24:27], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[2:3], a[24:27], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[12:13], a[20:23], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[4:5], a[20:23], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[14:15], a[16:19], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[6:7], a[16:19], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[10:11], v[16:19], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[2:3], v[16:19], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v[8:9], v[0:3], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[0:1], v[8:11], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v0, s8 ; HEURRC-NEXT: v_mov_b32_e32 v1, s9 ; HEURRC-NEXT: v_mov_b32_e32 v2, s10 ; HEURRC-NEXT: v_mov_b32_e32 v3, s11 -; HEURRC-NEXT: global_store_dwordx4 v[14:15], v[0:3], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[6:7], v[0:3], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v0, s12 ; HEURRC-NEXT: v_mov_b32_e32 v1, s13 ; HEURRC-NEXT: v_mov_b32_e32 v2, s14 ; HEURRC-NEXT: v_mov_b32_e32 v3, s15 -; HEURRC-NEXT: global_store_dwordx4 v[12:13], v[0:3], off sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v[4:5], v[0:3], off sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_endpgm ; @@ -958,15 +954,15 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__flags(<8 x half> %arg0, < ; VGPRRC: ; %bb.0: ; VGPRRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; VGPRRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 -; VGPRRC-NEXT: v_mov_b64_e32 v[40:41], 48 -; VGPRRC-NEXT: v_mov_b64_e32 v[42:43], 32 -; VGPRRC-NEXT: v_mov_b64_e32 v[44:45], 16 +; VGPRRC-NEXT: v_mov_b64_e32 v[32:33], 48 +; VGPRRC-NEXT: v_mov_b64_e32 v[34:35], 32 +; VGPRRC-NEXT: v_mov_b64_e32 v[36:37], 16 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; VGPRRC-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; VGPRRC-NEXT: v_mov_b64_e32 v[42:43], s[26:27] +; VGPRRC-NEXT: v_mov_b64_e32 v[40:41], s[24:25] +; VGPRRC-NEXT: v_mov_b64_e32 v[46:47], s[30:31] ; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; VGPRRC-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; VGPRRC-NEXT: v_mov_b64_e32 v[44:45], s[28:29] ; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] ; VGPRRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] ; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] @@ -976,40 +972,40 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__flags(<8 x half> %arg0, < ; VGPRRC-NEXT: v_mov_b64_e32 v[14:15], s[22:23] ; VGPRRC-NEXT: v_mov_b32_e32 v48, s16 ; VGPRRC-NEXT: v_mov_b32_e32 v49, s17 -; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[16:31], v[32:35], v[36:39], v[0:15] cbsz:2 abid:3 blgp:1 +; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[16:31], v[40:43], v[44:47], v[0:15] cbsz:2 abid:3 blgp:1 ; VGPRRC-NEXT: v_mov_b32_e32 v50, s18 ; VGPRRC-NEXT: v_mov_b32_e32 v51, s19 -; VGPRRC-NEXT: v_mov_b64_e32 v[46:47], 0 +; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], 0 ; VGPRRC-NEXT: s_nop 8 -; VGPRRC-NEXT: global_store_dwordx4 v[40:41], v[28:31], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[32:33], v[28:31], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[42:43], v[24:27], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[34:35], v[24:27], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[44:45], v[20:23], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[36:37], v[20:23], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[46:47], v[16:19], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[38:39], v[16:19], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: v_mov_b32_e32 v0, s20 ; VGPRRC-NEXT: v_mov_b32_e32 v1, s21 ; VGPRRC-NEXT: v_mov_b32_e32 v2, s22 ; VGPRRC-NEXT: v_mov_b32_e32 v3, s23 -; VGPRRC-NEXT: global_store_dwordx4 v[42:43], v[48:51], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[34:35], v[48:51], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v[40:41], v[0:3], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[32:33], v[0:3], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v0, s8 ; VGPRRC-NEXT: v_mov_b32_e32 v1, s9 ; VGPRRC-NEXT: v_mov_b32_e32 v2, s10 ; VGPRRC-NEXT: v_mov_b32_e32 v3, s11 -; VGPRRC-NEXT: global_store_dwordx4 v[46:47], v[0:3], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[38:39], v[0:3], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v0, s12 ; VGPRRC-NEXT: v_mov_b32_e32 v1, s13 ; VGPRRC-NEXT: v_mov_b32_e32 v2, s14 ; VGPRRC-NEXT: v_mov_b32_e32 v3, s15 -; VGPRRC-NEXT: global_store_dwordx4 v[44:45], v[0:3], off sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v[36:37], v[0:3], off sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_32x32x16_f16__flags: @@ -1489,13 +1485,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0, ; SDAG-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; SDAG-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; SDAG-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; SDAG-NEXT: v_mov_b32_e32 v44, 0 +; SDAG-NEXT: v_mov_b32_e32 v36, 0 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; SDAG-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; SDAG-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; SDAG-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; SDAG-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; SDAG-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; SDAG-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; SDAG-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; SDAG-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; SDAG-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; SDAG-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; SDAG-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -1503,41 +1499,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0, ; SDAG-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; SDAG-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; SDAG-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; SDAG-NEXT: v_mov_b32_e32 v40, s20 -; SDAG-NEXT: v_mov_b32_e32 v41, s21 -; SDAG-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] -; SDAG-NEXT: v_mov_b32_e32 v42, s22 -; SDAG-NEXT: v_mov_b32_e32 v43, s23 -; SDAG-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; SDAG-NEXT: v_mov_b32_e32 v32, s20 +; SDAG-NEXT: v_mov_b32_e32 v33, s21 +; SDAG-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[38:41], v[42:45], v[16:31] +; SDAG-NEXT: v_mov_b32_e32 v34, s22 +; SDAG-NEXT: v_mov_b32_e32 v35, s23 +; SDAG-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 2 ; SDAG-NEXT: v_mov_b32_e32 v16, s16 ; SDAG-NEXT: v_mov_b32_e32 v17, s17 ; SDAG-NEXT: v_mov_b32_e32 v18, s18 ; SDAG-NEXT: v_mov_b32_e32 v19, s19 -; SDAG-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v16, s12 ; SDAG-NEXT: v_mov_b32_e32 v17, s13 ; SDAG-NEXT: v_mov_b32_e32 v18, s14 ; SDAG-NEXT: v_mov_b32_e32 v19, s15 -; SDAG-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v16, s8 ; SDAG-NEXT: v_mov_b32_e32 v17, s9 ; SDAG-NEXT: v_mov_b32_e32 v18, s10 ; SDAG-NEXT: v_mov_b32_e32 v19, s11 -; SDAG-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_endpgm ; @@ -1592,13 +1588,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0, ; HEURRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; HEURRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; HEURRC-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; HEURRC-NEXT: v_mov_b32_e32 v44, 0 +; HEURRC-NEXT: v_mov_b32_e32 v36, 0 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; HEURRC-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; HEURRC-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; HEURRC-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; HEURRC-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; HEURRC-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; HEURRC-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; HEURRC-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; HEURRC-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; HEURRC-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; HEURRC-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; HEURRC-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -1606,41 +1602,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0, ; HEURRC-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; HEURRC-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; HEURRC-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; HEURRC-NEXT: v_mov_b32_e32 v40, s20 -; HEURRC-NEXT: v_mov_b32_e32 v41, s21 -; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] -; HEURRC-NEXT: v_mov_b32_e32 v42, s22 -; HEURRC-NEXT: v_mov_b32_e32 v43, s23 -; HEURRC-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; HEURRC-NEXT: v_mov_b32_e32 v32, s20 +; HEURRC-NEXT: v_mov_b32_e32 v33, s21 +; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[38:41], v[42:45], v[16:31] +; HEURRC-NEXT: v_mov_b32_e32 v34, s22 +; HEURRC-NEXT: v_mov_b32_e32 v35, s23 +; HEURRC-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 2 ; HEURRC-NEXT: v_mov_b32_e32 v16, s16 ; HEURRC-NEXT: v_mov_b32_e32 v17, s17 ; HEURRC-NEXT: v_mov_b32_e32 v18, s18 ; HEURRC-NEXT: v_mov_b32_e32 v19, s19 -; HEURRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v16, s12 ; HEURRC-NEXT: v_mov_b32_e32 v17, s13 ; HEURRC-NEXT: v_mov_b32_e32 v18, s14 ; HEURRC-NEXT: v_mov_b32_e32 v19, s15 -; HEURRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v16, s8 ; HEURRC-NEXT: v_mov_b32_e32 v17, s9 ; HEURRC-NEXT: v_mov_b32_e32 v18, s10 ; HEURRC-NEXT: v_mov_b32_e32 v19, s11 -; HEURRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_endpgm ; @@ -1649,13 +1645,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0, ; VGPRRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; VGPRRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; VGPRRC-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; VGPRRC-NEXT: v_mov_b32_e32 v44, 0 +; VGPRRC-NEXT: v_mov_b32_e32 v36, 0 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; VGPRRC-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; VGPRRC-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; VGPRRC-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; VGPRRC-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; VGPRRC-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; VGPRRC-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; VGPRRC-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; VGPRRC-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; VGPRRC-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -1663,41 +1659,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd(<8 x half> %arg0, ; VGPRRC-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; VGPRRC-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; VGPRRC-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; VGPRRC-NEXT: v_mov_b32_e32 v40, s20 -; VGPRRC-NEXT: v_mov_b32_e32 v41, s21 -; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] -; VGPRRC-NEXT: v_mov_b32_e32 v42, s22 -; VGPRRC-NEXT: v_mov_b32_e32 v43, s23 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; VGPRRC-NEXT: v_mov_b32_e32 v32, s20 +; VGPRRC-NEXT: v_mov_b32_e32 v33, s21 +; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[38:41], v[42:45], v[16:31] +; VGPRRC-NEXT: v_mov_b32_e32 v34, s22 +; VGPRRC-NEXT: v_mov_b32_e32 v35, s23 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 2 ; VGPRRC-NEXT: v_mov_b32_e32 v16, s16 ; VGPRRC-NEXT: v_mov_b32_e32 v17, s17 ; VGPRRC-NEXT: v_mov_b32_e32 v18, s18 ; VGPRRC-NEXT: v_mov_b32_e32 v19, s19 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v16, s12 ; VGPRRC-NEXT: v_mov_b32_e32 v17, s13 ; VGPRRC-NEXT: v_mov_b32_e32 v18, s14 ; VGPRRC-NEXT: v_mov_b32_e32 v19, s15 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v16, s8 ; VGPRRC-NEXT: v_mov_b32_e32 v17, s9 ; VGPRRC-NEXT: v_mov_b32_e32 v18, s10 ; VGPRRC-NEXT: v_mov_b32_e32 v19, s11 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_32x32x16_f16__vgprcd: @@ -1831,13 +1827,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half> ; SDAG-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; SDAG-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; SDAG-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; SDAG-NEXT: v_mov_b32_e32 v44, 0 +; SDAG-NEXT: v_mov_b32_e32 v36, 0 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; SDAG-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; SDAG-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; SDAG-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; SDAG-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; SDAG-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; SDAG-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; SDAG-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; SDAG-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; SDAG-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; SDAG-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; SDAG-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -1845,41 +1841,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half> ; SDAG-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; SDAG-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; SDAG-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; SDAG-NEXT: v_mov_b32_e32 v40, s20 -; SDAG-NEXT: v_mov_b32_e32 v41, s21 -; SDAG-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3 -; SDAG-NEXT: v_mov_b32_e32 v42, s22 -; SDAG-NEXT: v_mov_b32_e32 v43, s23 -; SDAG-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; SDAG-NEXT: v_mov_b32_e32 v32, s20 +; SDAG-NEXT: v_mov_b32_e32 v33, s21 +; SDAG-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[38:41], v[42:45], v[16:31] cbsz:1 abid:2 blgp:3 +; SDAG-NEXT: v_mov_b32_e32 v34, s22 +; SDAG-NEXT: v_mov_b32_e32 v35, s23 +; SDAG-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 2 ; SDAG-NEXT: v_mov_b32_e32 v16, s16 ; SDAG-NEXT: v_mov_b32_e32 v17, s17 ; SDAG-NEXT: v_mov_b32_e32 v18, s18 ; SDAG-NEXT: v_mov_b32_e32 v19, s19 -; SDAG-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v16, s12 ; SDAG-NEXT: v_mov_b32_e32 v17, s13 ; SDAG-NEXT: v_mov_b32_e32 v18, s14 ; SDAG-NEXT: v_mov_b32_e32 v19, s15 -; SDAG-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 ; SDAG-NEXT: v_mov_b32_e32 v16, s8 ; SDAG-NEXT: v_mov_b32_e32 v17, s9 ; SDAG-NEXT: v_mov_b32_e32 v18, s10 ; SDAG-NEXT: v_mov_b32_e32 v19, s11 -; SDAG-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) -; SDAG-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; SDAG-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_endpgm ; @@ -1934,13 +1930,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half> ; HEURRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; HEURRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; HEURRC-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; HEURRC-NEXT: v_mov_b32_e32 v44, 0 +; HEURRC-NEXT: v_mov_b32_e32 v36, 0 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; HEURRC-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; HEURRC-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; HEURRC-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; HEURRC-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; HEURRC-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; HEURRC-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; HEURRC-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; HEURRC-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; HEURRC-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; HEURRC-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; HEURRC-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -1948,41 +1944,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half> ; HEURRC-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; HEURRC-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; HEURRC-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; HEURRC-NEXT: v_mov_b32_e32 v40, s20 -; HEURRC-NEXT: v_mov_b32_e32 v41, s21 -; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3 -; HEURRC-NEXT: v_mov_b32_e32 v42, s22 -; HEURRC-NEXT: v_mov_b32_e32 v43, s23 -; HEURRC-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; HEURRC-NEXT: v_mov_b32_e32 v32, s20 +; HEURRC-NEXT: v_mov_b32_e32 v33, s21 +; HEURRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[38:41], v[42:45], v[16:31] cbsz:1 abid:2 blgp:3 +; HEURRC-NEXT: v_mov_b32_e32 v34, s22 +; HEURRC-NEXT: v_mov_b32_e32 v35, s23 +; HEURRC-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 2 ; HEURRC-NEXT: v_mov_b32_e32 v16, s16 ; HEURRC-NEXT: v_mov_b32_e32 v17, s17 ; HEURRC-NEXT: v_mov_b32_e32 v18, s18 ; HEURRC-NEXT: v_mov_b32_e32 v19, s19 -; HEURRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v16, s12 ; HEURRC-NEXT: v_mov_b32_e32 v17, s13 ; HEURRC-NEXT: v_mov_b32_e32 v18, s14 ; HEURRC-NEXT: v_mov_b32_e32 v19, s15 -; HEURRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_nop 0 ; HEURRC-NEXT: v_mov_b32_e32 v16, s8 ; HEURRC-NEXT: v_mov_b32_e32 v17, s9 ; HEURRC-NEXT: v_mov_b32_e32 v18, s10 ; HEURRC-NEXT: v_mov_b32_e32 v19, s11 -; HEURRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) -; HEURRC-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; HEURRC-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; HEURRC-NEXT: s_waitcnt vmcnt(0) ; HEURRC-NEXT: s_endpgm ; @@ -1991,13 +1987,13 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half> ; VGPRRC-NEXT: s_load_dwordx8 s[24:31], s[4:5], 0x24 ; VGPRRC-NEXT: s_load_dwordx16 s[8:23], s[4:5], 0x64 ; VGPRRC-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xa4 -; VGPRRC-NEXT: v_mov_b32_e32 v44, 0 +; VGPRRC-NEXT: v_mov_b32_e32 v36, 0 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[34:35], s[26:27] -; VGPRRC-NEXT: v_mov_b64_e32 v[32:33], s[24:25] -; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], s[30:31] +; VGPRRC-NEXT: v_mov_b64_e32 v[40:41], s[26:27] +; VGPRRC-NEXT: v_mov_b64_e32 v[38:39], s[24:25] +; VGPRRC-NEXT: v_mov_b64_e32 v[44:45], s[30:31] ; VGPRRC-NEXT: v_mov_b64_e32 v[30:31], s[22:23] -; VGPRRC-NEXT: v_mov_b64_e32 v[36:37], s[28:29] +; VGPRRC-NEXT: v_mov_b64_e32 v[42:43], s[28:29] ; VGPRRC-NEXT: v_mov_b64_e32 v[28:29], s[20:21] ; VGPRRC-NEXT: v_mov_b64_e32 v[26:27], s[18:19] ; VGPRRC-NEXT: v_mov_b64_e32 v[24:25], s[16:17] @@ -2005,41 +2001,41 @@ define amdgpu_kernel void @test_mfma_f32_32x32x16_f16__vgprcd__flags(<8 x half> ; VGPRRC-NEXT: v_mov_b64_e32 v[20:21], s[12:13] ; VGPRRC-NEXT: v_mov_b64_e32 v[18:19], s[10:11] ; VGPRRC-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; VGPRRC-NEXT: v_mov_b32_e32 v40, s20 -; VGPRRC-NEXT: v_mov_b32_e32 v41, s21 -; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[32:35], v[36:39], v[16:31] cbsz:1 abid:2 blgp:3 -; VGPRRC-NEXT: v_mov_b32_e32 v42, s22 -; VGPRRC-NEXT: v_mov_b32_e32 v43, s23 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[40:43], s[0:1] offset:48 sc0 sc1 +; VGPRRC-NEXT: v_mov_b32_e32 v32, s20 +; VGPRRC-NEXT: v_mov_b32_e32 v33, s21 +; VGPRRC-NEXT: v_mfma_f32_32x32x16_f16 v[0:15], v[38:41], v[42:45], v[16:31] cbsz:1 abid:2 blgp:3 +; VGPRRC-NEXT: v_mov_b32_e32 v34, s22 +; VGPRRC-NEXT: v_mov_b32_e32 v35, s23 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] offset:48 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 2 ; VGPRRC-NEXT: v_mov_b32_e32 v16, s16 ; VGPRRC-NEXT: v_mov_b32_e32 v17, s17 ; VGPRRC-NEXT: v_mov_b32_e32 v18, s18 ; VGPRRC-NEXT: v_mov_b32_e32 v19, s19 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:32 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:32 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v16, s12 ; VGPRRC-NEXT: v_mov_b32_e32 v17, s13 ; VGPRRC-NEXT: v_mov_b32_e32 v18, s14 ; VGPRRC-NEXT: v_mov_b32_e32 v19, s15 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] offset:16 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] offset:16 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_nop 0 ; VGPRRC-NEXT: v_mov_b32_e32 v16, s8 ; VGPRRC-NEXT: v_mov_b32_e32 v17, s9 ; VGPRRC-NEXT: v_mov_b32_e32 v18, s10 ; VGPRRC-NEXT: v_mov_b32_e32 v19, s11 -; VGPRRC-NEXT: global_store_dwordx4 v44, v[16:19], s[0:1] sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[16:19], s[0:1] sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[8:11], s[0:1] offset:32 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[8:11], s[0:1] offset:32 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[12:15], s[0:1] offset:48 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[12:15], s[0:1] offset:48 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[0:3], s[0:1] sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[0:3], s[0:1] sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) -; VGPRRC-NEXT: global_store_dwordx4 v44, v[4:7], s[0:1] offset:16 sc0 sc1 +; VGPRRC-NEXT: global_store_dwordx4 v36, v[4:7], s[0:1] offset:16 sc0 sc1 ; VGPRRC-NEXT: s_waitcnt vmcnt(0) ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_32x32x16_f16__vgprcd__flags: @@ -5425,18 +5421,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd(ptr addrs ; GCN-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; GCN-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; GCN-NEXT: v_mov_b32_e32 v12, 0 +; GCN-NEXT: v_mov_b32_e32 v4, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; GCN-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; GCN-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; GCN-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GCN-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; GCN-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GCN-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; GCN-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; GCN-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; GCN-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; GCN-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; GCN-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; GCN-NEXT: s_nop 1 -; GCN-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] +; GCN-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[6:9], v[10:13], v[0:3] ; GCN-NEXT: s_nop 7 -; GCN-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; GCN-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; GCN-NEXT: s_endpgm ; ; HEURRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd: @@ -5444,18 +5440,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd(ptr addrs ; HEURRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; HEURRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; HEURRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; HEURRC-NEXT: v_mov_b32_e32 v12, 0 +; HEURRC-NEXT: v_mov_b32_e32 v4, 0 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; HEURRC-NEXT: s_nop 1 -; HEURRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] +; HEURRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[6:9], v[10:13], v[0:3] ; HEURRC-NEXT: s_nop 7 -; HEURRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; HEURRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; HEURRC-NEXT: s_endpgm ; ; VGPRRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd: @@ -5463,18 +5459,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd(ptr addrs ; VGPRRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; VGPRRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; VGPRRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; VGPRRC-NEXT: v_mov_b32_e32 v12, 0 +; VGPRRC-NEXT: v_mov_b32_e32 v4, 0 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; VGPRRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; VGPRRC-NEXT: s_nop 1 -; VGPRRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] +; VGPRRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[6:9], v[10:13], v[0:3] ; VGPRRC-NEXT: s_nop 7 -; VGPRRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; VGPRRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd: ; AGPR: ; %bb.0: @@ -5525,18 +5521,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags(pt ; GCN-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; GCN-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; GCN-NEXT: v_mov_b32_e32 v12, 0 +; GCN-NEXT: v_mov_b32_e32 v4, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; GCN-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; GCN-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; GCN-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GCN-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; GCN-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GCN-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; GCN-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; GCN-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; GCN-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; GCN-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; GCN-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; GCN-NEXT: s_nop 1 -; GCN-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1 +; GCN-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1 ; GCN-NEXT: s_nop 7 -; GCN-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; GCN-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; GCN-NEXT: s_endpgm ; ; HEURRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags: @@ -5544,18 +5540,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags(pt ; HEURRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; HEURRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; HEURRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; HEURRC-NEXT: v_mov_b32_e32 v12, 0 +; HEURRC-NEXT: v_mov_b32_e32 v4, 0 ; HEURRC-NEXT: s_waitcnt lgkmcnt(0) -; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; HEURRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; HEURRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; HEURRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; HEURRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; HEURRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; HEURRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; HEURRC-NEXT: s_nop 1 -; HEURRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1 +; HEURRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1 ; HEURRC-NEXT: s_nop 7 -; HEURRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; HEURRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; HEURRC-NEXT: s_endpgm ; ; VGPRRC-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags: @@ -5563,18 +5559,18 @@ define amdgpu_kernel void @test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags(pt ; VGPRRC-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; VGPRRC-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x54 ; VGPRRC-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24 -; VGPRRC-NEXT: v_mov_b32_e32 v12, 0 +; VGPRRC-NEXT: v_mov_b32_e32 v4, 0 ; VGPRRC-NEXT: s_waitcnt lgkmcnt(0) -; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; VGPRRC-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; VGPRRC-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; VGPRRC-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; VGPRRC-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; VGPRRC-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; VGPRRC-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; VGPRRC-NEXT: s_nop 1 -; VGPRRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[4:7], v[8:11] cbsz:3 abid:2 blgp:1 +; VGPRRC-NEXT: v_mfma_f32_16x16x32_bf16 v[0:3], v[6:9], v[10:13], v[0:3] cbsz:3 abid:2 blgp:1 ; VGPRRC-NEXT: s_nop 7 -; VGPRRC-NEXT: global_store_dwordx4 v12, v[0:3], s[6:7] +; VGPRRC-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; VGPRRC-NEXT: s_endpgm ; AGPR-LABEL: test_mfma_f32_16x16x32_bf16_no_agpr__vgprcd__flags: ; AGPR: ; %bb.0: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll index 1e44a09..dbea832 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll @@ -15,7 +15,7 @@ define amdgpu_kernel void @rcp_bf16(ptr addrspace(1) %out, bfloat %src) #1 { ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 ; SDAG-TRUE16-NEXT: v_rcp_bf16_e32 v0.l, s2 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_bf16: @@ -35,10 +35,10 @@ define amdgpu_kernel void @rcp_bf16_constant_4(ptr addrspace(1) %out) #1 { ; SDAG-TRUE16-LABEL: rcp_bf16_constant_4: ; SDAG-TRUE16: ; %bb.0: ; SDAG-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 -; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3e80 ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 +; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3e80 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_bf16_constant_4: @@ -57,10 +57,10 @@ define amdgpu_kernel void @rcp_bf16_constant_100(ptr addrspace(1) %out) #1 { ; SDAG-TRUE16-LABEL: rcp_bf16_constant_100: ; SDAG-TRUE16: ; %bb.0: ; SDAG-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 -; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3c24 ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 +; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3c24 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_bf16_constant_100: @@ -79,10 +79,10 @@ define amdgpu_kernel void @rcp_undef_bf16(ptr addrspace(1) %out) #1 { ; SDAG-TRUE16-LABEL: rcp_undef_bf16: ; SDAG-TRUE16: ; %bb.0: ; SDAG-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 -; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7fc0 ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 +; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7fc0 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_undef_bf16: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll index 42d12fd..662dc613 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll @@ -15,7 +15,7 @@ define amdgpu_kernel void @rsq_bf16(ptr addrspace(1) %out, bfloat %src) #1 { ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 ; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, s2 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rsq_bf16: @@ -38,7 +38,7 @@ define amdgpu_kernel void @rsq_bf16_constant_4(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, 4.0 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rsq_bf16_constant_4: @@ -61,7 +61,7 @@ define amdgpu_kernel void @rsq_bf16_constant_100(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, 0x42c8 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rsq_bf16_constant_100: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll index 6eb9449..ee11b92 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll @@ -17,24 +17,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x64_f16__vgpr(ptr addrspace(1) % ; SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x34 ; SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; SDAG-NEXT: v_lshlrev_b32_e32 v0, 4, v0 -; SDAG-NEXT: v_mov_b32_e32 v16, 0 +; SDAG-NEXT: v_mov_b32_e32 v4, 0 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: global_load_dwordx4 v[8:11], v0, s[6:7] +; SDAG-NEXT: global_load_dwordx4 v[0:3], v0, s[6:7] ; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44 ; SDAG-NEXT: s_load_dword s16, s[4:5], 0x64 -; SDAG-NEXT: v_mov_b64_e32 v[14:15], s[2:3] -; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[0:1] +; SDAG-NEXT: v_mov_b64_e32 v[16:17], s[2:3] +; SDAG-NEXT: v_mov_b64_e32 v[14:15], s[0:1] ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; SDAG-NEXT: v_mov_b32_e32 v17, s16 +; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; SDAG-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; SDAG-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; SDAG-NEXT: v_mov_b32_e32 v5, s16 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 -; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2 +; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[0:3], v[14:17], v[6:13], v5 cbsz:1 abid:2 ; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: global_store_dwordx4 v16, v[8:11], s[6:7] +; SDAG-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; SDAG-NEXT: s_endpgm ; ; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__vgpr: @@ -120,30 +120,25 @@ define <4 x float> @test_smfmac_f32_16x16x64_f16__sgpr(<8 x half> inreg %arg0, < ; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v10, s0 -; SDAG-NEXT: v_mov_b32_e32 v11, s1 -; SDAG-NEXT: v_mov_b32_e32 v12, s2 -; SDAG-NEXT: v_mov_b32_e32 v13, s3 -; SDAG-NEXT: v_mov_b32_e32 v2, s16 -; SDAG-NEXT: v_mov_b32_e32 v3, s17 -; SDAG-NEXT: v_mov_b32_e32 v4, s18 -; SDAG-NEXT: v_mov_b32_e32 v5, s19 -; SDAG-NEXT: v_mov_b32_e32 v6, s20 -; SDAG-NEXT: v_mov_b32_e32 v7, s21 -; SDAG-NEXT: v_mov_b32_e32 v8, s22 -; SDAG-NEXT: v_mov_b32_e32 v9, s23 -; SDAG-NEXT: v_accvgpr_write_b32 a0, s24 -; SDAG-NEXT: v_accvgpr_write_b32 a1, s25 -; SDAG-NEXT: v_accvgpr_write_b32 a2, s26 -; SDAG-NEXT: v_accvgpr_write_b32 a3, s27 -; SDAG-NEXT: v_mov_b32_e32 v0, s28 +; SDAG-NEXT: v_mov_b32_e32 v14, s0 +; SDAG-NEXT: v_mov_b32_e32 v15, s1 +; SDAG-NEXT: v_mov_b32_e32 v16, s2 +; SDAG-NEXT: v_mov_b32_e32 v17, s3 +; SDAG-NEXT: v_mov_b32_e32 v6, s16 +; SDAG-NEXT: v_mov_b32_e32 v7, s17 +; SDAG-NEXT: v_mov_b32_e32 v8, s18 +; SDAG-NEXT: v_mov_b32_e32 v9, s19 +; SDAG-NEXT: v_mov_b32_e32 v10, s20 +; SDAG-NEXT: v_mov_b32_e32 v11, s21 +; SDAG-NEXT: v_mov_b32_e32 v12, s22 +; SDAG-NEXT: v_mov_b32_e32 v13, s23 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 a[0:3], v[10:13], v[2:9], v0 -; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: v_accvgpr_read_b32 v0, a0 -; SDAG-NEXT: v_accvgpr_read_b32 v1, a1 -; SDAG-NEXT: v_accvgpr_read_b32 v2, a2 -; SDAG-NEXT: v_accvgpr_read_b32 v3, a3 +; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[0:3], v[14:17], v[6:13], v4 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__sgpr: @@ -187,17 +182,17 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x32_f16__vgpr(ptr addrspace(1) % ; SDAG-NEXT: global_load_dwordx4 v[0:3], v16, s[6:7] ; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44 ; SDAG-NEXT: s_load_dword s16, s[4:5], 0x64 -; SDAG-NEXT: v_mov_b64_e32 v[26:27], s[2:3] -; SDAG-NEXT: v_mov_b64_e32 v[24:25], s[0:1] +; SDAG-NEXT: v_mov_b64_e32 v[28:29], s[2:3] +; SDAG-NEXT: v_mov_b64_e32 v[26:27], s[0:1] ; SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SDAG-NEXT: v_mov_b64_e32 v[22:23], s[14:15] -; SDAG-NEXT: v_mov_b64_e32 v[20:21], s[12:13] -; SDAG-NEXT: v_mov_b64_e32 v[18:19], s[10:11] -; SDAG-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; SDAG-NEXT: v_mov_b32_e32 v28, s16 +; SDAG-NEXT: v_mov_b64_e32 v[24:25], s[14:15] +; SDAG-NEXT: v_mov_b64_e32 v[22:23], s[12:13] +; SDAG-NEXT: v_mov_b64_e32 v[20:21], s[10:11] +; SDAG-NEXT: v_mov_b64_e32 v[18:19], s[8:9] +; SDAG-NEXT: v_mov_b32_e32 v16, s16 ; SDAG-NEXT: s_waitcnt vmcnt(0) ; SDAG-NEXT: s_nop 0 -; SDAG-NEXT: v_smfmac_f32_32x32x32_f16 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2 +; SDAG-NEXT: v_smfmac_f32_32x32x32_f16 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2 ; SDAG-NEXT: v_mov_b32_e32 v16, 0 ; SDAG-NEXT: s_nop 10 ; SDAG-NEXT: global_store_dwordx4 v16, v[8:11], s[6:7] offset:32 @@ -436,53 +431,37 @@ define <16 x float> @test_smfmac_f32_32x32x32_f16__sgpr(<8 x half> inreg %arg0, ; SDAG-LABEL: test_smfmac_f32_32x32x32_f16__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v36, s0 -; SDAG-NEXT: v_mov_b32_e32 v37, s1 -; SDAG-NEXT: v_mov_b32_e32 v38, s2 -; SDAG-NEXT: v_mov_b32_e32 v39, s3 -; SDAG-NEXT: v_mov_b32_e32 v13, s25 -; SDAG-NEXT: v_mov_b32_e32 v14, s26 -; SDAG-NEXT: v_mov_b32_e32 v15, s27 -; SDAG-NEXT: v_mov_b32_e32 v16, s28 -; SDAG-NEXT: v_mov_b32_e32 v17, s29 -; SDAG-NEXT: v_mov_b32_e32 v28, s16 -; SDAG-NEXT: v_mov_b32_e32 v29, s17 -; SDAG-NEXT: v_mov_b32_e32 v30, s18 -; SDAG-NEXT: v_mov_b32_e32 v31, s19 -; SDAG-NEXT: v_mov_b32_e32 v32, s20 -; SDAG-NEXT: v_mov_b32_e32 v33, s21 -; SDAG-NEXT: v_mov_b32_e32 v34, s22 -; SDAG-NEXT: v_mov_b32_e32 v35, s23 -; SDAG-NEXT: v_mov_b32_e32 v12, s24 -; SDAG-NEXT: v_mov_b32_e32 v18, v0 -; SDAG-NEXT: v_mov_b32_e32 v19, v1 -; SDAG-NEXT: v_mov_b32_e32 v20, v2 -; SDAG-NEXT: v_mov_b32_e32 v21, v3 -; SDAG-NEXT: v_mov_b32_e32 v22, v4 -; SDAG-NEXT: v_mov_b32_e32 v23, v5 -; SDAG-NEXT: v_mov_b32_e32 v24, v6 -; SDAG-NEXT: v_mov_b32_e32 v25, v7 -; SDAG-NEXT: v_mov_b32_e32 v26, v8 -; SDAG-NEXT: v_mov_b32_e32 v27, v9 +; SDAG-NEXT: v_mov_b32_e32 v26, s0 +; SDAG-NEXT: v_mov_b32_e32 v27, s1 +; SDAG-NEXT: v_mov_b32_e32 v28, s2 +; SDAG-NEXT: v_mov_b32_e32 v29, s3 +; SDAG-NEXT: v_mov_b32_e32 v16, v10 +; SDAG-NEXT: v_mov_b32_e32 v15, v9 +; SDAG-NEXT: v_mov_b32_e32 v14, v8 +; SDAG-NEXT: v_mov_b32_e32 v13, v7 +; SDAG-NEXT: v_mov_b32_e32 v12, v6 +; SDAG-NEXT: v_mov_b32_e32 v11, v5 +; SDAG-NEXT: v_mov_b32_e32 v10, v4 +; SDAG-NEXT: v_mov_b32_e32 v9, v3 +; SDAG-NEXT: v_mov_b32_e32 v8, v2 +; SDAG-NEXT: v_mov_b32_e32 v7, v1 +; SDAG-NEXT: v_mov_b32_e32 v6, v0 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 +; SDAG-NEXT: v_mov_b32_e32 v5, s29 +; SDAG-NEXT: v_mov_b32_e32 v18, s16 +; SDAG-NEXT: v_mov_b32_e32 v19, s17 +; SDAG-NEXT: v_mov_b32_e32 v20, s18 +; SDAG-NEXT: v_mov_b32_e32 v21, s19 +; SDAG-NEXT: v_mov_b32_e32 v22, s20 +; SDAG-NEXT: v_mov_b32_e32 v23, s21 +; SDAG-NEXT: v_mov_b32_e32 v24, s22 +; SDAG-NEXT: v_mov_b32_e32 v25, s23 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_32x32x32_f16 v[12:27], v[36:39], v[28:35], v10 -; SDAG-NEXT: s_nop 11 -; SDAG-NEXT: v_mov_b32_e32 v0, v12 -; SDAG-NEXT: v_mov_b32_e32 v1, v13 -; SDAG-NEXT: v_mov_b32_e32 v2, v14 -; SDAG-NEXT: v_mov_b32_e32 v3, v15 -; SDAG-NEXT: v_mov_b32_e32 v4, v16 -; SDAG-NEXT: v_mov_b32_e32 v5, v17 -; SDAG-NEXT: v_mov_b32_e32 v6, v18 -; SDAG-NEXT: v_mov_b32_e32 v7, v19 -; SDAG-NEXT: v_mov_b32_e32 v8, v20 -; SDAG-NEXT: v_mov_b32_e32 v9, v21 -; SDAG-NEXT: v_mov_b32_e32 v10, v22 -; SDAG-NEXT: v_mov_b32_e32 v11, v23 -; SDAG-NEXT: v_mov_b32_e32 v12, v24 -; SDAG-NEXT: v_mov_b32_e32 v13, v25 -; SDAG-NEXT: v_mov_b32_e32 v14, v26 -; SDAG-NEXT: v_mov_b32_e32 v15, v27 +; SDAG-NEXT: v_smfmac_f32_32x32x32_f16 v[0:15], v[26:29], v[18:25], v16 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_32x32x32_f16__sgpr: @@ -541,24 +520,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x64_bf16__vgpr(ptr addrspace(1) ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x34 ; GCN-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GCN-NEXT: v_lshlrev_b32_e32 v0, 4, v0 -; GCN-NEXT: v_mov_b32_e32 v16, 0 +; GCN-NEXT: v_mov_b32_e32 v4, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: global_load_dwordx4 v[8:11], v0, s[6:7] +; GCN-NEXT: global_load_dwordx4 v[0:3], v0, s[6:7] ; GCN-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44 ; GCN-NEXT: s_load_dword s16, s[4:5], 0x64 -; GCN-NEXT: v_mov_b64_e32 v[14:15], s[2:3] -; GCN-NEXT: v_mov_b64_e32 v[12:13], s[0:1] +; GCN-NEXT: v_mov_b64_e32 v[16:17], s[2:3] +; GCN-NEXT: v_mov_b64_e32 v[14:15], s[0:1] ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[0:1], s[8:9] -; GCN-NEXT: v_mov_b64_e32 v[2:3], s[10:11] -; GCN-NEXT: v_mov_b64_e32 v[4:5], s[12:13] -; GCN-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; GCN-NEXT: v_mov_b32_e32 v17, s16 +; GCN-NEXT: v_mov_b64_e32 v[6:7], s[8:9] +; GCN-NEXT: v_mov_b64_e32 v[8:9], s[10:11] +; GCN-NEXT: v_mov_b64_e32 v[10:11], s[12:13] +; GCN-NEXT: v_mov_b64_e32 v[12:13], s[14:15] +; GCN-NEXT: v_mov_b32_e32 v5, s16 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 -; GCN-NEXT: v_smfmac_f32_16x16x64_bf16 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2 +; GCN-NEXT: v_smfmac_f32_16x16x64_bf16 v[0:3], v[14:17], v[6:13], v5 cbsz:1 abid:2 ; GCN-NEXT: s_nop 7 -; GCN-NEXT: global_store_dwordx4 v16, v[8:11], s[6:7] +; GCN-NEXT: global_store_dwordx4 v4, v[0:3], s[6:7] ; GCN-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() @@ -618,30 +597,25 @@ define <4 x float> @test_smfmac_f32_16x16x64_bf16__sgpr(<8 x bfloat> inreg %arg0 ; GCN-LABEL: test_smfmac_f32_16x16x64_bf16__sgpr: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_mov_b32_e32 v10, s0 -; GCN-NEXT: v_mov_b32_e32 v11, s1 -; GCN-NEXT: v_mov_b32_e32 v12, s2 -; GCN-NEXT: v_mov_b32_e32 v13, s3 -; GCN-NEXT: v_mov_b32_e32 v2, s16 -; GCN-NEXT: v_mov_b32_e32 v3, s17 -; GCN-NEXT: v_mov_b32_e32 v4, s18 -; GCN-NEXT: v_mov_b32_e32 v5, s19 -; GCN-NEXT: v_mov_b32_e32 v6, s20 -; GCN-NEXT: v_mov_b32_e32 v7, s21 -; GCN-NEXT: v_mov_b32_e32 v8, s22 -; GCN-NEXT: v_mov_b32_e32 v9, s23 -; GCN-NEXT: v_accvgpr_write_b32 a0, s24 -; GCN-NEXT: v_accvgpr_write_b32 a1, s25 -; GCN-NEXT: v_accvgpr_write_b32 a2, s26 -; GCN-NEXT: v_accvgpr_write_b32 a3, s27 -; GCN-NEXT: v_mov_b32_e32 v0, s28 +; GCN-NEXT: v_mov_b32_e32 v14, s0 +; GCN-NEXT: v_mov_b32_e32 v15, s1 +; GCN-NEXT: v_mov_b32_e32 v16, s2 +; GCN-NEXT: v_mov_b32_e32 v17, s3 +; GCN-NEXT: v_mov_b32_e32 v6, s16 +; GCN-NEXT: v_mov_b32_e32 v7, s17 +; GCN-NEXT: v_mov_b32_e32 v8, s18 +; GCN-NEXT: v_mov_b32_e32 v9, s19 +; GCN-NEXT: v_mov_b32_e32 v10, s20 +; GCN-NEXT: v_mov_b32_e32 v11, s21 +; GCN-NEXT: v_mov_b32_e32 v12, s22 +; GCN-NEXT: v_mov_b32_e32 v13, s23 +; GCN-NEXT: v_mov_b32_e32 v0, s24 +; GCN-NEXT: v_mov_b32_e32 v1, s25 +; GCN-NEXT: v_mov_b32_e32 v2, s26 +; GCN-NEXT: v_mov_b32_e32 v3, s27 +; GCN-NEXT: v_mov_b32_e32 v4, s28 ; GCN-NEXT: s_nop 1 -; GCN-NEXT: v_smfmac_f32_16x16x64_bf16 a[0:3], v[10:13], v[2:9], v0 -; GCN-NEXT: s_nop 7 -; GCN-NEXT: v_accvgpr_read_b32 v0, a0 -; GCN-NEXT: v_accvgpr_read_b32 v1, a1 -; GCN-NEXT: v_accvgpr_read_b32 v2, a2 -; GCN-NEXT: v_accvgpr_read_b32 v3, a3 +; GCN-NEXT: v_smfmac_f32_16x16x64_bf16 v[0:3], v[14:17], v[6:13], v4 ; GCN-NEXT: s_setpc_b64 s[30:31] %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.bf16(<8 x bfloat> %arg0, <16 x bfloat> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0) ret <4 x float> %result @@ -667,17 +641,17 @@ define amdgpu_kernel void @test_smfmac_f32_32x32x32_bf16__vgpr(ptr addrspace(1) ; GCN-NEXT: global_load_dwordx4 v[0:3], v16, s[6:7] ; GCN-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44 ; GCN-NEXT: s_load_dword s16, s[4:5], 0x64 -; GCN-NEXT: v_mov_b64_e32 v[26:27], s[2:3] -; GCN-NEXT: v_mov_b64_e32 v[24:25], s[0:1] +; GCN-NEXT: v_mov_b64_e32 v[28:29], s[2:3] +; GCN-NEXT: v_mov_b64_e32 v[26:27], s[0:1] ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mov_b64_e32 v[22:23], s[14:15] -; GCN-NEXT: v_mov_b64_e32 v[20:21], s[12:13] -; GCN-NEXT: v_mov_b64_e32 v[18:19], s[10:11] -; GCN-NEXT: v_mov_b64_e32 v[16:17], s[8:9] -; GCN-NEXT: v_mov_b32_e32 v28, s16 +; GCN-NEXT: v_mov_b64_e32 v[24:25], s[14:15] +; GCN-NEXT: v_mov_b64_e32 v[22:23], s[12:13] +; GCN-NEXT: v_mov_b64_e32 v[20:21], s[10:11] +; GCN-NEXT: v_mov_b64_e32 v[18:19], s[8:9] +; GCN-NEXT: v_mov_b32_e32 v16, s16 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_nop 0 -; GCN-NEXT: v_smfmac_f32_32x32x32_bf16 v[0:15], v[24:27], v[16:23], v28 cbsz:1 abid:2 +; GCN-NEXT: v_smfmac_f32_32x32x32_bf16 v[0:15], v[26:29], v[18:25], v16 cbsz:1 abid:2 ; GCN-NEXT: v_mov_b32_e32 v16, 0 ; GCN-NEXT: s_nop 10 ; GCN-NEXT: global_store_dwordx4 v16, v[8:11], s[6:7] offset:32 @@ -779,53 +753,37 @@ define <16 x float> @test_smfmac_f32_32x32x32_bf16__sgpr(<8 x bfloat> inreg %arg ; GCN-LABEL: test_smfmac_f32_32x32x32_bf16__sgpr: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_mov_b32_e32 v36, s0 -; GCN-NEXT: v_mov_b32_e32 v37, s1 -; GCN-NEXT: v_mov_b32_e32 v38, s2 -; GCN-NEXT: v_mov_b32_e32 v39, s3 -; GCN-NEXT: v_mov_b32_e32 v13, s25 -; GCN-NEXT: v_mov_b32_e32 v14, s26 -; GCN-NEXT: v_mov_b32_e32 v15, s27 -; GCN-NEXT: v_mov_b32_e32 v16, s28 -; GCN-NEXT: v_mov_b32_e32 v17, s29 -; GCN-NEXT: v_mov_b32_e32 v28, s16 -; GCN-NEXT: v_mov_b32_e32 v29, s17 -; GCN-NEXT: v_mov_b32_e32 v30, s18 -; GCN-NEXT: v_mov_b32_e32 v31, s19 -; GCN-NEXT: v_mov_b32_e32 v32, s20 -; GCN-NEXT: v_mov_b32_e32 v33, s21 -; GCN-NEXT: v_mov_b32_e32 v34, s22 -; GCN-NEXT: v_mov_b32_e32 v35, s23 -; GCN-NEXT: v_mov_b32_e32 v12, s24 -; GCN-NEXT: v_mov_b32_e32 v18, v0 -; GCN-NEXT: v_mov_b32_e32 v19, v1 -; GCN-NEXT: v_mov_b32_e32 v20, v2 -; GCN-NEXT: v_mov_b32_e32 v21, v3 -; GCN-NEXT: v_mov_b32_e32 v22, v4 -; GCN-NEXT: v_mov_b32_e32 v23, v5 -; GCN-NEXT: v_mov_b32_e32 v24, v6 -; GCN-NEXT: v_mov_b32_e32 v25, v7 -; GCN-NEXT: v_mov_b32_e32 v26, v8 -; GCN-NEXT: v_mov_b32_e32 v27, v9 +; GCN-NEXT: v_mov_b32_e32 v26, s0 +; GCN-NEXT: v_mov_b32_e32 v27, s1 +; GCN-NEXT: v_mov_b32_e32 v28, s2 +; GCN-NEXT: v_mov_b32_e32 v29, s3 +; GCN-NEXT: v_mov_b32_e32 v16, v10 +; GCN-NEXT: v_mov_b32_e32 v15, v9 +; GCN-NEXT: v_mov_b32_e32 v14, v8 +; GCN-NEXT: v_mov_b32_e32 v13, v7 +; GCN-NEXT: v_mov_b32_e32 v12, v6 +; GCN-NEXT: v_mov_b32_e32 v11, v5 +; GCN-NEXT: v_mov_b32_e32 v10, v4 +; GCN-NEXT: v_mov_b32_e32 v9, v3 +; GCN-NEXT: v_mov_b32_e32 v8, v2 +; GCN-NEXT: v_mov_b32_e32 v7, v1 +; GCN-NEXT: v_mov_b32_e32 v6, v0 +; GCN-NEXT: v_mov_b32_e32 v0, s24 +; GCN-NEXT: v_mov_b32_e32 v1, s25 +; GCN-NEXT: v_mov_b32_e32 v2, s26 +; GCN-NEXT: v_mov_b32_e32 v3, s27 +; GCN-NEXT: v_mov_b32_e32 v4, s28 +; GCN-NEXT: v_mov_b32_e32 v5, s29 +; GCN-NEXT: v_mov_b32_e32 v18, s16 +; GCN-NEXT: v_mov_b32_e32 v19, s17 +; GCN-NEXT: v_mov_b32_e32 v20, s18 +; GCN-NEXT: v_mov_b32_e32 v21, s19 +; GCN-NEXT: v_mov_b32_e32 v22, s20 +; GCN-NEXT: v_mov_b32_e32 v23, s21 +; GCN-NEXT: v_mov_b32_e32 v24, s22 +; GCN-NEXT: v_mov_b32_e32 v25, s23 ; GCN-NEXT: s_nop 1 -; GCN-NEXT: v_smfmac_f32_32x32x32_bf16 v[12:27], v[36:39], v[28:35], v10 -; GCN-NEXT: s_nop 11 -; GCN-NEXT: v_mov_b32_e32 v0, v12 -; GCN-NEXT: v_mov_b32_e32 v1, v13 -; GCN-NEXT: v_mov_b32_e32 v2, v14 -; GCN-NEXT: v_mov_b32_e32 v3, v15 -; GCN-NEXT: v_mov_b32_e32 v4, v16 -; GCN-NEXT: v_mov_b32_e32 v5, v17 -; GCN-NEXT: v_mov_b32_e32 v6, v18 -; GCN-NEXT: v_mov_b32_e32 v7, v19 -; GCN-NEXT: v_mov_b32_e32 v8, v20 -; GCN-NEXT: v_mov_b32_e32 v9, v21 -; GCN-NEXT: v_mov_b32_e32 v10, v22 -; GCN-NEXT: v_mov_b32_e32 v11, v23 -; GCN-NEXT: v_mov_b32_e32 v12, v24 -; GCN-NEXT: v_mov_b32_e32 v13, v25 -; GCN-NEXT: v_mov_b32_e32 v14, v26 -; GCN-NEXT: v_mov_b32_e32 v15, v27 +; GCN-NEXT: v_smfmac_f32_32x32x32_bf16 v[0:15], v[26:29], v[18:25], v16 ; GCN-NEXT: s_setpc_b64 s[30:31] %result = call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x32.bf16(<8 x bfloat> %arg0, <16 x bfloat> %arg1, <16 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0) ret <16 x float> %result @@ -953,30 +911,25 @@ define <4 x i32> @test_smfmac_i32_16x16x128_i8__sgpr(<4 x i32> inreg %arg0, <8 x ; SDAG-LABEL: test_smfmac_i32_16x16x128_i8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v10, s0 -; SDAG-NEXT: v_mov_b32_e32 v11, s1 -; SDAG-NEXT: v_mov_b32_e32 v12, s2 -; SDAG-NEXT: v_mov_b32_e32 v13, s3 -; SDAG-NEXT: v_mov_b32_e32 v2, s16 -; SDAG-NEXT: v_mov_b32_e32 v3, s17 -; SDAG-NEXT: v_mov_b32_e32 v4, s18 -; SDAG-NEXT: v_mov_b32_e32 v5, s19 -; SDAG-NEXT: v_mov_b32_e32 v6, s20 -; SDAG-NEXT: v_mov_b32_e32 v7, s21 -; SDAG-NEXT: v_mov_b32_e32 v8, s22 -; SDAG-NEXT: v_mov_b32_e32 v9, s23 -; SDAG-NEXT: v_accvgpr_write_b32 a0, s24 -; SDAG-NEXT: v_accvgpr_write_b32 a1, s25 -; SDAG-NEXT: v_accvgpr_write_b32 a2, s26 -; SDAG-NEXT: v_accvgpr_write_b32 a3, s27 -; SDAG-NEXT: v_mov_b32_e32 v0, s28 +; SDAG-NEXT: v_mov_b32_e32 v14, s0 +; SDAG-NEXT: v_mov_b32_e32 v15, s1 +; SDAG-NEXT: v_mov_b32_e32 v16, s2 +; SDAG-NEXT: v_mov_b32_e32 v17, s3 +; SDAG-NEXT: v_mov_b32_e32 v6, s16 +; SDAG-NEXT: v_mov_b32_e32 v7, s17 +; SDAG-NEXT: v_mov_b32_e32 v8, s18 +; SDAG-NEXT: v_mov_b32_e32 v9, s19 +; SDAG-NEXT: v_mov_b32_e32 v10, s20 +; SDAG-NEXT: v_mov_b32_e32 v11, s21 +; SDAG-NEXT: v_mov_b32_e32 v12, s22 +; SDAG-NEXT: v_mov_b32_e32 v13, s23 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_i32_16x16x128_i8 a[0:3], v[10:13], v[2:9], v0 -; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: v_accvgpr_read_b32 v0, a0 -; SDAG-NEXT: v_accvgpr_read_b32 v1, a1 -; SDAG-NEXT: v_accvgpr_read_b32 v2, a2 -; SDAG-NEXT: v_accvgpr_read_b32 v3, a3 +; SDAG-NEXT: v_smfmac_i32_16x16x128_i8 v[0:3], v[14:17], v[6:13], v4 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_i32_16x16x128_i8__sgpr: @@ -1275,53 +1228,37 @@ define <16 x i32> @test_smfmac_i32_32x32x64_i8__sgpr(<4 x i32> inreg %arg0, <8 x ; SDAG-LABEL: test_smfmac_i32_32x32x64_i8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v36, s0 -; SDAG-NEXT: v_mov_b32_e32 v37, s1 -; SDAG-NEXT: v_mov_b32_e32 v38, s2 -; SDAG-NEXT: v_mov_b32_e32 v39, s3 -; SDAG-NEXT: v_mov_b32_e32 v13, s25 -; SDAG-NEXT: v_mov_b32_e32 v14, s26 -; SDAG-NEXT: v_mov_b32_e32 v15, s27 -; SDAG-NEXT: v_mov_b32_e32 v16, s28 -; SDAG-NEXT: v_mov_b32_e32 v17, s29 -; SDAG-NEXT: v_mov_b32_e32 v28, s16 -; SDAG-NEXT: v_mov_b32_e32 v29, s17 -; SDAG-NEXT: v_mov_b32_e32 v30, s18 -; SDAG-NEXT: v_mov_b32_e32 v31, s19 -; SDAG-NEXT: v_mov_b32_e32 v32, s20 -; SDAG-NEXT: v_mov_b32_e32 v33, s21 -; SDAG-NEXT: v_mov_b32_e32 v34, s22 -; SDAG-NEXT: v_mov_b32_e32 v35, s23 -; SDAG-NEXT: v_mov_b32_e32 v12, s24 -; SDAG-NEXT: v_mov_b32_e32 v18, v0 -; SDAG-NEXT: v_mov_b32_e32 v19, v1 -; SDAG-NEXT: v_mov_b32_e32 v20, v2 -; SDAG-NEXT: v_mov_b32_e32 v21, v3 -; SDAG-NEXT: v_mov_b32_e32 v22, v4 -; SDAG-NEXT: v_mov_b32_e32 v23, v5 -; SDAG-NEXT: v_mov_b32_e32 v24, v6 -; SDAG-NEXT: v_mov_b32_e32 v25, v7 -; SDAG-NEXT: v_mov_b32_e32 v26, v8 -; SDAG-NEXT: v_mov_b32_e32 v27, v9 +; SDAG-NEXT: v_mov_b32_e32 v26, s0 +; SDAG-NEXT: v_mov_b32_e32 v27, s1 +; SDAG-NEXT: v_mov_b32_e32 v28, s2 +; SDAG-NEXT: v_mov_b32_e32 v29, s3 +; SDAG-NEXT: v_mov_b32_e32 v16, v10 +; SDAG-NEXT: v_mov_b32_e32 v15, v9 +; SDAG-NEXT: v_mov_b32_e32 v14, v8 +; SDAG-NEXT: v_mov_b32_e32 v13, v7 +; SDAG-NEXT: v_mov_b32_e32 v12, v6 +; SDAG-NEXT: v_mov_b32_e32 v11, v5 +; SDAG-NEXT: v_mov_b32_e32 v10, v4 +; SDAG-NEXT: v_mov_b32_e32 v9, v3 +; SDAG-NEXT: v_mov_b32_e32 v8, v2 +; SDAG-NEXT: v_mov_b32_e32 v7, v1 +; SDAG-NEXT: v_mov_b32_e32 v6, v0 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 +; SDAG-NEXT: v_mov_b32_e32 v5, s29 +; SDAG-NEXT: v_mov_b32_e32 v18, s16 +; SDAG-NEXT: v_mov_b32_e32 v19, s17 +; SDAG-NEXT: v_mov_b32_e32 v20, s18 +; SDAG-NEXT: v_mov_b32_e32 v21, s19 +; SDAG-NEXT: v_mov_b32_e32 v22, s20 +; SDAG-NEXT: v_mov_b32_e32 v23, s21 +; SDAG-NEXT: v_mov_b32_e32 v24, s22 +; SDAG-NEXT: v_mov_b32_e32 v25, s23 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_i32_32x32x64_i8 v[12:27], v[36:39], v[28:35], v10 -; SDAG-NEXT: s_nop 11 -; SDAG-NEXT: v_mov_b32_e32 v0, v12 -; SDAG-NEXT: v_mov_b32_e32 v1, v13 -; SDAG-NEXT: v_mov_b32_e32 v2, v14 -; SDAG-NEXT: v_mov_b32_e32 v3, v15 -; SDAG-NEXT: v_mov_b32_e32 v4, v16 -; SDAG-NEXT: v_mov_b32_e32 v5, v17 -; SDAG-NEXT: v_mov_b32_e32 v6, v18 -; SDAG-NEXT: v_mov_b32_e32 v7, v19 -; SDAG-NEXT: v_mov_b32_e32 v8, v20 -; SDAG-NEXT: v_mov_b32_e32 v9, v21 -; SDAG-NEXT: v_mov_b32_e32 v10, v22 -; SDAG-NEXT: v_mov_b32_e32 v11, v23 -; SDAG-NEXT: v_mov_b32_e32 v12, v24 -; SDAG-NEXT: v_mov_b32_e32 v13, v25 -; SDAG-NEXT: v_mov_b32_e32 v14, v26 -; SDAG-NEXT: v_mov_b32_e32 v15, v27 +; SDAG-NEXT: v_smfmac_i32_32x32x64_i8 v[0:15], v[26:29], v[18:25], v16 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_i32_32x32x64_i8__sgpr: @@ -1489,30 +1426,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_bf8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v10, s0 -; SDAG-NEXT: v_mov_b32_e32 v11, s1 -; SDAG-NEXT: v_mov_b32_e32 v12, s2 -; SDAG-NEXT: v_mov_b32_e32 v13, s3 -; SDAG-NEXT: v_mov_b32_e32 v2, s16 -; SDAG-NEXT: v_mov_b32_e32 v3, s17 -; SDAG-NEXT: v_mov_b32_e32 v4, s18 -; SDAG-NEXT: v_mov_b32_e32 v5, s19 -; SDAG-NEXT: v_mov_b32_e32 v6, s20 -; SDAG-NEXT: v_mov_b32_e32 v7, s21 -; SDAG-NEXT: v_mov_b32_e32 v8, s22 -; SDAG-NEXT: v_mov_b32_e32 v9, s23 -; SDAG-NEXT: v_accvgpr_write_b32 a0, s24 -; SDAG-NEXT: v_accvgpr_write_b32 a1, s25 -; SDAG-NEXT: v_accvgpr_write_b32 a2, s26 -; SDAG-NEXT: v_accvgpr_write_b32 a3, s27 -; SDAG-NEXT: v_mov_b32_e32 v0, s28 +; SDAG-NEXT: v_mov_b32_e32 v14, s0 +; SDAG-NEXT: v_mov_b32_e32 v15, s1 +; SDAG-NEXT: v_mov_b32_e32 v16, s2 +; SDAG-NEXT: v_mov_b32_e32 v17, s3 +; SDAG-NEXT: v_mov_b32_e32 v6, s16 +; SDAG-NEXT: v_mov_b32_e32 v7, s17 +; SDAG-NEXT: v_mov_b32_e32 v8, s18 +; SDAG-NEXT: v_mov_b32_e32 v9, s19 +; SDAG-NEXT: v_mov_b32_e32 v10, s20 +; SDAG-NEXT: v_mov_b32_e32 v11, s21 +; SDAG-NEXT: v_mov_b32_e32 v12, s22 +; SDAG-NEXT: v_mov_b32_e32 v13, s23 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 a[0:3], v[10:13], v[2:9], v0 -; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: v_accvgpr_read_b32 v0, a0 -; SDAG-NEXT: v_accvgpr_read_b32 v1, a1 -; SDAG-NEXT: v_accvgpr_read_b32 v2, a2 -; SDAG-NEXT: v_accvgpr_read_b32 v3, a3 +; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[0:3], v[14:17], v[6:13], v4 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_bf8__sgpr: @@ -1658,30 +1590,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_bf8_fp8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v10, s0 -; SDAG-NEXT: v_mov_b32_e32 v11, s1 -; SDAG-NEXT: v_mov_b32_e32 v12, s2 -; SDAG-NEXT: v_mov_b32_e32 v13, s3 -; SDAG-NEXT: v_mov_b32_e32 v2, s16 -; SDAG-NEXT: v_mov_b32_e32 v3, s17 -; SDAG-NEXT: v_mov_b32_e32 v4, s18 -; SDAG-NEXT: v_mov_b32_e32 v5, s19 -; SDAG-NEXT: v_mov_b32_e32 v6, s20 -; SDAG-NEXT: v_mov_b32_e32 v7, s21 -; SDAG-NEXT: v_mov_b32_e32 v8, s22 -; SDAG-NEXT: v_mov_b32_e32 v9, s23 -; SDAG-NEXT: v_accvgpr_write_b32 a0, s24 -; SDAG-NEXT: v_accvgpr_write_b32 a1, s25 -; SDAG-NEXT: v_accvgpr_write_b32 a2, s26 -; SDAG-NEXT: v_accvgpr_write_b32 a3, s27 -; SDAG-NEXT: v_mov_b32_e32 v0, s28 +; SDAG-NEXT: v_mov_b32_e32 v14, s0 +; SDAG-NEXT: v_mov_b32_e32 v15, s1 +; SDAG-NEXT: v_mov_b32_e32 v16, s2 +; SDAG-NEXT: v_mov_b32_e32 v17, s3 +; SDAG-NEXT: v_mov_b32_e32 v6, s16 +; SDAG-NEXT: v_mov_b32_e32 v7, s17 +; SDAG-NEXT: v_mov_b32_e32 v8, s18 +; SDAG-NEXT: v_mov_b32_e32 v9, s19 +; SDAG-NEXT: v_mov_b32_e32 v10, s20 +; SDAG-NEXT: v_mov_b32_e32 v11, s21 +; SDAG-NEXT: v_mov_b32_e32 v12, s22 +; SDAG-NEXT: v_mov_b32_e32 v13, s23 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 a[0:3], v[10:13], v[2:9], v0 -; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: v_accvgpr_read_b32 v0, a0 -; SDAG-NEXT: v_accvgpr_read_b32 v1, a1 -; SDAG-NEXT: v_accvgpr_read_b32 v2, a2 -; SDAG-NEXT: v_accvgpr_read_b32 v3, a3 +; SDAG-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[0:3], v[14:17], v[6:13], v4 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_16x16x128_bf8_fp8__sgpr: @@ -1827,30 +1754,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_bf8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v10, s0 -; SDAG-NEXT: v_mov_b32_e32 v11, s1 -; SDAG-NEXT: v_mov_b32_e32 v12, s2 -; SDAG-NEXT: v_mov_b32_e32 v13, s3 -; SDAG-NEXT: v_mov_b32_e32 v2, s16 -; SDAG-NEXT: v_mov_b32_e32 v3, s17 -; SDAG-NEXT: v_mov_b32_e32 v4, s18 -; SDAG-NEXT: v_mov_b32_e32 v5, s19 -; SDAG-NEXT: v_mov_b32_e32 v6, s20 -; SDAG-NEXT: v_mov_b32_e32 v7, s21 -; SDAG-NEXT: v_mov_b32_e32 v8, s22 -; SDAG-NEXT: v_mov_b32_e32 v9, s23 -; SDAG-NEXT: v_accvgpr_write_b32 a0, s24 -; SDAG-NEXT: v_accvgpr_write_b32 a1, s25 -; SDAG-NEXT: v_accvgpr_write_b32 a2, s26 -; SDAG-NEXT: v_accvgpr_write_b32 a3, s27 -; SDAG-NEXT: v_mov_b32_e32 v0, s28 +; SDAG-NEXT: v_mov_b32_e32 v14, s0 +; SDAG-NEXT: v_mov_b32_e32 v15, s1 +; SDAG-NEXT: v_mov_b32_e32 v16, s2 +; SDAG-NEXT: v_mov_b32_e32 v17, s3 +; SDAG-NEXT: v_mov_b32_e32 v6, s16 +; SDAG-NEXT: v_mov_b32_e32 v7, s17 +; SDAG-NEXT: v_mov_b32_e32 v8, s18 +; SDAG-NEXT: v_mov_b32_e32 v9, s19 +; SDAG-NEXT: v_mov_b32_e32 v10, s20 +; SDAG-NEXT: v_mov_b32_e32 v11, s21 +; SDAG-NEXT: v_mov_b32_e32 v12, s22 +; SDAG-NEXT: v_mov_b32_e32 v13, s23 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 a[0:3], v[10:13], v[2:9], v0 -; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: v_accvgpr_read_b32 v0, a0 -; SDAG-NEXT: v_accvgpr_read_b32 v1, a1 -; SDAG-NEXT: v_accvgpr_read_b32 v2, a2 -; SDAG-NEXT: v_accvgpr_read_b32 v3, a3 +; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[0:3], v[14:17], v[6:13], v4 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_bf8__sgpr: @@ -1996,30 +1918,25 @@ define <4 x float> @test_smfmac_f32_16x16x128_fp8_fp8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v10, s0 -; SDAG-NEXT: v_mov_b32_e32 v11, s1 -; SDAG-NEXT: v_mov_b32_e32 v12, s2 -; SDAG-NEXT: v_mov_b32_e32 v13, s3 -; SDAG-NEXT: v_mov_b32_e32 v2, s16 -; SDAG-NEXT: v_mov_b32_e32 v3, s17 -; SDAG-NEXT: v_mov_b32_e32 v4, s18 -; SDAG-NEXT: v_mov_b32_e32 v5, s19 -; SDAG-NEXT: v_mov_b32_e32 v6, s20 -; SDAG-NEXT: v_mov_b32_e32 v7, s21 -; SDAG-NEXT: v_mov_b32_e32 v8, s22 -; SDAG-NEXT: v_mov_b32_e32 v9, s23 -; SDAG-NEXT: v_accvgpr_write_b32 a0, s24 -; SDAG-NEXT: v_accvgpr_write_b32 a1, s25 -; SDAG-NEXT: v_accvgpr_write_b32 a2, s26 -; SDAG-NEXT: v_accvgpr_write_b32 a3, s27 -; SDAG-NEXT: v_mov_b32_e32 v0, s28 +; SDAG-NEXT: v_mov_b32_e32 v14, s0 +; SDAG-NEXT: v_mov_b32_e32 v15, s1 +; SDAG-NEXT: v_mov_b32_e32 v16, s2 +; SDAG-NEXT: v_mov_b32_e32 v17, s3 +; SDAG-NEXT: v_mov_b32_e32 v6, s16 +; SDAG-NEXT: v_mov_b32_e32 v7, s17 +; SDAG-NEXT: v_mov_b32_e32 v8, s18 +; SDAG-NEXT: v_mov_b32_e32 v9, s19 +; SDAG-NEXT: v_mov_b32_e32 v10, s20 +; SDAG-NEXT: v_mov_b32_e32 v11, s21 +; SDAG-NEXT: v_mov_b32_e32 v12, s22 +; SDAG-NEXT: v_mov_b32_e32 v13, s23 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 a[0:3], v[10:13], v[2:9], v0 -; SDAG-NEXT: s_nop 7 -; SDAG-NEXT: v_accvgpr_read_b32 v0, a0 -; SDAG-NEXT: v_accvgpr_read_b32 v1, a1 -; SDAG-NEXT: v_accvgpr_read_b32 v2, a2 -; SDAG-NEXT: v_accvgpr_read_b32 v3, a3 +; SDAG-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[0:3], v[14:17], v[6:13], v4 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_16x16x128_fp8_fp8__sgpr: @@ -2318,53 +2235,37 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_bf8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_32x32x64_bf8_bf8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v36, s0 -; SDAG-NEXT: v_mov_b32_e32 v37, s1 -; SDAG-NEXT: v_mov_b32_e32 v38, s2 -; SDAG-NEXT: v_mov_b32_e32 v39, s3 -; SDAG-NEXT: v_mov_b32_e32 v13, s25 -; SDAG-NEXT: v_mov_b32_e32 v14, s26 -; SDAG-NEXT: v_mov_b32_e32 v15, s27 -; SDAG-NEXT: v_mov_b32_e32 v16, s28 -; SDAG-NEXT: v_mov_b32_e32 v17, s29 -; SDAG-NEXT: v_mov_b32_e32 v28, s16 -; SDAG-NEXT: v_mov_b32_e32 v29, s17 -; SDAG-NEXT: v_mov_b32_e32 v30, s18 -; SDAG-NEXT: v_mov_b32_e32 v31, s19 -; SDAG-NEXT: v_mov_b32_e32 v32, s20 -; SDAG-NEXT: v_mov_b32_e32 v33, s21 -; SDAG-NEXT: v_mov_b32_e32 v34, s22 -; SDAG-NEXT: v_mov_b32_e32 v35, s23 -; SDAG-NEXT: v_mov_b32_e32 v12, s24 -; SDAG-NEXT: v_mov_b32_e32 v18, v0 -; SDAG-NEXT: v_mov_b32_e32 v19, v1 -; SDAG-NEXT: v_mov_b32_e32 v20, v2 -; SDAG-NEXT: v_mov_b32_e32 v21, v3 -; SDAG-NEXT: v_mov_b32_e32 v22, v4 -; SDAG-NEXT: v_mov_b32_e32 v23, v5 -; SDAG-NEXT: v_mov_b32_e32 v24, v6 -; SDAG-NEXT: v_mov_b32_e32 v25, v7 -; SDAG-NEXT: v_mov_b32_e32 v26, v8 -; SDAG-NEXT: v_mov_b32_e32 v27, v9 +; SDAG-NEXT: v_mov_b32_e32 v26, s0 +; SDAG-NEXT: v_mov_b32_e32 v27, s1 +; SDAG-NEXT: v_mov_b32_e32 v28, s2 +; SDAG-NEXT: v_mov_b32_e32 v29, s3 +; SDAG-NEXT: v_mov_b32_e32 v16, v10 +; SDAG-NEXT: v_mov_b32_e32 v15, v9 +; SDAG-NEXT: v_mov_b32_e32 v14, v8 +; SDAG-NEXT: v_mov_b32_e32 v13, v7 +; SDAG-NEXT: v_mov_b32_e32 v12, v6 +; SDAG-NEXT: v_mov_b32_e32 v11, v5 +; SDAG-NEXT: v_mov_b32_e32 v10, v4 +; SDAG-NEXT: v_mov_b32_e32 v9, v3 +; SDAG-NEXT: v_mov_b32_e32 v8, v2 +; SDAG-NEXT: v_mov_b32_e32 v7, v1 +; SDAG-NEXT: v_mov_b32_e32 v6, v0 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 +; SDAG-NEXT: v_mov_b32_e32 v5, s29 +; SDAG-NEXT: v_mov_b32_e32 v18, s16 +; SDAG-NEXT: v_mov_b32_e32 v19, s17 +; SDAG-NEXT: v_mov_b32_e32 v20, s18 +; SDAG-NEXT: v_mov_b32_e32 v21, s19 +; SDAG-NEXT: v_mov_b32_e32 v22, s20 +; SDAG-NEXT: v_mov_b32_e32 v23, s21 +; SDAG-NEXT: v_mov_b32_e32 v24, s22 +; SDAG-NEXT: v_mov_b32_e32 v25, s23 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[12:27], v[36:39], v[28:35], v10 -; SDAG-NEXT: s_nop 11 -; SDAG-NEXT: v_mov_b32_e32 v0, v12 -; SDAG-NEXT: v_mov_b32_e32 v1, v13 -; SDAG-NEXT: v_mov_b32_e32 v2, v14 -; SDAG-NEXT: v_mov_b32_e32 v3, v15 -; SDAG-NEXT: v_mov_b32_e32 v4, v16 -; SDAG-NEXT: v_mov_b32_e32 v5, v17 -; SDAG-NEXT: v_mov_b32_e32 v6, v18 -; SDAG-NEXT: v_mov_b32_e32 v7, v19 -; SDAG-NEXT: v_mov_b32_e32 v8, v20 -; SDAG-NEXT: v_mov_b32_e32 v9, v21 -; SDAG-NEXT: v_mov_b32_e32 v10, v22 -; SDAG-NEXT: v_mov_b32_e32 v11, v23 -; SDAG-NEXT: v_mov_b32_e32 v12, v24 -; SDAG-NEXT: v_mov_b32_e32 v13, v25 -; SDAG-NEXT: v_mov_b32_e32 v14, v26 -; SDAG-NEXT: v_mov_b32_e32 v15, v27 +; SDAG-NEXT: v_smfmac_f32_32x32x64_bf8_bf8 v[0:15], v[26:29], v[18:25], v16 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_bf8__sgpr: @@ -2685,53 +2586,37 @@ define <16 x float> @test_smfmac_f32_32x32x64_bf8_fp8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_32x32x64_bf8_fp8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v36, s0 -; SDAG-NEXT: v_mov_b32_e32 v37, s1 -; SDAG-NEXT: v_mov_b32_e32 v38, s2 -; SDAG-NEXT: v_mov_b32_e32 v39, s3 -; SDAG-NEXT: v_mov_b32_e32 v13, s25 -; SDAG-NEXT: v_mov_b32_e32 v14, s26 -; SDAG-NEXT: v_mov_b32_e32 v15, s27 -; SDAG-NEXT: v_mov_b32_e32 v16, s28 -; SDAG-NEXT: v_mov_b32_e32 v17, s29 -; SDAG-NEXT: v_mov_b32_e32 v28, s16 -; SDAG-NEXT: v_mov_b32_e32 v29, s17 -; SDAG-NEXT: v_mov_b32_e32 v30, s18 -; SDAG-NEXT: v_mov_b32_e32 v31, s19 -; SDAG-NEXT: v_mov_b32_e32 v32, s20 -; SDAG-NEXT: v_mov_b32_e32 v33, s21 -; SDAG-NEXT: v_mov_b32_e32 v34, s22 -; SDAG-NEXT: v_mov_b32_e32 v35, s23 -; SDAG-NEXT: v_mov_b32_e32 v12, s24 -; SDAG-NEXT: v_mov_b32_e32 v18, v0 -; SDAG-NEXT: v_mov_b32_e32 v19, v1 -; SDAG-NEXT: v_mov_b32_e32 v20, v2 -; SDAG-NEXT: v_mov_b32_e32 v21, v3 -; SDAG-NEXT: v_mov_b32_e32 v22, v4 -; SDAG-NEXT: v_mov_b32_e32 v23, v5 -; SDAG-NEXT: v_mov_b32_e32 v24, v6 -; SDAG-NEXT: v_mov_b32_e32 v25, v7 -; SDAG-NEXT: v_mov_b32_e32 v26, v8 -; SDAG-NEXT: v_mov_b32_e32 v27, v9 +; SDAG-NEXT: v_mov_b32_e32 v26, s0 +; SDAG-NEXT: v_mov_b32_e32 v27, s1 +; SDAG-NEXT: v_mov_b32_e32 v28, s2 +; SDAG-NEXT: v_mov_b32_e32 v29, s3 +; SDAG-NEXT: v_mov_b32_e32 v16, v10 +; SDAG-NEXT: v_mov_b32_e32 v15, v9 +; SDAG-NEXT: v_mov_b32_e32 v14, v8 +; SDAG-NEXT: v_mov_b32_e32 v13, v7 +; SDAG-NEXT: v_mov_b32_e32 v12, v6 +; SDAG-NEXT: v_mov_b32_e32 v11, v5 +; SDAG-NEXT: v_mov_b32_e32 v10, v4 +; SDAG-NEXT: v_mov_b32_e32 v9, v3 +; SDAG-NEXT: v_mov_b32_e32 v8, v2 +; SDAG-NEXT: v_mov_b32_e32 v7, v1 +; SDAG-NEXT: v_mov_b32_e32 v6, v0 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 +; SDAG-NEXT: v_mov_b32_e32 v5, s29 +; SDAG-NEXT: v_mov_b32_e32 v18, s16 +; SDAG-NEXT: v_mov_b32_e32 v19, s17 +; SDAG-NEXT: v_mov_b32_e32 v20, s18 +; SDAG-NEXT: v_mov_b32_e32 v21, s19 +; SDAG-NEXT: v_mov_b32_e32 v22, s20 +; SDAG-NEXT: v_mov_b32_e32 v23, s21 +; SDAG-NEXT: v_mov_b32_e32 v24, s22 +; SDAG-NEXT: v_mov_b32_e32 v25, s23 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[12:27], v[36:39], v[28:35], v10 -; SDAG-NEXT: s_nop 11 -; SDAG-NEXT: v_mov_b32_e32 v0, v12 -; SDAG-NEXT: v_mov_b32_e32 v1, v13 -; SDAG-NEXT: v_mov_b32_e32 v2, v14 -; SDAG-NEXT: v_mov_b32_e32 v3, v15 -; SDAG-NEXT: v_mov_b32_e32 v4, v16 -; SDAG-NEXT: v_mov_b32_e32 v5, v17 -; SDAG-NEXT: v_mov_b32_e32 v6, v18 -; SDAG-NEXT: v_mov_b32_e32 v7, v19 -; SDAG-NEXT: v_mov_b32_e32 v8, v20 -; SDAG-NEXT: v_mov_b32_e32 v9, v21 -; SDAG-NEXT: v_mov_b32_e32 v10, v22 -; SDAG-NEXT: v_mov_b32_e32 v11, v23 -; SDAG-NEXT: v_mov_b32_e32 v12, v24 -; SDAG-NEXT: v_mov_b32_e32 v13, v25 -; SDAG-NEXT: v_mov_b32_e32 v14, v26 -; SDAG-NEXT: v_mov_b32_e32 v15, v27 +; SDAG-NEXT: v_smfmac_f32_32x32x64_bf8_fp8 v[0:15], v[26:29], v[18:25], v16 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_32x32x64_bf8_fp8__sgpr: @@ -3052,53 +2937,37 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_bf8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_32x32x64_fp8_bf8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v36, s0 -; SDAG-NEXT: v_mov_b32_e32 v37, s1 -; SDAG-NEXT: v_mov_b32_e32 v38, s2 -; SDAG-NEXT: v_mov_b32_e32 v39, s3 -; SDAG-NEXT: v_mov_b32_e32 v13, s25 -; SDAG-NEXT: v_mov_b32_e32 v14, s26 -; SDAG-NEXT: v_mov_b32_e32 v15, s27 -; SDAG-NEXT: v_mov_b32_e32 v16, s28 -; SDAG-NEXT: v_mov_b32_e32 v17, s29 -; SDAG-NEXT: v_mov_b32_e32 v28, s16 -; SDAG-NEXT: v_mov_b32_e32 v29, s17 -; SDAG-NEXT: v_mov_b32_e32 v30, s18 -; SDAG-NEXT: v_mov_b32_e32 v31, s19 -; SDAG-NEXT: v_mov_b32_e32 v32, s20 -; SDAG-NEXT: v_mov_b32_e32 v33, s21 -; SDAG-NEXT: v_mov_b32_e32 v34, s22 -; SDAG-NEXT: v_mov_b32_e32 v35, s23 -; SDAG-NEXT: v_mov_b32_e32 v12, s24 -; SDAG-NEXT: v_mov_b32_e32 v18, v0 -; SDAG-NEXT: v_mov_b32_e32 v19, v1 -; SDAG-NEXT: v_mov_b32_e32 v20, v2 -; SDAG-NEXT: v_mov_b32_e32 v21, v3 -; SDAG-NEXT: v_mov_b32_e32 v22, v4 -; SDAG-NEXT: v_mov_b32_e32 v23, v5 -; SDAG-NEXT: v_mov_b32_e32 v24, v6 -; SDAG-NEXT: v_mov_b32_e32 v25, v7 -; SDAG-NEXT: v_mov_b32_e32 v26, v8 -; SDAG-NEXT: v_mov_b32_e32 v27, v9 +; SDAG-NEXT: v_mov_b32_e32 v26, s0 +; SDAG-NEXT: v_mov_b32_e32 v27, s1 +; SDAG-NEXT: v_mov_b32_e32 v28, s2 +; SDAG-NEXT: v_mov_b32_e32 v29, s3 +; SDAG-NEXT: v_mov_b32_e32 v16, v10 +; SDAG-NEXT: v_mov_b32_e32 v15, v9 +; SDAG-NEXT: v_mov_b32_e32 v14, v8 +; SDAG-NEXT: v_mov_b32_e32 v13, v7 +; SDAG-NEXT: v_mov_b32_e32 v12, v6 +; SDAG-NEXT: v_mov_b32_e32 v11, v5 +; SDAG-NEXT: v_mov_b32_e32 v10, v4 +; SDAG-NEXT: v_mov_b32_e32 v9, v3 +; SDAG-NEXT: v_mov_b32_e32 v8, v2 +; SDAG-NEXT: v_mov_b32_e32 v7, v1 +; SDAG-NEXT: v_mov_b32_e32 v6, v0 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 +; SDAG-NEXT: v_mov_b32_e32 v5, s29 +; SDAG-NEXT: v_mov_b32_e32 v18, s16 +; SDAG-NEXT: v_mov_b32_e32 v19, s17 +; SDAG-NEXT: v_mov_b32_e32 v20, s18 +; SDAG-NEXT: v_mov_b32_e32 v21, s19 +; SDAG-NEXT: v_mov_b32_e32 v22, s20 +; SDAG-NEXT: v_mov_b32_e32 v23, s21 +; SDAG-NEXT: v_mov_b32_e32 v24, s22 +; SDAG-NEXT: v_mov_b32_e32 v25, s23 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[12:27], v[36:39], v[28:35], v10 -; SDAG-NEXT: s_nop 11 -; SDAG-NEXT: v_mov_b32_e32 v0, v12 -; SDAG-NEXT: v_mov_b32_e32 v1, v13 -; SDAG-NEXT: v_mov_b32_e32 v2, v14 -; SDAG-NEXT: v_mov_b32_e32 v3, v15 -; SDAG-NEXT: v_mov_b32_e32 v4, v16 -; SDAG-NEXT: v_mov_b32_e32 v5, v17 -; SDAG-NEXT: v_mov_b32_e32 v6, v18 -; SDAG-NEXT: v_mov_b32_e32 v7, v19 -; SDAG-NEXT: v_mov_b32_e32 v8, v20 -; SDAG-NEXT: v_mov_b32_e32 v9, v21 -; SDAG-NEXT: v_mov_b32_e32 v10, v22 -; SDAG-NEXT: v_mov_b32_e32 v11, v23 -; SDAG-NEXT: v_mov_b32_e32 v12, v24 -; SDAG-NEXT: v_mov_b32_e32 v13, v25 -; SDAG-NEXT: v_mov_b32_e32 v14, v26 -; SDAG-NEXT: v_mov_b32_e32 v15, v27 +; SDAG-NEXT: v_smfmac_f32_32x32x64_fp8_bf8 v[0:15], v[26:29], v[18:25], v16 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_bf8__sgpr: @@ -3419,53 +3288,37 @@ define <16 x float> @test_smfmac_f32_32x32x64_fp8_fp8__sgpr(<4 x i32> inreg %arg ; SDAG-LABEL: test_smfmac_f32_32x32x64_fp8_fp8__sgpr: ; SDAG: ; %bb.0: ; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SDAG-NEXT: v_mov_b32_e32 v36, s0 -; SDAG-NEXT: v_mov_b32_e32 v37, s1 -; SDAG-NEXT: v_mov_b32_e32 v38, s2 -; SDAG-NEXT: v_mov_b32_e32 v39, s3 -; SDAG-NEXT: v_mov_b32_e32 v13, s25 -; SDAG-NEXT: v_mov_b32_e32 v14, s26 -; SDAG-NEXT: v_mov_b32_e32 v15, s27 -; SDAG-NEXT: v_mov_b32_e32 v16, s28 -; SDAG-NEXT: v_mov_b32_e32 v17, s29 -; SDAG-NEXT: v_mov_b32_e32 v28, s16 -; SDAG-NEXT: v_mov_b32_e32 v29, s17 -; SDAG-NEXT: v_mov_b32_e32 v30, s18 -; SDAG-NEXT: v_mov_b32_e32 v31, s19 -; SDAG-NEXT: v_mov_b32_e32 v32, s20 -; SDAG-NEXT: v_mov_b32_e32 v33, s21 -; SDAG-NEXT: v_mov_b32_e32 v34, s22 -; SDAG-NEXT: v_mov_b32_e32 v35, s23 -; SDAG-NEXT: v_mov_b32_e32 v12, s24 -; SDAG-NEXT: v_mov_b32_e32 v18, v0 -; SDAG-NEXT: v_mov_b32_e32 v19, v1 -; SDAG-NEXT: v_mov_b32_e32 v20, v2 -; SDAG-NEXT: v_mov_b32_e32 v21, v3 -; SDAG-NEXT: v_mov_b32_e32 v22, v4 -; SDAG-NEXT: v_mov_b32_e32 v23, v5 -; SDAG-NEXT: v_mov_b32_e32 v24, v6 -; SDAG-NEXT: v_mov_b32_e32 v25, v7 -; SDAG-NEXT: v_mov_b32_e32 v26, v8 -; SDAG-NEXT: v_mov_b32_e32 v27, v9 +; SDAG-NEXT: v_mov_b32_e32 v26, s0 +; SDAG-NEXT: v_mov_b32_e32 v27, s1 +; SDAG-NEXT: v_mov_b32_e32 v28, s2 +; SDAG-NEXT: v_mov_b32_e32 v29, s3 +; SDAG-NEXT: v_mov_b32_e32 v16, v10 +; SDAG-NEXT: v_mov_b32_e32 v15, v9 +; SDAG-NEXT: v_mov_b32_e32 v14, v8 +; SDAG-NEXT: v_mov_b32_e32 v13, v7 +; SDAG-NEXT: v_mov_b32_e32 v12, v6 +; SDAG-NEXT: v_mov_b32_e32 v11, v5 +; SDAG-NEXT: v_mov_b32_e32 v10, v4 +; SDAG-NEXT: v_mov_b32_e32 v9, v3 +; SDAG-NEXT: v_mov_b32_e32 v8, v2 +; SDAG-NEXT: v_mov_b32_e32 v7, v1 +; SDAG-NEXT: v_mov_b32_e32 v6, v0 +; SDAG-NEXT: v_mov_b32_e32 v0, s24 +; SDAG-NEXT: v_mov_b32_e32 v1, s25 +; SDAG-NEXT: v_mov_b32_e32 v2, s26 +; SDAG-NEXT: v_mov_b32_e32 v3, s27 +; SDAG-NEXT: v_mov_b32_e32 v4, s28 +; SDAG-NEXT: v_mov_b32_e32 v5, s29 +; SDAG-NEXT: v_mov_b32_e32 v18, s16 +; SDAG-NEXT: v_mov_b32_e32 v19, s17 +; SDAG-NEXT: v_mov_b32_e32 v20, s18 +; SDAG-NEXT: v_mov_b32_e32 v21, s19 +; SDAG-NEXT: v_mov_b32_e32 v22, s20 +; SDAG-NEXT: v_mov_b32_e32 v23, s21 +; SDAG-NEXT: v_mov_b32_e32 v24, s22 +; SDAG-NEXT: v_mov_b32_e32 v25, s23 ; SDAG-NEXT: s_nop 1 -; SDAG-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[12:27], v[36:39], v[28:35], v10 -; SDAG-NEXT: s_nop 11 -; SDAG-NEXT: v_mov_b32_e32 v0, v12 -; SDAG-NEXT: v_mov_b32_e32 v1, v13 -; SDAG-NEXT: v_mov_b32_e32 v2, v14 -; SDAG-NEXT: v_mov_b32_e32 v3, v15 -; SDAG-NEXT: v_mov_b32_e32 v4, v16 -; SDAG-NEXT: v_mov_b32_e32 v5, v17 -; SDAG-NEXT: v_mov_b32_e32 v6, v18 -; SDAG-NEXT: v_mov_b32_e32 v7, v19 -; SDAG-NEXT: v_mov_b32_e32 v8, v20 -; SDAG-NEXT: v_mov_b32_e32 v9, v21 -; SDAG-NEXT: v_mov_b32_e32 v10, v22 -; SDAG-NEXT: v_mov_b32_e32 v11, v23 -; SDAG-NEXT: v_mov_b32_e32 v12, v24 -; SDAG-NEXT: v_mov_b32_e32 v13, v25 -; SDAG-NEXT: v_mov_b32_e32 v14, v26 -; SDAG-NEXT: v_mov_b32_e32 v15, v27 +; SDAG-NEXT: v_smfmac_f32_32x32x64_fp8_fp8 v[0:15], v[26:29], v[18:25], v16 ; SDAG-NEXT: s_setpc_b64 s[30:31] ; ; GISEL-LABEL: test_smfmac_f32_32x32x64_fp8_fp8__sgpr: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll index dd89f80..ba769ef 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll @@ -100,7 +100,7 @@ define amdgpu_kernel void @tanh_f16(ptr addrspace(1) %out, half %src) #1 { ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 ; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, s2 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_f16: @@ -123,7 +123,7 @@ define amdgpu_kernel void @tanh_f16_constant_4.0(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, 4.0 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_f16_constant_4.0: @@ -146,7 +146,7 @@ define amdgpu_kernel void @tanh_f16_constant_100.0(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, 0x5640 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_f16_constant_100.0: @@ -182,7 +182,7 @@ define amdgpu_kernel void @tanh_bf16(ptr addrspace(1) %out, bfloat %src) #1 { ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 ; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, s2 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_bf16: @@ -205,7 +205,7 @@ define amdgpu_kernel void @tanh_bf16_constant_4(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, 4.0 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_bf16_constant_4: @@ -228,7 +228,7 @@ define amdgpu_kernel void @tanh_bf16_constant_100(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, 0x42c8 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_bf16_constant_100: diff --git a/llvm/test/CodeGen/AMDGPU/mfma-no-register-aliasing.ll b/llvm/test/CodeGen/AMDGPU/mfma-no-register-aliasing.ll index 51cd564..f46116e 100644 --- a/llvm/test/CodeGen/AMDGPU/mfma-no-register-aliasing.ll +++ b/llvm/test/CodeGen/AMDGPU/mfma-no-register-aliasing.ll @@ -95,66 +95,66 @@ define amdgpu_kernel void @test_mfma_f32_32x32x1f32(ptr addrspace(1) %arg) #0 { ; GREEDY908-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v3, v0, a[0:31] ; GREEDY908-NEXT: s_nop 15 ; GREEDY908-NEXT: s_nop 1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a32 -; GREEDY908-NEXT: v_accvgpr_read_b32 v5, a61 -; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a60 -; GREEDY908-NEXT: v_accvgpr_write_b32 a2, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a33 -; GREEDY908-NEXT: v_accvgpr_read_b32 v7, a59 -; GREEDY908-NEXT: v_accvgpr_read_b32 v8, a58 -; GREEDY908-NEXT: v_accvgpr_write_b32 a3, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a32 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a33 ; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a34 -; GREEDY908-NEXT: v_accvgpr_read_b32 v9, a57 -; GREEDY908-NEXT: v_accvgpr_read_b32 v10, a56 +; GREEDY908-NEXT: v_accvgpr_write_b32 a2, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a3, v6 ; GREEDY908-NEXT: v_accvgpr_write_b32 a4, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a35 -; GREEDY908-NEXT: v_accvgpr_read_b32 v11, a55 -; GREEDY908-NEXT: v_accvgpr_read_b32 v12, a54 -; GREEDY908-NEXT: v_accvgpr_write_b32 a5, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a36 -; GREEDY908-NEXT: v_accvgpr_read_b32 v13, a53 -; GREEDY908-NEXT: v_accvgpr_read_b32 v14, a52 -; GREEDY908-NEXT: v_accvgpr_write_b32 a6, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a35 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a36 ; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a37 -; GREEDY908-NEXT: v_accvgpr_read_b32 v15, a51 -; GREEDY908-NEXT: v_accvgpr_read_b32 v16, a50 +; GREEDY908-NEXT: v_accvgpr_write_b32 a5, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a6, v6 ; GREEDY908-NEXT: v_accvgpr_write_b32 a7, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a38 -; GREEDY908-NEXT: v_accvgpr_read_b32 v17, a49 -; GREEDY908-NEXT: v_accvgpr_read_b32 v18, a48 -; GREEDY908-NEXT: v_accvgpr_write_b32 a8, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a39 -; GREEDY908-NEXT: v_accvgpr_read_b32 v19, a47 -; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a46 -; GREEDY908-NEXT: v_accvgpr_write_b32 a9, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a38 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a39 ; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a40 -; GREEDY908-NEXT: v_accvgpr_write_b32 a16, v2 -; GREEDY908-NEXT: v_accvgpr_write_b32 a17, v19 +; GREEDY908-NEXT: v_accvgpr_write_b32 a8, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a9, v6 ; GREEDY908-NEXT: v_accvgpr_write_b32 a10, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a41 -; GREEDY908-NEXT: v_accvgpr_write_b32 a18, v18 -; GREEDY908-NEXT: v_accvgpr_write_b32 a19, v17 -; GREEDY908-NEXT: v_accvgpr_write_b32 a11, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a42 -; GREEDY908-NEXT: v_accvgpr_write_b32 a20, v16 -; GREEDY908-NEXT: v_accvgpr_write_b32 a21, v15 -; GREEDY908-NEXT: v_accvgpr_write_b32 a12, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a41 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a42 ; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a43 -; GREEDY908-NEXT: v_accvgpr_write_b32 a22, v14 -; GREEDY908-NEXT: v_accvgpr_write_b32 a23, v13 +; GREEDY908-NEXT: v_accvgpr_write_b32 a11, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a12, v6 ; GREEDY908-NEXT: v_accvgpr_write_b32 a13, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a44 -; GREEDY908-NEXT: v_accvgpr_write_b32 a24, v12 -; GREEDY908-NEXT: v_accvgpr_write_b32 a25, v11 -; GREEDY908-NEXT: v_accvgpr_write_b32 a14, v1 -; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a45 -; GREEDY908-NEXT: v_accvgpr_write_b32 a26, v10 -; GREEDY908-NEXT: v_accvgpr_write_b32 a27, v9 -; GREEDY908-NEXT: v_accvgpr_write_b32 a15, v1 -; GREEDY908-NEXT: v_accvgpr_write_b32 a28, v8 -; GREEDY908-NEXT: v_accvgpr_write_b32 a29, v7 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a44 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a45 +; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a46 +; GREEDY908-NEXT: v_accvgpr_write_b32 a14, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a15, v6 +; GREEDY908-NEXT: v_accvgpr_write_b32 a16, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a47 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a48 +; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a49 +; GREEDY908-NEXT: v_accvgpr_write_b32 a17, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a18, v6 +; GREEDY908-NEXT: v_accvgpr_write_b32 a19, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a50 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a51 +; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a52 +; GREEDY908-NEXT: v_accvgpr_write_b32 a20, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a21, v6 +; GREEDY908-NEXT: v_accvgpr_write_b32 a22, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a53 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a54 +; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a55 +; GREEDY908-NEXT: v_accvgpr_write_b32 a23, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a24, v6 +; GREEDY908-NEXT: v_accvgpr_write_b32 a25, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a56 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a57 +; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a58 +; GREEDY908-NEXT: v_accvgpr_write_b32 a26, v2 +; GREEDY908-NEXT: v_accvgpr_write_b32 a27, v6 +; GREEDY908-NEXT: v_accvgpr_write_b32 a28, v1 +; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a59 +; GREEDY908-NEXT: v_accvgpr_read_b32 v6, a60 +; GREEDY908-NEXT: v_accvgpr_read_b32 v1, a61 +; GREEDY908-NEXT: v_accvgpr_write_b32 a29, v2 ; GREEDY908-NEXT: v_accvgpr_write_b32 a30, v6 -; GREEDY908-NEXT: v_accvgpr_write_b32 a31, v5 +; GREEDY908-NEXT: v_accvgpr_write_b32 a31, v1 ; GREEDY908-NEXT: s_nop 0 ; GREEDY908-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v3, v0, a[0:31] ; GREEDY908-NEXT: s_nop 15 @@ -667,11 +667,11 @@ define amdgpu_kernel void @test_mfma_f32_16x16x1f32(ptr addrspace(1) %arg) #0 { ; GREEDY908-NEXT: v_mfma_f32_16x16x1f32 a[18:33], v0, v1, a[18:33] ; GREEDY908-NEXT: v_mfma_f32_16x16x1f32 a[2:17], v0, v1, a[18:33] ; GREEDY908-NEXT: s_nop 8 +; GREEDY908-NEXT: v_accvgpr_read_b32 v5, a18 ; GREEDY908-NEXT: v_accvgpr_read_b32 v2, a19 -; GREEDY908-NEXT: v_accvgpr_read_b32 v3, a18 ; GREEDY908-NEXT: s_nop 0 +; GREEDY908-NEXT: v_accvgpr_write_b32 a0, v5 ; GREEDY908-NEXT: v_accvgpr_write_b32 a1, v2 -; GREEDY908-NEXT: v_accvgpr_write_b32 a0, v3 ; GREEDY908-NEXT: s_nop 0 ; GREEDY908-NEXT: v_mfma_f32_16x16x1f32 a[0:15], v0, v1, a[0:15] ; GREEDY908-NEXT: s_nop 9 diff --git a/llvm/test/CodeGen/AMDGPU/min.ll b/llvm/test/CodeGen/AMDGPU/min.ll index 6a3d31f..0458a64 100644 --- a/llvm/test/CodeGen/AMDGPU/min.ll +++ b/llvm/test/CodeGen/AMDGPU/min.ll @@ -6,9 +6,7 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck --check-prefix=GFX10 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX11,GFX11-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX11,GFX11-FAKE16 %s -; TODO: FIXME-TRUE16 - Enable this llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-TRUE16 %s -; Crashing on v_test_imin_slt_i16 -; LLVM ERROR: Cannot select: 0x5f895f65b050: i16,ch = load<(load (s16) from %ir.b.gep, addrspace 1)> +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-FAKE16 %s define amdgpu_kernel void @v_test_imin_sle_i32(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 { @@ -1482,20 +1480,35 @@ define amdgpu_kernel void @v_test_imin_slt_i16(ptr addrspace(1) %out, ptr addrsp ; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] ; GFX11-FAKE16-NEXT: s_endpgm ; -; GFX1250-LABEL: v_test_imin_slt_i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX1250-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 -; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: global_load_u16 v1, v0, s[2:3] scale_offset -; GFX1250-NEXT: global_load_u16 v2, v0, s[6:7] scale_offset -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_min_i16 v1, v1, v2 -; GFX1250-NEXT: global_store_b16 v0, v1, s[0:1] scale_offset -; GFX1250-NEXT: s_endpgm +; GFX1250-TRUE16-LABEL: v_test_imin_slt_i16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-TRUE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-TRUE16-NEXT: v_and_b32_e32 v1, 0x3ff, v0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3] scale_offset +; GFX1250-TRUE16-NEXT: global_load_u16 v2, v1, s[6:7] scale_offset +; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-TRUE16-NEXT: v_min_i16 v0.l, v0.l, v2.l +; GFX1250-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] scale_offset +; GFX1250-TRUE16-NEXT: s_endpgm +; +; GFX1250-FAKE16-LABEL: v_test_imin_slt_i16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-FAKE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3] scale_offset +; GFX1250-FAKE16-NEXT: global_load_u16 v2, v0, s[6:7] scale_offset +; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-FAKE16-NEXT: v_min_i16 v1, v1, v2 +; GFX1250-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] scale_offset +; GFX1250-FAKE16-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %a.gep = getelementptr inbounds i16, ptr addrspace(1) %aptr, i32 %tid %b.gep = getelementptr inbounds i16, ptr addrspace(1) %bptr, i32 %tid @@ -2769,20 +2782,35 @@ define amdgpu_kernel void @v_test_umin_ult_i8(ptr addrspace(1) %out, ptr addrspa ; GFX11-FAKE16-NEXT: global_store_b8 v0, v1, s[0:1] ; GFX11-FAKE16-NEXT: s_endpgm ; -; GFX1250-LABEL: v_test_umin_ult_i8: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX1250-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 -; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: global_load_u8 v1, v0, s[2:3] -; GFX1250-NEXT: global_load_u8 v2, v0, s[6:7] -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_min_u16 v1, v1, v2 -; GFX1250-NEXT: global_store_b8 v0, v1, s[0:1] -; GFX1250-NEXT: s_endpgm +; GFX1250-TRUE16-LABEL: v_test_umin_ult_i8: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-TRUE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-TRUE16-NEXT: v_and_b32_e32 v1, 0x3ff, v0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: global_load_u8 v0, v1, s[2:3] +; GFX1250-TRUE16-NEXT: global_load_u8 v2, v1, s[6:7] +; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-TRUE16-NEXT: v_min_u16 v0.l, v0.l, v2.l +; GFX1250-TRUE16-NEXT: global_store_b8 v1, v0, s[0:1] +; GFX1250-TRUE16-NEXT: s_endpgm +; +; GFX1250-FAKE16-LABEL: v_test_umin_ult_i8: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-FAKE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: global_load_u8 v1, v0, s[2:3] +; GFX1250-FAKE16-NEXT: global_load_u8 v2, v0, s[6:7] +; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-FAKE16-NEXT: v_min_u16 v1, v1, v2 +; GFX1250-FAKE16-NEXT: global_store_b8 v0, v1, s[0:1] +; GFX1250-FAKE16-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %a.gep = getelementptr inbounds i8, ptr addrspace(1) %a.ptr, i32 %tid %b.gep = getelementptr inbounds i8, ptr addrspace(1) %b.ptr, i32 %tid @@ -5069,5 +5097,3 @@ declare i32 @llvm.amdgcn.workitem.id.x() #1 attributes #0 = { nounwind } attributes #1 = { nounwind readnone } -;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: -; GFX1250-FAKE16: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/minmax.ll b/llvm/test/CodeGen/AMDGPU/minmax.ll index 57e6943..56f9c5d 100644 --- a/llvm/test/CodeGen/AMDGPU/minmax.ll +++ b/llvm/test/CodeGen/AMDGPU/minmax.ll @@ -638,6 +638,14 @@ define void @test_med3_minimumnum_maximumnum_f32(ptr addrspace(1) %arg, float %x ; GFX12-NEXT: v_med3_num_f32 v2, v2, v3, v4 ; GFX12-NEXT: global_store_b32 v[0:1], v2, off ; GFX12-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: test_med3_minimumnum_maximumnum_f32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_med3_num_f32 v2, v2, v3, v4 +; GFX1250-NEXT: global_store_b32 v[0:1], v2, off +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %tmp0 = call float @llvm.minimumnum.f32(float %x, float %y) %tmp1 = call float @llvm.maximumnum.f32(float %x, float %y) %tmp2 = call float @llvm.minimumnum.f32(float %tmp1, float %z) @@ -798,7 +806,7 @@ define amdgpu_ps void @s_test_minmax_f16_ieee_false(half inreg %a, half inreg %b ; SDAG-GFX1250-TRUE16-NEXT: s_mov_b32 s5, s4 ; SDAG-GFX1250-TRUE16-NEXT: s_mov_b32 s4, s3 ; SDAG-GFX1250-TRUE16-NEXT: v_maxmin_num_f16 v0.l, s0, s1, v0.l -; SDAG-GFX1250-TRUE16-NEXT: flat_store_b16 v1, v0, s[4:5] +; SDAG-GFX1250-TRUE16-NEXT: global_store_b16 v1, v0, s[4:5] ; SDAG-GFX1250-TRUE16-NEXT: s_endpgm ; ; SDAG-GFX1250-FAKE16-LABEL: s_test_minmax_f16_ieee_false: @@ -813,12 +821,12 @@ define amdgpu_ps void @s_test_minmax_f16_ieee_false(half inreg %a, half inreg %b ; GISEL-GFX1250-TRUE16-LABEL: s_test_minmax_f16_ieee_false: ; GISEL-GFX1250-TRUE16: ; %bb.0: ; GISEL-GFX1250-TRUE16-NEXT: s_max_num_f16 s0, s0, s1 +; GISEL-GFX1250-TRUE16-NEXT: v_mov_b32_e32 v1, 0 ; GISEL-GFX1250-TRUE16-NEXT: s_mov_b32 s6, s3 ; GISEL-GFX1250-TRUE16-NEXT: s_mov_b32 s7, s4 -; GISEL-GFX1250-TRUE16-NEXT: v_mov_b32_e32 v1, 0 ; GISEL-GFX1250-TRUE16-NEXT: s_min_num_f16 s0, s0, s2 -; GISEL-GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, s0 -; GISEL-GFX1250-TRUE16-NEXT: flat_store_b16 v1, v0, s[6:7] +; GISEL-GFX1250-TRUE16-NEXT: v_mov_b16_e32 v0.l, s0 +; GISEL-GFX1250-TRUE16-NEXT: global_store_b16 v1, v0, s[6:7] ; GISEL-GFX1250-TRUE16-NEXT: s_endpgm ; ; GISEL-GFX1250-FAKE16-LABEL: s_test_minmax_f16_ieee_false: @@ -1246,7 +1254,7 @@ define void @test_med3_f16(ptr addrspace(1) %arg, half %x, half %y, half %z) #0 ; SDAG-GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 ; SDAG-GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 ; SDAG-GFX1250-TRUE16-NEXT: v_med3_num_f16 v2.l, v2.l, v3.l, v4.l -; SDAG-GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v2 +; SDAG-GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v2, off ; SDAG-GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] ; ; SDAG-GFX1250-FAKE16-LABEL: test_med3_f16: @@ -1262,7 +1270,7 @@ define void @test_med3_f16(ptr addrspace(1) %arg, half %x, half %y, half %z) #0 ; GISEL-GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 ; GISEL-GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GISEL-GFX1250-TRUE16-NEXT: v_med3_num_f16 v2.l, v2.l, v3.l, v4.l -; GISEL-GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v2 +; GISEL-GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v2, off ; GISEL-GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] ; ; GISEL-GFX1250-FAKE16-LABEL: test_med3_f16: diff --git a/llvm/test/CodeGen/AMDGPU/no-fold-accvgpr-mov.ll b/llvm/test/CodeGen/AMDGPU/no-fold-accvgpr-mov.ll index cf244f0..be1788c 100644 --- a/llvm/test/CodeGen/AMDGPU/no-fold-accvgpr-mov.ll +++ b/llvm/test/CodeGen/AMDGPU/no-fold-accvgpr-mov.ll @@ -54,19 +54,20 @@ define amdgpu_kernel void @matmul_kernel(i32 %a0, i32 %a1) { ; GFX908-NEXT: s_branch .LBB0_2 ; GFX908-NEXT: .LBB0_1: ; %bb2 ; GFX908-NEXT: ; in Loop: Header=BB0_2 Depth=1 +; GFX908-NEXT: s_nop 6 +; GFX908-NEXT: v_accvgpr_read_b32 v3, a2 ; GFX908-NEXT: s_or_b32 s4, s3, 1 ; GFX908-NEXT: s_ashr_i32 s5, s3, 31 ; GFX908-NEXT: s_mov_b32 s3, s2 ; GFX908-NEXT: v_mov_b32_e32 v1, s2 -; GFX908-NEXT: s_nop 2 -; GFX908-NEXT: v_accvgpr_read_b32 v0, a2 ; GFX908-NEXT: v_mov_b32_e32 v2, s3 +; GFX908-NEXT: v_accvgpr_write_b32 a0, v3 ; GFX908-NEXT: v_accvgpr_read_b32 v4, a1 ; GFX908-NEXT: v_accvgpr_read_b32 v3, a1 -; GFX908-NEXT: v_accvgpr_write_b32 a0, v0 +; GFX908-NEXT: s_and_b32 s3, s5, s4 ; GFX908-NEXT: v_accvgpr_write_b32 a2, v4 ; GFX908-NEXT: v_accvgpr_write_b32 a3, v3 -; GFX908-NEXT: s_and_b32 s3, s5, s4 +; GFX908-NEXT: s_nop 0 ; GFX908-NEXT: v_mfma_f32_16x16x16f16 a[2:5], v[1:2], v[1:2], a[0:3] ; GFX908-NEXT: s_cbranch_execz .LBB0_4 ; GFX908-NEXT: .LBB0_2: ; %bb diff --git a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll index 6b7d704..ede470b 100644 --- a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll +++ b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll @@ -1,13 +1,11 @@ ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 < %s | FileCheck --check-prefixes=CHECK,GFX11 %s ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 < %s | FileCheck --check-prefixes=CHECK,GFX12 %s -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=+dynamic-vgpr < %s | FileCheck --check-prefixes=CHECK,GFX12,DVGPR %s ; CHECK: .amdgpu_pal_metadata ; CHECK-NEXT: --- ; CHECK-NEXT: amdpal.pipelines: ; CHECK-NEXT: - .api: Vulkan ; CHECK-NEXT: .compute_registers: -; DVGPR-NEXT: .dynamic_vgpr_en: true ; CHECK-NEXT: .tg_size_en: true ; CHECK-NEXT: .tgid_x_en: false ; CHECK-NEXT: .tgid_y_en: false diff --git a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0.ll b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0.ll index 5c0c366..5325499 100644 --- a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0.ll +++ b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0.ll @@ -1,17 +1,14 @@ -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 <%s | FileCheck %s --check-prefixes=CHECK,GFX11,NODVGPR -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 <%s | FileCheck %s --check-prefixes=CHECK,NODVGPR -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=+dynamic-vgpr <%s | FileCheck %s --check-prefixes=CHECK,DVGPR +; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 <%s | FileCheck %s --check-prefixes=CHECK,GFX11 +; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 <%s | FileCheck %s --check-prefixes=CHECK ; CHECK-LABEL: {{^}}_amdgpu_cs_main: -; NODVGPR: ; TotalNumSgprs: 4 -; DVGPR: ; TotalNumSgprs: 34 +; CHECK: ; TotalNumSgprs: 4 ; CHECK: ; NumVgprs: 2 ; CHECK: .amdgpu_pal_metadata ; CHECK-NEXT: --- ; CHECK-NEXT: amdpal.pipelines: ; CHECK-NEXT: - .api: Vulkan ; CHECK-NEXT: .compute_registers: -; DVGPR-NEXT: .dynamic_vgpr_en: true ; CHECK-NEXT: .tg_size_en: true ; CHECK-NEXT: .tgid_x_en: false ; CHECK-NEXT: .tgid_y_en: false @@ -57,7 +54,6 @@ ; CHECK-NEXT: .cs: ; CHECK-NEXT: .checksum_value: 0x9444d7d0 ; CHECK-NEXT: .debug_mode: false -; DVGPR-NEXT: .dynamic_vgpr_saved_count: 0x70 ; CHECK-NEXT: .entry_point: _amdgpu_cs_main ; CHECK-NEXT: .entry_point_symbol: _amdgpu_cs_main ; CHECK-NEXT: .excp_en: 0 @@ -69,8 +65,7 @@ ; CHECK-NEXT: .mem_ordered: true ; CHECK-NEXT: .scratch_en: false ; CHECK-NEXT: .scratch_memory_size: 0 -; NODVGPR-NEXT: .sgpr_count: 0x4 -; DVGPR-NEXT: .sgpr_count: 0x22 +; CHECK-NEXT: .sgpr_count: 0x4 ; CHECK-NEXT: .sgpr_limit: 0x6a ; CHECK-NEXT: .threadgroup_dimensions: ; CHECK-NEXT: - 0x1 diff --git a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6-dvgpr.ll b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6-dvgpr.ll new file mode 100644 index 0000000..e598b0c --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6-dvgpr.ll @@ -0,0 +1,204 @@ +; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 <%s | FileCheck %s --check-prefixes=CHECK + +; CHECK-LABEL: {{^}}_amdgpu_cs_main: +; CHECK: ; TotalNumSgprs: 34 +; CHECK: ; NumVgprs: 2 +; CHECK: .amdgpu_pal_metadata +; CHECK-NEXT: --- +; CHECK-NEXT: amdpal.pipelines: +; CHECK-NEXT: - .api: Vulkan +; CHECK-NEXT: .compute_registers: +; CHECK-NEXT: .dynamic_vgpr_en: true +; CHECK-NEXT: .tg_size_en: true +; CHECK-NEXT: .tgid_x_en: false +; CHECK-NEXT: .tgid_y_en: false +; CHECK-NEXT: .tgid_z_en: false +; CHECK-NEXT: .tidig_comp_cnt: 0x1 +; CHECK-NEXT: .graphics_registers: +; CHECK-NEXT: .ps_extra_lds_size: 0 +; CHECK-NEXT: .spi_ps_input_addr: +; CHECK-NEXT: .ancillary_ena: false +; CHECK-NEXT: .front_face_ena: true +; CHECK-NEXT: .line_stipple_tex_ena: false +; CHECK-NEXT: .linear_center_ena: true +; CHECK-NEXT: .linear_centroid_ena: true +; CHECK-NEXT: .linear_sample_ena: true +; CHECK-NEXT: .persp_center_ena: true +; CHECK-NEXT: .persp_centroid_ena: true +; CHECK-NEXT: .persp_pull_model_ena: false +; CHECK-NEXT: .persp_sample_ena: true +; CHECK-NEXT: .pos_fixed_pt_ena: true +; CHECK-NEXT: .pos_w_float_ena: false +; CHECK-NEXT: .pos_x_float_ena: false +; CHECK-NEXT: .pos_y_float_ena: false +; CHECK-NEXT: .pos_z_float_ena: false +; CHECK-NEXT: .sample_coverage_ena: false +; CHECK-NEXT: .spi_ps_input_ena: +; CHECK-NEXT: .ancillary_ena: false +; CHECK-NEXT: .front_face_ena: false +; CHECK-NEXT: .line_stipple_tex_ena: false +; CHECK-NEXT: .linear_center_ena: false +; CHECK-NEXT: .linear_centroid_ena: false +; CHECK-NEXT: .linear_sample_ena: false +; CHECK-NEXT: .persp_center_ena: false +; CHECK-NEXT: .persp_centroid_ena: false +; CHECK-NEXT: .persp_pull_model_ena: false +; CHECK-NEXT: .persp_sample_ena: true +; CHECK-NEXT: .pos_fixed_pt_ena: false +; CHECK-NEXT: .pos_w_float_ena: false +; CHECK-NEXT: .pos_x_float_ena: false +; CHECK-NEXT: .pos_y_float_ena: false +; CHECK-NEXT: .pos_z_float_ena: false +; CHECK-NEXT: .sample_coverage_ena: false +; CHECK-NEXT: .hardware_stages: +; CHECK-NEXT: .cs: +; CHECK-NEXT: .checksum_value: 0x9444d7d0 +; CHECK-NEXT: .debug_mode: false +; CHECK-NEXT: .dynamic_vgpr_saved_count: 0x70 +; CHECK-NOT: .entry_point: _amdgpu_cs_main +; CHECK-NEXT: .entry_point_symbol: _amdgpu_cs_main +; CHECK-NEXT: .excp_en: 0 +; CHECK-NEXT: .float_mode: 0xc0 +; CHECK-NEXT: .forward_progress: true +; GFX11-NEXT: .ieee_mode: false +; CHECK-NEXT: .image_op: false +; CHECK-NEXT: .lds_size: 0 +; CHECK-NEXT: .mem_ordered: true +; CHECK-NEXT: .scratch_en: false +; CHECK-NEXT: .scratch_memory_size: 0 +; CHECK-NEXT: .sgpr_count: 0x22 +; CHECK-NEXT: .sgpr_limit: 0x6a +; CHECK-NEXT: .threadgroup_dimensions: +; CHECK-NEXT: - 0x1 +; CHECK-NEXT: - 0x400 +; CHECK-NEXT: - 0x1 +; CHECK-NEXT: .trap_present: false +; CHECK-NEXT: .user_data_reg_map: +; CHECK-NEXT: - 0x10000000 +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0 +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: - 0xffffffff +; CHECK-NEXT: .user_sgprs: 0x3 +; CHECK-NEXT: .vgpr_count: 0x2 +; CHECK-NEXT: .vgpr_limit: 0x100 +; CHECK-NEXT: .wavefront_size: 0x40 +; CHECK-NEXT: .wgp_mode: false +; CHECK-NEXT: .gs: +; CHECK-NEXT: .debug_mode: false +; CHECK-NOT: .entry_point: _amdgpu_gs_main +; CHECK-NEXT: .entry_point_symbol: gs_shader +; CHECK-NEXT: .forward_progress: true +; GFX11-NEXT: .ieee_mode: false +; CHECK-NEXT: .lds_size: 0x200 +; CHECK-NEXT: .mem_ordered: true +; CHECK-NEXT: .scratch_en: false +; CHECK-NEXT: .scratch_memory_size: 0 +; CHECK-NEXT: .sgpr_count: 0x1 +; CHECK-NEXT: .vgpr_count: 0x1 +; CHECK-NEXT: .wgp_mode: true +; CHECK-NEXT: .hs: +; CHECK-NEXT: .debug_mode: false +; CHECK-NOT: .entry_point: _amdgpu_hs_main +; CHECK-NEXT: .entry_point_symbol: hs_shader +; CHECK-NEXT: .forward_progress: true +; GFX11-NEXT: .ieee_mode: false +; CHECK-NEXT: .lds_size: 0x1000 +; CHECK-NEXT: .mem_ordered: true +; CHECK-NEXT: .scratch_en: false +; CHECK-NEXT: .scratch_memory_size: 0 +; CHECK-NEXT: .sgpr_count: 0x1 +; CHECK-NEXT: .vgpr_count: 0x1 +; CHECK-NEXT: .wgp_mode: true +; CHECK-NEXT: .ps: +; CHECK-NEXT: .debug_mode: false +; CHECK-NOT: .entry_point: _amdgpu_ps_main +; CHECK-NEXT: .entry_point_symbol: ps_shader +; CHECK-NEXT: .forward_progress: true +; GFX11-NEXT: .ieee_mode: false +; CHECK-NEXT: .lds_size: 0 +; CHECK-NEXT: .mem_ordered: true +; CHECK-NEXT: .scratch_en: false +; CHECK-NEXT: .scratch_memory_size: 0 +; CHECK-NEXT: .sgpr_count: 0x1 +; CHECK-NEXT: .vgpr_count: 0x1 +; CHECK-NEXT: .wgp_mode: true +; CHECK: .registers: {} +; CHECK:amdpal.version: +; CHECK-NEXT: - 0x3 +; CHECK-NEXT: - 0x6 +; CHECK-NEXT:... +; CHECK-NEXT: .end_amdgpu_pal_metadata + +define dllexport amdgpu_cs void @_amdgpu_cs_main(i32 inreg %arg1, i32 %arg2) #0 !lgc.shaderstage !1 { +.entry: + %i = call i64 @llvm.amdgcn.s.getpc() + %i1 = and i64 %i, -4294967296 + %i2 = zext i32 %arg1 to i64 + %i3 = or i64 %i1, %i2 + %i4 = inttoptr i64 %i3 to ptr addrspace(4) + %i5 = and i32 %arg2, 1023 + %i6 = lshr i32 %arg2, 10 + %i7 = and i32 %i6, 1023 + %i8 = add nuw nsw i32 %i7, %i5 + %i9 = load <4 x i32>, ptr addrspace(4) %i4, align 16 + %.idx = shl nuw nsw i32 %i8, 2 + call void @llvm.amdgcn.raw.buffer.store.i32(i32 1, <4 x i32> %i9, i32 %.idx, i32 0, i32 0) + ret void +} + +define dllexport amdgpu_ps void @ps_shader() #1 { + ret void +} + +@LDS.GS = external addrspace(3) global [1 x i32], align 4 + +define dllexport amdgpu_gs void @gs_shader() { + %ptr = getelementptr i32, ptr addrspace(3) @LDS.GS, i32 0 + store i32 0, ptr addrspace(3) %ptr, align 4 + ret void +} + +@LDS.HS = external addrspace(3) global [1024 x i32], align 4 + +define dllexport amdgpu_hs void @hs_shader() { + %ptr = getelementptr i32, ptr addrspace(3) @LDS.HS, i32 0 + store i32 0, ptr addrspace(3) %ptr, align 4 + ret void +} + +!amdgpu.pal.metadata.msgpack = !{!0} + +attributes #0 = { nounwind memory(readwrite) "target-features"=",+wavefrontsize64,+cumode" "amdgpu-dynamic-vgpr-block-size"="16" } + +attributes #1 = { nounwind memory(readwrite) "InitialPSInputAddr"="36983" "amdgpu-dynamic-vgpr-block-size"="16" } + +!0 = !{!"\82\B0amdpal.pipelines\91\8A\A4.api\A6Vulkan\B2.compute_registers\85\AB.tg_size_en\C3\AA.tgid_x_en\C2\AA.tgid_y_en\C2\AA.tgid_z_en\C2\AF.tidig_comp_cnt\01\B0.hardware_stages\81\A3.cs\8C\AF.checksum_value\CE\94D\D7\D0\AB.debug_mode\00\AB.float_mode\CC\C0\A9.image_op\C2\AC.mem_ordered\C3\AB.sgpr_limitj\B7.threadgroup_dimensions\93\01\CD\04\00\01\AD.trap_present\00\B2.user_data_reg_map\DC\00 \CE\10\00\00\00\CE\FF\FF\FF\FF\00\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\CE\FF\FF\FF\FF\AB.user_sgprs\03\AB.vgpr_limit\CD\01\00\AF.wavefront_size@\B7.internal_pipeline_hash\92\CF\E7\10k\A6:\A6%\F7\CF\B2\1F\1A\D4{\DA\E1T\AA.registers\80\A8.shaders\81\A8.compute\82\B0.api_shader_hash\92\CF\E9Zn7}\1E\B9\E7\00\B1.hardware_mapping\91\A3.cs\B0.spill_threshold\CE\FF\FF\FF\FF\A5.type\A2Cs\B0.user_data_limit\01\AF.xgl_cache_info\82\B3.128_bit_cache_hash\92\CF\B4X\B8\11[\A4\88P\CF\A0;\B0\AF\FF\B4\BE\C0\AD.llpc_version\A461.1\AEamdpal.version\92\03\06"} +!1 = !{i32 7} diff --git a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6.ll b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6.ll index 830872a..d2f26e8 100644 --- a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6.ll +++ b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.6.ll @@ -1,17 +1,14 @@ -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 <%s | FileCheck %s --check-prefixes=CHECK,GFX11,NODVGPR -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 <%s | FileCheck %s --check-prefixes=CHECK,NODVGPR -; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=+dynamic-vgpr <%s | FileCheck %s --check-prefixes=CHECK,DVGPR +; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 <%s | FileCheck %s --check-prefixes=CHECK,GFX11 +; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 <%s | FileCheck %s --check-prefixes=CHECK ; CHECK-LABEL: {{^}}_amdgpu_cs_main: -; NODVGPR: ; TotalNumSgprs: 4 -; DVGPR: ; TotalNumSgprs: 34 +; CHECK: ; TotalNumSgprs: 4 ; CHECK: ; NumVgprs: 2 ; CHECK: .amdgpu_pal_metadata ; CHECK-NEXT: --- ; CHECK-NEXT: amdpal.pipelines: ; CHECK-NEXT: - .api: Vulkan ; CHECK-NEXT: .compute_registers: -; DVGPR-NEXT: .dynamic_vgpr_en: true ; CHECK-NEXT: .tg_size_en: true ; CHECK-NEXT: .tgid_x_en: false ; CHECK-NEXT: .tgid_y_en: false @@ -57,7 +54,6 @@ ; CHECK-NEXT: .cs: ; CHECK-NEXT: .checksum_value: 0x9444d7d0 ; CHECK-NEXT: .debug_mode: false -; DVGPR-NEXT: .dynamic_vgpr_saved_count: 0x70 ; CHECK-NOT: .entry_point: _amdgpu_cs_main ; CHECK-NEXT: .entry_point_symbol: _amdgpu_cs_main ; CHECK-NEXT: .excp_en: 0 @@ -69,8 +65,7 @@ ; CHECK-NEXT: .mem_ordered: true ; CHECK-NEXT: .scratch_en: false ; CHECK-NEXT: .scratch_memory_size: 0 -; NODVGPR-NEXT: .sgpr_count: 0x4 -; DVGPR-NEXT: .sgpr_count: 0x22 +; CHECK-NEXT: .sgpr_count: 0x4 ; CHECK-NEXT: .sgpr_limit: 0x6a ; CHECK-NEXT: .threadgroup_dimensions: ; CHECK-NEXT: - 0x1 diff --git a/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll b/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll index 42469c8..23e90b3 100644 --- a/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll +++ b/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll @@ -202,13 +202,13 @@ attributes #5 = { "amdgpu-flat-work-group-size"="128,512" } attributes #6 = { "amdgpu-flat-work-group-size"="512,512" } attributes #7 = { "amdgpu-flat-work-group-size"="64,256" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="64,128" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="64,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="128,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="512,1024" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR7]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="64,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR8]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-flat-work-group-size"="64,128" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-flat-work-group-size"="64,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-flat-work-group-size"="128,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-flat-work-group-size"="512,1024" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { "amdgpu-flat-work-group-size"="64,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll b/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll index 06533b4..0be3147 100644 --- a/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll +++ b/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll @@ -399,25 +399,25 @@ attributes #17 = { "amdgpu-waves-per-eu"="5,8" } attributes #18 = { "amdgpu-waves-per-eu"="9,10" } attributes #19 = { "amdgpu-waves-per-eu"="8,9" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,2" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,4" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,9" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,1" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR7]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,9" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR8]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR9]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR10]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR11]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,123" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR12]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR13]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR14]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR15]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR16]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="6,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR17]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,5" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR18]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR19]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR20]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="8,9" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,2" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,4" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,9" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,1" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,9" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR9]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR10]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR11]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,123" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR12]] = { "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR13]] = { "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR14]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR15]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR16]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="6,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR17]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,5" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR18]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR19]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR20]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="8,9" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll b/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll index 8930626..33da671 100644 --- a/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll +++ b/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll @@ -19,5 +19,5 @@ define void @hoge() { ret void } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll b/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll index 3dfb0e1..f847d66 100644 --- a/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll +++ b/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll @@ -191,12 +191,12 @@ define amdgpu_kernel void @kernel_lds_recursion() { !1 = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR3]] = { "amdgpu-lds-size"="4" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-lds-size"="4" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-lds-size"="4" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR6:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) } ; CHECK: attributes #[[ATTR7:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } ;. diff --git a/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll b/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll index b9e9893..9a23788 100644 --- a/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll +++ b/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll @@ -369,7 +369,7 @@ define amdgpu_kernel void @illegal_mfma_after_rewrite() #1 { ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_mov_b32 s0, 0 ; CHECK-NEXT: s_mov_b32 s1, s0 -; CHECK-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; CHECK-NEXT: v_mov_b64_e32 v[28:29], s[0:1] ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; def s[0:3] ; CHECK-NEXT: ;;#ASMEND @@ -378,73 +378,66 @@ define amdgpu_kernel void @illegal_mfma_after_rewrite() #1 { ; CHECK-NEXT: v_mov_b64_e32 v[4:5], s[0:1] ; CHECK-NEXT: s_mov_b32 s0, 0x3c003c00 ; CHECK-NEXT: s_mov_b32 s1, s0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[8:9], v[8:9], v[4:7] -; CHECK-NEXT: v_mov_b64_e32 v[12:13], s[0:1] +; CHECK-NEXT: v_mov_b64_e32 v[30:31], s[0:1] ; CHECK-NEXT: s_mov_b32 s0, 0x7e007e00 ; CHECK-NEXT: s_mov_b32 s1, s0 -; CHECK-NEXT: v_mov_b64_e32 v[10:11], s[0:1] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[14:17], v[8:9], v[12:13], v[4:7] -; CHECK-NEXT: s_nop 1 -; CHECK-NEXT: v_accvgpr_write_b32 a0, v0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[18:21], v[8:9], v[10:11], v[4:7] -; CHECK-NEXT: v_accvgpr_write_b32 a1, v1 -; CHECK-NEXT: v_accvgpr_write_b32 a2, v2 -; CHECK-NEXT: v_accvgpr_write_b32 a3, v3 +; CHECK-NEXT: v_accvgpr_write_b32 a0, s0 +; CHECK-NEXT: v_accvgpr_write_b32 a1, s1 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[28:29], v[28:29], v[4:7] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[30:31], v[4:7] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[12:15], v[28:29], a[0:1], v[4:7] +; CHECK-NEXT: s_nop 2 ; CHECK-NEXT: v_mov_b32_e32 v4, 0x7fc00000 ; CHECK-NEXT: v_mov_b32_e32 v5, v4 ; CHECK-NEXT: v_mov_b32_e32 v6, v4 ; CHECK-NEXT: v_mov_b32_e32 v7, v4 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[14:17], v[8:9], v[8:9], v[14:17] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[28:29], v[8:11] ; CHECK-NEXT: s_nop 0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[22:25], v[8:9], v[8:9], v[4:7] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[16:19], v[28:29], v[28:29], v[4:7] ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; def v[4:7] ; CHECK-NEXT: ;;#ASMEND -; CHECK-NEXT: s_nop 0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[8:9], v[12:13], v[4:7] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[26:29], v[8:9], v[8:9], v[4:7] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[8:9], v[8:9], v[0:3] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[22:25], v[8:9], v[8:9], v[22:25] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[4:7], v[8:9], v[8:9], v[26:29] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[16:19], v[28:29], v[28:29], v[16:19] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[24:27], v[28:29], v[30:31], v[4:7] ; CHECK-NEXT: s_nop 5 -; CHECK-NEXT: v_cvt_f16_f32_e32 v23, v14 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[14:17], v[8:9], v[8:9], v[18:21] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[12:13], v[8:9], v[0:3] -; CHECK-NEXT: s_nop 1 -; CHECK-NEXT: v_accvgpr_read_b32 v19, a3 -; CHECK-NEXT: v_accvgpr_read_b32 v18, a2 -; CHECK-NEXT: v_mov_b64_e32 v[20:21], 0 -; CHECK-NEXT: s_nop 0 -; CHECK-NEXT: v_accvgpr_read_b32 v17, a1 -; CHECK-NEXT: v_accvgpr_read_b32 v16, a0 -; CHECK-NEXT: v_cvt_f16_f32_e32 v15, v22 -; CHECK-NEXT: v_cvt_f16_f32_e32 v14, v14 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[16:19], v[8:9], v[8:9], v[16:19] -; CHECK-NEXT: v_cvt_f16_f32_e32 v12, v0 -; CHECK-NEXT: global_store_short v[20:21], v23, off +; CHECK-NEXT: v_cvt_f16_f32_e32 v17, v8 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[28:29], v[12:15] +; CHECK-NEXT: s_nop 2 +; CHECK-NEXT: v_mov_b64_e32 v[12:13], 0 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[28:29], v[28:29], v[0:3] +; CHECK-NEXT: global_store_short v[12:13], v17, off ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[10:11], v[8:9], v[4:7] -; CHECK-NEXT: global_store_short v[20:21], v15, off +; CHECK-NEXT: v_cvt_f16_f32_e32 v9, v16 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[20:23], v[28:29], v[28:29], v[4:7] +; CHECK-NEXT: global_store_short v[12:13], v9, off +; CHECK-NEXT: v_cvt_f16_f32_e32 v1, v8 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[28:29], v[24:27] ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: global_store_short v[20:21], v14, off -; CHECK-NEXT: v_cvt_f16_f32_e32 v14, v16 +; CHECK-NEXT: v_cvt_f16_f32_e32 v14, v0 +; CHECK-NEXT: global_store_short v[12:13], v1, off +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[4:7], v[28:29], v[28:29], v[20:23] ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: global_store_short v[20:21], v14, off -; CHECK-NEXT: v_cvt_f16_f32_e32 v0, v0 +; CHECK-NEXT: global_store_short v[12:13], v14, off ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: global_store_short v[20:21], v12, off +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[30:31], v[28:29], v[8:11] +; CHECK-NEXT: s_nop 6 +; CHECK-NEXT: v_cvt_f16_f32_e32 v8, v0 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], a[0:1], v[28:29], v[4:7] +; CHECK-NEXT: global_store_short v[12:13], v8, off ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: global_store_short v[20:21], v0, off +; CHECK-NEXT: s_nop 2 +; CHECK-NEXT: v_cvt_f16_f32_e32 v0, v0 +; CHECK-NEXT: global_store_short v[12:13], v0, off ; CHECK-NEXT: s_endpgm entry: %k0 = call <4 x float> asm sideeffect "; def $0", "=s"() diff --git a/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll b/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll index f1cadea..0868148 100644 --- a/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll +++ b/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll @@ -63,7 +63,7 @@ define amdgpu_kernel void @foo(ptr noundef %fp) { ; OW-NEXT: ret void ; ; CW-LABEL: define {{[^@]+}}@foo -; CW-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR1:[0-9]+]] { +; CW-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR0]] { ; CW-NEXT: entry: ; CW-NEXT: [[FP_ADDR:%.*]] = alloca ptr, align 8, addrspace(5) ; CW-NEXT: store ptr [[FP]], ptr addrspace(5) [[FP_ADDR]], align 8 @@ -84,7 +84,7 @@ define amdgpu_kernel void @foo(ptr noundef %fp) { ; CW-NEXT: ret void ; ; NO-LABEL: define {{[^@]+}}@foo -; NO-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR1:[0-9]+]] { +; NO-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR0]] { ; NO-NEXT: entry: ; NO-NEXT: [[FP_ADDR:%.*]] = alloca ptr, align 8, addrspace(5) ; NO-NEXT: store ptr [[FP]], ptr addrspace(5) [[FP_ADDR]], align 8 @@ -101,14 +101,12 @@ entry: } ;. -; NO: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; NO: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; NO: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. -; OW: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; OW: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; OW: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. -; CW: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CW: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CW: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. ; NO: [[META0]] = !{ptr @bar1, ptr @bar2} ;. diff --git a/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll b/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll index 775d2f9..8fcaf5e 100644 --- a/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll +++ b/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll @@ -58,7 +58,7 @@ define amdgpu_kernel void @test_simple_indirect_call() { ;. -; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_GCN: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. ; ATTRIBUTOR_GCN: [[META0]] = !{i32 1, i32 5, i32 6, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/smfmac_alloc_failure_no_agpr_O0.ll b/llvm/test/CodeGen/AMDGPU/smfmac_alloc_failure_no_agpr_O0.ll new file mode 100644 index 0000000..ba0fdc68 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/smfmac_alloc_failure_no_agpr_O0.ll @@ -0,0 +1,119 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx950 -amdgpu-mfma-vgpr-form=0 < %s | FileCheck %s +; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx950 -amdgpu-mfma-vgpr-form=1 < %s | FileCheck %s + +declare <16 x float> @llvm.amdgcn.smfmac.f32.32x32x32.f16(<8 x half>, <16 x half>, <16 x float>, i32, i32 immarg, i32 immarg) + +define amdgpu_kernel void @test_smfmac_f32_32x32x32_f16__vgpr(ptr addrspace(1) %arg, <8 x half> %a, <16 x half> %b, i32 %idx) #0 { +; CHECK-LABEL: test_smfmac_f32_32x32x32_f16__vgpr: +; CHECK: ; %bb.0: ; %bb +; CHECK-NEXT: s_mov_b64 s[2:3], s[4:5] +; CHECK-NEXT: v_mov_b32_e32 v1, v0 +; CHECK-NEXT: v_mov_b32_e32 v0, 0 +; CHECK-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x24 +; CHECK-NEXT: s_load_dwordx4 s[12:15], s[2:3], 0x34 +; CHECK-NEXT: s_load_dwordx8 s[4:11], s[2:3], 0x44 +; CHECK-NEXT: s_nop 0 +; CHECK-NEXT: s_load_dword s2, s[2:3], 0x64 +; CHECK-NEXT: s_mov_b32 s3, 0x3ff +; CHECK-NEXT: v_and_b32_e64 v1, v1, s3 +; CHECK-NEXT: s_mov_b32 s3, 6 +; CHECK-NEXT: v_lshlrev_b32_e64 v8, s3, v1 +; CHECK-NEXT: s_waitcnt lgkmcnt(0) +; CHECK-NEXT: global_load_dwordx4 v[4:7], v8, s[0:1] offset:48 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: v_mov_b32_e32 v1, v7 +; CHECK-NEXT: v_mov_b32_e32 v2, v6 +; CHECK-NEXT: v_mov_b32_e32 v3, v5 +; CHECK-NEXT: ; kill: def $vgpr4 killed $vgpr4 killed $vgpr4_vgpr5_vgpr6_vgpr7 killed $exec +; CHECK-NEXT: global_load_dwordx4 v[10:13], v8, s[0:1] offset:32 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: v_mov_b32_e32 v5, v13 +; CHECK-NEXT: v_mov_b32_e32 v6, v12 +; CHECK-NEXT: v_mov_b32_e32 v7, v11 +; CHECK-NEXT: v_mov_b32_e32 v24, v10 +; CHECK-NEXT: global_load_dwordx4 v[10:13], v8, s[0:1] offset:16 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: v_mov_b32_e32 v25, v13 +; CHECK-NEXT: v_mov_b32_e32 v26, v12 +; CHECK-NEXT: v_mov_b32_e32 v27, v11 +; CHECK-NEXT: v_mov_b32_e32 v28, v10 +; CHECK-NEXT: global_load_dwordx4 v[8:11], v8, s[0:1] +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: v_mov_b32_e32 v29, v11 +; CHECK-NEXT: v_mov_b32_e32 v30, v10 +; CHECK-NEXT: v_mov_b32_e32 v31, v9 +; CHECK-NEXT: ; kill: def $vgpr8 killed $vgpr8 killed $vgpr8_vgpr9_vgpr10_vgpr11 killed $exec +; CHECK-NEXT: ; kill: def $vgpr8 killed $vgpr8 def $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 killed $exec +; CHECK-NEXT: v_mov_b32_e32 v9, v31 +; CHECK-NEXT: v_mov_b32_e32 v10, v30 +; CHECK-NEXT: v_mov_b32_e32 v11, v29 +; CHECK-NEXT: v_mov_b32_e32 v12, v28 +; CHECK-NEXT: v_mov_b32_e32 v13, v27 +; CHECK-NEXT: v_mov_b32_e32 v14, v26 +; CHECK-NEXT: v_mov_b32_e32 v15, v25 +; CHECK-NEXT: v_mov_b32_e32 v16, v24 +; CHECK-NEXT: v_mov_b32_e32 v17, v7 +; CHECK-NEXT: v_mov_b32_e32 v18, v6 +; CHECK-NEXT: v_mov_b32_e32 v19, v5 +; CHECK-NEXT: v_mov_b32_e32 v20, v4 +; CHECK-NEXT: v_mov_b32_e32 v21, v3 +; CHECK-NEXT: v_mov_b32_e32 v22, v2 +; CHECK-NEXT: v_mov_b32_e32 v23, v1 +; CHECK-NEXT: v_mov_b64_e32 v[2:3], s[12:13] +; CHECK-NEXT: v_mov_b64_e32 v[4:5], s[14:15] +; CHECK-NEXT: v_mov_b64_e32 v[30:31], s[10:11] +; CHECK-NEXT: v_mov_b64_e32 v[28:29], s[8:9] +; CHECK-NEXT: v_mov_b64_e32 v[26:27], s[6:7] +; CHECK-NEXT: v_mov_b64_e32 v[24:25], s[4:5] +; CHECK-NEXT: v_mov_b32_e32 v1, s2 +; CHECK-NEXT: s_nop 1 +; CHECK-NEXT: v_smfmac_f32_32x32x32_f16 v[8:23], v[2:5], v[24:31], v1 cbsz:1 abid:2 +; CHECK-NEXT: s_nop 11 +; CHECK-NEXT: v_mov_b32_e32 v1, v23 +; CHECK-NEXT: v_mov_b32_e32 v6, v22 +; CHECK-NEXT: v_mov_b32_e32 v7, v21 +; CHECK-NEXT: v_mov_b32_e32 v2, v20 +; CHECK-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3_vgpr4_vgpr5 killed $exec +; CHECK-NEXT: v_mov_b32_e32 v3, v7 +; CHECK-NEXT: v_mov_b32_e32 v4, v6 +; CHECK-NEXT: v_mov_b32_e32 v5, v1 +; CHECK-NEXT: global_store_dwordx4 v0, v[2:5], s[0:1] offset:48 +; CHECK-NEXT: v_mov_b32_e32 v1, v19 +; CHECK-NEXT: v_mov_b32_e32 v6, v18 +; CHECK-NEXT: v_mov_b32_e32 v7, v17 +; CHECK-NEXT: v_mov_b32_e32 v2, v16 +; CHECK-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3_vgpr4_vgpr5 killed $exec +; CHECK-NEXT: v_mov_b32_e32 v3, v7 +; CHECK-NEXT: v_mov_b32_e32 v4, v6 +; CHECK-NEXT: v_mov_b32_e32 v5, v1 +; CHECK-NEXT: global_store_dwordx4 v0, v[2:5], s[0:1] offset:32 +; CHECK-NEXT: v_mov_b32_e32 v1, v15 +; CHECK-NEXT: v_mov_b32_e32 v6, v14 +; CHECK-NEXT: v_mov_b32_e32 v7, v13 +; CHECK-NEXT: v_mov_b32_e32 v2, v12 +; CHECK-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3_vgpr4_vgpr5 killed $exec +; CHECK-NEXT: v_mov_b32_e32 v3, v7 +; CHECK-NEXT: v_mov_b32_e32 v4, v6 +; CHECK-NEXT: v_mov_b32_e32 v5, v1 +; CHECK-NEXT: global_store_dwordx4 v0, v[2:5], s[0:1] offset:16 +; CHECK-NEXT: v_mov_b32_e32 v1, v11 +; CHECK-NEXT: v_mov_b32_e32 v6, v10 +; CHECK-NEXT: v_mov_b32_e32 v7, v9 +; CHECK-NEXT: v_mov_b32_e32 v2, v8 +; CHECK-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3_vgpr4_vgpr5 killed $exec +; CHECK-NEXT: v_mov_b32_e32 v3, v7 +; CHECK-NEXT: v_mov_b32_e32 v4, v6 +; CHECK-NEXT: v_mov_b32_e32 v5, v1 +; CHECK-NEXT: global_store_dwordx4 v0, v[2:5], s[0:1] +; CHECK-NEXT: s_endpgm +bb: + %id = call i32 @llvm.amdgcn.workitem.id.x() + %gep = getelementptr <16 x float>, ptr addrspace(1) %arg, i32 %id + %in.1 = load <16 x float>, ptr addrspace(1) %gep + %mai.1 = tail call <16 x float> @llvm.amdgcn.smfmac.f32.32x32x32.f16(<8 x half> %a, <16 x half> %b, <16 x float> %in.1, i32 %idx, i32 1, i32 2) + store <16 x float> %mai.1, ptr addrspace(1) %arg + ret void +} + +attributes #0 = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-agpr-alloc"="0,0" } diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll index a1557418..8dfd3b7 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll @@ -31,5 +31,5 @@ define amdgpu_kernel void @kernel1() #1 { attributes #0 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll index fb225a9..fa01ee9 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll @@ -98,7 +98,7 @@ define amdgpu_kernel void @kernel2() #0 { attributes #0 = { "uniform-work-group-size"="true" } ;. ; CHECK: attributes #[[ATTR0]] = { "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR2]] = { "uniform-work-group-size"="true" } -; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll index cfede0c..09001ca 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll @@ -41,6 +41,6 @@ define amdgpu_kernel void @kernel3() #2 { attributes #2 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll index 854b724..4dede21 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll @@ -41,6 +41,6 @@ define amdgpu_kernel void @kernel2() #2 { attributes #1 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll index c4e0a60..08e1556 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll @@ -52,8 +52,8 @@ attributes #0 = { nounwind } attributes #1 = { "uniform-work-group-size"="false" } attributes #2 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR2]] = { nounwind "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR3]] = { "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll index 05af74d..9090d605 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll @@ -101,7 +101,7 @@ define amdgpu_kernel void @kernel(ptr addrspace(1) %m) #1 { attributes #0 = { nounwind readnone } attributes #1 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { nounwind memory(none) "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { nounwind memory(none) "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR0]] = { nounwind memory(none) "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { nounwind memory(none) "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll index cdbca7f..5e109f4 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll @@ -61,5 +61,5 @@ define amdgpu_kernel void @kernel3() #0 { attributes #0 = { "uniform-work-group-size"="false" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. |