diff options
Diffstat (limited to 'llvm/test/CodeGen')
35 files changed, 2435 insertions, 861 deletions
diff --git a/llvm/test/CodeGen/AArch64/sve-streaming-mode-cvt-fp-int-fp.ll b/llvm/test/CodeGen/AArch64/sve-streaming-mode-cvt-fp-int-fp.ll index f4ae66a..4ad5b38 100644 --- a/llvm/test/CodeGen/AArch64/sve-streaming-mode-cvt-fp-int-fp.ll +++ b/llvm/test/CodeGen/AArch64/sve-streaming-mode-cvt-fp-int-fp.ll @@ -70,6 +70,12 @@ define half @t3(half %x) { ; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0 ; CHECK-NEXT: ret ; +; USE-NEON-NO-GPRS-LABEL: t3: +; USE-NEON-NO-GPRS: // %bb.0: // %entry +; USE-NEON-NO-GPRS-NEXT: fcvtzs h0, h0 +; USE-NEON-NO-GPRS-NEXT: scvtf h0, h0 +; USE-NEON-NO-GPRS-NEXT: ret +; ; NONEON-NOSVE-LABEL: t3: ; NONEON-NOSVE: // %bb.0: // %entry ; NONEON-NOSVE-NEXT: fcvt s0, h0 @@ -147,6 +153,12 @@ define half @t6(half %x) { ; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0 ; CHECK-NEXT: ret ; +; USE-NEON-NO-GPRS-LABEL: t6: +; USE-NEON-NO-GPRS: // %bb.0: // %entry +; USE-NEON-NO-GPRS-NEXT: fcvtzu h0, h0 +; USE-NEON-NO-GPRS-NEXT: ucvtf h0, h0 +; USE-NEON-NO-GPRS-NEXT: ret +; ; NONEON-NOSVE-LABEL: t6: ; NONEON-NOSVE: // %bb.0: // %entry ; NONEON-NOSVE-NEXT: fcvt s0, h0 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll index c92e5c5..edb3607 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll @@ -1,5 +1,5 @@ ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=fiji -stop-after=irtranslator -o - %s | FileCheck %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=fiji -stop-after=irtranslator -o - %s | FileCheck %s declare align(8) dereferenceable(8) ptr @declared_with_ret_deref() #0 declare align(8) ptr @unknown_decl() #0 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll index e6a8bac..2356dad 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll @@ -1,15 +1,15 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefix=GFX9 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 -global-isel -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -global-isel -mattr=-promote-alloca < %s | FileCheck -check-prefix=GFX942 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -global-isel -mattr=-promote-alloca < %s | FileCheck -check-prefix=GFX11 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -global-isel -mattr=-promote-alloca < %s | FileCheck -check-prefix=GFX12 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel -new-reg-bank-select -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefix=GFX9 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 -global-isel -new-reg-bank-select -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefix=GFX10 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -global-isel -new-reg-bank-select -mattr=-promote-alloca < %s | FileCheck -check-prefix=GFX942 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -global-isel -new-reg-bank-select -mattr=-promote-alloca < %s | FileCheck -check-prefix=GFX11 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -global-isel -new-reg-bank-select -mattr=-promote-alloca < %s | FileCheck -check-prefix=GFX12 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel -mattr=-unaligned-access-mode -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefixes=UNALIGNED_GFX9 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 -global-isel -mattr=-unaligned-access-mode -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefixes=UNALIGNED_GFX10 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -global-isel -mattr=-unaligned-access-mode -mattr=-promote-alloca < %s | FileCheck -check-prefixes=UNALIGNED_GFX942 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -global-isel -mattr=-unaligned-access-mode -mattr=-promote-alloca < %s | FileCheck -check-prefixes=UNALIGNED_GFX11 %s -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -global-isel -mattr=-unaligned-access-mode -mattr=-promote-alloca < %s | FileCheck -check-prefixes=UNALIGNED_GFX12 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel -new-reg-bank-select -mattr=-unaligned-access-mode -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefixes=UNALIGNED_GFX9 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 -global-isel -new-reg-bank-select -mattr=-unaligned-access-mode -mattr=-promote-alloca -mattr=+enable-flat-scratch < %s | FileCheck -check-prefixes=UNALIGNED_GFX10 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -global-isel -new-reg-bank-select -mattr=-unaligned-access-mode -mattr=-promote-alloca < %s | FileCheck -check-prefixes=UNALIGNED_GFX942 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -global-isel -new-reg-bank-select -mattr=-unaligned-access-mode -mattr=-promote-alloca < %s | FileCheck -check-prefixes=UNALIGNED_GFX11 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -global-isel -new-reg-bank-select -mattr=-unaligned-access-mode -mattr=-promote-alloca < %s | FileCheck -check-prefixes=UNALIGNED_GFX12 %s define amdgpu_kernel void @store_load_sindex_kernel(i32 %idx) { ; GFX9-LABEL: store_load_sindex_kernel: diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.large.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.large.ll index 920d8fa..ae7f6ec 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.large.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.large.ll @@ -1,7 +1,7 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s -; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s -; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s define amdgpu_kernel void @v_insert_v64i32_37(ptr addrspace(1) %ptr.in, ptr addrspace(1) %ptr.out) #0 { ; GCN-LABEL: v_insert_v64i32_37: diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/lds-misaligned-bug.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/lds-misaligned-bug.ll index cfbb429..aabf256 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/lds-misaligned-bug.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/lds-misaligned-bug.ll @@ -1,11 +1,11 @@ -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-WGP %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1011 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-WGP %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1012 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-WGP %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1010 -mattr=+cumode < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-CU %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1010 -mattr=+cumode,+unaligned-access-mode < %s | FileCheck -check-prefixes=GCN,UNALIGNED %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-CU %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1100 -mattr=+cumode < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-CU %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1100 -mattr=+cumode,+unaligned-access-mode < %s | FileCheck -check-prefixes=GCN,UNALIGNED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-WGP %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1011 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-WGP %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1012 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-WGP %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 -mattr=+cumode < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-CU %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 -mattr=+cumode,+unaligned-access-mode < %s | FileCheck -check-prefixes=GCN,UNALIGNED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-CU %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+cumode < %s | FileCheck -check-prefixes=GCN,ALIGNED,ALIGNED-CU %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+cumode,+unaligned-access-mode < %s | FileCheck -check-prefixes=GCN,UNALIGNED %s ; GCN-LABEL: test_local_misaligned_v2: ; GCN-DAG: ds_{{read2|load_2addr}}_b32 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.workitem.id.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.workitem.id.ll index 66cdfc2..7b923f4 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.workitem.id.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.workitem.id.ll @@ -1,14 +1,14 @@ ; RUN: sed 's/CODE_OBJECT_VERSION/400/g' %s | opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-attributor -o %t.v4.ll ; RUN: sed 's/CODE_OBJECT_VERSION/600/g' %s | opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-attributor -o %t.v6.ll -; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa < %t.v4.ll | FileCheck --check-prefixes=ALL,HSA,UNPACKED %s -; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa < %t.v4.ll | FileCheck --check-prefixes=ALL,HSA,UNPACKED %s -; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=hawaii -mattr=+flat-for-global < %t.v4.ll | FileCheck --check-prefixes=ALL,MESA,UNPACKED %s -; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=tonga -mattr=+flat-for-global < %t.v4.ll | FileCheck --check-prefixes=ALL,MESA,UNPACKED %s -; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=+flat-for-global -mcpu=hawaii < %t.v4.ll | FileCheck -check-prefixes=ALL,MESA3D,UNPACKED %s -; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mcpu=tonga < %t.v4.ll | FileCheck -check-prefixes=ALL,MESA3D,UNPACKED %s -; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx90a < %t.v4.ll | FileCheck -check-prefixes=ALL,PACKED-TID %s -; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx1100 -amdgpu-enable-vopd=0 < %t.v4.ll | FileCheck -check-prefixes=ALL,PACKED-TID %s -; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa --amdhsa-code-object-version=6 -mcpu=gfx11-generic -amdgpu-enable-vopd=0 < %t.v6.ll | FileCheck -check-prefixes=ALL,PACKED-TID %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-amdhsa < %t.v4.ll | FileCheck --check-prefixes=ALL,HSA,UNPACKED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-amdhsa < %t.v4.ll | FileCheck --check-prefixes=ALL,HSA,UNPACKED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-- -mcpu=hawaii -mattr=+flat-for-global < %t.v4.ll | FileCheck --check-prefixes=ALL,MESA,UNPACKED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-- -mcpu=tonga -mattr=+flat-for-global < %t.v4.ll | FileCheck --check-prefixes=ALL,MESA,UNPACKED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-mesa3d -mattr=+flat-for-global -mcpu=hawaii < %t.v4.ll | FileCheck -check-prefixes=ALL,MESA3D,UNPACKED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-mesa3d -mcpu=tonga < %t.v4.ll | FileCheck -check-prefixes=ALL,MESA3D,UNPACKED %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx90a < %t.v4.ll | FileCheck -check-prefixes=ALL,PACKED-TID %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx1100 -amdgpu-enable-vopd=0 < %t.v4.ll | FileCheck -check-prefixes=ALL,PACKED-TID %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-unknown-amdhsa --amdhsa-code-object-version=6 -mcpu=gfx11-generic -amdgpu-enable-vopd=0 < %t.v6.ll | FileCheck -check-prefixes=ALL,PACKED-TID %s declare i32 @llvm.amdgcn.workitem.id.x() #0 declare i32 @llvm.amdgcn.workitem.id.y() #0 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-sext.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-sext.mir new file mode 100644 index 0000000..87836e2 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-sext.mir @@ -0,0 +1,170 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py +# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - | FileCheck %s + +--- +name: assert_sext_vgpr +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $vgpr0 + + ; CHECK-LABEL: name: assert_sext_vgpr + ; CHECK: liveins: $vgpr0 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:vgpr(s32) = COPY $vgpr0 + ; CHECK-NEXT: %assert_sext:vgpr(s32) = G_ASSERT_SEXT %copy, 4 + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s32) + %copy:_(s32) = COPY $vgpr0 + %assert_sext:_(s32) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_sgpr +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr8 + + ; CHECK-LABEL: name: assert_sext_sgpr + ; CHECK: liveins: $sgpr8 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:sgpr(s32) = COPY $sgpr8 + ; CHECK-NEXT: %assert_sext:sgpr(s32) = G_ASSERT_SEXT %copy, 4 + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s32) + %copy:_(s32) = COPY $sgpr8 + %assert_sext:_(s32) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_agpr +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $agpr0 + + ; CHECK-LABEL: name: assert_sext_agpr + ; CHECK: liveins: $agpr0 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:vgpr(s32) = COPY $agpr0 + ; CHECK-NEXT: %assert_sext:vgpr(s32) = G_ASSERT_SEXT %copy, 4 + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s32) + %copy:_(s32) = COPY $agpr0 + %assert_sext:_(s32) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_vgpr_regclass +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $vgpr0 + + ; CHECK-LABEL: name: assert_sext_vgpr_regclass + ; CHECK: liveins: $vgpr0 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:vgpr_32(s32) = COPY $vgpr0 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY %copy(s32) + ; CHECK-NEXT: %assert_sext:vgpr(s32) = G_ASSERT_SEXT [[COPY]], 4 + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s32) + %copy:vgpr_32(s32) = COPY $vgpr0 + %assert_sext:_(s32) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_sgpr_regcllass +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr8 + + ; CHECK-LABEL: name: assert_sext_sgpr_regcllass + ; CHECK: liveins: $sgpr8 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:sgpr_32(s32) = COPY $sgpr8 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY %copy(s32) + ; CHECK-NEXT: %assert_sext:sgpr(s32) = G_ASSERT_SEXT [[COPY]], 4 + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s32) + %copy:sgpr_32(s32) = COPY $sgpr8 + %assert_sext:_(s32) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_vgpr_64 +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $vgpr0_vgpr1 + + ; CHECK-LABEL: name: assert_sext_vgpr_64 + ; CHECK: liveins: $vgpr0_vgpr1 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:vreg_64(s64) = COPY $vgpr0_vgpr1 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY %copy(s64) + ; CHECK-NEXT: [[ASSERT_SEXT:%[0-9]+]]:vgpr(s64) = G_ASSERT_SEXT [[COPY]], 4 + ; CHECK-NEXT: %assert_sext:vreg_64(s64) = COPY [[ASSERT_SEXT]](s64) + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s64) + %copy:vreg_64(s64) = COPY $vgpr0_vgpr1 + %assert_sext:vreg_64(s64) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_sgpr_64 +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr0_sgpr1 + + ; CHECK-LABEL: name: assert_sext_sgpr_64 + ; CHECK: liveins: $sgpr0_sgpr1 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:sreg_64(s64) = COPY $sgpr0_sgpr1 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY %copy(s64) + ; CHECK-NEXT: [[ASSERT_SEXT:%[0-9]+]]:sgpr(s64) = G_ASSERT_SEXT [[COPY]], 4 + ; CHECK-NEXT: %assert_sext:sreg_64(s64) = COPY [[ASSERT_SEXT]](s64) + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s64) + %copy:sreg_64(s64) = COPY $sgpr0_sgpr1 + %assert_sext:sreg_64(s64) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... + +--- +name: assert_sext_agpr_64 +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $agpr0_agpr1 + + ; CHECK-LABEL: name: assert_sext_agpr_64 + ; CHECK: liveins: $agpr0_agpr1 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:areg_64(s64) = COPY $agpr0_agpr1 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY %copy(s64) + ; CHECK-NEXT: [[ASSERT_SEXT:%[0-9]+]]:vgpr(s64) = G_ASSERT_SEXT [[COPY]], 4 + ; CHECK-NEXT: %assert_sext:areg_64(s64) = COPY [[ASSERT_SEXT]](s64) + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_sext(s64) + %copy:areg_64(s64) = COPY $agpr0_agpr1 + %assert_sext:areg_64(s64) = G_ASSERT_SEXT %copy, 4 + S_ENDPGM 0, implicit %assert_sext +... diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-zext.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-zext.mir index 0bce908..c64a8ec 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-zext.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-assert-zext.mir @@ -1,5 +1,5 @@ # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=regbankselect %s -verify-machineinstrs -o - | FileCheck %s +# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - | FileCheck %s --- name: assert_zext_vgpr @@ -53,8 +53,8 @@ body: | ; CHECK-LABEL: name: assert_zext_agpr ; CHECK: liveins: $agpr0 ; CHECK-NEXT: {{ $}} - ; CHECK-NEXT: %copy:agpr(s32) = COPY $agpr0 - ; CHECK-NEXT: %assert_zext:agpr(s32) = G_ASSERT_ZEXT %copy, 4 + ; CHECK-NEXT: %copy:vgpr(s32) = COPY $agpr0 + ; CHECK-NEXT: %assert_zext:vgpr(s32) = G_ASSERT_ZEXT %copy, 4 ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_zext(s32) %copy:_(s32) = COPY $agpr0 %assert_zext:_(s32) = G_ASSERT_ZEXT %copy, 4 @@ -74,7 +74,8 @@ body: | ; CHECK: liveins: $vgpr0 ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: %copy:vgpr_32(s32) = COPY $vgpr0 - ; CHECK-NEXT: %assert_zext:vgpr(s32) = G_ASSERT_ZEXT %copy, 4 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY %copy(s32) + ; CHECK-NEXT: %assert_zext:vgpr(s32) = G_ASSERT_ZEXT [[COPY]], 4 ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_zext(s32) %copy:vgpr_32(s32) = COPY $vgpr0 %assert_zext:_(s32) = G_ASSERT_ZEXT %copy, 4 @@ -94,9 +95,76 @@ body: | ; CHECK: liveins: $sgpr8 ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: %copy:sgpr_32(s32) = COPY $sgpr8 - ; CHECK-NEXT: %assert_zext:sgpr(s32) = G_ASSERT_ZEXT %copy, 4 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY %copy(s32) + ; CHECK-NEXT: %assert_zext:sgpr(s32) = G_ASSERT_ZEXT [[COPY]], 4 ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_zext(s32) %copy:sgpr_32(s32) = COPY $sgpr8 %assert_zext:_(s32) = G_ASSERT_ZEXT %copy, 4 S_ENDPGM 0, implicit %assert_zext ... + +--- +name: assert_zext_vgpr_64 +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $vgpr0_vgpr1 + + ; CHECK-LABEL: name: assert_zext_vgpr_64 + ; CHECK: liveins: $vgpr0_vgpr1 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:vreg_64(s64) = COPY $vgpr0_vgpr1 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY %copy(s64) + ; CHECK-NEXT: [[ASSERT_ZEXT:%[0-9]+]]:vgpr(s64) = G_ASSERT_ZEXT [[COPY]], 4 + ; CHECK-NEXT: %assert_zext:vreg_64(s64) = COPY [[ASSERT_ZEXT]](s64) + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_zext(s64) + %copy:vreg_64(s64) = COPY $vgpr0_vgpr1 + %assert_zext:vreg_64(s64) = G_ASSERT_ZEXT %copy, 4 + S_ENDPGM 0, implicit %assert_zext +... + +--- +name: assert_zext_sgpr_64 +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr0_sgpr1 + + ; CHECK-LABEL: name: assert_zext_sgpr_64 + ; CHECK: liveins: $sgpr0_sgpr1 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:sreg_64(s64) = COPY $sgpr0_sgpr1 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY %copy(s64) + ; CHECK-NEXT: [[ASSERT_ZEXT:%[0-9]+]]:sgpr(s64) = G_ASSERT_ZEXT [[COPY]], 4 + ; CHECK-NEXT: %assert_zext:sreg_64(s64) = COPY [[ASSERT_ZEXT]](s64) + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_zext(s64) + %copy:sreg_64(s64) = COPY $sgpr0_sgpr1 + %assert_zext:sreg_64(s64) = G_ASSERT_ZEXT %copy, 4 + S_ENDPGM 0, implicit %assert_zext +... + +--- +name: assert_zext_agpr_64 +alignment: 4 +legalized: true +tracksRegLiveness: true +body: | + bb.0: + liveins: $agpr0_agpr1 + + ; CHECK-LABEL: name: assert_zext_agpr_64 + ; CHECK: liveins: $agpr0_agpr1 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: %copy:areg_64(s64) = COPY $agpr0_agpr1 + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY %copy(s64) + ; CHECK-NEXT: [[ASSERT_ZEXT:%[0-9]+]]:vgpr(s64) = G_ASSERT_ZEXT [[COPY]], 4 + ; CHECK-NEXT: %assert_zext:areg_64(s64) = COPY [[ASSERT_ZEXT]](s64) + ; CHECK-NEXT: S_ENDPGM 0, implicit %assert_zext(s64) + %copy:areg_64(s64) = COPY $agpr0_agpr1 + %assert_zext:areg_64(s64) = G_ASSERT_ZEXT %copy, 4 + S_ENDPGM 0, implicit %assert_zext +... diff --git a/llvm/test/CodeGen/AMDGPU/bf16.ll b/llvm/test/CodeGen/AMDGPU/bf16.ll index 4b14dc6..7ee0015f 100644 --- a/llvm/test/CodeGen/AMDGPU/bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/bf16.ll @@ -21204,18 +21204,14 @@ define bfloat @v_fabs_bf16(bfloat %a) { ; 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_and_b32_e32 v0, 0xffff0000, v0 -; GCN-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 -; GCN-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0x7fff0000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] ; ; GFX7-LABEL: v_fabs_bf16: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v0 -; GFX7-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX7-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 -; GFX7-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX7-NEXT: v_and_b32_e32 v0, 0x7fff0000, v0 ; GFX7-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: v_fabs_bf16: @@ -21440,10 +21436,7 @@ define bfloat @v_fneg_fabs_bf16(bfloat %a) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_mul_f32_e32 v0, 1.0, v0 ; GCN-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GCN-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 -; GCN-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 -; GCN-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GCN-NEXT: v_or_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] ; ; GFX7-LABEL: v_fneg_fabs_bf16: @@ -21451,10 +21444,7 @@ define bfloat @v_fneg_fabs_bf16(bfloat %a) { ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v0 ; GFX7-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX7-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 -; GFX7-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX7-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 -; GFX7-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX7-NEXT: v_or_b32_e32 v0, 0x80000000, v0 ; GFX7-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: v_fneg_fabs_bf16: @@ -21510,23 +21500,17 @@ define amdgpu_ps i32 @s_fneg_fabs_bf16(bfloat inreg %a) { ; GCN-LABEL: s_fneg_fabs_bf16: ; GCN: ; %bb.0: ; GCN-NEXT: v_mul_f32_e64 v0, 1.0, s0 +; GCN-NEXT: v_lshrrev_b32_e32 v0, 16, v0 +; GCN-NEXT: v_or_b32_e32 v0, 0x8000, v0 ; GCN-NEXT: v_readfirstlane_b32 s0, v0 -; GCN-NEXT: s_and_b32 s0, s0, 0xffff0000 -; GCN-NEXT: s_bitset0_b32 s0, 31 -; GCN-NEXT: s_and_b32 s0, s0, 0xffff0000 -; GCN-NEXT: s_xor_b32 s0, s0, 0x80000000 -; GCN-NEXT: s_lshr_b32 s0, s0, 16 ; GCN-NEXT: ; return to shader part epilog ; ; GFX7-LABEL: s_fneg_fabs_bf16: ; GFX7: ; %bb.0: ; GFX7-NEXT: v_mul_f32_e64 v0, 1.0, s0 +; GFX7-NEXT: v_lshrrev_b32_e32 v0, 16, v0 +; GFX7-NEXT: v_or_b32_e32 v0, 0x8000, v0 ; GFX7-NEXT: v_readfirstlane_b32 s0, v0 -; GFX7-NEXT: s_and_b32 s0, s0, 0xffff0000 -; GFX7-NEXT: s_bitset0_b32 s0, 31 -; GFX7-NEXT: s_and_b32 s0, s0, 0xffff0000 -; GFX7-NEXT: s_xor_b32 s0, s0, 0x80000000 -; GFX7-NEXT: s_lshr_b32 s0, s0, 16 ; GFX7-NEXT: ; return to shader part epilog ; ; GFX8-LABEL: s_fneg_fabs_bf16: diff --git a/llvm/test/CodeGen/AMDGPU/fabs.bf16.ll b/llvm/test/CodeGen/AMDGPU/fabs.bf16.ll index 5d184b1..c46fcde 100644 --- a/llvm/test/CodeGen/AMDGPU/fabs.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/fabs.bf16.ll @@ -218,19 +218,11 @@ define amdgpu_kernel void @s_fabs_v4bf16(ptr addrspace(1) %out, <4 x bfloat> %in ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_and_b32 s4, s3, 0xffff0000 -; CI-NEXT: s_lshl_b32 s3, s3, 16 -; CI-NEXT: s_and_b32 s5, s2, 0xffff0000 -; CI-NEXT: v_mul_f32_e64 v0, 1.0, |s4| -; CI-NEXT: v_mul_f32_e64 v1, 1.0, |s3| -; CI-NEXT: v_mul_f32_e64 v2, 1.0, |s5| -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_alignbit_b32 v1, v0, v1, 16 -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v2 -; CI-NEXT: v_mul_f32_e64 v2, 1.0, |s2| -; CI-NEXT: v_alignbit_b32 v0, v0, v2, 16 +; CI-NEXT: s_and_b32 s3, s3, 0x7fff7fff +; CI-NEXT: s_and_b32 s2, s2, 0x7fff7fff ; CI-NEXT: v_mov_b32_e32 v3, s1 +; CI-NEXT: v_mov_b32_e32 v0, s2 +; CI-NEXT: v_mov_b32_e32 v1, s3 ; CI-NEXT: v_mov_b32_e32 v2, s0 ; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; CI-NEXT: s_endpgm @@ -537,16 +529,15 @@ define amdgpu_kernel void @v_fabs_fold_self_v2bf16(ptr addrspace(1) %out, ptr ad ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 -; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_mul_f32_e64 v4, 1.0, |v3| -; CI-NEXT: v_mul_f32_e64 v5, 1.0, |v2| -; CI-NEXT: v_and_b32_e32 v4, 0xffff0000, v4 -; CI-NEXT: v_and_b32_e32 v5, 0xffff0000, v5 -; CI-NEXT: v_mul_f32_e32 v3, v4, v3 -; CI-NEXT: v_mul_f32_e32 v2, v5, v2 -; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v3 -; CI-NEXT: v_alignbit_b32 v2, v3, v2, 16 +; CI-NEXT: v_and_b32_e32 v3, 0x7fff, v2 +; CI-NEXT: v_lshlrev_b32_e32 v4, 16, v2 +; CI-NEXT: v_and_b32_e32 v5, 0xffff0000, v2 +; CI-NEXT: v_and_b32_e32 v2, 0x7fff0000, v2 +; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; CI-NEXT: v_mul_f32_e32 v2, v2, v5 +; CI-NEXT: v_mul_f32_e32 v3, v3, v4 +; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 +; CI-NEXT: v_alignbit_b32 v2, v2, v3, 16 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -898,16 +889,13 @@ define amdgpu_kernel void @v_extract_fabs_fold_v2bf16(ptr addrspace(1) %in) #0 { ; CI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; CI-NEXT: flat_load_dword v0, v[0:1] ; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; CI-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; CI-NEXT: v_mul_f32_e64 v1, 1.0, |v1| -; CI-NEXT: v_mul_f32_e64 v0, 1.0, |v0| -; CI-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 -; CI-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; CI-NEXT: v_mul_f32_e32 v1, 4.0, v1 +; CI-NEXT: v_and_b32_e32 v1, 0x7fff, v0 +; CI-NEXT: v_and_b32_e32 v0, 0x7fff0000, v0 +; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1 ; CI-NEXT: v_add_f32_e32 v0, 2.0, v0 -; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; CI-NEXT: v_mul_f32_e32 v1, 4.0, v1 ; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 +; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1 ; CI-NEXT: flat_store_short v[0:1], v1 ; CI-NEXT: s_waitcnt vmcnt(0) ; CI-NEXT: flat_store_short v[0:1], v0 diff --git a/llvm/test/CodeGen/AMDGPU/fneg-fabs.bf16.ll b/llvm/test/CodeGen/AMDGPU/fneg-fabs.bf16.ll index 64a9727..76da0aa 100644 --- a/llvm/test/CodeGen/AMDGPU/fneg-fabs.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/fneg-fabs.bf16.ll @@ -107,12 +107,10 @@ define amdgpu_kernel void @fneg_fabs_fmul_bf16(ptr addrspace(1) %out, bfloat %x, ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_and_b32 s3, s2, 0x7fff -; CI-NEXT: s_lshl_b32 s3, s3, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s3 +; CI-NEXT: s_lshl_b32 s3, s2, 16 ; CI-NEXT: s_and_b32 s2, s2, 0xffff0000 -; CI-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; CI-NEXT: v_mul_f32_e32 v0, s2, v0 +; CI-NEXT: v_mov_b32_e32 v0, s3 +; CI-NEXT: v_mul_f32_e64 v0, s2, -|v0| ; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 @@ -204,12 +202,10 @@ define amdgpu_kernel void @fneg_fabs_free_bf16(ptr addrspace(1) %out, i16 %in) { ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_and_b32 s2, s2, 0x7fff -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s2 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; CI-NEXT: s_bitset1_b32 s2, 15 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_short v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -279,12 +275,10 @@ define amdgpu_kernel void @fneg_fabs_bf16(ptr addrspace(1) %out, bfloat %in) { ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_and_b32 s2, s2, 0x7fff -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s2 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; CI-NEXT: s_bitset1_b32 s2, 15 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_short v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -345,43 +339,22 @@ define amdgpu_kernel void @fneg_fabs_bf16(ptr addrspace(1) %out, bfloat %in) { } define amdgpu_kernel void @v_fneg_fabs_bf16(ptr addrspace(1) %out, ptr addrspace(1) %in) { -; CI-LABEL: v_fneg_fabs_bf16: -; CI: ; %bb.0: -; CI-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; CI-NEXT: s_add_i32 s12, s12, s17 -; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 -; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 -; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: v_mov_b32_e32 v0, s2 -; CI-NEXT: v_mov_b32_e32 v1, s3 -; CI-NEXT: flat_load_ushort v2, v[0:1] -; CI-NEXT: v_mov_b32_e32 v0, s0 -; CI-NEXT: v_mov_b32_e32 v1, s1 -; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_mul_f32_e64 v2, 1.0, |v2| -; CI-NEXT: v_and_b32_e32 v2, 0xffff0000, v2 -; CI-NEXT: v_xor_b32_e32 v2, 0x80000000, v2 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; CI-NEXT: flat_store_short v[0:1], v2 -; CI-NEXT: s_endpgm -; -; VI-LABEL: v_fneg_fabs_bf16: -; VI: ; %bb.0: -; VI-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; VI-NEXT: s_add_i32 s12, s12, s17 -; VI-NEXT: s_mov_b32 flat_scratch_lo, s13 -; VI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 -; VI-NEXT: s_waitcnt lgkmcnt(0) -; VI-NEXT: v_mov_b32_e32 v0, s2 -; VI-NEXT: v_mov_b32_e32 v1, s3 -; VI-NEXT: flat_load_ushort v2, v[0:1] -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: v_mov_b32_e32 v1, s1 -; VI-NEXT: s_waitcnt vmcnt(0) -; VI-NEXT: v_or_b32_e32 v2, 0x8000, v2 -; VI-NEXT: flat_store_short v[0:1], v2 -; VI-NEXT: s_endpgm +; CIVI-LABEL: v_fneg_fabs_bf16: +; CIVI: ; %bb.0: +; CIVI-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 +; CIVI-NEXT: s_add_i32 s12, s12, s17 +; CIVI-NEXT: s_mov_b32 flat_scratch_lo, s13 +; CIVI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 +; CIVI-NEXT: s_waitcnt lgkmcnt(0) +; CIVI-NEXT: v_mov_b32_e32 v0, s2 +; CIVI-NEXT: v_mov_b32_e32 v1, s3 +; CIVI-NEXT: flat_load_ushort v2, v[0:1] +; CIVI-NEXT: v_mov_b32_e32 v0, s0 +; CIVI-NEXT: v_mov_b32_e32 v1, s1 +; CIVI-NEXT: s_waitcnt vmcnt(0) +; CIVI-NEXT: v_or_b32_e32 v2, 0x8000, v2 +; CIVI-NEXT: flat_store_short v[0:1], v2 +; CIVI-NEXT: s_endpgm ; ; GFX9-LABEL: v_fneg_fabs_bf16: ; GFX9: ; %bb.0: @@ -431,21 +404,13 @@ define amdgpu_kernel void @s_fneg_fabs_v2bf16_non_bc_src(ptr addrspace(1) %out, ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_and_b32 s3, s2, 0xffff0000 -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_add_f32_e64 v0, s3, 2.0 -; CI-NEXT: v_add_f32_e64 v1, s2, 1.0 -; CI-NEXT: v_readfirstlane_b32 s2, v0 +; CI-NEXT: s_lshl_b32 s3, s2, 16 ; CI-NEXT: s_and_b32 s2, s2, 0xffff0000 -; CI-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 -; CI-NEXT: s_bitset0_b32 s2, 31 -; CI-NEXT: v_and_b32_e32 v0, 0x7fffffff, v1 -; CI-NEXT: s_and_b32 s2, s2, 0xffff0000 -; CI-NEXT: s_xor_b32 s2, s2, 0x80000000 -; CI-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; CI-NEXT: s_lshr_b32 s2, s2, 16 -; CI-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 -; CI-NEXT: v_alignbit_b32 v2, s2, v0, 16 +; CI-NEXT: v_add_f32_e64 v1, s2, 2.0 +; CI-NEXT: v_add_f32_e64 v0, s3, 1.0 +; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; CI-NEXT: v_alignbit_b32 v0, v1, v0, 16 +; CI-NEXT: v_or_b32_e32 v2, 0x80008000, v0 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: flat_store_dword v[0:1], v2 @@ -566,15 +531,10 @@ define amdgpu_kernel void @s_fneg_fabs_v2bf16_bc_src(ptr addrspace(1) %out, <2 x ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_and_b32 s3, s2, 0x7fff -; CI-NEXT: s_and_b32 s2, s2, 0x7fff0000 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s2 -; CI-NEXT: s_lshl_b32 s2, s3, 16 -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; CI-NEXT: v_mul_f32_e64 v1, -1.0, s2 -; CI-NEXT: v_alignbit_b32 v2, v0, v1, 16 +; CI-NEXT: s_or_b32 s2, s2, 0x80008000 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -629,27 +589,11 @@ define amdgpu_kernel void @fneg_fabs_v4bf16(ptr addrspace(1) %out, <4 x bfloat> ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_lshl_b32 s4, s2, 16 -; CI-NEXT: s_and_b32 s2, s2, 0xffff0000 -; CI-NEXT: v_mul_f32_e64 v2, 1.0, |s2| -; CI-NEXT: s_and_b32 s2, s3, 0xffff0000 -; CI-NEXT: s_lshl_b32 s5, s3, 16 -; CI-NEXT: v_mul_f32_e64 v3, 1.0, |s2| -; CI-NEXT: v_mul_f32_e64 v0, 1.0, |s4| -; CI-NEXT: v_mul_f32_e64 v1, 1.0, |s5| -; CI-NEXT: v_and_b32_e32 v3, 0xffff0000, v3 -; CI-NEXT: v_and_b32_e32 v2, 0xffff0000, v2 -; CI-NEXT: v_xor_b32_e32 v3, 0x80000000, v3 -; CI-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 -; CI-NEXT: v_xor_b32_e32 v2, 0x80000000, v2 -; CI-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v3 -; CI-NEXT: v_xor_b32_e32 v1, 0x80000000, v1 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 -; CI-NEXT: v_alignbit_b32 v1, v3, v1, 16 -; CI-NEXT: v_alignbit_b32 v0, v2, v0, 16 +; CI-NEXT: s_or_b32 s3, s3, 0x80008000 +; CI-NEXT: s_or_b32 s2, s2, 0x80008000 ; CI-NEXT: v_mov_b32_e32 v3, s1 +; CI-NEXT: v_mov_b32_e32 v0, s2 +; CI-NEXT: v_mov_b32_e32 v1, s3 ; CI-NEXT: v_mov_b32_e32 v2, s0 ; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; CI-NEXT: s_endpgm @@ -860,21 +804,20 @@ define amdgpu_kernel void @s_fneg_multi_use_fabs_v2bf16(ptr addrspace(1) %out0, ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: v_mov_b32_e32 v1, s1 -; CI-NEXT: v_mov_b32_e32 v2, s2 -; CI-NEXT: s_and_b32 s1, s4, 0x7fff -; CI-NEXT: s_and_b32 s2, s4, 0x7fff0000 -; CI-NEXT: v_mul_f32_e64 v4, -1.0, s2 -; CI-NEXT: s_lshl_b32 s1, s1, 16 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: s_and_b32 s0, s4, 0x7fff7fff -; CI-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; CI-NEXT: v_mul_f32_e64 v5, -1.0, s1 -; CI-NEXT: v_alignbit_b32 v4, v4, v5, 16 -; CI-NEXT: v_mov_b32_e32 v5, s0 +; CI-NEXT: v_mov_b32_e32 v2, s2 +; CI-NEXT: s_or_b32 s2, s0, 0x8000 +; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: s_and_b32 s1, s4, 0x7fff0000 +; CI-NEXT: s_and_b32 s2, s2, 0xffff +; CI-NEXT: s_or_b32 s1, s1, s2 +; CI-NEXT: s_bitset1_b32 s1, 31 +; CI-NEXT: v_mov_b32_e32 v4, s0 ; CI-NEXT: v_mov_b32_e32 v3, s3 -; CI-NEXT: flat_store_dword v[0:1], v5 -; CI-NEXT: flat_store_dword v[2:3], v4 +; CI-NEXT: flat_store_dword v[0:1], v4 +; CI-NEXT: v_mov_b32_e32 v0, s1 +; CI-NEXT: flat_store_dword v[2:3], v0 ; CI-NEXT: s_endpgm ; ; VI-LABEL: s_fneg_multi_use_fabs_v2bf16: @@ -1086,5 +1029,3 @@ declare <4 x bfloat> @llvm.fabs.v4bf16(<4 x bfloat>) #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: -; CIVI: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/fneg.bf16.ll b/llvm/test/CodeGen/AMDGPU/fneg.bf16.ll index d232693..98044a7 100644 --- a/llvm/test/CodeGen/AMDGPU/fneg.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/fneg.bf16.ll @@ -14,11 +14,10 @@ define amdgpu_kernel void @s_fneg_bf16(ptr addrspace(1) %out, bfloat %in) #0 { ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s2 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; CI-NEXT: s_xor_b32 s2, s2, 0x8000 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_short v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -93,9 +92,7 @@ define amdgpu_kernel void @v_fneg_bf16(ptr addrspace(1) %out, ptr addrspace(1) % ; CI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; CI-NEXT: flat_load_ushort v2, v[0:1] ; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_mul_f32_e32 v2, -1.0, v2 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 +; CI-NEXT: v_xor_b32_e32 v2, 0x8000, v2 ; CI-NEXT: flat_store_short v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -170,11 +167,10 @@ define amdgpu_kernel void @s_fneg_free_bf16(ptr addrspace(1) %out, i16 %in) #0 { ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s2 -; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; CI-NEXT: s_xor_b32 s2, s2, 0x8000 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_short v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -248,9 +244,9 @@ define amdgpu_kernel void @v_fneg_fold_bf16(ptr addrspace(1) %out, ptr addrspace ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: s_waitcnt vmcnt(0) +; CI-NEXT: v_xor_b32_e32 v3, 0x8000, v2 ; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_mul_f32_e32 v3, -1.0, v2 -; CI-NEXT: v_and_b32_e32 v3, 0xffff0000, v3 +; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; CI-NEXT: v_mul_f32_e32 v2, v3, v2 ; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 ; CI-NEXT: flat_store_short v[0:1], v2 @@ -365,13 +361,13 @@ define amdgpu_kernel void @s_fneg_v2bf16(ptr addrspace(1) %out, <2 x bfloat> %in ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_and_b32 s3, s2, 0xffff0000 -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s3 -; CI-NEXT: v_mul_f32_e64 v1, -1.0, s2 -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; CI-NEXT: v_alignbit_b32 v2, v0, v1, 16 +; CI-NEXT: s_xor_b32 s2, s2, 0x8000 +; CI-NEXT: s_and_b32 s2, s2, 0xffff +; CI-NEXT: s_or_b32 s2, s2, s3 +; CI-NEXT: s_add_i32 s2, s2, 0x80000000 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -426,16 +422,16 @@ define amdgpu_kernel void @s_fneg_v2bf16_nonload(ptr addrspace(1) %out) #0 { ; CI-NEXT: ; def s2 ; CI-NEXT: ;;#ASMEND ; CI-NEXT: s_and_b32 s3, s2, 0xffff0000 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s3 -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; CI-NEXT: v_mul_f32_e64 v1, -1.0, s2 -; CI-NEXT: v_alignbit_b32 v2, v0, v1, 16 +; CI-NEXT: s_xor_b32 s2, s2, 0x8000 +; CI-NEXT: s_and_b32 s2, s2, 0xffff +; CI-NEXT: s_or_b32 s2, s2, s3 +; CI-NEXT: s_add_i32 s2, s2, 0x80000000 ; CI-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: s_mov_b32 flat_scratch_lo, s13 ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -501,13 +497,11 @@ define amdgpu_kernel void @v_fneg_v2bf16(ptr addrspace(1) %out, ptr addrspace(1) ; CI-NEXT: v_add_i32_e32 v0, vcc, s0, v0 ; CI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; CI-NEXT: flat_load_dword v2, v[0:1] +; CI-NEXT: s_mov_b32 s0, 0xffff ; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 -; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_mul_f32_e32 v3, -1.0, v3 -; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v3 -; CI-NEXT: v_mul_f32_e32 v2, -1.0, v2 -; CI-NEXT: v_alignbit_b32 v2, v3, v2, 16 +; CI-NEXT: v_xor_b32_e32 v3, 0x8000, v2 +; CI-NEXT: v_bfi_b32 v2, s0, v3, v2 +; CI-NEXT: v_add_i32_e32 v2, vcc, 0x80000000, v2 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -570,13 +564,13 @@ define amdgpu_kernel void @fneg_free_v2bf16(ptr addrspace(1) %out, i32 %in) #0 { ; CI-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8 ; CI-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_and_b32 s3, s2, 0xffff0000 -; CI-NEXT: s_lshl_b32 s2, s2, 16 -; CI-NEXT: v_mul_f32_e64 v0, -1.0, s3 -; CI-NEXT: v_mul_f32_e64 v1, -1.0, s2 -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; CI-NEXT: v_alignbit_b32 v2, v0, v1, 16 +; CI-NEXT: s_xor_b32 s2, s2, 0x8000 +; CI-NEXT: s_and_b32 s2, s2, 0xffff +; CI-NEXT: s_or_b32 s2, s2, s3 +; CI-NEXT: s_add_i32 s2, s2, 0x80000000 ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 +; CI-NEXT: v_mov_b32_e32 v2, s2 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -637,16 +631,14 @@ define amdgpu_kernel void @v_fneg_fold_v2bf16(ptr addrspace(1) %out, ptr addrspa ; CI-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 -; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2 -; CI-NEXT: v_mul_f32_e32 v4, -1.0, v3 -; CI-NEXT: v_mul_f32_e32 v5, -1.0, v2 -; CI-NEXT: v_and_b32_e32 v4, 0xffff0000, v4 -; CI-NEXT: v_and_b32_e32 v5, 0xffff0000, v5 -; CI-NEXT: v_mul_f32_e32 v3, v4, v3 -; CI-NEXT: v_mul_f32_e32 v2, v5, v2 -; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v3 -; CI-NEXT: v_alignbit_b32 v2, v3, v2, 16 +; CI-NEXT: v_xor_b32_e32 v3, 0x8000, v2 +; CI-NEXT: v_lshlrev_b32_e32 v4, 16, v2 +; CI-NEXT: v_and_b32_e32 v2, 0xffff0000, v2 +; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; CI-NEXT: v_mul_f32_e64 v2, -v2, v2 +; CI-NEXT: v_mul_f32_e32 v3, v3, v4 +; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 +; CI-NEXT: v_alignbit_b32 v2, v2, v3, 16 ; CI-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: s_endpgm ; @@ -912,12 +904,9 @@ define amdgpu_kernel void @v_extract_fneg_no_fold_v2bf16(ptr addrspace(1) %in) # ; CI-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: flat_load_dword v0, v[0:1] ; CI-NEXT: s_waitcnt vmcnt(0) -; CI-NEXT: v_and_b32_e32 v1, 0xffff0000, v0 -; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; CI-NEXT: v_mul_f32_e32 v1, -1.0, v1 -; CI-NEXT: v_mul_f32_e32 v0, -1.0, v0 -; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1 -; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0 +; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; CI-NEXT: v_xor_b32_e32 v0, 0x8000, v0 +; CI-NEXT: v_xor_b32_e32 v1, 0x8000, v1 ; CI-NEXT: flat_store_short v[0:1], v0 ; CI-NEXT: s_waitcnt vmcnt(0) ; CI-NEXT: flat_store_short v[0:1], v1 diff --git a/llvm/test/CodeGen/AMDGPU/fptoui_uitofp.ll b/llvm/test/CodeGen/AMDGPU/fptoui_uitofp.ll new file mode 100644 index 0000000..49204f8 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/fptoui_uitofp.ll @@ -0,0 +1,296 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=amdgcn -mcpu=gfx600 < %s | FileCheck -check-prefix=GFX6 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s + +define amdgpu_kernel void @fptoui_f32_to_i16_to_f32(ptr addrspace(1) %out, float %x) { +; GFX6-LABEL: fptoui_f32_to_i16_to_f32: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dword s6, s[4:5], 0xb +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_trunc_f32_e64 v0, |s6| +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f32_to_i16_to_f32: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f32_e64 v1, |s2| +; GFX9-NEXT: global_store_dword v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui float %x to i16 + %fp = uitofp i16 %ui to float + store float %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f32_to_i32_to_f32(ptr addrspace(1) %out, float %x) { +; GFX6-LABEL: fptoui_f32_to_i32_to_f32: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dword s6, s[4:5], 0xb +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_trunc_f32_e64 v0, |s6| +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f32_to_i32_to_f32: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f32_e64 v1, |s2| +; GFX9-NEXT: global_store_dword v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui float %x to i32 + %fp = uitofp i32 %ui to float + store float %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f32_to_i64_to_f32(ptr addrspace(1) %out, float %x) { +; GFX6-LABEL: fptoui_f32_to_i64_to_f32: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dword s6, s[4:5], 0xb +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_trunc_f32_e64 v0, |s6| +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f32_to_i64_to_f32: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f32_e64 v1, |s2| +; GFX9-NEXT: global_store_dword v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui float %x to i64 + %fp = uitofp i64 %ui to float + store float %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f16_to_i16_to_f16(ptr addrspace(1) %out, half %x) { +; GFX6-LABEL: fptoui_f16_to_i16_to_f16: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dword s0, s[4:5], 0xb +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_cvt_f32_f16_e32 v0, s0 +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: buffer_store_short v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f16_to_i16_to_f16: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f16_e64 v1, |s2| +; GFX9-NEXT: global_store_short v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui half %x to i16 + %fp = uitofp i16 %ui to half + store half %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f16_to_i32_to_f16(ptr addrspace(1) %out, half %x) { +; GFX6-LABEL: fptoui_f16_to_i32_to_f16: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dword s0, s[4:5], 0xb +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_cvt_f32_f16_e64 v0, |s0| +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GFX6-NEXT: v_trunc_f32_e32 v0, v0 +; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: buffer_store_short v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f16_to_i32_to_f16: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f16_e64 v1, |s2| +; GFX9-NEXT: global_store_short v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui half %x to i32 + %fp = uitofp i32 %ui to half + store half %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f16_to_i64_to_f16(ptr addrspace(1) %out, half %x) { +; GFX6-LABEL: fptoui_f16_to_i64_to_f16: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dword s0, s[4:5], 0xb +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_cvt_f32_f16_e64 v0, |s0| +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GFX6-NEXT: v_trunc_f32_e32 v0, v0 +; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: buffer_store_short v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f16_to_i64_to_f16: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dword s2, s[4:5], 0x2c +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f16_e64 v1, |s2| +; GFX9-NEXT: global_store_short v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui half %x to i64 + %fp = uitofp i64 %ui to half + store half %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f64_to_i16_to_f64(ptr addrspace(1) %out, double %x) { +; GFX6-LABEL: fptoui_f64_to_i16_to_f64: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_cvt_u32_f64_e32 v0, s[2:3] +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_cvt_f64_u32_e32 v[0:1], v0 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f64_to_i16_to_f64: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f64_e64 v[0:1], |s[2:3]| +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui double %x to i16 + %fp = uitofp i16 %ui to double + store double %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f64_to_i32_to_f64(ptr addrspace(1) %out, double %x) { +; GFX6-LABEL: fptoui_f64_to_i32_to_f64: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: v_cvt_u32_f64_e32 v0, s[2:3] +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_cvt_f64_u32_e32 v[0:1], v0 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f64_to_i32_to_f64: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f64_e64 v[0:1], |s[2:3]| +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui double %x to i32 + %fp = uitofp i32 %ui to double + store double %fp, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @fptoui_f64_to_i64_to_f64(ptr addrspace(1) %out, double %x) { +; GFX6-LABEL: fptoui_f64_to_i64_to_f64: +; GFX6: ; %bb.0: ; %entry +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s5, 0xfffff +; GFX6-NEXT: s_mov_b32 s4, s6 +; GFX6-NEXT: v_not_b32_e32 v0, 31 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_bfe_u32 s7, s3, 0xb0014 +; GFX6-NEXT: s_addk_i32 s7, 0xfc01 +; GFX6-NEXT: s_lshr_b64 s[4:5], s[4:5], s7 +; GFX6-NEXT: s_and_b32 s8, s3, 0x80000000 +; GFX6-NEXT: s_andn2_b64 s[4:5], s[2:3], s[4:5] +; GFX6-NEXT: s_cmp_lt_i32 s7, 0 +; GFX6-NEXT: s_cselect_b32 s4, 0, s4 +; GFX6-NEXT: s_cselect_b32 s5, s8, s5 +; GFX6-NEXT: s_cmp_gt_i32 s7, 51 +; GFX6-NEXT: s_cselect_b32 s3, s3, s5 +; GFX6-NEXT: s_cselect_b32 s2, s2, s4 +; GFX6-NEXT: v_ldexp_f64 v[0:1], s[2:3], v0 +; GFX6-NEXT: v_mov_b32_e32 v4, -1 +; GFX6-NEXT: v_fract_f64_e32 v[2:3], v[0:1] +; GFX6-NEXT: v_mov_b32_e32 v5, 0x3fefffff +; GFX6-NEXT: v_min_f64 v[2:3], v[2:3], v[4:5] +; GFX6-NEXT: v_cmp_class_f64_e64 vcc, v[0:1], 3 +; GFX6-NEXT: s_mov_b32 s4, 0 +; GFX6-NEXT: v_cndmask_b32_e32 v2, v2, v0, vcc +; GFX6-NEXT: v_cndmask_b32_e32 v3, v3, v1, vcc +; GFX6-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3] +; GFX6-NEXT: v_mov_b32_e32 v2, s2 +; GFX6-NEXT: s_mov_b32 s5, 0xc1f00000 +; GFX6-NEXT: v_mov_b32_e32 v3, s3 +; GFX6-NEXT: v_fma_f64 v[2:3], v[0:1], s[4:5], v[2:3] +; GFX6-NEXT: v_cvt_u32_f64_e32 v0, v[0:1] +; GFX6-NEXT: v_cvt_u32_f64_e32 v2, v[2:3] +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: v_cvt_f64_u32_e32 v[0:1], v0 +; GFX6-NEXT: v_cvt_f64_u32_e32 v[2:3], v2 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: v_ldexp_f64 v[0:1], v[0:1], 32 +; GFX6-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3] +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX9-LABEL: fptoui_f64_to_i64_to_f64: +; GFX9: ; %bb.0: ; %entry +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: v_trunc_f64_e64 v[0:1], |s[2:3]| +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] +; GFX9-NEXT: s_endpgm +entry: + %ui = fptoui double %x to i64 + %fp = uitofp i64 %ui to double + store double %fp, ptr addrspace(1) %out + ret void +} diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.private.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.private.ll index 67d0410..3324018 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.private.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.private.ll @@ -3,11 +3,11 @@ ; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-SDAG %s ; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s ; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s define amdgpu_kernel void @is_private_vgpr(ptr addrspace(1) %ptr.ptr) { ; SI-LABEL: is_private_vgpr: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.shared.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.shared.ll index 63333ed..355d002 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.shared.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.is.shared.ll @@ -3,11 +3,11 @@ ; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-SDAG %s ; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s ; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s -; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-GISEL %s define amdgpu_kernel void @is_local_vgpr(ptr addrspace(1) %ptr.ptr) { ; CIT-LABEL: is_local_vgpr: diff --git a/llvm/test/CodeGen/AMDGPU/s_cmp_0.ll b/llvm/test/CodeGen/AMDGPU/s_cmp_0.ll new file mode 100644 index 0000000..f53aaaa --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/s_cmp_0.ll @@ -0,0 +1,625 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck %s + +declare i32 @llvm.ctpop.i32(i32) +declare i64 @llvm.ctpop.i64(i64) +declare i32 @llvm.amdgcn.s.quadmask.i32(i32) +declare i64 @llvm.amdgcn.s.quadmask.i64(i64) + +define amdgpu_ps i32 @shl32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: shl32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_lshl_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = shl i32 %val0, %val1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @shl64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: shl64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_lshl_b64 s[0:1], s[0:1], s2 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = shl i64 %val0, %val1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @lshr32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: lshr32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_lshr_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = lshr i32 %val0, %val1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @lshr64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: lshr64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_lshr_b64 s[0:1], s[0:1], s2 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = lshr i64 %val0, %val1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @ashr32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: ashr32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_ashr_i32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = ashr i32 %val0, %val1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @ashr64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: ashr64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_ashr_i64 s[0:1], s[0:1], s2 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = ashr i64 %val0, %val1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @abs32(i32 inreg %val0) { +; CHECK-LABEL: abs32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_abs_i32 s0, s0 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %neg = sub i32 0, %val0 + %cond = icmp sgt i32 %val0, %neg + %result = select i1 %cond, i32 %val0, i32 %neg + call void asm "; use $0", "s"(i32 %result) + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @and32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: and32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_and_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = and i32 %val0, %val1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @and64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: and64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_and_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = and i64 %val0, %val1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @or32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: or32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_or_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = or i32 %val0, %val1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @or64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: or64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = or i64 %val0, %val1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @xor32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: xor32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_xor_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = xor i32 %val0, %val1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @xor64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: xor64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_xor_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = xor i64 %val0, %val1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @nand32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: nand32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_nand_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = and i32 %val0, %val1 + %result2 = xor i32 %result, -1 + call void asm "; use $0", "s"(i32 %result2) + %cmp = icmp ne i32 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @nand64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: nand64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_nand_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = and i64 %val0, %val1 + %result2 = xor i64 %result, -1 + call void asm "; use $0", "s"(i64 %result2) + %cmp = icmp ne i64 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @nor32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: nor32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_nor_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = or i32 %val0, %val1 + %result2 = xor i32 %result, -1 + call void asm "; use $0", "s"(i32 %result2) + %cmp = icmp ne i32 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @nor64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: nor64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_nor_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = or i64 %val0, %val1 + %result2 = xor i64 %result, -1 + call void asm "; use $0", "s"(i64 %result2) + %cmp = icmp ne i64 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @xnor32(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: xnor32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_xnor_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = xor i32 %val0, %val1 + %result2 = xor i32 %result, -1 + call void asm "; use $0", "s"(i32 %result2) + %cmp = icmp ne i32 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @xnor64(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: xnor64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_xnor_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = xor i64 %val0, %val1 + %result2 = xor i64 %result, -1 + call void asm "; use $0", "s"(i64 %result2) + %cmp = icmp ne i64 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @andn232(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: andn232: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_andn2_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %nval1 = xor i32 %val1, -1 + %result = and i32 %val0, %nval1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @nandn264(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: nandn264: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_andn2_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %nval1 = xor i64 %val1, -1 + %result = and i64 %val0, %nval1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @orn232(i32 inreg %val0, i32 inreg %val1) { +; CHECK-LABEL: orn232: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_orn2_b32 s0, s0, s1 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %nval1 = xor i32 %val1, -1 + %result = or i32 %val0, %nval1 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @orn264(i64 inreg %val0, i64 inreg %val1) { +; CHECK-LABEL: orn264: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_orn2_b64 s[0:1], s[0:1], s[2:3] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %nval1 = xor i64 %val1, -1 + %result = or i64 %val0, %nval1 + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bfe_i32(i32 inreg %val0) { +; CHECK-LABEL: bfe_i32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bfe_i32 s0, s0, 0x80010 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %shl = shl i32 %val0, 8 + %result = ashr i32 %shl, 24 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bfe_i64(i64 inreg %val0) { +; CHECK-LABEL: bfe_i64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bfe_i64 s[2:3], s[0:1], 0x80000 +; CHECK-NEXT: s_and_b32 s0, s0, 0xff +; CHECK-NEXT: s_mov_b32 s1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[2:3] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: ; return to shader part epilog + %shl = shl i64 %val0, 56 + %result = ashr i64 %shl, 56 + call void asm "; use $0", "s"(i64 %result) + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bfe_u32(i32 inreg %val0) { +; CHECK-LABEL: bfe_u32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bfe_u32 s0, s0, 0x80010 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %shl = shl i32 %val0, 8 + %result = lshr i32 %shl, 24 + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bfe_u64(i64 inreg %val0) { +; CHECK-LABEL: bfe_u64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_and_b32 s0, s0, 0xff +; CHECK-NEXT: s_mov_b32 s1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %shl = shl i64 %val0, 56 + %result = lshr i64 %shl, 56 + call void asm "; use $0", "s"(i64 %result) + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bcnt032(i32 inreg %val0) { +; CHECK-LABEL: bcnt032: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bcnt1_i32_b32 s0, s0 +; CHECK-NEXT: s_sub_i32 s0, 32, s0 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = call i32 @llvm.ctpop.i32(i32 %val0) nounwind readnone + %result2 = sub i32 32, %result + call void asm "; use $0", "s"(i32 %result2) + %cmp = icmp ne i32 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bcnt064(i64 inreg %val0) { +; CHECK-LABEL: bcnt064: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bcnt1_i32_b64 s0, s[0:1] +; CHECK-NEXT: s_sub_u32 s0, 64, s0 +; CHECK-NEXT: s_subb_u32 s1, 0, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = call i64 @llvm.ctpop.i64(i64 %val0) nounwind readnone + %result2 = sub i64 64, %result + call void asm "; use $0", "s"(i64 %result2) + %cmp = icmp ne i64 %result2, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bcnt132(i32 inreg %val0) { +; CHECK-LABEL: bcnt132: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bcnt1_i32_b32 s0, s0 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = call i32 @llvm.ctpop.i32(i32 %val0) nounwind readnone + call void asm "; use $0", "s"(i32 %result) + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @bcnt164(i64 inreg %val0) { +; CHECK-LABEL: bcnt164: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_bcnt1_i32_b64 s0, s[0:1] +; CHECK-NEXT: s_mov_b32 s1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = call i64 @llvm.ctpop.i64(i64 %val0) nounwind readnone + call void asm "; use $0", "s"(i64 %result) + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @quadmask32(i32 inreg %val0) { +; CHECK-LABEL: quadmask32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_quadmask_b32 s0, s0 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = call i32 @llvm.amdgcn.s.quadmask.i32(i32 %val0) nounwind readnone + call void asm "; use $0", "s"(i32 %result) + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @quadmask64(i64 inreg %val0) { +; CHECK-LABEL: quadmask64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_quadmask_b64 s[0:1], s[0:1] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = call i64 @llvm.amdgcn.s.quadmask.i64(i64 %val0) nounwind readnone + call void asm "; use $0", "s"(i64 %result) + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @not32(i32 inreg %val0) { +; CHECK-LABEL: not32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_not_b32 s0, s0 +; CHECK-NEXT: s_cmp_lg_u32 s0, 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s0 +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = xor i32 %val0, -1 + call void asm "; use $0", "s"(i32 %result) + %cmp = icmp ne i32 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} + +define amdgpu_ps i32 @not64(i64 inreg %val0) { +; CHECK-LABEL: not64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_not_b64 s[0:1], s[0:1] +; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use s[0:1] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v0 +; CHECK-NEXT: ; return to shader part epilog + %result = xor i64 %val0, -1 + call void asm "; use $0", "s"(i64 %result) + %cmp = icmp ne i64 %result, 0 + %zext = zext i1 %cmp to i32 + ret i32 %zext +} diff --git a/llvm/test/CodeGen/AMDGPU/unpack-non-coissue-insts-post-ra-scheduler.mir b/llvm/test/CodeGen/AMDGPU/unpack-non-coissue-insts-post-ra-scheduler.mir index 8b467eb..75ae76f 100644 --- a/llvm/test/CodeGen/AMDGPU/unpack-non-coissue-insts-post-ra-scheduler.mir +++ b/llvm/test/CodeGen/AMDGPU/unpack-non-coissue-insts-post-ra-scheduler.mir @@ -1,7 +1,7 @@ -# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5 +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6 # RUN: llc -mtriple amdgcn-amd-amdhsa -mcpu=gfx950 -run-pass=si-pre-emit-peephole -o - %s | FileCheck -check-prefix=GFX950 %s # RUN: llc -mtriple amdgcn-amd-amdhsa -mcpu=gfx942 -run-pass=si-pre-emit-peephole -o - %s | FileCheck -check-prefix=GFX942 %s -# RUN: llc -mtriple amdgcn-amd-amdhsa -mcpu=gfx90a -run-pass=si-pre-emit-peephole -o - %s | FileCheck -check-prefix=GFX90a %s +# RUN: llc -mtriple amdgcn-amd-amdhsa -mcpu=gfx90a -run-pass=si-pre-emit-peephole -o - %s | FileCheck -check-prefix=GFX90A %s --- name: test_pk_mul_unpacking_f32 @@ -57,26 +57,26 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_MUL_F32_e64 0, killed $sgpr31, 0, killed $vgpr5, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_pk_mul_unpacking_f32 - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_pk_mul_unpacking_f32 + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec S_WAITCNT 49279 @@ -150,26 +150,26 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_MUL_F32_e64 0, killed $sgpr31, 0, killed $vgpr5, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_op_sel_selection_unpacking_f32 - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_op_sel_selection_unpacking_f32 + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec S_WAITCNT 49279 @@ -243,26 +243,26 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_MUL_F32_e64 0, killed $sgpr30, 0, killed $vgpr5, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_op_sel_hi_selection_unpacking_f32 - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 0, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_op_sel_hi_selection_unpacking_f32 + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 0, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec S_WAITCNT 49279 @@ -370,41 +370,41 @@ body: | ; GFX942-NEXT: renamable $vgpr10_vgpr11 = nofpexcept V_PK_ADD_F32 8, killed $sgpr10_sgpr11, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_pk_add_unpacking_f32 - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr4 = V_MOV_B32_e32 2, implicit $exec - ; GFX90a-NEXT: renamable $vgpr5 = V_MOV_B32_e32 1, implicit $exec - ; GFX90a-NEXT: renamable $vgpr2 = V_MOV_B32_e32 4, implicit $exec - ; GFX90a-NEXT: renamable $vgpr3 = V_MOV_B32_e32 3, implicit $exec - ; GFX90a-NEXT: renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: renamable $vgpr16 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 $sgpr15, implicit $exec, implicit-def $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 $sgpr14, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 $sgpr13, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 $sgpr12, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 $sgpr11, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 $sgpr10, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 $sgpr9, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 $sgpr8, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 $sgpr7, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 $sgpr6, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 $sgpr5, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 $sgpr4, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 $sgpr3, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 $sgpr2, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 $sgpr1, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GFX90a-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 $sgpr0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, implicit $exec - ; GFX90a-NEXT: $vgpr0 = V_MOV_B32_e32 killed $sgpr18, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr1 = V_MOV_B32_e32 killed $sgpr19, implicit $exec, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_BF8_BF8_e64 killed $vgpr4_vgpr5, killed $vgpr2_vgpr3, killed $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31, 1, 2, 3, implicit $mode, implicit $exec, implicit $mode, implicit $exec - ; GFX90a-NEXT: renamable $vgpr2_vgpr3 = nofpexcept V_PK_ADD_F32 8, killed $sgpr2_sgpr3, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: renamable $vgpr6_vgpr7 = nofpexcept V_PK_ADD_F32 8, killed $sgpr6_sgpr7, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_ADD_F32 8, killed $sgpr4_sgpr5, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: renamable $vgpr10_vgpr11 = nofpexcept V_PK_ADD_F32 8, killed $sgpr10_sgpr11, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_pk_add_unpacking_f32 + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr4 = V_MOV_B32_e32 2, implicit $exec + ; GFX90A-NEXT: renamable $vgpr5 = V_MOV_B32_e32 1, implicit $exec + ; GFX90A-NEXT: renamable $vgpr2 = V_MOV_B32_e32 4, implicit $exec + ; GFX90A-NEXT: renamable $vgpr3 = V_MOV_B32_e32 3, implicit $exec + ; GFX90A-NEXT: renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: renamable $vgpr16 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 $sgpr15, implicit $exec, implicit-def $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 $sgpr14, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr29 = V_ACCVGPR_WRITE_B32_e64 $sgpr13, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr28 = V_ACCVGPR_WRITE_B32_e64 $sgpr12, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr27 = V_ACCVGPR_WRITE_B32_e64 $sgpr11, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr26 = V_ACCVGPR_WRITE_B32_e64 $sgpr10, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr25 = V_ACCVGPR_WRITE_B32_e64 $sgpr9, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr24 = V_ACCVGPR_WRITE_B32_e64 $sgpr8, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr23 = V_ACCVGPR_WRITE_B32_e64 $sgpr7, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr22 = V_ACCVGPR_WRITE_B32_e64 $sgpr6, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr21 = V_ACCVGPR_WRITE_B32_e64 $sgpr5, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr20 = V_ACCVGPR_WRITE_B32_e64 $sgpr4, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr19 = V_ACCVGPR_WRITE_B32_e64 $sgpr3, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr18 = V_ACCVGPR_WRITE_B32_e64 $sgpr2, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr17 = V_ACCVGPR_WRITE_B32_e64 $sgpr1, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GFX90A-NEXT: $agpr16 = V_ACCVGPR_WRITE_B32_e64 $sgpr0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, implicit $exec + ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 killed $sgpr18, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr1 = V_MOV_B32_e32 killed $sgpr19, implicit $exec, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_BF8_BF8_e64 killed $vgpr4_vgpr5, killed $vgpr2_vgpr3, killed $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31, 1, 2, 3, implicit $mode, implicit $exec, implicit $mode, implicit $exec + ; GFX90A-NEXT: renamable $vgpr2_vgpr3 = nofpexcept V_PK_ADD_F32 8, killed $sgpr2_sgpr3, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: renamable $vgpr6_vgpr7 = nofpexcept V_PK_ADD_F32 8, killed $sgpr6_sgpr7, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_ADD_F32 8, killed $sgpr4_sgpr5, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: renamable $vgpr10_vgpr11 = nofpexcept V_PK_ADD_F32 8, killed $sgpr10_sgpr11, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr4 = V_MOV_B32_e32 2, implicit $exec renamable $vgpr5 = V_MOV_B32_e32 1, implicit $exec @@ -438,7 +438,6 @@ body: | renamable $vgpr10_vgpr11 = nofpexcept V_PK_ADD_F32 8, killed $sgpr10_sgpr11, 8, $vgpr0_vgpr1, 0, 0, 0, 0, 0, implicit $mode, implicit $exec S_ENDPGM 0 - ... --- name: test_pk_fma_unpacking_f32 @@ -490,24 +489,24 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_FMA_F32_e64 0, killed $sgpr30, 0, killed $vgpr5, 0, $vgpr5, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_pk_fma_unpacking_f32 - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 8, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_pk_fma_unpacking_f32 + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 8, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 @@ -577,25 +576,25 @@ body: | ; GFX942-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_FMA_F32 8, killed $sgpr30_sgpr31, 8, $vgpr4_vgpr5, 0, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_unpacking_does_not_introduce_rw_dependency - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_MUL_F32 8, $sgpr30_sgpr31, 12, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_FMA_F32 8, killed $sgpr30_sgpr31, 8, $vgpr4_vgpr5, 0, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_unpacking_does_not_introduce_rw_dependency + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_MUL_F32 8, $sgpr30_sgpr31, 12, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: renamable $vgpr4_vgpr5 = nofpexcept V_PK_FMA_F32 8, killed $sgpr30_sgpr31, 8, $vgpr4_vgpr5, 0, $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 @@ -670,28 +669,28 @@ body: | ; GFX942-NEXT: $vgpr10_vgpr11 = V_PK_MOV_B32 12, killed $vgpr6_vgpr7, 8, killed $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_opcodes_not_supported_for_unpacking_are_skipped - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_dpp $vgpr4, $vgpr4, 228, 15, 15, -1, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_CVT_PK_BF8_F32_e64 0, killed $vgpr4, 0, $vgpr4, $vgpr5, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: $vgpr8_vgpr9 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: $vgpr10_vgpr11 = V_PK_MOV_B32 12, killed $vgpr6_vgpr7, 8, killed $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_opcodes_not_supported_for_unpacking_are_skipped + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_dpp $vgpr4, $vgpr4, 228, 15, 15, -1, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_CVT_PK_BF8_F32_e64 0, killed $vgpr4, 0, $vgpr4, $vgpr5, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: $vgpr8_vgpr9 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: $vgpr10_vgpr11 = V_PK_MOV_B32 12, killed $vgpr6_vgpr7, 8, killed $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 @@ -705,14 +704,11 @@ body: | early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - $vgpr4 = V_MOV_B32_dpp $vgpr4, $vgpr4, 228, 15, 15, -1, implicit $exec $vgpr5 = V_CVT_PK_BF8_F32_e64 0, killed $vgpr4, 0, $vgpr4, $vgpr5, 0, implicit $mode, implicit $exec - $vgpr6_vgpr7 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec $vgpr8_vgpr9 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec $vgpr10_vgpr11 = V_PK_MOV_B32 12, killed $vgpr6_vgpr7, 8, killed $vgpr8_vgpr9, 0, 0, 0, 0, 0, implicit $exec - S_ENDPGM 0 ... @@ -770,26 +766,26 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_FMA_F32_e64 0, killed $sgpr30, 0, killed $vgpr5, 0, killed $vgpr7, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_opsel_register_is_correctly_marked_as_killed - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr6 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr7 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 8, killed $vgpr6_vgpr7, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_opsel_register_is_correctly_marked_as_killed + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr6 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr7 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 8, killed $vgpr6_vgpr7, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 @@ -861,26 +857,26 @@ body: | ; GFX942-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 8, killed $vgpr6_vgpr7, 0, 0, 0, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_inst_dependent_on_mfma_are_not_unpacked - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec - ; GFX90a-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec - ; GFX90a-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 $agpr3, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 8, killed $vgpr6_vgpr7, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_inst_dependent_on_mfma_are_not_unpacked + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec + ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec + ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 $agpr3, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 0, killed $sgpr30_sgpr31, 12, killed $vgpr4_vgpr5, 8, killed $vgpr6_vgpr7, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 @@ -950,25 +946,25 @@ body: | ; GFX942-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 8, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_mfma_def_using_instr_blocks_unpacking - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 $agpr3, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 8, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_mfma_def_using_instr_blocks_unpacking + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 $agpr3, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_FMA_F32 8, killed $sgpr30_sgpr31, 8, killed $vgpr4_vgpr5, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 @@ -1041,26 +1037,26 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, killed $vgpr5, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_unpacking_with_imm_input - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, 1065353216, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_unpacking_with_imm_input + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, 1065353216, 8, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec S_WAITCNT 49279 @@ -1134,26 +1130,26 @@ body: | ; GFX942-NEXT: $vgpr17 = nofpexcept V_MUL_F32_e64 0, killed $sgpr31, 1, killed $vgpr5, 0, 0, implicit $mode, implicit $exec ; GFX942-NEXT: S_ENDPGM 0 ; - ; GFX90a-LABEL: name: test_neg_lo_hi_post_unpacking - ; GFX90a: liveins: $sgpr4_sgpr5 - ; GFX90a-NEXT: {{ $}} - ; GFX90a-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 - ; GFX90a-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 - ; GFX90a-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 - ; GFX90a-NEXT: S_WAITCNT 49279 - ; GFX90a-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 - ; GFX90a-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec - ; GFX90a-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 - ; GFX90a-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec - ; GFX90a-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec - ; GFX90a-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec - ; GFX90a-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, killed $sgpr30_sgpr31, 11, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec - ; GFX90a-NEXT: S_ENDPGM 0 + ; GFX90A-LABEL: name: test_neg_lo_hi_post_unpacking + ; GFX90A: liveins: $sgpr4_sgpr5 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 + ; GFX90A-NEXT: renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47 = S_LOAD_DWORDX4_IMM renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: renamable $sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX4_IMM renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr42_sgpr43, 0, 0 + ; GFX90A-NEXT: early-clobber renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31 = S_LOAD_DWORDX16_IMM_ec killed renamable $sgpr40_sgpr41, 0, 0 + ; GFX90A-NEXT: S_WAITCNT 49279 + ; GFX90A-NEXT: $vgpr0_vgpr1 = V_MOV_B64_e32 $sgpr44_sgpr45, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $sgpr44_sgpr45_sgpr46_sgpr47 + ; GFX90A-NEXT: $vgpr2_vgpr3 = V_MOV_B64_e32 killed $sgpr46_sgpr47, implicit $exec, implicit $sgpr44_sgpr45_sgpr46_sgpr47, implicit $exec + ; GFX90A-NEXT: $vgpr4_vgpr5 = V_MOV_B64_e32 $sgpr48_sgpr49, implicit $exec, implicit-def $vgpr4_vgpr5_vgpr6_vgpr7, implicit $sgpr48_sgpr49_sgpr50_sgpr51 + ; GFX90A-NEXT: $vgpr6_vgpr7 = V_MOV_B64_e32 killed $sgpr50_sgpr51, implicit $exec, implicit $sgpr48_sgpr49_sgpr50_sgpr51, implicit $exec + ; GFX90A-NEXT: early-clobber renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_32X32X16_F16_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3, killed $vgpr4_vgpr5_vgpr6_vgpr7, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: $vgpr4 = V_MOV_B32_e32 killed $sgpr14, implicit $exec, implicit $exec + ; GFX90A-NEXT: $vgpr5 = V_MOV_B32_e32 killed $sgpr15, implicit $exec, implicit $exec + ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = nofpexcept V_PK_MUL_F32 8, killed $sgpr30_sgpr31, 11, killed $vgpr4_vgpr5, 0, 0, 0, 0, 0, implicit $mode, implicit $exec + ; GFX90A-NEXT: S_ENDPGM 0 early-clobber renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM_ec killed renamable $sgpr4_sgpr5, 0, 0 renamable $vgpr18 = V_MOV_B32_e32 0, implicit $exec S_WAITCNT 49279 diff --git a/llvm/test/CodeGen/ARM/call-graph-section-addrtaken.ll b/llvm/test/CodeGen/ARM/call-graph-section-addrtaken.ll new file mode 100644 index 0000000..a2d6ca9 --- /dev/null +++ b/llvm/test/CodeGen/ARM/call-graph-section-addrtaken.ll @@ -0,0 +1,39 @@ +;; Test if a potential indirect call target function which has internal linkage and +;; address taken has its type ID emitted to callgraph section. +;; This test also makes sure that callback functions which meet the above constraint +;; are handled correctly. + +; RUN: llc -mtriple=arm-unknown-linux --call-graph-section -o - < %s | FileCheck %s + +declare !type !0 void @_Z6doWorkPFviE(ptr) + +define i32 @_Z4testv() !type !1 { +entry: + call void @_Z6doWorkPFviE(ptr nonnull @_ZL10myCallbacki) + ret i32 0 +} + +; CHECK: _ZL10myCallbacki: +; CHECK-NEXT: [[LABEL_FUNC:\.Lfunc_begin[0-9]+]]: +define internal void @_ZL10myCallbacki(i32 %value) !type !2 { +entry: + %sink = alloca i32, align 4 + store volatile i32 %value, ptr %sink, align 4 + %i1 = load volatile i32, ptr %sink, align 4 + ret void +} + +!0 = !{i64 0, !"_ZTSFvPFviEE.generalized"} +!1 = !{i64 0, !"_ZTSFivE.generalized"} +!2 = !{i64 0, !"_ZTSFviE.generalized"} + +; CHECK: .section .callgraph,"o",%progbits,.text +;; Version +; CHECK-NEXT: .byte 0 +;; Flags -- Potential indirect target so LSB is set to 1. Other bits are 0. +; CHECK-NEXT: .byte 1 +;; Function Entry PC +; CHECK-NEXT: .long [[LABEL_FUNC]] +;; Function type ID -5212364466660467813 +; CHECK-NEXT: .long 1154849691 +; CHECK-NEXT: .long 3081369122 diff --git a/llvm/test/CodeGen/ARM/call-graph-section-assembly.ll b/llvm/test/CodeGen/ARM/call-graph-section-assembly.ll new file mode 100644 index 0000000..bf5249e --- /dev/null +++ b/llvm/test/CodeGen/ARM/call-graph-section-assembly.ll @@ -0,0 +1,63 @@ +;; Test if temporary labels are generated for each indirect callsite. +;; Test if the .callgraph section contains the MD5 hash of callees' type (type id) +;; is correctly paired with its corresponding temporary label generated for indirect +;; call sites annotated with !callee_type metadata. +;; Test if the .callgraph section contains unique direct callees. + +; RUN: llc -mtriple=arm-unknown-linux --call-graph-section -o - < %s | FileCheck %s + +declare !type !0 void @direct_foo() +declare !type !1 i32 @direct_bar(i8) +declare !type !2 ptr @direct_baz(ptr) + +; CHECK: ball: +; CHECK-NEXT: [[LABEL_FUNC:\.Lfunc_begin[0-9]+]]: +define ptr @ball() { +entry: + call void @direct_foo() + %fp_foo_val = load ptr, ptr null, align 8 + call void (...) %fp_foo_val(), !callee_type !0 + call void @direct_foo() + %fp_bar_val = load ptr, ptr null, align 8 + %call_fp_bar = call i32 %fp_bar_val(i8 0), !callee_type !2 + %call_fp_bar_direct = call i32 @direct_bar(i8 1) + %fp_baz_val = load ptr, ptr null, align 8 + %call_fp_baz = call ptr %fp_baz_val(ptr null), !callee_type !4 + call void @direct_foo() + %call_fp_baz_direct = call ptr @direct_baz(ptr null) + call void @direct_foo() + ret ptr %call_fp_baz +} + +!0 = !{!1} +!1 = !{i64 0, !"_ZTSFvE.generalized"} +!2 = !{!3} +!3 = !{i64 0, !"_ZTSFicE.generalized"} +!4 = !{!5} +!5 = !{i64 0, !"_ZTSFPvS_E.generalized"} + +; CHECK: .section .callgraph,"o",%progbits,.text +;; Version +; CHECK-NEXT: .byte 0 +;; Flags +; CHECK-NEXT: .byte 7 +;; Function Entry PC +; CHECK-NEXT: .long [[LABEL_FUNC]] +;; Function type ID -- set to 0 as no type metadata attached to function. +; CHECK-NEXT: .long 0 +; CHECK-NEXT: .long 0 +;; Number of unique direct callees. +; CHECK-NEXT: .byte 3 +;; Direct callees. +; CHECK-NEXT: .long direct_foo +; CHECK-NEXT: .long direct_bar +; CHECK-NEXT: .long direct_baz +;; Number of unique indirect target type IDs. +; CHECK-NEXT: .byte 3 +;; Indirect type IDs. +; CHECK-NEXT: .long 838288420 +; CHECK-NEXT: .long 1053552373 +; CHECK-NEXT: .long 1505527380 +; CHECK-NEXT: .long 814631809 +; CHECK-NEXT: .long 342417018 +; CHECK-NEXT: .long 2013108216 diff --git a/llvm/test/CodeGen/ARM/call-graph-section-tailcall.ll b/llvm/test/CodeGen/ARM/call-graph-section-tailcall.ll new file mode 100644 index 0000000..d577603 --- /dev/null +++ b/llvm/test/CodeGen/ARM/call-graph-section-tailcall.ll @@ -0,0 +1,34 @@ +;; Tests that we store the type identifiers in .callgraph section of the object file for tailcalls. + +; RUN: llc -mtriple=arm-unknown-linux --call-graph-section -filetype=obj -o - < %s | \ +; RUN: llvm-readelf -x .callgraph - | FileCheck %s + +define i32 @check_tailcall(ptr %func, i8 %x) !type !0 { +entry: + %call = tail call i32 %func(i8 signext %x), !callee_type !1 + ret i32 %call +} + +define i32 @main(i32 %argc) !type !3 { +entry: + %andop = and i32 %argc, 1 + %cmp = icmp eq i32 %andop, 0 + %foo.bar = select i1 %cmp, ptr @foo, ptr @bar + %call.i = tail call i32 %foo.bar(i8 signext 97), !callee_type !1 + ret i32 %call.i +} + +declare !type !2 i32 @foo(i8 signext) + +declare !type !2 i32 @bar(i8 signext) + +!0 = !{i64 0, !"_ZTSFiPvcE.generalized"} +!1 = !{!2} +!2 = !{i64 0, !"_ZTSFicE.generalized"} +!3 = !{i64 0, !"_ZTSFiiE.generalized"} + +; CHECK: Hex dump of section '.callgraph': +; CHECK-NEXT: 0x00000000 00050000 00008e19 0b7f3326 e3000154 +; CHECK-NEXT: 0x00000010 86bc5981 4b8e3000 05100000 00a150b8 +;; Verify that the type id 0x308e4b8159bc8654 is in section. +; CHECK-NEXT: 0x00000020 3e0cfe3c b2015486 bc59814b 8e30 diff --git a/llvm/test/CodeGen/ARM/call-graph-section.ll b/llvm/test/CodeGen/ARM/call-graph-section.ll new file mode 100644 index 0000000..928a1067 --- /dev/null +++ b/llvm/test/CodeGen/ARM/call-graph-section.ll @@ -0,0 +1,37 @@ +;; Tests that we store the type identifiers in .callgraph section of the object file. + +; RUN: llc -mtriple=arm-unknown-linux --call-graph-section -filetype=obj -o - < %s | \ +; RUN: llvm-readelf -x .callgraph - | FileCheck %s + +declare !type !0 void @foo() + +declare !type !1 i32 @bar(i8) + +declare !type !2 ptr @baz(ptr) + +define void @main() { +entry: + %fp_foo_val = load ptr, ptr null, align 8 + call void (...) %fp_foo_val(), !callee_type !1 + %fp_bar_val = load ptr, ptr null, align 8 + %call_fp_bar = call i32 %fp_bar_val(i8 0), !callee_type !3 + %fp_baz_val = load ptr, ptr null, align 8 + %call_fp_baz = call ptr %fp_baz_val(ptr null), !callee_type !4 + ret void +} + +;; Check that the numeric type id (md5 hash) for the below type ids are emitted +;; to the callgraph section. +!0 = !{i64 0, !"_ZTSFvE.generalized"} +!1 = !{!0} +!2 = !{i64 0, !"_ZTSFicE.generalized"} +!3 = !{!2} +!4 = !{!5} +!5 = !{i64 0, !"_ZTSFPvS_E.generalized"} + +;; Make sure following type IDs are in call graph section +;; 0x5eecb3e2444f731f, 0x814b8e305486bc59, 0xf897fd777ade6814 +; CHECK: Hex dump of section '.callgraph': +; CHECK-NEXT: 0x00000000 00050000 00000000 00000000 00000324 +; CHECK-NEXT: 0x00000010 44f731f5 eecb3e54 86bc5981 4b8e307a +; CHECK-NEXT: 0x00000020 de6814f8 97fd77 diff --git a/llvm/test/CodeGen/ARM/fp16-promote.ll b/llvm/test/CodeGen/ARM/fp16-promote.ll index 800ee87..8230e47 100644 --- a/llvm/test/CodeGen/ARM/fp16-promote.ll +++ b/llvm/test/CodeGen/ARM/fp16-promote.ll @@ -1572,26 +1572,11 @@ define void @test_fma(ptr %p, ptr %q, ptr %r) #0 { } define void @test_fabs(ptr %p) { -; CHECK-FP16-LABEL: test_fabs: -; CHECK-FP16: ldrh r1, [r0] -; CHECK-FP16-NEXT: vmov s0, r1 -; CHECK-FP16-NEXT: vcvtb.f32.f16 s0, s0 -; CHECK-FP16-NEXT: vabs.f32 s0, s0 -; CHECK-FP16-NEXT: vcvtb.f16.f32 s0, s0 -; CHECK-FP16-NEXT: vmov r1, s0 -; CHECK-FP16-NEXT: strh r1, [r0] -; CHECK-FP16-NEXT: bx lr -; -; CHECK-LIBCALL-LABEL: test_fabs: -; CHECK-LIBCALL: .save {r4, lr} -; CHECK-LIBCALL-NEXT: push {r4, lr} -; CHECK-LIBCALL-NEXT: mov r4, r0 -; CHECK-LIBCALL-NEXT: ldrh r0, [r0] -; CHECK-LIBCALL-NEXT: bl __aeabi_h2f -; CHECK-LIBCALL-NEXT: bic r0, r0, #-2147483648 -; CHECK-LIBCALL-NEXT: bl __aeabi_f2h -; CHECK-LIBCALL-NEXT: strh r0, [r4] -; CHECK-LIBCALL-NEXT: pop {r4, pc} +; CHECK-ALL-LABEL: test_fabs: +; CHECK-ALL: ldrh r1, [r0] +; CHECK-ALL-NEXT: bfc r1, #15, #17 +; CHECK-ALL-NEXT: strh r1, [r0] +; CHECK-ALL-NEXT: bx lr %a = load half, ptr %p, align 2 %r = call half @llvm.fabs.f16(half %a) store half %r, ptr %p @@ -2454,26 +2439,11 @@ define half @test_sitofp_i32_fadd(i32 %a, half %b) #0 { } define void @test_fneg(ptr %p1, ptr %p2) #0 { -; CHECK-FP16-LABEL: test_fneg: -; CHECK-FP16: ldrh r0, [r0] -; CHECK-FP16-NEXT: vmov s0, r0 -; CHECK-FP16-NEXT: vcvtb.f32.f16 s0, s0 -; CHECK-FP16-NEXT: vneg.f32 s0, s0 -; CHECK-FP16-NEXT: vcvtb.f16.f32 s0, s0 -; CHECK-FP16-NEXT: vmov r0, s0 -; CHECK-FP16-NEXT: strh r0, [r1] -; CHECK-FP16-NEXT: bx lr -; -; CHECK-LIBCALL-LABEL: test_fneg: -; CHECK-LIBCALL: .save {r4, lr} -; CHECK-LIBCALL-NEXT: push {r4, lr} -; CHECK-LIBCALL-NEXT: ldrh r0, [r0] -; CHECK-LIBCALL-NEXT: mov r4, r1 -; CHECK-LIBCALL-NEXT: bl __aeabi_h2f -; CHECK-LIBCALL-NEXT: eor r0, r0, #-2147483648 -; CHECK-LIBCALL-NEXT: bl __aeabi_f2h -; CHECK-LIBCALL-NEXT: strh r0, [r4] -; CHECK-LIBCALL-NEXT: pop {r4, pc} +; CHECK-ALL-LABEL: test_fneg: +; CHECK-ALL: ldrh r0, [r0] +; CHECK-ALL-NEXT: eor r0, r0, #32768 +; CHECK-ALL-NEXT: strh r0, [r1] +; CHECK-ALL-NEXT: bx lr %v = load half, ptr %p1, align 2 %res = fneg half %v store half %res, ptr %p2, align 2 diff --git a/llvm/test/CodeGen/Generic/bfloat-op.ll b/llvm/test/CodeGen/Generic/bfloat-op.ll new file mode 100644 index 0000000..d593328 --- /dev/null +++ b/llvm/test/CodeGen/Generic/bfloat-op.ll @@ -0,0 +1,104 @@ +; Same as `bfloat.ll`, but for `fneg`, `fabs`, `copysign` and `fma`. +; Can be merged back into `bfloat.ll` once they have the same platform coverage. +; Once all targets are fixed, the `CHECK-*` prefixes should all be merged into a single `CHECK` prefix and the `BAD-*` prefixes should be removed. + +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-apple-darwin | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=arm64ec-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if amdgpu-registered-target %{ llc %s -o - -mtriple=amdgcn-amd-amdhsa | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if arc-registered-target %{ llc %s -o - -mtriple=arc-elf | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if arm-registered-target %{ llc %s -o - -mtriple=arm-unknown-linux-gnueabi | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if arm-registered-target %{ llc %s -o - -mtriple=thumbv7em-none-eabi | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if avr-registered-target %{ llc %s -o - -mtriple=avr-none | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; FIXME: BPF has a compiler error +; RUN: %if csky-registered-target %{ llc %s -o - -mtriple=csky-unknown-linux-gnuabiv2 | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; FIXME: hard float csky crashes +; FIXME: directx has a compiler error +; FIXME: hexagon crashes +; RUN: %if lanai-registered-target %{ llc %s -o - -mtriple=lanai-unknown-unknown | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch32-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch64-unknown-linux-gnu -mattr=+f | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if m68k-registered-target %{ llc %s -o - -mtriple=m68k-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; FIXME: mips crashes +; RUN: %if msp430-registered-target %{ llc %s -o - -mtriple=msp430-none-elf | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if nvptx-registered-target %{ llc %s -o - -mtriple=nvptx64-nvidia-cuda | FileCheck %s --check-prefixes=NOCRASH %} +; FIXME: powerpc crashes +; RUN: %if riscv-registered-target %{ llc %s -o - -mtriple=riscv32-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if riscv-registered-target %{ llc %s -o - -mtriple=riscv64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; FIXME: sparc crashes +; FIXME: spirv crashes +; FIXME: s390x crashes +; FIXME: ve crashes +; FIXME: wasm crashes +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=i686-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=x86_64-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=x86_64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if xcore-registered-target %{ llc %s -o - -mtriple=xcore-unknown-unknown | FileCheck %s --check-prefixes=ALL,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if xtensa-registered-target %{ llc %s -o - -mtriple=xtensa-none-elf | FileCheck %s --check-prefixes=ALL,BAD-COPYSIGN,CHECK-FMA %} + +; Note that arm64ec labels are quoted, hence the `{{"?}}:`. + +; Codegen tests don't work the same for graphics targets. Add a dummy directive +; for filecheck, just make sure we don't crash. +; NOCRASH: {{.*}} + +; fneg, fabs and copysign all need to not quieten signalling NaNs, so should not call any conversion functions which do. +; These tests won't catch cases where the everything is done using native instructions instead of builtins. + +define void @test_fneg(ptr %p1, ptr %p2) #0 { +; ALL-LABEL: test_fneg{{"?}}: +; ALL-NEG-NOT: __extend +; ALL-NEG-NOT: __trunc +; ALL-NEG-NOT: __gnu +; ALL-NEG-NOT: __aeabi + %v = load bfloat, ptr %p1 + %res = fneg bfloat %v + store bfloat %res, ptr %p2 + ret void +} + +define void @test_fabs(ptr %p1, ptr %p2) { +; ALL-LABEL: test_fabs{{"?}}: +; ALL-ABS-NOT: __extend +; ALL-ABS-NOT: __trunc +; ALL-ABS-NOT: __gnu +; ALL-ABS-NOT: __aeabi + %a = load bfloat, ptr %p1 + %r = call bfloat @llvm.fabs.f16(bfloat %a) + store bfloat %r, ptr %p2 + ret void +} + +define void @test_copysign(ptr %p1, ptr %p2, ptr %p3) { +; ALL-LABEL: test_copysign{{"?}}: +; CHECK-COPYSIGN-NOT: __extend +; CHECK-COPYSIGN-NOT: __trunc +; CHECK-COPYSIGN-NOT: __gnu +; CHECK-COPYSIGN-NOT: __aeabi +; BAD-COPYSIGN: __truncsfbf2 + %a = load bfloat, ptr %p1 + %b = load bfloat, ptr %p2 + %r = call bfloat @llvm.copysign.f16(bfloat %a, bfloat %b) + store bfloat %r, ptr %p3 + ret void +} + +; There is no floating-point type LLVM supports that is large enough to promote bfloat FMA to +; without causing double rounding issues. This checks for libcalls to f32/f64 fma and truncating +; f32/f64 to bf16. See https://github.com/llvm/llvm-project/issues/131531 + +define void @test_fma(ptr %p1, ptr %p2, ptr %p3, ptr %p4) { +; ALL-LABEL: test_fma{{"?}}: +; CHECK-FMA-NOT: {{\bfmaf?\b}} +; CHECK-FMA-NOT: __truncsfbf2 +; CHECK-FMA-NOT: __truncdfbf2 +; BAD-FMA: {{__truncsfbf2|\bfmaf?\b}} + %a = load bfloat, ptr %p1 + %b = load bfloat, ptr %p2 + %c = load bfloat, ptr %p3 + %r = call bfloat @llvm.fma.f16(bfloat %a, bfloat %b, bfloat %c) + store bfloat %r, ptr %p4 + ret void +} diff --git a/llvm/test/CodeGen/Generic/bfloat.ll b/llvm/test/CodeGen/Generic/bfloat.ll new file mode 100644 index 0000000..83c6711 --- /dev/null +++ b/llvm/test/CodeGen/Generic/bfloat.ll @@ -0,0 +1,75 @@ +; Simple cross-platform smoke checks for basic bf16 operations. +; +; There shouldn't be any architectures that crash when trying to use `bfloat`; +; check that here. Additionally do a small handful of smoke tests that work +; well cross-platform. + +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-apple-darwin | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; FIXME: arm64ec crashes when passing/returning bfloat +; RUN: %if amdgpu-registered-target %{ llc %s -o - -mtriple=amdgcn-amd-amdhsa | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if arc-registered-target %{ llc %s -o - -mtriple=arc-elf | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if arm-registered-target %{ llc %s -o - -mtriple=arm-unknown-linux-gnueabi | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if arm-registered-target %{ llc %s -o - -mtriple=thumbv7em-none-eabi | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if avr-registered-target %{ llc %s -o - -mtriple=avr-none | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if bpf-registered-target %{ llc %s -o - -mtriple=bpfel | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if csky-registered-target %{ llc %s -o - -mtriple=csky-unknown-linux-gnuabiv2 | FileCheck %s --check-prefixes=ALL,CHECK %} +; FIXME: hard float csky crashes +; RUN: %if directx-registered-target %{ llc %s -o - -mtriple=dxil-pc-shadermodel6.3-library | FileCheck %s --check-prefixes=NOCRASH %} +; FIXME: hexagon crashes +; RUN: %if lanai-registered-target %{ llc %s -o - -mtriple=lanai-unknown-unknown | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch32-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch64-unknown-linux-gnu -mattr=+f | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if m68k-registered-target %{ llc %s -o - -mtriple=m68k-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; FIXME: mips crashes +; RUN: %if msp430-registered-target %{ llc %s -o - -mtriple=msp430-none-elf | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if nvptx-registered-target %{ llc %s -o - -mtriple=nvptx64-nvidia-cuda | FileCheck %s --check-prefixes=NOCRASH %} +; FIXME: powerpc crashes +; RUN: %if riscv-registered-target %{ llc %s -o - -mtriple=riscv32-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if riscv-registered-target %{ llc %s -o - -mtriple=riscv64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; FIXME: sparc crashes +; FIXME: spirv crashes +; FIXME: s390x crashes +; FIXME: ve crashes +; FIXME: wasm crashes +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=i686-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,BAD %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=x86_64-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=x86_64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if xcore-registered-target %{ llc %s -o - -mtriple=xcore-unknown-unknown | FileCheck %s --check-prefixes=ALL,CHECK %} +; RUN: %if xtensa-registered-target %{ llc %s -o - -mtriple=xtensa-none-elf | FileCheck %s --check-prefixes=ALL,CHECK %} + +; Note that arm64ec labels are quoted, hence the `{{"?}}:`. + +; Codegen tests don't work the same for graphics targets. Add a dummy directive +; for filecheck, just make sure we don't crash. +; NOCRASH: {{.*}} + +; All backends need to be able to bitcast without converting to another format, +; so we assert against libcalls (specifically __truncsfbf2). This won't catch hardware conversions. + +define bfloat @from_bits(i16 %bits) nounwind { +; ALL-LABEL: from_bits{{"?}}: +; ALL-NOT: __extend +; ALL-NOT: __trunc +; ALL-NOT: __gnu + %f = bitcast i16 %bits to bfloat + ret bfloat %f +} + +define i16 @to_bits(bfloat %f) nounwind { +; ALL-LABEL: to_bits{{"?}}: +; CHECK-NOT: __extend +; CHECK-NOT: __trunc +; CHECK-NOT: __gnu +; BAD: __truncsfbf2 + %bits = bitcast bfloat %f to i16 + ret i16 %bits +} + +define bfloat @check_freeze(bfloat %f) nounwind { +; ALL-LABEL: check_freeze{{"?}}: + %t0 = freeze bfloat %f + ret bfloat %t0 +} diff --git a/llvm/test/CodeGen/Generic/half-op.ll b/llvm/test/CodeGen/Generic/half-op.ll new file mode 100644 index 0000000..1037d8e --- /dev/null +++ b/llvm/test/CodeGen/Generic/half-op.ll @@ -0,0 +1,115 @@ +; Same as `half.ll`, but for `fneg`, `fabs`, `copysign` and `fma`. +; Can be merged back into `half.ll` once BPF doesn't have a compiler error. +; Once all targets are fixed, the `CHECK-*` prefixes should all be merged into a single `CHECK` prefix and the `BAD-*` prefixes should be removed. + +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-apple-darwin | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=aarch64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if aarch64-registered-target %{ llc %s -o - -mtriple=arm64ec-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if amdgpu-registered-target %{ llc %s -o - -mtriple=amdgcn-amd-amdhsa | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,CHECK-FMA %} +; RUN: %if arc-registered-target %{ llc %s -o - -mtriple=arc-elf | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if arm-registered-target %{ llc %s -o - -mtriple=arm-unknown-linux-gnueabi | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if arm-registered-target %{ llc %s -o - -mtriple=thumbv7em-none-eabi | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if avr-registered-target %{ llc %s -o - -mtriple=avr-none | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; FIXME: BPF has a compiler error +; RUN: %if csky-registered-target %{ llc %s -o - -mtriple=csky-unknown-linux-gnuabiv2 | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if csky-registered-target %{ llc %s -o - -mtriple=csky-unknown-linux-gnuabiv2 -mcpu=ck860fv -mattr=+hard-float | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; FIXME: directx has a compiler error +; RUN: %if hexagon-registered-target %{ llc %s -o - -mtriple=hexagon-unknown-linux-musl | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if lanai-registered-target %{ llc %s -o - -mtriple=lanai-unknown-unknown | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch32-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if loongarch-registered-target %{ llc %s -o - -mtriple=loongarch64-unknown-linux-gnu -mattr=+f | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if m68k-registered-target %{ llc %s -o - -mtriple=m68k-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if mips-registered-target %{ llc %s -o - -mtriple=mips-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if mips-registered-target %{ llc %s -o - -mtriple=mips64-unknown-linux-gnuabi64 | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if mips-registered-target %{ llc %s -o - -mtriple=mips64el-unknown-linux-gnuabi64 | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if mips-registered-target %{ llc %s -o - -mtriple=mipsel-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if msp430-registered-target %{ llc %s -o - -mtriple=msp430-none-elf | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if nvptx-registered-target %{ llc %s -o - -mtriple=nvptx64-nvidia-cuda | FileCheck %s --check-prefixes=NOCRASH %} +; RUN: %if powerpc-registered-target %{ llc %s -o - -mtriple=powerpc-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if powerpc-registered-target %{ llc %s -o - -mtriple=powerpc64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if powerpc-registered-target %{ llc %s -o - -mtriple=powerpc64le-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if riscv-registered-target %{ llc %s -o - -mtriple=riscv32-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if riscv-registered-target %{ llc %s -o - -mtriple=riscv64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if sparc-registered-target %{ llc %s -o - -mtriple=sparc-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if sparc-registered-target %{ llc %s -o - -mtriple=sparc64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if spirv-registered-target %{ llc %s -o - -mtriple=spirv-unknown-unknown | FileCheck %s --check-prefixes=NOCRASH %} +; RUN: %if systemz-registered-target %{ llc %s -o - -mtriple=s390x-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if ve-registered-target %{ llc %s -o - -mtriple=ve-unknown-unknown | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if webassembly-registered-target %{ llc %s -o - -mtriple=wasm32-unknown-unknown | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=i686-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=x86_64-pc-windows-msvc | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if x86-registered-target %{ llc %s -o - -mtriple=x86_64-unknown-linux-gnu | FileCheck %s --check-prefixes=ALL,CHECK-NEG-ABS,CHECK-COPYSIGN,BAD-FMA %} +; RUN: %if xcore-registered-target %{ llc %s -o - -mtriple=xcore-unknown-unknown | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,BAD-FMA %} +; RUN: %if xtensa-registered-target %{ llc %s -o - -mtriple=xtensa-none-elf | FileCheck %s --check-prefixes=ALL,BAD-NEG-ABS,BAD-COPYSIGN,CHECK-FMA %} + +; Note that arm64ec labels are quoted, hence the `{{"?}}:`. + +; Codegen tests don't work the same for graphics targets. Add a dummy directive +; for filecheck, just make sure we don't crash. +; NOCRASH: {{.*}} + +; fneg, fabs and copysign all need to not quieten signalling NaNs, so should not call any conversion functions which do. +; These tests won't catch cases where the everything is done using native instructions instead of builtins. +; See https://github.com/llvm/llvm-project/issues/104915 + +define void @test_fneg(ptr %p1, ptr %p2) #0 { +; ALL-LABEL: test_fneg{{"?}}: +; CHECK-NEG-ABS-NOT: __extend +; CHECK-NEG-ABS-NOT: __trunc +; CHECK-NEG-ABS-NOT: __gnu +; CHECK-NEG-ABS-NOT: __aeabi +; BAD-NEG-ABS: {{__extendhfsf2|__gnu_h2f_ieee|__aeabi_h2f}} + %v = load half, ptr %p1 + %res = fneg half %v + store half %res, ptr %p2 + ret void +} + +define void @test_fabs(ptr %p1, ptr %p2) { +; ALL-LABEL: test_fabs{{"?}}: +; CHECK-NEG-ABS-NOT: __extend +; CHECK-NEG-ABS-NOT: __trunc +; CHECK-NEG-ABS-NOT: __gnu +; CHECK-NEG-ABS-NOT: __aeabi +; BAD-NEG-ABS: {{__extendhfsf2|__gnu_h2f_ieee|__aeabi_h2f}} + %a = load half, ptr %p1 + %r = call half @llvm.fabs.f16(half %a) + store half %r, ptr %p2 + ret void +} + +define void @test_copysign(ptr %p1, ptr %p2, ptr %p3) { +; ALL-LABEL: test_copysign{{"?}}: +; CHECK-COPYSIGN-NOT: __extend +; CHECK-COPYSIGN-NOT: __trunc +; CHECK-COPYSIGN-NOT: __gnu +; CHECK-COPYSIGN-NOT: __aeabi +; BAD-COPYSIGN: {{__extendhfsf2|__gnu_h2f_ieee}} + %a = load half, ptr %p1 + %b = load half, ptr %p2 + %r = call half @llvm.copysign.f16(half %a, half %b) + store half %r, ptr %p3 + ret void +} + +; If promoting, fma must promote at least to f64 to avoid double rounding issues. +; This checks for calls to f32 fmaf and truncating f32 to f16. +; See https://github.com/llvm/llvm-project/issues/98389 + +define void @test_fma(ptr %p1, ptr %p2, ptr %p3, ptr %p4) { +; ALL-LABEL: test_fma{{"?}}: +; Allow fmaf16 +; CHECK-FMA-NOT: fmaf{{\b}} +; CHECK-FMA-NOT: __truncsfhf2 +; CHECK-FMA-NOT: __gnu_f2h_ieee +; CHECK-FMA-NOT: __aeabi_f2h +; BAD-FMA: {{__truncsfhf2|__gnu_f2h_ieee|__aeabi_f2h|fmaf\b}} + %a = load half, ptr %p1 + %b = load half, ptr %p2 + %c = load half, ptr %p3 + %r = call half @llvm.fma.f16(half %a, half %b, half %c) + store half %r, ptr %p4 + ret void +} diff --git a/llvm/test/CodeGen/PowerPC/memcmp.ll b/llvm/test/CodeGen/PowerPC/memcmp.ll index 39f9269..4998d87 100644 --- a/llvm/test/CodeGen/PowerPC/memcmp.ll +++ b/llvm/test/CodeGen/PowerPC/memcmp.ll @@ -6,12 +6,10 @@ define signext i32 @memcmp8(ptr nocapture readonly %buffer1, ptr nocapture reado ; CHECK: # %bb.0: ; CHECK-NEXT: ldbrx 3, 0, 3 ; CHECK-NEXT: ldbrx 4, 0, 4 -; CHECK-NEXT: cmpld 3, 4 -; CHECK-NEXT: subc 3, 4, 3 -; CHECK-NEXT: subfe 3, 4, 4 -; CHECK-NEXT: li 4, -1 -; CHECK-NEXT: neg 3, 3 -; CHECK-NEXT: isellt 3, 4, 3 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: extsw 3, 3 ; CHECK-NEXT: blr %call = tail call signext i32 @memcmp(ptr %buffer1, ptr %buffer2, i64 8) @@ -23,11 +21,11 @@ define signext i32 @memcmp4(ptr nocapture readonly %buffer1, ptr nocapture reado ; CHECK: # %bb.0: ; CHECK-NEXT: lwbrx 3, 0, 3 ; CHECK-NEXT: lwbrx 4, 0, 4 -; CHECK-NEXT: cmplw 3, 4 -; CHECK-NEXT: sub 5, 4, 3 -; CHECK-NEXT: li 3, -1 -; CHECK-NEXT: rldicl 5, 5, 1, 63 -; CHECK-NEXT: isellt 3, 3, 5 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 +; CHECK-NEXT: extsw 3, 3 ; CHECK-NEXT: blr %call = tail call signext i32 @memcmp(ptr %buffer1, ptr %buffer2, i64 4) ret i32 %call diff --git a/llvm/test/CodeGen/PowerPC/ucmp.ll b/llvm/test/CodeGen/PowerPC/ucmp.ll index d2dff6e..4d393dd 100644 --- a/llvm/test/CodeGen/PowerPC/ucmp.ll +++ b/llvm/test/CodeGen/PowerPC/ucmp.ll @@ -4,12 +4,10 @@ define i8 @ucmp_8_8(i8 zeroext %x, i8 zeroext %y) nounwind { ; CHECK-LABEL: ucmp_8_8: ; CHECK: # %bb.0: -; CHECK-NEXT: cmplw 3, 4 -; CHECK-NEXT: sub 5, 4, 3 -; CHECK-NEXT: li 3, -1 -; CHECK-NEXT: rldicl 5, 5, 1, 63 -; CHECK-NEXT: rldic 3, 3, 0, 32 -; CHECK-NEXT: isellt 3, 3, 5 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i8 @llvm.ucmp(i8 %x, i8 %y) ret i8 %1 @@ -18,12 +16,10 @@ define i8 @ucmp_8_8(i8 zeroext %x, i8 zeroext %y) nounwind { define i8 @ucmp_8_16(i16 zeroext %x, i16 zeroext %y) nounwind { ; CHECK-LABEL: ucmp_8_16: ; CHECK: # %bb.0: -; CHECK-NEXT: cmplw 3, 4 -; CHECK-NEXT: sub 5, 4, 3 -; CHECK-NEXT: li 3, -1 -; CHECK-NEXT: rldicl 5, 5, 1, 63 -; CHECK-NEXT: rldic 3, 3, 0, 32 -; CHECK-NEXT: isellt 3, 3, 5 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i8 @llvm.ucmp(i16 %x, i16 %y) ret i8 %1 @@ -32,14 +28,10 @@ define i8 @ucmp_8_16(i16 zeroext %x, i16 zeroext %y) nounwind { define i8 @ucmp_8_32(i32 %x, i32 %y) nounwind { ; CHECK-LABEL: ucmp_8_32: ; CHECK: # %bb.0: -; CHECK-NEXT: clrldi 5, 4, 32 -; CHECK-NEXT: clrldi 6, 3, 32 -; CHECK-NEXT: sub 5, 5, 6 -; CHECK-NEXT: cmplw 3, 4 -; CHECK-NEXT: li 3, -1 -; CHECK-NEXT: rldic 3, 3, 0, 32 -; CHECK-NEXT: rldicl 5, 5, 1, 63 -; CHECK-NEXT: isellt 3, 3, 5 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i8 @llvm.ucmp(i32 %x, i32 %y) ret i8 %1 @@ -48,12 +40,10 @@ define i8 @ucmp_8_32(i32 %x, i32 %y) nounwind { define i8 @ucmp_8_64(i64 %x, i64 %y) nounwind { ; CHECK-LABEL: ucmp_8_64: ; CHECK: # %bb.0: -; CHECK-NEXT: cmpld 3, 4 -; CHECK-NEXT: subc 3, 4, 3 -; CHECK-NEXT: subfe 3, 4, 4 -; CHECK-NEXT: li 4, -1 -; CHECK-NEXT: neg 3, 3 -; CHECK-NEXT: isellt 3, 4, 3 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i8 @llvm.ucmp(i64 %x, i64 %y) ret i8 %1 @@ -82,14 +72,10 @@ define i8 @ucmp_8_128(i128 %x, i128 %y) nounwind { define i32 @ucmp_32_32(i32 %x, i32 %y) nounwind { ; CHECK-LABEL: ucmp_32_32: ; CHECK: # %bb.0: -; CHECK-NEXT: clrldi 5, 4, 32 -; CHECK-NEXT: clrldi 6, 3, 32 -; CHECK-NEXT: sub 5, 5, 6 -; CHECK-NEXT: cmplw 3, 4 -; CHECK-NEXT: li 3, -1 -; CHECK-NEXT: rldic 3, 3, 0, 32 -; CHECK-NEXT: rldicl 5, 5, 1, 63 -; CHECK-NEXT: isellt 3, 3, 5 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i32 @llvm.ucmp(i32 %x, i32 %y) ret i32 %1 @@ -98,12 +84,10 @@ define i32 @ucmp_32_32(i32 %x, i32 %y) nounwind { define i32 @ucmp_32_64(i64 %x, i64 %y) nounwind { ; CHECK-LABEL: ucmp_32_64: ; CHECK: # %bb.0: -; CHECK-NEXT: cmpld 3, 4 -; CHECK-NEXT: subc 3, 4, 3 -; CHECK-NEXT: subfe 3, 4, 4 -; CHECK-NEXT: li 4, -1 -; CHECK-NEXT: neg 3, 3 -; CHECK-NEXT: isellt 3, 4, 3 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i32 @llvm.ucmp(i64 %x, i64 %y) ret i32 %1 @@ -112,12 +96,10 @@ define i32 @ucmp_32_64(i64 %x, i64 %y) nounwind { define i64 @ucmp_64_64(i64 %x, i64 %y) nounwind { ; CHECK-LABEL: ucmp_64_64: ; CHECK: # %bb.0: -; CHECK-NEXT: subc 5, 4, 3 -; CHECK-NEXT: cmpld 3, 4 -; CHECK-NEXT: li 3, -1 -; CHECK-NEXT: subfe 5, 4, 4 -; CHECK-NEXT: neg 5, 5 -; CHECK-NEXT: isellt 3, 3, 5 +; CHECK-NEXT: subc 6, 4, 3 +; CHECK-NEXT: sub 5, 3, 4 +; CHECK-NEXT: subfe 3, 4, 3 +; CHECK-NEXT: subfe 3, 3, 5 ; CHECK-NEXT: blr %1 = call i64 @llvm.ucmp(i64 %x, i64 %y) ret i64 %1 diff --git a/llvm/test/CodeGen/RISCV/half-arith.ll b/llvm/test/CodeGen/RISCV/half-arith.ll index 2ebb6e9..d089e36 100644 --- a/llvm/test/CodeGen/RISCV/half-arith.ll +++ b/llvm/test/CodeGen/RISCV/half-arith.ll @@ -514,6 +514,7 @@ define i32 @fneg_h(half %a, half %b) nounwind { ; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill ; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill ; RV32I-NEXT: sw s1, 4(sp) # 4-byte Folded Spill +; RV32I-NEXT: sw s2, 0(sp) # 4-byte Folded Spill ; RV32I-NEXT: lui a1, 16 ; RV32I-NEXT: addi s1, a1, -1 ; RV32I-NEXT: and a0, a0, s1 @@ -521,13 +522,12 @@ define i32 @fneg_h(half %a, half %b) nounwind { ; RV32I-NEXT: mv a1, a0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s2, a0, a1 ; RV32I-NEXT: and a0, a0, s1 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 -; RV32I-NEXT: lui a0, 524288 -; RV32I-NEXT: xor a0, s0, a0 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s1 +; RV32I-NEXT: and a0, s2, s1 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv a1, a0 ; RV32I-NEXT: mv a0, s0 @@ -536,6 +536,7 @@ define i32 @fneg_h(half %a, half %b) nounwind { ; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload ; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload ; RV32I-NEXT: lw s1, 4(sp) # 4-byte Folded Reload +; RV32I-NEXT: lw s2, 0(sp) # 4-byte Folded Reload ; RV32I-NEXT: addi sp, sp, 16 ; RV32I-NEXT: ret ; @@ -545,6 +546,7 @@ define i32 @fneg_h(half %a, half %b) nounwind { ; RV64I-NEXT: sd ra, 24(sp) # 8-byte Folded Spill ; RV64I-NEXT: sd s0, 16(sp) # 8-byte Folded Spill ; RV64I-NEXT: sd s1, 8(sp) # 8-byte Folded Spill +; RV64I-NEXT: sd s2, 0(sp) # 8-byte Folded Spill ; RV64I-NEXT: lui a1, 16 ; RV64I-NEXT: addi s1, a1, -1 ; RV64I-NEXT: and a0, a0, s1 @@ -552,13 +554,12 @@ define i32 @fneg_h(half %a, half %b) nounwind { ; RV64I-NEXT: mv a1, a0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s2, a0, a1 ; RV64I-NEXT: and a0, a0, s1 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 -; RV64I-NEXT: lui a0, 524288 -; RV64I-NEXT: xor a0, s0, a0 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s1 +; RV64I-NEXT: and a0, s2, s1 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv a1, a0 ; RV64I-NEXT: mv a0, s0 @@ -567,6 +568,7 @@ define i32 @fneg_h(half %a, half %b) nounwind { ; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload ; RV64I-NEXT: ld s0, 16(sp) # 8-byte Folded Reload ; RV64I-NEXT: ld s1, 8(sp) # 8-byte Folded Reload +; RV64I-NEXT: ld s2, 0(sp) # 8-byte Folded Reload ; RV64I-NEXT: addi sp, sp, 32 ; RV64I-NEXT: ret ; @@ -638,11 +640,7 @@ define half @fsgnjn_h(half %a, half %b) nounwind { ; RV32I-NEXT: mv a0, s2 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui a1, 524288 -; RV32I-NEXT: xor a0, a0, a1 -; RV32I-NEXT: call __truncsfhf2 +; RV32I-NEXT: not a0, a0 ; RV32I-NEXT: lui a1, 1048568 ; RV32I-NEXT: slli s1, s1, 17 ; RV32I-NEXT: and a0, a0, a1 @@ -677,11 +675,7 @@ define half @fsgnjn_h(half %a, half %b) nounwind { ; RV64I-NEXT: mv a0, s2 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui a1, 524288 -; RV64I-NEXT: xor a0, a0, a1 -; RV64I-NEXT: call __truncsfhf2 +; RV64I-NEXT: not a0, a0 ; RV64I-NEXT: lui a1, 1048568 ; RV64I-NEXT: slli s1, s1, 49 ; RV64I-NEXT: and a0, a0, a1 @@ -804,15 +798,14 @@ define half @fabs_h(half %a, half %b) nounwind { ; RV32I-NEXT: mv a0, s1 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 +; RV32I-NEXT: slli s0, a0, 17 +; RV32I-NEXT: srli s0, s0, 17 ; RV32I-NEXT: and a0, a0, s2 ; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: mv s0, a0 -; RV32I-NEXT: slli a0, a0, 1 -; RV32I-NEXT: srli a0, a0, 1 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s2 +; RV32I-NEXT: mv s1, a0 +; RV32I-NEXT: mv a0, s0 ; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: mv a1, s0 +; RV32I-NEXT: mv a1, s1 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 ; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload @@ -841,15 +834,14 @@ define half @fabs_h(half %a, half %b) nounwind { ; RV64I-NEXT: mv a0, s1 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 +; RV64I-NEXT: slli s0, a0, 49 +; RV64I-NEXT: srli s0, s0, 49 ; RV64I-NEXT: and a0, a0, s2 ; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: mv s0, a0 -; RV64I-NEXT: slli a0, a0, 33 -; RV64I-NEXT: srli a0, a0, 33 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s2 +; RV64I-NEXT: mv s1, a0 +; RV64I-NEXT: mv a0, s0 ; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: mv a1, s0 +; RV64I-NEXT: mv a1, s1 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 ; RV64I-NEXT: ld ra, 24(sp) # 8-byte Folded Reload @@ -1217,25 +1209,21 @@ define half @fmsub_h(half %a, half %b, half %c) nounwind { ; RV32I-NEXT: mv s0, a1 ; RV32I-NEXT: mv s1, a0 ; RV32I-NEXT: lui a0, 16 -; RV32I-NEXT: addi s3, a0, -1 -; RV32I-NEXT: and a0, a2, s3 +; RV32I-NEXT: addi s2, a0, -1 +; RV32I-NEXT: and a0, a2, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui a1, 524288 -; RV32I-NEXT: xor a0, a0, a1 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s3, a0, a1 +; RV32I-NEXT: and a0, s1, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s0, s3 +; RV32I-NEXT: and a0, s0, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 -; RV32I-NEXT: and a0, s2, s3 +; RV32I-NEXT: and a0, s3, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv a2, a0 ; RV32I-NEXT: mv a0, s1 @@ -1261,25 +1249,21 @@ define half @fmsub_h(half %a, half %b, half %c) nounwind { ; RV64I-NEXT: mv s0, a1 ; RV64I-NEXT: mv s1, a0 ; RV64I-NEXT: lui a0, 16 -; RV64I-NEXT: addi s3, a0, -1 -; RV64I-NEXT: and a0, a2, s3 +; RV64I-NEXT: addi s2, a0, -1 +; RV64I-NEXT: and a0, a2, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui a1, 524288 -; RV64I-NEXT: xor a0, a0, a1 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s3, a0, a1 +; RV64I-NEXT: and a0, s1, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s0, s3 +; RV64I-NEXT: and a0, s0, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 -; RV64I-NEXT: and a0, s2, s3 +; RV64I-NEXT: and a0, s3, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv a2, a0 ; RV64I-NEXT: mv a0, s1 @@ -1355,43 +1339,34 @@ define half @fnmadd_h(half %a, half %b, half %c) nounwind { ; RV32I-NEXT: sw s2, 16(sp) # 4-byte Folded Spill ; RV32I-NEXT: sw s3, 12(sp) # 4-byte Folded Spill ; RV32I-NEXT: sw s4, 8(sp) # 4-byte Folded Spill -; RV32I-NEXT: mv s1, a2 -; RV32I-NEXT: mv s0, a1 -; RV32I-NEXT: lui s3, 16 -; RV32I-NEXT: addi s3, s3, -1 +; RV32I-NEXT: mv s0, a2 +; RV32I-NEXT: mv s1, a1 +; RV32I-NEXT: lui a1, 16 +; RV32I-NEXT: addi s3, a1, -1 ; RV32I-NEXT: and a0, a0, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 ; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: and a0, s0, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s2, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui s4, 524288 -; RV32I-NEXT: xor a0, a0, s4 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s2, a0 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s2, s2, a1 +; RV32I-NEXT: xor s4, a0, a1 ; RV32I-NEXT: and a0, s1, s3 ; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: xor a0, a0, s4 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s0, s3 -; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 ; RV32I-NEXT: and a0, s2, s3 ; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: mv s1, a0 +; RV32I-NEXT: and a0, s4, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv a2, a0 -; RV32I-NEXT: mv a0, s2 +; RV32I-NEXT: mv a0, s1 ; RV32I-NEXT: mv a1, s0 ; RV32I-NEXT: call fmaf ; RV32I-NEXT: call __truncsfhf2 @@ -1413,43 +1388,34 @@ define half @fnmadd_h(half %a, half %b, half %c) nounwind { ; RV64I-NEXT: sd s2, 16(sp) # 8-byte Folded Spill ; RV64I-NEXT: sd s3, 8(sp) # 8-byte Folded Spill ; RV64I-NEXT: sd s4, 0(sp) # 8-byte Folded Spill -; RV64I-NEXT: mv s1, a2 -; RV64I-NEXT: mv s0, a1 -; RV64I-NEXT: lui s3, 16 -; RV64I-NEXT: addi s3, s3, -1 +; RV64I-NEXT: mv s0, a2 +; RV64I-NEXT: mv s1, a1 +; RV64I-NEXT: lui a1, 16 +; RV64I-NEXT: addi s3, a1, -1 ; RV64I-NEXT: and a0, a0, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 ; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: and a0, s0, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s2, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui s4, 524288 -; RV64I-NEXT: xor a0, a0, s4 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s2, a0 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s2, s2, a1 +; RV64I-NEXT: xor s4, a0, a1 ; RV64I-NEXT: and a0, s1, s3 ; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: xor a0, a0, s4 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s0, s3 -; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 ; RV64I-NEXT: and a0, s2, s3 ; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: mv s1, a0 +; RV64I-NEXT: and a0, s4, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv a2, a0 -; RV64I-NEXT: mv a0, s2 +; RV64I-NEXT: mv a0, s1 ; RV64I-NEXT: mv a1, s0 ; RV64I-NEXT: call fmaf ; RV64I-NEXT: call __truncsfhf2 @@ -1535,44 +1501,35 @@ define half @fnmadd_h_2(half %a, half %b, half %c) nounwind { ; RV32I-NEXT: sw s2, 16(sp) # 4-byte Folded Spill ; RV32I-NEXT: sw s3, 12(sp) # 4-byte Folded Spill ; RV32I-NEXT: sw s4, 8(sp) # 4-byte Folded Spill -; RV32I-NEXT: mv s1, a2 -; RV32I-NEXT: mv s0, a0 -; RV32I-NEXT: lui s3, 16 -; RV32I-NEXT: addi s3, s3, -1 +; RV32I-NEXT: mv s0, a2 +; RV32I-NEXT: mv s1, a0 +; RV32I-NEXT: lui a0, 16 +; RV32I-NEXT: addi s3, a0, -1 ; RV32I-NEXT: and a0, a1, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 ; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: and a0, s0, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s2, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui s4, 524288 -; RV32I-NEXT: xor a0, a0, s4 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s2, a0 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s2, s2, a1 +; RV32I-NEXT: xor s4, a0, a1 ; RV32I-NEXT: and a0, s1, s3 ; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: xor a0, a0, s4 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s0, s3 -; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 ; RV32I-NEXT: and a0, s2, s3 ; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: mv s1, a0 +; RV32I-NEXT: and a0, s4, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv a2, a0 ; RV32I-NEXT: mv a0, s0 -; RV32I-NEXT: mv a1, s2 +; RV32I-NEXT: mv a1, s1 ; RV32I-NEXT: call fmaf ; RV32I-NEXT: call __truncsfhf2 ; RV32I-NEXT: lw ra, 28(sp) # 4-byte Folded Reload @@ -1593,44 +1550,35 @@ define half @fnmadd_h_2(half %a, half %b, half %c) nounwind { ; RV64I-NEXT: sd s2, 16(sp) # 8-byte Folded Spill ; RV64I-NEXT: sd s3, 8(sp) # 8-byte Folded Spill ; RV64I-NEXT: sd s4, 0(sp) # 8-byte Folded Spill -; RV64I-NEXT: mv s1, a2 -; RV64I-NEXT: mv s0, a0 -; RV64I-NEXT: lui s3, 16 -; RV64I-NEXT: addi s3, s3, -1 +; RV64I-NEXT: mv s0, a2 +; RV64I-NEXT: mv s1, a0 +; RV64I-NEXT: lui a0, 16 +; RV64I-NEXT: addi s3, a0, -1 ; RV64I-NEXT: and a0, a1, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 ; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: and a0, s0, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s2, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui s4, 524288 -; RV64I-NEXT: xor a0, a0, s4 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s2, a0 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s2, s2, a1 +; RV64I-NEXT: xor s4, a0, a1 ; RV64I-NEXT: and a0, s1, s3 ; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: xor a0, a0, s4 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s0, s3 -; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 ; RV64I-NEXT: and a0, s2, s3 ; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: mv s1, a0 +; RV64I-NEXT: and a0, s4, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv a2, a0 ; RV64I-NEXT: mv a0, s0 -; RV64I-NEXT: mv a1, s2 +; RV64I-NEXT: mv a1, s1 ; RV64I-NEXT: call fmaf ; RV64I-NEXT: call __truncsfhf2 ; RV64I-NEXT: ld ra, 40(sp) # 8-byte Folded Reload @@ -1960,25 +1908,21 @@ define half @fnmsub_h(half %a, half %b, half %c) nounwind { ; RV32I-NEXT: mv s0, a2 ; RV32I-NEXT: mv s1, a1 ; RV32I-NEXT: lui a1, 16 -; RV32I-NEXT: addi s3, a1, -1 -; RV32I-NEXT: and a0, a0, s3 +; RV32I-NEXT: addi s2, a1, -1 +; RV32I-NEXT: and a0, a0, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui a1, 524288 -; RV32I-NEXT: xor a0, a0, a1 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s3, a0, a1 +; RV32I-NEXT: and a0, s1, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s0, s3 +; RV32I-NEXT: and a0, s0, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 -; RV32I-NEXT: and a0, s2, s3 +; RV32I-NEXT: and a0, s3, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv a1, s1 ; RV32I-NEXT: mv a2, s0 @@ -2003,25 +1947,21 @@ define half @fnmsub_h(half %a, half %b, half %c) nounwind { ; RV64I-NEXT: mv s0, a2 ; RV64I-NEXT: mv s1, a1 ; RV64I-NEXT: lui a1, 16 -; RV64I-NEXT: addi s3, a1, -1 -; RV64I-NEXT: and a0, a0, s3 +; RV64I-NEXT: addi s2, a1, -1 +; RV64I-NEXT: and a0, a0, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui a1, 524288 -; RV64I-NEXT: xor a0, a0, a1 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s3, a0, a1 +; RV64I-NEXT: and a0, s1, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s0, s3 +; RV64I-NEXT: and a0, s0, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 -; RV64I-NEXT: and a0, s2, s3 +; RV64I-NEXT: and a0, s3, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv a1, s1 ; RV64I-NEXT: mv a2, s0 @@ -2096,25 +2036,21 @@ define half @fnmsub_h_2(half %a, half %b, half %c) nounwind { ; RV32I-NEXT: mv s0, a2 ; RV32I-NEXT: mv s1, a0 ; RV32I-NEXT: lui a0, 16 -; RV32I-NEXT: addi s3, a0, -1 -; RV32I-NEXT: and a0, a1, s3 +; RV32I-NEXT: addi s2, a0, -1 +; RV32I-NEXT: and a0, a1, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: li a1, 0 ; RV32I-NEXT: call __addsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui a1, 524288 -; RV32I-NEXT: xor a0, a0, a1 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s2, a0 -; RV32I-NEXT: and a0, s1, s3 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s3, a0, a1 +; RV32I-NEXT: and a0, s1, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s1, a0 -; RV32I-NEXT: and a0, s0, s3 +; RV32I-NEXT: and a0, s0, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 -; RV32I-NEXT: and a0, s2, s3 +; RV32I-NEXT: and a0, s3, s2 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv a1, a0 ; RV32I-NEXT: mv a0, s1 @@ -2140,25 +2076,21 @@ define half @fnmsub_h_2(half %a, half %b, half %c) nounwind { ; RV64I-NEXT: mv s0, a2 ; RV64I-NEXT: mv s1, a0 ; RV64I-NEXT: lui a0, 16 -; RV64I-NEXT: addi s3, a0, -1 -; RV64I-NEXT: and a0, a1, s3 +; RV64I-NEXT: addi s2, a0, -1 +; RV64I-NEXT: and a0, a1, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: li a1, 0 ; RV64I-NEXT: call __addsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui a1, 524288 -; RV64I-NEXT: xor a0, a0, a1 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s2, a0 -; RV64I-NEXT: and a0, s1, s3 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s3, a0, a1 +; RV64I-NEXT: and a0, s1, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s1, a0 -; RV64I-NEXT: and a0, s0, s3 +; RV64I-NEXT: and a0, s0, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 -; RV64I-NEXT: and a0, s2, s3 +; RV64I-NEXT: and a0, s3, s2 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv a1, a0 ; RV64I-NEXT: mv a0, s1 @@ -2519,12 +2451,8 @@ define half @fnmadd_h_contract(half %a, half %b, half %c) nounwind { ; RV32I-NEXT: mv a0, s2 ; RV32I-NEXT: call __mulsf3 ; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: and a0, a0, s3 -; RV32I-NEXT: call __extendhfsf2 -; RV32I-NEXT: lui a1, 524288 -; RV32I-NEXT: xor a0, a0, a1 -; RV32I-NEXT: call __truncsfhf2 -; RV32I-NEXT: mv s1, a0 +; RV32I-NEXT: lui a1, 8 +; RV32I-NEXT: xor s1, a0, a1 ; RV32I-NEXT: and a0, s0, s3 ; RV32I-NEXT: call __extendhfsf2 ; RV32I-NEXT: mv s0, a0 @@ -2580,12 +2508,8 @@ define half @fnmadd_h_contract(half %a, half %b, half %c) nounwind { ; RV64I-NEXT: mv a0, s2 ; RV64I-NEXT: call __mulsf3 ; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: and a0, a0, s3 -; RV64I-NEXT: call __extendhfsf2 -; RV64I-NEXT: lui a1, 524288 -; RV64I-NEXT: xor a0, a0, a1 -; RV64I-NEXT: call __truncsfhf2 -; RV64I-NEXT: mv s1, a0 +; RV64I-NEXT: lui a1, 8 +; RV64I-NEXT: xor s1, a0, a1 ; RV64I-NEXT: and a0, s0, s3 ; RV64I-NEXT: call __extendhfsf2 ; RV64I-NEXT: mv s0, a0 diff --git a/llvm/test/CodeGen/Thumb2/mve-vabd.ll b/llvm/test/CodeGen/Thumb2/mve-vabd.ll index 8d52fe5..3c35a29 100644 --- a/llvm/test/CodeGen/Thumb2/mve-vabd.ll +++ b/llvm/test/CodeGen/Thumb2/mve-vabd.ll @@ -63,34 +63,30 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: .vsave {d8, d9, d10, d11, d12, d13} ; CHECK-MVE-NEXT: vpush {d8, d9, d10, d11, d12, d13} ; CHECK-MVE-NEXT: mov r4, r0 -; CHECK-MVE-NEXT: vmov.u16 r0, q1[0] +; CHECK-MVE-NEXT: vmov.u16 r0, q1[1] ; CHECK-MVE-NEXT: vmov q5, q1 ; CHECK-MVE-NEXT: vmov q4, q0 ; CHECK-MVE-NEXT: bl __aeabi_h2f ; CHECK-MVE-NEXT: mov r5, r0 -; CHECK-MVE-NEXT: vmov.u16 r0, q4[0] +; CHECK-MVE-NEXT: vmov.u16 r0, q4[1] ; CHECK-MVE-NEXT: bl __aeabi_h2f ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h ; CHECK-MVE-NEXT: mov r5, r0 -; CHECK-MVE-NEXT: vmov.u16 r0, q5[1] +; CHECK-MVE-NEXT: vmov.u16 r0, q5[0] ; CHECK-MVE-NEXT: bl __aeabi_h2f ; CHECK-MVE-NEXT: mov r6, r0 -; CHECK-MVE-NEXT: vmov.u16 r0, q4[1] +; CHECK-MVE-NEXT: vmov.u16 r0, q4[0] ; CHECK-MVE-NEXT: bl __aeabi_h2f ; CHECK-MVE-NEXT: mov r1, r6 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: vmov.16 q6[0], r5 -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: vmov.16 q6[1], r0 +; CHECK-MVE-NEXT: bfc r0, #15, #17 +; CHECK-MVE-NEXT: bfc r5, #15, #17 +; CHECK-MVE-NEXT: vmov.16 q6[0], r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q5[2] +; CHECK-MVE-NEXT: vmov.16 q6[1], r5 ; CHECK-MVE-NEXT: bl __aeabi_h2f ; CHECK-MVE-NEXT: mov r5, r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q4[2] @@ -98,9 +94,7 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h +; CHECK-MVE-NEXT: bfc r0, #15, #17 ; CHECK-MVE-NEXT: vmov.16 q6[2], r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q5[3] ; CHECK-MVE-NEXT: bl __aeabi_h2f @@ -110,9 +104,7 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h +; CHECK-MVE-NEXT: bfc r0, #15, #17 ; CHECK-MVE-NEXT: vmov.16 q6[3], r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q5[4] ; CHECK-MVE-NEXT: bl __aeabi_h2f @@ -122,9 +114,7 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h +; CHECK-MVE-NEXT: bfc r0, #15, #17 ; CHECK-MVE-NEXT: vmov.16 q6[4], r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q5[5] ; CHECK-MVE-NEXT: bl __aeabi_h2f @@ -134,9 +124,7 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h +; CHECK-MVE-NEXT: bfc r0, #15, #17 ; CHECK-MVE-NEXT: vmov.16 q6[5], r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q5[6] ; CHECK-MVE-NEXT: bl __aeabi_h2f @@ -146,9 +134,7 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h +; CHECK-MVE-NEXT: bfc r0, #15, #17 ; CHECK-MVE-NEXT: vmov.16 q6[6], r0 ; CHECK-MVE-NEXT: vmov.u16 r0, q5[7] ; CHECK-MVE-NEXT: bl __aeabi_h2f @@ -158,9 +144,7 @@ define arm_aapcs_vfpcc void @vabd_v8f16(<8 x half> %x, <8 x half> %y, ptr %z) { ; CHECK-MVE-NEXT: mov r1, r5 ; CHECK-MVE-NEXT: bl __aeabi_fsub ; CHECK-MVE-NEXT: bl __aeabi_f2h -; CHECK-MVE-NEXT: bl __aeabi_h2f -; CHECK-MVE-NEXT: bic r0, r0, #-2147483648 -; CHECK-MVE-NEXT: bl __aeabi_f2h +; CHECK-MVE-NEXT: bfc r0, #15, #17 ; CHECK-MVE-NEXT: vmov.16 q6[7], r0 ; CHECK-MVE-NEXT: vstrw.32 q6, [r4] ; CHECK-MVE-NEXT: vpop {d8, d9, d10, d11, d12, d13} diff --git a/llvm/test/CodeGen/X86/call-graph-section-addrtaken.ll b/llvm/test/CodeGen/X86/call-graph-section-addrtaken.ll new file mode 100644 index 0000000..2aea9c1 --- /dev/null +++ b/llvm/test/CodeGen/X86/call-graph-section-addrtaken.ll @@ -0,0 +1,38 @@ +;; Test if a potential indirect call target function which has internal linkage and +;; address taken has its type ID emitted to callgraph section. +;; This test also makes sure that callback functions which meet the above constraint +;; are handled correctly. + +; RUN: llc -mtriple=x86_64-unknown-linux --call-graph-section -o - < %s | FileCheck %s + +declare !type !0 void @_Z6doWorkPFviE(ptr) + +define i32 @_Z4testv() !type !1 { +entry: + call void @_Z6doWorkPFviE(ptr nonnull @_ZL10myCallbacki) + ret i32 0 +} + +; CHECK: _ZL10myCallbacki: +; CHECK-NEXT: [[LABEL_FUNC:\.Lfunc_begin[0-9]+]]: +define internal void @_ZL10myCallbacki(i32 %value) !type !2 { +entry: + %sink = alloca i32, align 4 + store volatile i32 %value, ptr %sink, align 4 + %i1 = load volatile i32, ptr %sink, align 4 + ret void +} + +!0 = !{i64 0, !"_ZTSFvPFviEE.generalized"} +!1 = !{i64 0, !"_ZTSFivE.generalized"} +!2 = !{i64 0, !"_ZTSFviE.generalized"} + +; CHECK: .section .callgraph,"o",@progbits,.text +;; Version +; CHECK-NEXT: .byte 0 +;; Flags -- Potential indirect target so LSB is set to 1. Other bits are 0. +; CHECK-NEXT: .byte 1 +;; Function Entry PC +; CHECK-NEXT: .quad [[LABEL_FUNC]] +;; Function type ID +; CHECK-NEXT: .quad -5212364466660467813 diff --git a/llvm/test/CodeGen/X86/call-graph-section-assembly.ll b/llvm/test/CodeGen/X86/call-graph-section-assembly.ll index f0dbc31..1aabf66 100644 --- a/llvm/test/CodeGen/X86/call-graph-section-assembly.ll +++ b/llvm/test/CodeGen/X86/call-graph-section-assembly.ll @@ -15,16 +15,13 @@ declare !type !2 ptr @direct_baz(ptr) define ptr @ball() { entry: call void @direct_foo() - %fp_foo_val = load ptr, ptr null, align 8 - ; CHECK: [[LABEL_TMP0:\.L.*]]: + %fp_foo_val = load ptr, ptr null, align 8 call void (...) %fp_foo_val(), !callee_type !0 call void @direct_foo() - %fp_bar_val = load ptr, ptr null, align 8 - ; CHECK: [[LABEL_TMP1:\.L.*]]: + %fp_bar_val = load ptr, ptr null, align 8 %call_fp_bar = call i32 %fp_bar_val(i8 0), !callee_type !2 %call_fp_bar_direct = call i32 @direct_bar(i8 1) %fp_baz_val = load ptr, ptr null, align 8 - ; CHECK: [[LABEL_TMP2:\.L.*]]: %call_fp_baz = call ptr %fp_baz_val(ptr null), !callee_type !4 call void @direct_foo() %call_fp_baz_direct = call ptr @direct_baz(ptr null) @@ -32,29 +29,31 @@ entry: ret ptr %call_fp_baz } -; CHECK: .section .callgraph,"o",@progbits,.text - -; CHECK-NEXT: .quad 0 -; CHECK-NEXT: .quad [[LABEL_FUNC]] -; CHECK-NEXT: .quad 1 -; CHECK-NEXT: .quad 3 !0 = !{!1} !1 = !{i64 0, !"_ZTSFvE.generalized"} -;; Test for MD5 hash of _ZTSFvE.generalized and the generated temporary callsite label. -; CHECK-NEXT: .quad 4524972987496481828 -; CHECK-NEXT: .quad [[LABEL_TMP0]] !2 = !{!3} !3 = !{i64 0, !"_ZTSFicE.generalized"} -;; Test for MD5 hash of _ZTSFicE.generalized and the generated temporary callsite label. -; CHECK-NEXT: .quad 3498816979441845844 -; CHECK-NEXT: .quad [[LABEL_TMP1]] !4 = !{!5} !5 = !{i64 0, !"_ZTSFPvS_E.generalized"} -;; Test for MD5 hash of _ZTSFPvS_E.generalized and the generated temporary callsite label. -; CHECK-NEXT: .quad 8646233951371320954 -; CHECK-NEXT: .quad [[LABEL_TMP2]] -;; Test for number of direct calls and {callsite_label, callee} pairs. -; CHECK-NEXT: .quad 3 + +; CHECK: .section .callgraph,"o",@progbits,.text +;; Version +; CHECK-NEXT: .byte 0 +;; Flags +; CHECK-NEXT: .byte 7 +;; Function Entry PC +; CHECK-NEXT: .quad [[LABEL_FUNC]] +;; Function type ID -- set to 0 as no type metadata attached to function. +; CHECK-NEXT: .quad 0 +;; Number of unique direct callees. +; CHECK-NEXT: .byte 3 +;; Direct callees. ; CHECK-NEXT: .quad direct_foo ; CHECK-NEXT: .quad direct_bar ; CHECK-NEXT: .quad direct_baz +;; Number of unique indirect target type IDs. +; CHECK-NEXT: .byte 3 +;; Indirect type IDs. +; CHECK-NEXT: .quad 4524972987496481828 +; CHECK-NEXT: .quad 3498816979441845844 +; CHECK-NEXT: .quad 8646233951371320954 diff --git a/llvm/test/CodeGen/X86/call-graph-section-tailcall.ll b/llvm/test/CodeGen/X86/call-graph-section-tailcall.ll index fa14a98..34dc5b8 100644 --- a/llvm/test/CodeGen/X86/call-graph-section-tailcall.ll +++ b/llvm/test/CodeGen/X86/call-graph-section-tailcall.ll @@ -22,13 +22,14 @@ declare !type !2 i32 @foo(i8 signext) declare !type !2 i32 @bar(i8 signext) -;; Check that the numeric type id (md5 hash) for the below type ids are emitted -;; to the callgraph section. - -; CHECK: Hex dump of section '.callgraph': - !0 = !{i64 0, !"_ZTSFiPvcE.generalized"} !1 = !{!2} -; CHECK-DAG: 5486bc59 814b8e30 !2 = !{i64 0, !"_ZTSFicE.generalized"} !3 = !{i64 0, !"_ZTSFiiE.generalized"} + +; CHECK: Hex dump of section '.callgraph': +; CHECK-NEXT: 0x00000000 00050000 00000000 00008e19 0b7f3326 +; CHECK-NEXT: 0x00000010 e3000154 86bc5981 4b8e3000 05000000 +;; Verify that the type id 0x308e4b8159bc8654 is in section. +; CHECK-NEXT: 0x00000020 00000000 00a150b8 3e0cfe3c b2015486 +; CHECK-NEXT: 0x00000030 bc59814b 8e30 diff --git a/llvm/test/CodeGen/X86/call-graph-section.ll b/llvm/test/CodeGen/X86/call-graph-section.ll index 66d009c..c144a24 100644 --- a/llvm/test/CodeGen/X86/call-graph-section.ll +++ b/llvm/test/CodeGen/X86/call-graph-section.ll @@ -22,15 +22,16 @@ entry: ;; Check that the numeric type id (md5 hash) for the below type ids are emitted ;; to the callgraph section. - -; CHECK: Hex dump of section '.callgraph': - -; CHECK-DAG: 2444f731 f5eecb3e !0 = !{i64 0, !"_ZTSFvE.generalized"} !1 = !{!0} -; CHECK-DAG: 5486bc59 814b8e30 !2 = !{i64 0, !"_ZTSFicE.generalized"} !3 = !{!2} -; CHECK-DAG: 7ade6814 f897fd77 !4 = !{!5} !5 = !{i64 0, !"_ZTSFPvS_E.generalized"} + +;; Make sure following type IDs are in call graph section +;; 0x5eecb3e2444f731f, 0x814b8e305486bc59, 0xf897fd777ade6814 +; CHECK: Hex dump of section '.callgraph': +; CHECK-NEXT: 0x00000000 00050000 00000000 00000000 00000000 +; CHECK-NEXT: 0x00000010 00000324 44f731f5 eecb3e54 86bc5981 +; CHECK-NEXT: 0x00000020 4b8e307a de6814f8 97fd77 diff --git a/llvm/test/CodeGen/X86/global-variable-partition-with-dap.ll b/llvm/test/CodeGen/X86/global-variable-partition-with-dap.ll new file mode 100644 index 0000000..a0c243b --- /dev/null +++ b/llvm/test/CodeGen/X86/global-variable-partition-with-dap.ll @@ -0,0 +1,43 @@ +target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128" +target triple = "x86_64-unknown-linux-gnu" + +;; A minimal test case. llc will crash if global variables already has a section +;; prefix. Subsequent PRs will expand on this test case to test the hotness +;; reconciliation implementation. + +; RUN: not llc -mtriple=x86_64-unknown-linux-gnu -relocation-model=pic \ +; RUN: -partition-static-data-sections=true \ +; RUN: -data-sections=true -unique-section-names=false \ +; RUN: %s -o - 2>&1 | FileCheck %s --check-prefix=ERR + +; ERR: Global variable hot_bss already has a section prefix hot + +@hot_bss = internal global i32 0, !section_prefix !17 + +define void @hot_func() !prof !14 { + %9 = load i32, ptr @hot_bss + %11 = call i32 (...) @func_taking_arbitrary_param(i32 %9) + ret void +} + +declare i32 @func_taking_arbitrary_param(...) + +!llvm.module.flags = !{!1} + +!1 = !{i32 1, !"ProfileSummary", !2} +!2 = !{!3, !4, !5, !6, !7, !8, !9, !10} +!3 = !{!"ProfileFormat", !"InstrProf"} +!4 = !{!"TotalCount", i64 1460183} +!5 = !{!"MaxCount", i64 849024} +!6 = !{!"MaxInternalCount", i64 32769} +!7 = !{!"MaxFunctionCount", i64 849024} +!8 = !{!"NumCounts", i64 23627} +!9 = !{!"NumFunctions", i64 3271} +!10 = !{!"DetailedSummary", !11} +!11 = !{!12, !13} +!12 = !{i32 990000, i64 166, i32 73} +!13 = !{i32 999999, i64 3, i32 1443} +!14 = !{!"function_entry_count", i64 100000} +!15 = !{!"function_entry_count", i64 1} +!16 = !{!"branch_weights", i32 1, i32 99999} +!17 = !{!"section_prefix", !"hot"} diff --git a/llvm/test/CodeGen/X86/pr162812.ll b/llvm/test/CodeGen/X86/pr162812.ll new file mode 100644 index 0000000..4ea3101 --- /dev/null +++ b/llvm/test/CodeGen/X86/pr162812.ll @@ -0,0 +1,98 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3 +; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64 | FileCheck %s --check-prefixes=SSE2 +; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE42 +; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX2 +; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX512 + +define <32 x i8> @PR162812(<32 x i8> %a, <32 x i8> %mask) { +; SSE2-LABEL: PR162812: +; SSE2: # %bb.0: +; SSE2-NEXT: psrlw $2, %xmm2 +; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [8224,8224,8224,8224,8224,8224,8224,8224] +; SSE2-NEXT: pand %xmm4, %xmm2 +; SSE2-NEXT: paddb %xmm2, %xmm2 +; SSE2-NEXT: paddb %xmm2, %xmm2 +; SSE2-NEXT: pxor %xmm5, %xmm5 +; SSE2-NEXT: pxor %xmm6, %xmm6 +; SSE2-NEXT: pcmpgtb %xmm2, %xmm6 +; SSE2-NEXT: movdqa %xmm6, %xmm2 +; SSE2-NEXT: pandn %xmm0, %xmm2 +; SSE2-NEXT: paddb %xmm0, %xmm0 +; SSE2-NEXT: pand %xmm6, %xmm0 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: psrlw $2, %xmm3 +; SSE2-NEXT: pand %xmm4, %xmm3 +; SSE2-NEXT: paddb %xmm3, %xmm3 +; SSE2-NEXT: paddb %xmm3, %xmm3 +; SSE2-NEXT: pcmpgtb %xmm3, %xmm5 +; SSE2-NEXT: movdqa %xmm5, %xmm2 +; SSE2-NEXT: pandn %xmm1, %xmm2 +; SSE2-NEXT: paddb %xmm1, %xmm1 +; SSE2-NEXT: pand %xmm5, %xmm1 +; SSE2-NEXT: por %xmm2, %xmm1 +; SSE2-NEXT: retq +; +; SSE42-LABEL: PR162812: +; SSE42: # %bb.0: +; SSE42-NEXT: movdqa %xmm2, %xmm5 +; SSE42-NEXT: movdqa %xmm0, %xmm2 +; SSE42-NEXT: movdqa %xmm0, %xmm6 +; SSE42-NEXT: psllw $2, %xmm6 +; SSE42-NEXT: movdqa {{.*#+}} xmm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252] +; SSE42-NEXT: pand %xmm7, %xmm6 +; SSE42-NEXT: psrlw $2, %xmm5 +; SSE42-NEXT: movdqa {{.*#+}} xmm4 = [8224,8224,8224,8224,8224,8224,8224,8224] +; SSE42-NEXT: pand %xmm4, %xmm5 +; SSE42-NEXT: paddb %xmm5, %xmm5 +; SSE42-NEXT: movdqa %xmm5, %xmm0 +; SSE42-NEXT: pblendvb %xmm0, %xmm6, %xmm2 +; SSE42-NEXT: movdqa %xmm2, %xmm6 +; SSE42-NEXT: paddb %xmm2, %xmm6 +; SSE42-NEXT: paddb %xmm5, %xmm5 +; SSE42-NEXT: movdqa %xmm5, %xmm0 +; SSE42-NEXT: pblendvb %xmm0, %xmm6, %xmm2 +; SSE42-NEXT: movdqa %xmm1, %xmm5 +; SSE42-NEXT: psllw $2, %xmm5 +; SSE42-NEXT: pand %xmm7, %xmm5 +; SSE42-NEXT: psrlw $2, %xmm3 +; SSE42-NEXT: pand %xmm3, %xmm4 +; SSE42-NEXT: paddb %xmm4, %xmm4 +; SSE42-NEXT: movdqa %xmm4, %xmm0 +; SSE42-NEXT: pblendvb %xmm0, %xmm5, %xmm1 +; SSE42-NEXT: movdqa %xmm1, %xmm3 +; SSE42-NEXT: paddb %xmm1, %xmm3 +; SSE42-NEXT: paddb %xmm4, %xmm4 +; SSE42-NEXT: movdqa %xmm4, %xmm0 +; SSE42-NEXT: pblendvb %xmm0, %xmm3, %xmm1 +; SSE42-NEXT: movdqa %xmm2, %xmm0 +; SSE42-NEXT: retq +; +; AVX2-LABEL: PR162812: +; AVX2: # %bb.0: +; AVX2-NEXT: vpsllw $2, %ymm0, %ymm2 +; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX2-NEXT: vpsrlw $2, %ymm1, %ymm1 +; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1 +; AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; AVX2-NEXT: vpaddb %ymm0, %ymm0, %ymm2 +; AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; AVX2-NEXT: retq +; +; AVX512-LABEL: PR162812: +; AVX512: # %bb.0: +; AVX512-NEXT: vpsllw $2, %ymm0, %ymm2 +; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm2, %ymm2 +; AVX512-NEXT: vpsrlw $2, %ymm1, %ymm1 +; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm1 +; AVX512-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; AVX512-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; AVX512-NEXT: vpaddb %ymm0, %ymm0, %ymm2 +; AVX512-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; AVX512-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; AVX512-NEXT: retq + %1 = lshr <32 x i8> %mask, splat (i8 7) + %ret = shl <32 x i8> %a, %1 + ret <32 x i8> %ret +} |