diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU')
22 files changed, 2876 insertions, 1190 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/ashr.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/ashr.ll index 353c09b..ecd7cc2 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/ashr.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/ashr.ll @@ -1778,7 +1778,7 @@ define i65 @v_ashr_i65_33(i65 %value) { ; GFX6-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX6-NEXT: v_lshl_b64 v[0:1], v[1:2], 31 ; GFX6-NEXT: v_lshrrev_b32_e32 v3, 1, v3 -; GFX6-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX6-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX6-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX6-NEXT: s_setpc_b64 s[30:31] ; @@ -1790,7 +1790,7 @@ define i65 @v_ashr_i65_33(i65 %value) { ; GFX8-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX8-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX8-NEXT: v_lshrrev_b32_e32 v3, 1, v3 -; GFX8-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX8-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX8-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX8-NEXT: s_setpc_b64 s[30:31] ; @@ -1802,7 +1802,7 @@ define i65 @v_ashr_i65_33(i65 %value) { ; GFX9-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX9-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX9-NEXT: v_lshrrev_b32_e32 v3, 1, v3 -; GFX9-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX9-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX9-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; @@ -1815,7 +1815,7 @@ define i65 @v_ashr_i65_33(i65 %value) { ; GFX10PLUS-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX10PLUS-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX10PLUS-NEXT: v_ashrrev_i32_e32 v2, 1, v2 -; GFX10PLUS-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX10PLUS-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] %result = ashr i65 %value, 33 ret i65 %result @@ -1875,21 +1875,19 @@ define amdgpu_ps i65 @s_ashr_i65_33(i65 inreg %value) { ; GCN-LABEL: s_ashr_i65_33: ; GCN: ; %bb.0: ; GCN-NEXT: s_bfe_i64 s[2:3], s[2:3], 0x10000 -; GCN-NEXT: s_lshr_b32 s0, s1, 1 -; GCN-NEXT: s_mov_b32 s1, 0 -; GCN-NEXT: s_lshl_b64 s[4:5], s[2:3], 31 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] +; GCN-NEXT: s_lshr_b32 s4, s1, 1 +; GCN-NEXT: s_lshl_b64 s[0:1], s[2:3], 31 +; GCN-NEXT: s_or_b32 s0, s0, s4 ; GCN-NEXT: s_ashr_i32 s2, s3, 1 ; GCN-NEXT: ; return to shader part epilog ; ; GFX10PLUS-LABEL: s_ashr_i65_33: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_bfe_i64 s[2:3], s[2:3], 0x10000 -; GFX10PLUS-NEXT: s_lshr_b32 s0, s1, 1 -; GFX10PLUS-NEXT: s_mov_b32 s1, 0 -; GFX10PLUS-NEXT: s_lshl_b64 s[4:5], s[2:3], 31 +; GFX10PLUS-NEXT: s_lshr_b32 s4, s1, 1 +; GFX10PLUS-NEXT: s_lshl_b64 s[0:1], s[2:3], 31 ; GFX10PLUS-NEXT: s_ashr_i32 s2, s3, 1 -; GFX10PLUS-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] +; GFX10PLUS-NEXT: s_or_b32 s0, s0, s4 ; GFX10PLUS-NEXT: ; return to shader part epilog %result = ashr i65 %value, 33 ret i65 %result diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-or-s64-s32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-or-s64-s32.mir new file mode 100644 index 0000000..48e9818 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-or-s64-s32.mir @@ -0,0 +1,97 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5 +# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=amdgpu-prelegalizer-combiner %s -o - | FileCheck %s + +--- +name: test_combine_or_s64_s32 +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr0_sgpr1, $sgpr2 + ; CHECK-LABEL: name: test_combine_or_s64_s32 + ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $sgpr0_sgpr1 + ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64) + ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[UV]], [[COPY1]] + ; CHECK-NEXT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[UV1]](s32) + ; CHECK-NEXT: $sgpr0_sgpr1 = COPY [[MV]](s64) + ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $sgpr0_sgpr1 + %0:_(s64) = COPY $sgpr0_sgpr1 + %1:_(s32) = COPY $sgpr2 + %2:_(s64) = G_ZEXT %1(s32) + %3:_(s64) = G_OR %0, %2 + $sgpr0_sgpr1 = COPY %3(s64) + SI_RETURN_TO_EPILOG implicit $sgpr0_sgpr1 +... +--- +name: test_combine_or_s64_s32_rhs +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr0_sgpr1, $sgpr2 + ; CHECK-LABEL: name: test_combine_or_s64_s32_rhs + ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $sgpr0_sgpr1 + ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64) + ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[UV]], [[COPY1]] + ; CHECK-NEXT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[OR]](s32), [[UV1]](s32) + ; CHECK-NEXT: $sgpr0_sgpr1 = COPY [[MV]](s64) + ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $sgpr0_sgpr1 + %0:_(s64) = COPY $sgpr0_sgpr1 + %1:_(s32) = COPY $sgpr2 + %2:_(s64) = G_ZEXT %1(s32) + %3:_(s64) = G_OR %2, %0 + $sgpr0_sgpr1 = COPY %3(s64) + SI_RETURN_TO_EPILOG implicit $sgpr0_sgpr1 +... +--- +name: test_combine_or_s64_s32_merge_unmerge +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr0, $sgpr1, $sgpr2 + ; CHECK-LABEL: name: test_combine_or_s64_s32_merge_unmerge + ; CHECK: liveins: $sgpr0, $sgpr1, $sgpr2 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr0 + ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr1 + ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[COPY]], [[COPY2]] + ; CHECK-NEXT: $sgpr0 = COPY [[OR]](s32) + ; CHECK-NEXT: $sgpr1 = COPY [[COPY1]](s32) + ; CHECK-NEXT: SI_RETURN_TO_EPILOG implicit $sgpr0, implicit $sgpr1 + %0:_(s32) = COPY $sgpr0 + %1:_(s32) = COPY $sgpr1 + %2:_(s32) = COPY $sgpr2 + %3:_(s64) = G_MERGE_VALUES %0(s32), %1(s32) + %4:_(s64) = G_ZEXT %2(s32) + %5:_(s64) = G_OR %3, %4 + %6:_(s32), %7:_(s32) = G_UNMERGE_VALUES %5(s64) + $sgpr0 = COPY %6(s32) + $sgpr1 = COPY %7(s32) + SI_RETURN_TO_EPILOG implicit $sgpr0, implicit $sgpr1 +... +--- +name: negative_test_incorrect_types +tracksRegLiveness: true +body: | + bb.0: + liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5 + ; CHECK-LABEL: name: negative_test_incorrect_types + ; CHECK: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s128) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 + ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5 + ; CHECK-NEXT: [[ZEXT:%[0-9]+]]:_(s128) = G_ZEXT [[COPY1]](s64) + ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s128) = G_OR [[COPY]], [[ZEXT]] + ; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[OR]](s128) + %0:_(s128) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 + %1:_(s64) = COPY $vgpr4_vgpr5 + %2:_(s128) = G_ZEXT %1 + %3:_(s128) = G_OR %0, %2 + $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3 +... + diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-phis-no-lane-mask-merging.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-phis-no-lane-mask-merging.ll index 5dff8c1..667fa98 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-phis-no-lane-mask-merging.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-phis-no-lane-mask-merging.ll @@ -227,39 +227,38 @@ exit: define amdgpu_cs void @single_lane_execution_attribute(i32 inreg %.userdata0, <3 x i32> inreg %.WorkgroupId, <3 x i32> %.LocalInvocationId) #0 { ; GFX10-LABEL: single_lane_execution_attribute: ; GFX10: ; %bb.0: ; %.entry -; GFX10-NEXT: s_getpc_b64 s[12:13] -; GFX10-NEXT: s_mov_b32 s12, 0 +; GFX10-NEXT: s_getpc_b64 s[4:5] ; GFX10-NEXT: s_mov_b32 s2, s0 -; GFX10-NEXT: s_mov_b32 s3, s12 +; GFX10-NEXT: s_mov_b32 s3, s5 ; GFX10-NEXT: v_mbcnt_lo_u32_b32 v1, -1, 0 -; GFX10-NEXT: s_or_b64 s[2:3], s[12:13], s[2:3] ; GFX10-NEXT: s_load_dwordx8 s[4:11], s[2:3], 0x0 ; GFX10-NEXT: v_mbcnt_hi_u32_b32 v1, -1, v1 ; GFX10-NEXT: v_lshlrev_b32_e32 v2, 2, v1 ; GFX10-NEXT: v_and_b32_e32 v3, 1, v1 ; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v3 ; GFX10-NEXT: s_xor_b32 s2, vcc_lo, exec_lo -; GFX10-NEXT: s_and_b32 vcc_lo, s2, exec_lo ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_load_dword v2, v2, s[4:7], 0 offen +; GFX10-NEXT: s_and_b32 vcc_lo, exec_lo, s2 +; GFX10-NEXT: s_mov_b32 s2, 0 ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_cmp_eq_u32_e64 s0, 0, v2 ; GFX10-NEXT: s_cbranch_vccnz .LBB4_4 ; GFX10-NEXT: ; %bb.1: ; %.preheader.preheader -; GFX10-NEXT: s_mov_b32 s2, 0 +; GFX10-NEXT: s_mov_b32 s3, 0 ; GFX10-NEXT: .LBB4_2: ; %.preheader ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: v_mov_b32_e32 v3, s12 +; GFX10-NEXT: v_mov_b32_e32 v3, s2 ; GFX10-NEXT: v_add_nc_u32_e32 v1, -1, v1 -; GFX10-NEXT: s_add_i32 s12, s12, 4 +; GFX10-NEXT: s_add_i32 s2, s2, 4 ; GFX10-NEXT: buffer_load_dword v3, v3, s[4:7], 0 offen ; GFX10-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_readfirstlane_b32 s3, v3 -; GFX10-NEXT: s_add_i32 s2, s3, s2 +; GFX10-NEXT: v_readfirstlane_b32 s12, v3 +; GFX10-NEXT: s_add_i32 s3, s12, s3 ; GFX10-NEXT: s_cbranch_vccnz .LBB4_2 ; GFX10-NEXT: ; %bb.3: ; %.preheader._crit_edge -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, s2, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, s3, v2 ; GFX10-NEXT: s_or_b32 s2, s0, vcc_lo ; GFX10-NEXT: v_cndmask_b32_e64 v1, 0, 1, s2 ; GFX10-NEXT: s_branch .LBB4_6 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fshl.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fshl.ll index bd53032..715a777 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fshl.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fshl.ll @@ -4934,17 +4934,15 @@ define amdgpu_ps i64 @s_fshl_i64_5(i64 inreg %lhs, i64 inreg %rhs) { ; GCN: ; %bb.0: ; GCN-NEXT: s_lshl_b64 s[0:1], s[0:1], 5 ; GCN-NEXT: s_lshr_b32 s2, s3, 27 -; GCN-NEXT: s_mov_b32 s3, 0 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GCN-NEXT: s_or_b32 s0, s0, s2 ; GCN-NEXT: ; return to shader part epilog ; ; GFX11-LABEL: s_fshl_i64_5: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 5 ; GFX11-NEXT: s_lshr_b32 s2, s3, 27 -; GFX11-NEXT: s_mov_b32 s3, 0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GFX11-NEXT: s_or_b32 s0, s0, s2 ; GFX11-NEXT: ; return to shader part epilog %result = call i64 @llvm.fshl.i64(i64 %lhs, i64 %rhs, i64 5) ret i64 %result @@ -4954,20 +4952,13 @@ define amdgpu_ps i64 @s_fshl_i64_32(i64 inreg %lhs, i64 inreg %rhs) { ; GCN-LABEL: s_fshl_i64_32: ; GCN: ; %bb.0: ; GCN-NEXT: s_mov_b32 s1, s0 -; GCN-NEXT: s_mov_b32 s0, 0 -; GCN-NEXT: s_mov_b32 s2, s3 -; GCN-NEXT: s_mov_b32 s3, s0 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GCN-NEXT: s_mov_b32 s0, s3 ; GCN-NEXT: ; return to shader part epilog ; ; GFX11-LABEL: s_fshl_i64_32: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_mov_b32 s1, s0 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: s_mov_b32 s2, s3 -; GFX11-NEXT: s_mov_b32 s3, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GFX11-NEXT: s_mov_b32 s0, s3 ; GFX11-NEXT: ; return to shader part epilog %result = call i64 @llvm.fshl.i64(i64 %lhs, i64 %rhs, i64 32) ret i64 %result @@ -6823,56 +6814,50 @@ define amdgpu_ps i128 @s_fshl_i128_65(i128 inreg %lhs, i128 inreg %rhs) { ; GFX6: ; %bb.0: ; GFX6-NEXT: s_lshl_b64 s[2:3], s[0:1], 1 ; GFX6-NEXT: s_lshr_b32 s4, s5, 31 -; GFX6-NEXT: s_mov_b32 s5, 0 ; GFX6-NEXT: s_lshl_b64 s[0:1], s[6:7], 1 -; GFX6-NEXT: s_or_b64 s[0:1], s[4:5], s[0:1] +; GFX6-NEXT: s_or_b32 s0, s0, s4 ; GFX6-NEXT: s_lshr_b32 s4, s7, 31 -; GFX6-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GFX6-NEXT: s_or_b32 s2, s2, s4 ; GFX6-NEXT: ; return to shader part epilog ; ; GFX8-LABEL: s_fshl_i128_65: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_lshl_b64 s[2:3], s[0:1], 1 ; GFX8-NEXT: s_lshr_b32 s4, s5, 31 -; GFX8-NEXT: s_mov_b32 s5, 0 ; GFX8-NEXT: s_lshl_b64 s[0:1], s[6:7], 1 -; GFX8-NEXT: s_or_b64 s[0:1], s[4:5], s[0:1] +; GFX8-NEXT: s_or_b32 s0, s0, s4 ; GFX8-NEXT: s_lshr_b32 s4, s7, 31 -; GFX8-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GFX8-NEXT: s_or_b32 s2, s2, s4 ; GFX8-NEXT: ; return to shader part epilog ; ; GFX9-LABEL: s_fshl_i128_65: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_lshl_b64 s[2:3], s[0:1], 1 ; GFX9-NEXT: s_lshr_b32 s4, s5, 31 -; GFX9-NEXT: s_mov_b32 s5, 0 ; GFX9-NEXT: s_lshl_b64 s[0:1], s[6:7], 1 -; GFX9-NEXT: s_or_b64 s[0:1], s[4:5], s[0:1] +; GFX9-NEXT: s_or_b32 s0, s0, s4 ; GFX9-NEXT: s_lshr_b32 s4, s7, 31 -; GFX9-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GFX9-NEXT: s_or_b32 s2, s2, s4 ; GFX9-NEXT: ; return to shader part epilog ; ; GFX10-LABEL: s_fshl_i128_65: ; GFX10: ; %bb.0: -; GFX10-NEXT: s_lshr_b32 s2, s5, 31 -; GFX10-NEXT: s_mov_b32 s3, 0 -; GFX10-NEXT: s_lshl_b64 s[4:5], s[6:7], 1 -; GFX10-NEXT: s_lshl_b64 s[8:9], s[0:1], 1 -; GFX10-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5] -; GFX10-NEXT: s_lshr_b32 s2, s7, 31 -; GFX10-NEXT: s_or_b64 s[2:3], s[8:9], s[2:3] +; GFX10-NEXT: s_lshl_b64 s[2:3], s[0:1], 1 +; GFX10-NEXT: s_lshr_b32 s4, s5, 31 +; GFX10-NEXT: s_lshl_b64 s[0:1], s[6:7], 1 +; GFX10-NEXT: s_lshr_b32 s5, s7, 31 +; GFX10-NEXT: s_or_b32 s0, s0, s4 +; GFX10-NEXT: s_or_b32 s2, s2, s5 ; GFX10-NEXT: ; return to shader part epilog ; ; GFX11-LABEL: s_fshl_i128_65: ; GFX11: ; %bb.0: -; GFX11-NEXT: s_lshr_b32 s2, s5, 31 -; GFX11-NEXT: s_mov_b32 s3, 0 -; GFX11-NEXT: s_lshl_b64 s[4:5], s[6:7], 1 -; GFX11-NEXT: s_lshl_b64 s[8:9], s[0:1], 1 -; GFX11-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5] -; GFX11-NEXT: s_lshr_b32 s2, s7, 31 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_or_b64 s[2:3], s[8:9], s[2:3] +; GFX11-NEXT: s_lshl_b64 s[2:3], s[0:1], 1 +; GFX11-NEXT: s_lshr_b32 s4, s5, 31 +; GFX11-NEXT: s_lshl_b64 s[0:1], s[6:7], 1 +; GFX11-NEXT: s_lshr_b32 s5, s7, 31 +; GFX11-NEXT: s_or_b32 s0, s0, s4 +; GFX11-NEXT: s_or_b32 s2, s2, s5 ; GFX11-NEXT: ; return to shader part epilog %result = call i128 @llvm.fshl.i128(i128 %lhs, i128 %rhs, i128 65) ret i128 %result @@ -6885,7 +6870,7 @@ define i128 @v_fshl_i128_65(i128 %lhs, i128 %rhs) { ; GFX6-NEXT: v_lshl_b64 v[2:3], v[0:1], 1 ; GFX6-NEXT: v_lshl_b64 v[0:1], v[6:7], 1 ; GFX6-NEXT: v_lshrrev_b32_e32 v4, 31, v5 -; GFX6-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX6-NEXT: v_or_b32_e32 v0, v0, v4 ; GFX6-NEXT: v_lshrrev_b32_e32 v4, 31, v7 ; GFX6-NEXT: v_or_b32_e32 v2, v2, v4 ; GFX6-NEXT: s_setpc_b64 s[30:31] @@ -6896,7 +6881,7 @@ define i128 @v_fshl_i128_65(i128 %lhs, i128 %rhs) { ; GFX8-NEXT: v_lshlrev_b64 v[2:3], 1, v[0:1] ; GFX8-NEXT: v_lshlrev_b64 v[0:1], 1, v[6:7] ; GFX8-NEXT: v_lshrrev_b32_e32 v4, 31, v5 -; GFX8-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX8-NEXT: v_or_b32_e32 v0, v0, v4 ; GFX8-NEXT: v_lshrrev_b32_e32 v4, 31, v7 ; GFX8-NEXT: v_or_b32_e32 v2, v2, v4 ; GFX8-NEXT: s_setpc_b64 s[30:31] @@ -6907,7 +6892,7 @@ define i128 @v_fshl_i128_65(i128 %lhs, i128 %rhs) { ; GFX9-NEXT: v_lshlrev_b64 v[2:3], 1, v[0:1] ; GFX9-NEXT: v_lshlrev_b64 v[0:1], 1, v[6:7] ; GFX9-NEXT: v_lshrrev_b32_e32 v4, 31, v5 -; GFX9-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX9-NEXT: v_or_b32_e32 v0, v0, v4 ; GFX9-NEXT: v_lshrrev_b32_e32 v4, 31, v7 ; GFX9-NEXT: v_or_b32_e32 v2, v2, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] @@ -6919,7 +6904,7 @@ define i128 @v_fshl_i128_65(i128 %lhs, i128 %rhs) { ; GFX10-NEXT: v_lshlrev_b64 v[0:1], 1, v[6:7] ; GFX10-NEXT: v_lshrrev_b32_e32 v4, 31, v5 ; GFX10-NEXT: v_lshrrev_b32_e32 v5, 31, v7 -; GFX10-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX10-NEXT: v_or_b32_e32 v0, v0, v4 ; GFX10-NEXT: v_or_b32_e32 v2, v2, v5 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; @@ -6931,7 +6916,7 @@ define i128 @v_fshl_i128_65(i128 %lhs, i128 %rhs) { ; GFX11-NEXT: v_lshrrev_b32_e32 v4, 31, v5 ; GFX11-NEXT: v_lshrrev_b32_e32 v5, 31, v7 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NEXT: v_or_b32_e32 v0, v4, v0 +; GFX11-NEXT: v_or_b32_e32 v0, v0, v4 ; GFX11-NEXT: v_or_b32_e32 v2, v2, v5 ; GFX11-NEXT: s_setpc_b64 s[30:31] %result = call i128 @llvm.fshl.i128(i128 %lhs, i128 %rhs, i128 65) diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fshr.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fshr.ll index ea6b3a3..5aa5a671 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fshr.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fshr.ll @@ -4715,20 +4715,13 @@ define amdgpu_ps i64 @s_fshr_i64_32(i64 inreg %lhs, i64 inreg %rhs) { ; GCN-LABEL: s_fshr_i64_32: ; GCN: ; %bb.0: ; GCN-NEXT: s_mov_b32 s1, s0 -; GCN-NEXT: s_mov_b32 s0, 0 -; GCN-NEXT: s_mov_b32 s2, s3 -; GCN-NEXT: s_mov_b32 s3, s0 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GCN-NEXT: s_mov_b32 s0, s3 ; GCN-NEXT: ; return to shader part epilog ; ; GFX11-LABEL: s_fshr_i64_32: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_mov_b32 s1, s0 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: s_mov_b32 s2, s3 -; GFX11-NEXT: s_mov_b32 s3, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GFX11-NEXT: s_mov_b32 s0, s3 ; GFX11-NEXT: ; return to shader part epilog %result = call i64 @llvm.fshr.i64(i64 %lhs, i64 %rhs, i64 32) ret i64 %result @@ -4739,17 +4732,15 @@ define amdgpu_ps i64 @s_fshr_i64_48(i64 inreg %lhs, i64 inreg %rhs) { ; GCN: ; %bb.0: ; GCN-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 ; GCN-NEXT: s_lshr_b32 s2, s3, 16 -; GCN-NEXT: s_mov_b32 s3, 0 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GCN-NEXT: s_or_b32 s0, s0, s2 ; GCN-NEXT: ; return to shader part epilog ; ; GFX11-LABEL: s_fshr_i64_48: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 16 ; GFX11-NEXT: s_lshr_b32 s2, s3, 16 -; GFX11-NEXT: s_mov_b32 s3, 0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GFX11-NEXT: s_or_b32 s0, s0, s2 ; GFX11-NEXT: ; return to shader part epilog %result = call i64 @llvm.fshr.i64(i64 %lhs, i64 %rhs, i64 48) ret i64 %result @@ -5293,34 +5284,33 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX6-NEXT: s_lshl_b64 s[10:11], s[0:1], 1 ; GFX6-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX6-NEXT: s_lshr_b32 s0, s1, 31 -; GFX6-NEXT: s_mov_b32 s1, 0 -; GFX6-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] -; GFX6-NEXT: s_andn2_b32 s2, 0x7f, s8 +; GFX6-NEXT: s_or_b32 s2, s2, s0 +; GFX6-NEXT: s_andn2_b32 s0, 0x7f, s8 ; GFX6-NEXT: s_not_b32 s9, s8 -; GFX6-NEXT: s_sub_i32 s16, s2, 64 -; GFX6-NEXT: s_sub_i32 s12, 64, s2 -; GFX6-NEXT: s_cmp_lt_u32 s2, 64 +; GFX6-NEXT: s_sub_i32 s16, s0, 64 +; GFX6-NEXT: s_sub_i32 s12, 64, s0 +; GFX6-NEXT: s_cmp_lt_u32 s0, 64 ; GFX6-NEXT: s_cselect_b32 s17, 1, 0 -; GFX6-NEXT: s_cmp_eq_u32 s2, 0 +; GFX6-NEXT: s_cmp_eq_u32 s0, 0 ; GFX6-NEXT: s_cselect_b32 s18, 1, 0 ; GFX6-NEXT: s_lshr_b64 s[12:13], s[10:11], s12 -; GFX6-NEXT: s_lshl_b64 s[14:15], s[0:1], s9 -; GFX6-NEXT: s_lshl_b64 s[2:3], s[10:11], s9 +; GFX6-NEXT: s_lshl_b64 s[14:15], s[2:3], s9 +; GFX6-NEXT: s_lshl_b64 s[0:1], s[10:11], s9 ; GFX6-NEXT: s_or_b64 s[12:13], s[12:13], s[14:15] ; GFX6-NEXT: s_lshl_b64 s[10:11], s[10:11], s16 ; GFX6-NEXT: s_cmp_lg_u32 s17, 0 -; GFX6-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 +; GFX6-NEXT: s_cselect_b64 s[0:1], s[0:1], 0 ; GFX6-NEXT: s_cselect_b64 s[10:11], s[12:13], s[10:11] ; GFX6-NEXT: s_cmp_lg_u32 s18, 0 -; GFX6-NEXT: s_cselect_b64 s[10:11], s[0:1], s[10:11] -; GFX6-NEXT: s_and_b32 s0, s8, 0x7f -; GFX6-NEXT: s_sub_i32 s14, s0, 64 -; GFX6-NEXT: s_sub_i32 s12, 64, s0 -; GFX6-NEXT: s_cmp_lt_u32 s0, 64 +; GFX6-NEXT: s_cselect_b64 s[2:3], s[2:3], s[10:11] +; GFX6-NEXT: s_and_b32 s9, s8, 0x7f +; GFX6-NEXT: s_sub_i32 s14, s9, 64 +; GFX6-NEXT: s_sub_i32 s12, 64, s9 +; GFX6-NEXT: s_cmp_lt_u32 s9, 64 ; GFX6-NEXT: s_cselect_b32 s15, 1, 0 -; GFX6-NEXT: s_cmp_eq_u32 s0, 0 +; GFX6-NEXT: s_cmp_eq_u32 s9, 0 ; GFX6-NEXT: s_cselect_b32 s16, 1, 0 -; GFX6-NEXT: s_lshr_b64 s[0:1], s[6:7], s8 +; GFX6-NEXT: s_lshr_b64 s[10:11], s[6:7], s8 ; GFX6-NEXT: s_lshr_b64 s[8:9], s[4:5], s8 ; GFX6-NEXT: s_lshl_b64 s[12:13], s[6:7], s12 ; GFX6-NEXT: s_or_b64 s[8:9], s[8:9], s[12:13] @@ -5330,9 +5320,9 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX6-NEXT: s_cmp_lg_u32 s16, 0 ; GFX6-NEXT: s_cselect_b64 s[4:5], s[4:5], s[6:7] ; GFX6-NEXT: s_cmp_lg_u32 s15, 0 -; GFX6-NEXT: s_cselect_b64 s[6:7], s[0:1], 0 -; GFX6-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5] -; GFX6-NEXT: s_or_b64 s[2:3], s[10:11], s[6:7] +; GFX6-NEXT: s_cselect_b64 s[6:7], s[10:11], 0 +; GFX6-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] +; GFX6-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7] ; GFX6-NEXT: ; return to shader part epilog ; ; GFX8-LABEL: s_fshr_i128: @@ -5340,34 +5330,33 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX8-NEXT: s_lshl_b64 s[10:11], s[0:1], 1 ; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX8-NEXT: s_lshr_b32 s0, s1, 31 -; GFX8-NEXT: s_mov_b32 s1, 0 -; GFX8-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] -; GFX8-NEXT: s_andn2_b32 s2, 0x7f, s8 +; GFX8-NEXT: s_or_b32 s2, s2, s0 +; GFX8-NEXT: s_andn2_b32 s0, 0x7f, s8 ; GFX8-NEXT: s_not_b32 s9, s8 -; GFX8-NEXT: s_sub_i32 s16, s2, 64 -; GFX8-NEXT: s_sub_i32 s12, 64, s2 -; GFX8-NEXT: s_cmp_lt_u32 s2, 64 +; GFX8-NEXT: s_sub_i32 s16, s0, 64 +; GFX8-NEXT: s_sub_i32 s12, 64, s0 +; GFX8-NEXT: s_cmp_lt_u32 s0, 64 ; GFX8-NEXT: s_cselect_b32 s17, 1, 0 -; GFX8-NEXT: s_cmp_eq_u32 s2, 0 +; GFX8-NEXT: s_cmp_eq_u32 s0, 0 ; GFX8-NEXT: s_cselect_b32 s18, 1, 0 ; GFX8-NEXT: s_lshr_b64 s[12:13], s[10:11], s12 -; GFX8-NEXT: s_lshl_b64 s[14:15], s[0:1], s9 -; GFX8-NEXT: s_lshl_b64 s[2:3], s[10:11], s9 +; GFX8-NEXT: s_lshl_b64 s[14:15], s[2:3], s9 +; GFX8-NEXT: s_lshl_b64 s[0:1], s[10:11], s9 ; GFX8-NEXT: s_or_b64 s[12:13], s[12:13], s[14:15] ; GFX8-NEXT: s_lshl_b64 s[10:11], s[10:11], s16 ; GFX8-NEXT: s_cmp_lg_u32 s17, 0 -; GFX8-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 +; GFX8-NEXT: s_cselect_b64 s[0:1], s[0:1], 0 ; GFX8-NEXT: s_cselect_b64 s[10:11], s[12:13], s[10:11] ; GFX8-NEXT: s_cmp_lg_u32 s18, 0 -; GFX8-NEXT: s_cselect_b64 s[10:11], s[0:1], s[10:11] -; GFX8-NEXT: s_and_b32 s0, s8, 0x7f -; GFX8-NEXT: s_sub_i32 s14, s0, 64 -; GFX8-NEXT: s_sub_i32 s12, 64, s0 -; GFX8-NEXT: s_cmp_lt_u32 s0, 64 +; GFX8-NEXT: s_cselect_b64 s[2:3], s[2:3], s[10:11] +; GFX8-NEXT: s_and_b32 s9, s8, 0x7f +; GFX8-NEXT: s_sub_i32 s14, s9, 64 +; GFX8-NEXT: s_sub_i32 s12, 64, s9 +; GFX8-NEXT: s_cmp_lt_u32 s9, 64 ; GFX8-NEXT: s_cselect_b32 s15, 1, 0 -; GFX8-NEXT: s_cmp_eq_u32 s0, 0 +; GFX8-NEXT: s_cmp_eq_u32 s9, 0 ; GFX8-NEXT: s_cselect_b32 s16, 1, 0 -; GFX8-NEXT: s_lshr_b64 s[0:1], s[6:7], s8 +; GFX8-NEXT: s_lshr_b64 s[10:11], s[6:7], s8 ; GFX8-NEXT: s_lshr_b64 s[8:9], s[4:5], s8 ; GFX8-NEXT: s_lshl_b64 s[12:13], s[6:7], s12 ; GFX8-NEXT: s_or_b64 s[8:9], s[8:9], s[12:13] @@ -5377,9 +5366,9 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX8-NEXT: s_cmp_lg_u32 s16, 0 ; GFX8-NEXT: s_cselect_b64 s[4:5], s[4:5], s[6:7] ; GFX8-NEXT: s_cmp_lg_u32 s15, 0 -; GFX8-NEXT: s_cselect_b64 s[6:7], s[0:1], 0 -; GFX8-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5] -; GFX8-NEXT: s_or_b64 s[2:3], s[10:11], s[6:7] +; GFX8-NEXT: s_cselect_b64 s[6:7], s[10:11], 0 +; GFX8-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] +; GFX8-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7] ; GFX8-NEXT: ; return to shader part epilog ; ; GFX9-LABEL: s_fshr_i128: @@ -5387,34 +5376,33 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX9-NEXT: s_lshl_b64 s[10:11], s[0:1], 1 ; GFX9-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX9-NEXT: s_lshr_b32 s0, s1, 31 -; GFX9-NEXT: s_mov_b32 s1, 0 -; GFX9-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] -; GFX9-NEXT: s_andn2_b32 s2, 0x7f, s8 +; GFX9-NEXT: s_or_b32 s2, s2, s0 +; GFX9-NEXT: s_andn2_b32 s0, 0x7f, s8 ; GFX9-NEXT: s_not_b32 s9, s8 -; GFX9-NEXT: s_sub_i32 s16, s2, 64 -; GFX9-NEXT: s_sub_i32 s12, 64, s2 -; GFX9-NEXT: s_cmp_lt_u32 s2, 64 +; GFX9-NEXT: s_sub_i32 s16, s0, 64 +; GFX9-NEXT: s_sub_i32 s12, 64, s0 +; GFX9-NEXT: s_cmp_lt_u32 s0, 64 ; GFX9-NEXT: s_cselect_b32 s17, 1, 0 -; GFX9-NEXT: s_cmp_eq_u32 s2, 0 +; GFX9-NEXT: s_cmp_eq_u32 s0, 0 ; GFX9-NEXT: s_cselect_b32 s18, 1, 0 ; GFX9-NEXT: s_lshr_b64 s[12:13], s[10:11], s12 -; GFX9-NEXT: s_lshl_b64 s[14:15], s[0:1], s9 -; GFX9-NEXT: s_lshl_b64 s[2:3], s[10:11], s9 +; GFX9-NEXT: s_lshl_b64 s[14:15], s[2:3], s9 +; GFX9-NEXT: s_lshl_b64 s[0:1], s[10:11], s9 ; GFX9-NEXT: s_or_b64 s[12:13], s[12:13], s[14:15] ; GFX9-NEXT: s_lshl_b64 s[10:11], s[10:11], s16 ; GFX9-NEXT: s_cmp_lg_u32 s17, 0 -; GFX9-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 +; GFX9-NEXT: s_cselect_b64 s[0:1], s[0:1], 0 ; GFX9-NEXT: s_cselect_b64 s[10:11], s[12:13], s[10:11] ; GFX9-NEXT: s_cmp_lg_u32 s18, 0 -; GFX9-NEXT: s_cselect_b64 s[10:11], s[0:1], s[10:11] -; GFX9-NEXT: s_and_b32 s0, s8, 0x7f -; GFX9-NEXT: s_sub_i32 s14, s0, 64 -; GFX9-NEXT: s_sub_i32 s12, 64, s0 -; GFX9-NEXT: s_cmp_lt_u32 s0, 64 +; GFX9-NEXT: s_cselect_b64 s[2:3], s[2:3], s[10:11] +; GFX9-NEXT: s_and_b32 s9, s8, 0x7f +; GFX9-NEXT: s_sub_i32 s14, s9, 64 +; GFX9-NEXT: s_sub_i32 s12, 64, s9 +; GFX9-NEXT: s_cmp_lt_u32 s9, 64 ; GFX9-NEXT: s_cselect_b32 s15, 1, 0 -; GFX9-NEXT: s_cmp_eq_u32 s0, 0 +; GFX9-NEXT: s_cmp_eq_u32 s9, 0 ; GFX9-NEXT: s_cselect_b32 s16, 1, 0 -; GFX9-NEXT: s_lshr_b64 s[0:1], s[6:7], s8 +; GFX9-NEXT: s_lshr_b64 s[10:11], s[6:7], s8 ; GFX9-NEXT: s_lshr_b64 s[8:9], s[4:5], s8 ; GFX9-NEXT: s_lshl_b64 s[12:13], s[6:7], s12 ; GFX9-NEXT: s_or_b64 s[8:9], s[8:9], s[12:13] @@ -5424,19 +5412,18 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX9-NEXT: s_cmp_lg_u32 s16, 0 ; GFX9-NEXT: s_cselect_b64 s[4:5], s[4:5], s[6:7] ; GFX9-NEXT: s_cmp_lg_u32 s15, 0 -; GFX9-NEXT: s_cselect_b64 s[6:7], s[0:1], 0 -; GFX9-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5] -; GFX9-NEXT: s_or_b64 s[2:3], s[10:11], s[6:7] +; GFX9-NEXT: s_cselect_b64 s[6:7], s[10:11], 0 +; GFX9-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] +; GFX9-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7] ; GFX9-NEXT: ; return to shader part epilog ; ; GFX10-LABEL: s_fshr_i128: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX10-NEXT: s_lshr_b32 s10, s1, 31 -; GFX10-NEXT: s_mov_b32 s11, 0 -; GFX10-NEXT: s_andn2_b32 s9, 0x7f, s8 +; GFX10-NEXT: s_lshr_b32 s9, s1, 31 ; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX10-NEXT: s_or_b64 s[2:3], s[2:3], s[10:11] +; GFX10-NEXT: s_or_b32 s2, s2, s9 +; GFX10-NEXT: s_andn2_b32 s9, 0x7f, s8 ; GFX10-NEXT: s_not_b32 s14, s8 ; GFX10-NEXT: s_sub_i32 s16, s9, 64 ; GFX10-NEXT: s_sub_i32 s10, 64, s9 @@ -5479,11 +5466,10 @@ define amdgpu_ps i128 @s_fshr_i128(i128 inreg %lhs, i128 inreg %rhs, i128 inreg ; GFX11-LABEL: s_fshr_i128: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX11-NEXT: s_lshr_b32 s10, s1, 31 -; GFX11-NEXT: s_mov_b32 s11, 0 -; GFX11-NEXT: s_and_not1_b32 s9, 0x7f, s8 +; GFX11-NEXT: s_lshr_b32 s9, s1, 31 ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX11-NEXT: s_or_b64 s[2:3], s[2:3], s[10:11] +; GFX11-NEXT: s_or_b32 s2, s2, s9 +; GFX11-NEXT: s_and_not1_b32 s9, 0x7f, s8 ; GFX11-NEXT: s_not_b32 s14, s8 ; GFX11-NEXT: s_sub_i32 s16, s9, 64 ; GFX11-NEXT: s_sub_i32 s10, 64, s9 @@ -5786,13 +5772,12 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX6-NEXT: s_lshl_b64 s[8:9], s[0:1], 1 ; GFX6-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX6-NEXT: s_lshr_b32 s0, s1, 31 -; GFX6-NEXT: s_mov_b32 s1, 0 ; GFX6-NEXT: v_bfi_b32 v7, v0, 0, v1 -; GFX6-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] +; GFX6-NEXT: s_or_b32 s2, s2, s0 ; GFX6-NEXT: v_sub_i32_e32 v1, vcc, 64, v7 ; GFX6-NEXT: v_not_b32_e32 v8, 63 ; GFX6-NEXT: v_lshr_b64 v[1:2], s[8:9], v1 -; GFX6-NEXT: v_lshl_b64 v[3:4], s[0:1], v7 +; GFX6-NEXT: v_lshl_b64 v[3:4], s[2:3], v7 ; GFX6-NEXT: v_add_i32_e32 v9, vcc, v7, v8 ; GFX6-NEXT: v_lshl_b64 v[5:6], s[8:9], v7 ; GFX6-NEXT: v_or_b32_e32 v3, v1, v3 @@ -5803,8 +5788,8 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX6-NEXT: v_cndmask_b32_e32 v6, 0, v6, vcc ; GFX6-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc ; GFX6-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc -; GFX6-NEXT: v_mov_b32_e32 v3, s0 -; GFX6-NEXT: v_mov_b32_e32 v4, s1 +; GFX6-NEXT: v_mov_b32_e32 v3, s2 +; GFX6-NEXT: v_mov_b32_e32 v4, s3 ; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 0, v7 ; GFX6-NEXT: v_and_b32_e32 v11, 0x7f, v0 ; GFX6-NEXT: v_cndmask_b32_e32 v7, v1, v3, vcc @@ -5839,13 +5824,12 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX8-NEXT: s_lshl_b64 s[8:9], s[0:1], 1 ; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX8-NEXT: s_lshr_b32 s0, s1, 31 -; GFX8-NEXT: s_mov_b32 s1, 0 ; GFX8-NEXT: v_bfi_b32 v7, v0, 0, v1 -; GFX8-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] +; GFX8-NEXT: s_or_b32 s2, s2, s0 ; GFX8-NEXT: v_sub_u32_e32 v1, vcc, 64, v7 ; GFX8-NEXT: v_not_b32_e32 v8, 63 ; GFX8-NEXT: v_lshrrev_b64 v[1:2], v1, s[8:9] -; GFX8-NEXT: v_lshlrev_b64 v[3:4], v7, s[0:1] +; GFX8-NEXT: v_lshlrev_b64 v[3:4], v7, s[2:3] ; GFX8-NEXT: v_add_u32_e32 v9, vcc, v7, v8 ; GFX8-NEXT: v_lshlrev_b64 v[5:6], v7, s[8:9] ; GFX8-NEXT: v_or_b32_e32 v3, v1, v3 @@ -5856,8 +5840,8 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX8-NEXT: v_cndmask_b32_e32 v6, 0, v6, vcc ; GFX8-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc ; GFX8-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc -; GFX8-NEXT: v_mov_b32_e32 v3, s0 -; GFX8-NEXT: v_mov_b32_e32 v4, s1 +; GFX8-NEXT: v_mov_b32_e32 v3, s2 +; GFX8-NEXT: v_mov_b32_e32 v4, s3 ; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v7 ; GFX8-NEXT: v_and_b32_e32 v11, 0x7f, v0 ; GFX8-NEXT: v_cndmask_b32_e32 v7, v1, v3, vcc @@ -5892,12 +5876,11 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX9-NEXT: s_lshl_b64 s[8:9], s[0:1], 1 ; GFX9-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX9-NEXT: s_lshr_b32 s0, s1, 31 -; GFX9-NEXT: s_mov_b32 s1, 0 ; GFX9-NEXT: v_bfi_b32 v7, v0, 0, v1 -; GFX9-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] +; GFX9-NEXT: s_or_b32 s2, s2, s0 ; GFX9-NEXT: v_sub_u32_e32 v1, 64, v7 ; GFX9-NEXT: v_lshrrev_b64 v[1:2], v1, s[8:9] -; GFX9-NEXT: v_lshlrev_b64 v[3:4], v7, s[0:1] +; GFX9-NEXT: v_lshlrev_b64 v[3:4], v7, s[2:3] ; GFX9-NEXT: v_add_u32_e32 v8, 0xffffffc0, v7 ; GFX9-NEXT: v_lshlrev_b64 v[5:6], v7, s[8:9] ; GFX9-NEXT: v_or_b32_e32 v3, v1, v3 @@ -5908,10 +5891,10 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX9-NEXT: v_cndmask_b32_e32 v6, 0, v6, vcc ; GFX9-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc ; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc -; GFX9-NEXT: v_mov_b32_e32 v4, s1 +; GFX9-NEXT: v_mov_b32_e32 v4, s3 ; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v7 ; GFX9-NEXT: v_and_b32_e32 v10, 0x7f, v0 -; GFX9-NEXT: v_mov_b32_e32 v3, s0 +; GFX9-NEXT: v_mov_b32_e32 v3, s2 ; GFX9-NEXT: v_cndmask_b32_e32 v9, v2, v4, vcc ; GFX9-NEXT: v_sub_u32_e32 v2, 64, v10 ; GFX9-NEXT: v_cndmask_b32_e32 v7, v1, v3, vcc @@ -5941,34 +5924,33 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX10-LABEL: v_fshr_i128_ssv: ; GFX10: ; %bb.0: ; GFX10-NEXT: v_bfi_b32 v11, v0, 0, 0x7f -; GFX10-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX10-NEXT: s_lshr_b32 s8, s1, 31 -; GFX10-NEXT: s_mov_b32 s9, 0 +; GFX10-NEXT: s_lshl_b64 s[8:9], s[2:3], 1 +; GFX10-NEXT: s_lshr_b32 s2, s1, 31 ; GFX10-NEXT: v_and_b32_e32 v12, 0x7f, v0 -; GFX10-NEXT: v_sub_nc_u32_e32 v1, 64, v11 ; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX10-NEXT: s_or_b64 s[8:9], s[2:3], s[8:9] +; GFX10-NEXT: v_sub_nc_u32_e32 v1, 64, v11 +; GFX10-NEXT: s_or_b32 s8, s8, s2 ; GFX10-NEXT: v_add_nc_u32_e32 v0, 0xffffffc0, v11 ; GFX10-NEXT: v_lshlrev_b64 v[3:4], v11, s[8:9] -; GFX10-NEXT: v_lshrrev_b64 v[1:2], v1, s[0:1] ; GFX10-NEXT: v_sub_nc_u32_e32 v9, 64, v12 +; GFX10-NEXT: v_lshrrev_b64 v[1:2], v1, s[0:1] ; GFX10-NEXT: v_cmp_gt_u32_e32 vcc_lo, 64, v11 ; GFX10-NEXT: v_add_nc_u32_e32 v13, 0xffffffc0, v12 ; GFX10-NEXT: v_lshrrev_b64 v[7:8], v12, s[4:5] +; GFX10-NEXT: v_lshlrev_b64 v[9:10], v9, s[6:7] ; GFX10-NEXT: v_lshlrev_b64 v[5:6], v11, s[0:1] ; GFX10-NEXT: v_or_b32_e32 v3, v1, v3 ; GFX10-NEXT: v_lshlrev_b64 v[0:1], v0, s[0:1] -; GFX10-NEXT: v_lshlrev_b64 v[9:10], v9, s[6:7] ; GFX10-NEXT: v_or_b32_e32 v4, v2, v4 ; GFX10-NEXT: v_cmp_gt_u32_e64 s1, 64, v12 ; GFX10-NEXT: v_cmp_eq_u32_e64 s0, 0, v11 ; GFX10-NEXT: v_cmp_eq_u32_e64 s2, 0, v12 +; GFX10-NEXT: v_cndmask_b32_e32 v5, 0, v5, vcc_lo ; GFX10-NEXT: v_cndmask_b32_e32 v14, v0, v3, vcc_lo ; GFX10-NEXT: v_lshrrev_b64 v[2:3], v13, s[6:7] ; GFX10-NEXT: v_or_b32_e32 v0, v7, v9 ; GFX10-NEXT: v_or_b32_e32 v7, v8, v10 ; GFX10-NEXT: v_cndmask_b32_e32 v4, v1, v4, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e32 v5, 0, v5, vcc_lo ; GFX10-NEXT: v_cndmask_b32_e32 v6, 0, v6, vcc_lo ; GFX10-NEXT: v_cndmask_b32_e64 v2, v2, v0, s1 ; GFX10-NEXT: v_lshrrev_b64 v[0:1], v12, s[6:7] @@ -5988,18 +5970,18 @@ define amdgpu_ps <4 x float> @v_fshr_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, ; GFX11-LABEL: v_fshr_i128_ssv: ; GFX11: ; %bb.0: ; GFX11-NEXT: v_bfi_b32 v11, v0, 0, 0x7f -; GFX11-NEXT: s_lshr_b32 s8, s1, 31 +; GFX11-NEXT: s_lshl_b64 s[8:9], s[2:3], 1 +; GFX11-NEXT: s_lshr_b32 s2, s1, 31 ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX11-NEXT: s_mov_b32 s9, 0 +; GFX11-NEXT: s_or_b32 s8, s8, s2 ; GFX11-NEXT: v_sub_nc_u32_e32 v1, 64, v11 ; GFX11-NEXT: v_lshlrev_b64 v[5:6], v11, s[0:1] ; GFX11-NEXT: v_cmp_gt_u32_e32 vcc_lo, 64, v11 ; GFX11-NEXT: v_and_b32_e32 v12, 0x7f, v0 -; GFX11-NEXT: s_or_b64 s[8:9], s[2:3], s[8:9] -; GFX11-NEXT: v_lshrrev_b64 v[1:2], v1, s[0:1] ; GFX11-NEXT: v_lshlrev_b64 v[3:4], v11, s[8:9] +; GFX11-NEXT: v_lshrrev_b64 v[1:2], v1, s[0:1] ; GFX11-NEXT: v_dual_cndmask_b32 v5, 0, v5 :: v_dual_add_nc_u32 v0, 0xffffffc0, v11 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) ; GFX11-NEXT: v_sub_nc_u32_e32 v9, 64, v12 ; GFX11-NEXT: v_lshrrev_b64 v[7:8], v12, s[4:5] ; GFX11-NEXT: v_cmp_eq_u32_e64 s2, 0, v12 @@ -6045,26 +6027,25 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX6-NEXT: s_lshl_b64 s[6:7], s[0:1], 1 ; GFX6-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX6-NEXT: s_lshr_b32 s0, s1, 31 -; GFX6-NEXT: s_mov_b32 s1, 0 -; GFX6-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] -; GFX6-NEXT: s_andn2_b32 s2, 0x7f, s4 +; GFX6-NEXT: s_or_b32 s2, s2, s0 +; GFX6-NEXT: s_andn2_b32 s0, 0x7f, s4 ; GFX6-NEXT: s_not_b32 s5, s4 -; GFX6-NEXT: s_sub_i32 s12, s2, 64 -; GFX6-NEXT: s_sub_i32 s8, 64, s2 -; GFX6-NEXT: s_cmp_lt_u32 s2, 64 +; GFX6-NEXT: s_sub_i32 s12, s0, 64 +; GFX6-NEXT: s_sub_i32 s8, 64, s0 +; GFX6-NEXT: s_cmp_lt_u32 s0, 64 ; GFX6-NEXT: s_cselect_b32 s13, 1, 0 -; GFX6-NEXT: s_cmp_eq_u32 s2, 0 +; GFX6-NEXT: s_cmp_eq_u32 s0, 0 ; GFX6-NEXT: s_cselect_b32 s14, 1, 0 ; GFX6-NEXT: s_lshr_b64 s[8:9], s[6:7], s8 -; GFX6-NEXT: s_lshl_b64 s[10:11], s[0:1], s5 -; GFX6-NEXT: s_lshl_b64 s[2:3], s[6:7], s5 +; GFX6-NEXT: s_lshl_b64 s[10:11], s[2:3], s5 +; GFX6-NEXT: s_lshl_b64 s[0:1], s[6:7], s5 ; GFX6-NEXT: s_or_b64 s[8:9], s[8:9], s[10:11] ; GFX6-NEXT: s_lshl_b64 s[6:7], s[6:7], s12 ; GFX6-NEXT: s_cmp_lg_u32 s13, 0 -; GFX6-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 -; GFX6-NEXT: s_cselect_b64 s[6:7], s[8:9], s[6:7] +; GFX6-NEXT: s_cselect_b64 s[10:11], s[0:1], 0 +; GFX6-NEXT: s_cselect_b64 s[0:1], s[8:9], s[6:7] ; GFX6-NEXT: s_cmp_lg_u32 s14, 0 -; GFX6-NEXT: s_cselect_b64 s[6:7], s[0:1], s[6:7] +; GFX6-NEXT: s_cselect_b64 s[2:3], s[2:3], s[0:1] ; GFX6-NEXT: s_and_b32 s0, s4, 0x7f ; GFX6-NEXT: s_sub_i32 s1, s0, 64 ; GFX6-NEXT: s_sub_i32 s4, 64, s0 @@ -6073,14 +6054,14 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX6-NEXT: s_cmp_eq_u32 s0, 0 ; GFX6-NEXT: v_lshr_b64 v[4:5], v[0:1], s0 ; GFX6-NEXT: v_lshl_b64 v[6:7], v[2:3], s4 -; GFX6-NEXT: s_cselect_b32 s8, 1, 0 +; GFX6-NEXT: s_cselect_b32 s6, 1, 0 ; GFX6-NEXT: v_lshr_b64 v[8:9], v[2:3], s0 ; GFX6-NEXT: v_lshr_b64 v[2:3], v[2:3], s1 ; GFX6-NEXT: s_and_b32 s0, 1, s5 ; GFX6-NEXT: v_or_b32_e32 v4, v4, v6 ; GFX6-NEXT: v_or_b32_e32 v5, v5, v7 ; GFX6-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 -; GFX6-NEXT: s_and_b32 s0, 1, s8 +; GFX6-NEXT: s_and_b32 s0, 1, s6 ; GFX6-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc ; GFX6-NEXT: v_cndmask_b32_e32 v3, v3, v5, vcc ; GFX6-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0 @@ -6088,10 +6069,10 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX6-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[0:1] ; GFX6-NEXT: v_cndmask_b32_e32 v2, 0, v8, vcc ; GFX6-NEXT: v_cndmask_b32_e32 v3, 0, v9, vcc -; GFX6-NEXT: v_or_b32_e32 v0, s2, v0 -; GFX6-NEXT: v_or_b32_e32 v1, s3, v1 -; GFX6-NEXT: v_or_b32_e32 v2, s6, v2 -; GFX6-NEXT: v_or_b32_e32 v3, s7, v3 +; GFX6-NEXT: v_or_b32_e32 v0, s10, v0 +; GFX6-NEXT: v_or_b32_e32 v1, s11, v1 +; GFX6-NEXT: v_or_b32_e32 v2, s2, v2 +; GFX6-NEXT: v_or_b32_e32 v3, s3, v3 ; GFX6-NEXT: ; return to shader part epilog ; ; GFX8-LABEL: v_fshr_i128_svs: @@ -6099,26 +6080,25 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX8-NEXT: s_lshl_b64 s[6:7], s[0:1], 1 ; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX8-NEXT: s_lshr_b32 s0, s1, 31 -; GFX8-NEXT: s_mov_b32 s1, 0 -; GFX8-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] -; GFX8-NEXT: s_andn2_b32 s2, 0x7f, s4 +; GFX8-NEXT: s_or_b32 s2, s2, s0 +; GFX8-NEXT: s_andn2_b32 s0, 0x7f, s4 ; GFX8-NEXT: s_not_b32 s5, s4 -; GFX8-NEXT: s_sub_i32 s12, s2, 64 -; GFX8-NEXT: s_sub_i32 s8, 64, s2 -; GFX8-NEXT: s_cmp_lt_u32 s2, 64 +; GFX8-NEXT: s_sub_i32 s12, s0, 64 +; GFX8-NEXT: s_sub_i32 s8, 64, s0 +; GFX8-NEXT: s_cmp_lt_u32 s0, 64 ; GFX8-NEXT: s_cselect_b32 s13, 1, 0 -; GFX8-NEXT: s_cmp_eq_u32 s2, 0 +; GFX8-NEXT: s_cmp_eq_u32 s0, 0 ; GFX8-NEXT: s_cselect_b32 s14, 1, 0 ; GFX8-NEXT: s_lshr_b64 s[8:9], s[6:7], s8 -; GFX8-NEXT: s_lshl_b64 s[10:11], s[0:1], s5 -; GFX8-NEXT: s_lshl_b64 s[2:3], s[6:7], s5 +; GFX8-NEXT: s_lshl_b64 s[10:11], s[2:3], s5 +; GFX8-NEXT: s_lshl_b64 s[0:1], s[6:7], s5 ; GFX8-NEXT: s_or_b64 s[8:9], s[8:9], s[10:11] ; GFX8-NEXT: s_lshl_b64 s[6:7], s[6:7], s12 ; GFX8-NEXT: s_cmp_lg_u32 s13, 0 -; GFX8-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 -; GFX8-NEXT: s_cselect_b64 s[6:7], s[8:9], s[6:7] +; GFX8-NEXT: s_cselect_b64 s[10:11], s[0:1], 0 +; GFX8-NEXT: s_cselect_b64 s[0:1], s[8:9], s[6:7] ; GFX8-NEXT: s_cmp_lg_u32 s14, 0 -; GFX8-NEXT: s_cselect_b64 s[6:7], s[0:1], s[6:7] +; GFX8-NEXT: s_cselect_b64 s[2:3], s[2:3], s[0:1] ; GFX8-NEXT: s_and_b32 s0, s4, 0x7f ; GFX8-NEXT: s_sub_i32 s1, s0, 64 ; GFX8-NEXT: s_sub_i32 s4, 64, s0 @@ -6127,14 +6107,14 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX8-NEXT: s_cmp_eq_u32 s0, 0 ; GFX8-NEXT: v_lshrrev_b64 v[4:5], s0, v[0:1] ; GFX8-NEXT: v_lshlrev_b64 v[6:7], s4, v[2:3] -; GFX8-NEXT: s_cselect_b32 s8, 1, 0 +; GFX8-NEXT: s_cselect_b32 s6, 1, 0 ; GFX8-NEXT: v_lshrrev_b64 v[8:9], s0, v[2:3] ; GFX8-NEXT: v_lshrrev_b64 v[2:3], s1, v[2:3] ; GFX8-NEXT: s_and_b32 s0, 1, s5 ; GFX8-NEXT: v_or_b32_e32 v4, v4, v6 ; GFX8-NEXT: v_or_b32_e32 v5, v5, v7 ; GFX8-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 -; GFX8-NEXT: s_and_b32 s0, 1, s8 +; GFX8-NEXT: s_and_b32 s0, 1, s6 ; GFX8-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc ; GFX8-NEXT: v_cndmask_b32_e32 v3, v3, v5, vcc ; GFX8-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0 @@ -6142,10 +6122,10 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX8-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[0:1] ; GFX8-NEXT: v_cndmask_b32_e32 v2, 0, v8, vcc ; GFX8-NEXT: v_cndmask_b32_e32 v3, 0, v9, vcc -; GFX8-NEXT: v_or_b32_e32 v0, s2, v0 -; GFX8-NEXT: v_or_b32_e32 v1, s3, v1 -; GFX8-NEXT: v_or_b32_e32 v2, s6, v2 -; GFX8-NEXT: v_or_b32_e32 v3, s7, v3 +; GFX8-NEXT: v_or_b32_e32 v0, s10, v0 +; GFX8-NEXT: v_or_b32_e32 v1, s11, v1 +; GFX8-NEXT: v_or_b32_e32 v2, s2, v2 +; GFX8-NEXT: v_or_b32_e32 v3, s3, v3 ; GFX8-NEXT: ; return to shader part epilog ; ; GFX9-LABEL: v_fshr_i128_svs: @@ -6153,26 +6133,25 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX9-NEXT: s_lshl_b64 s[6:7], s[0:1], 1 ; GFX9-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GFX9-NEXT: s_lshr_b32 s0, s1, 31 -; GFX9-NEXT: s_mov_b32 s1, 0 -; GFX9-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1] -; GFX9-NEXT: s_andn2_b32 s2, 0x7f, s4 +; GFX9-NEXT: s_or_b32 s2, s2, s0 +; GFX9-NEXT: s_andn2_b32 s0, 0x7f, s4 ; GFX9-NEXT: s_not_b32 s5, s4 -; GFX9-NEXT: s_sub_i32 s12, s2, 64 -; GFX9-NEXT: s_sub_i32 s8, 64, s2 -; GFX9-NEXT: s_cmp_lt_u32 s2, 64 +; GFX9-NEXT: s_sub_i32 s12, s0, 64 +; GFX9-NEXT: s_sub_i32 s8, 64, s0 +; GFX9-NEXT: s_cmp_lt_u32 s0, 64 ; GFX9-NEXT: s_cselect_b32 s13, 1, 0 -; GFX9-NEXT: s_cmp_eq_u32 s2, 0 +; GFX9-NEXT: s_cmp_eq_u32 s0, 0 ; GFX9-NEXT: s_cselect_b32 s14, 1, 0 ; GFX9-NEXT: s_lshr_b64 s[8:9], s[6:7], s8 -; GFX9-NEXT: s_lshl_b64 s[10:11], s[0:1], s5 -; GFX9-NEXT: s_lshl_b64 s[2:3], s[6:7], s5 +; GFX9-NEXT: s_lshl_b64 s[10:11], s[2:3], s5 +; GFX9-NEXT: s_lshl_b64 s[0:1], s[6:7], s5 ; GFX9-NEXT: s_or_b64 s[8:9], s[8:9], s[10:11] ; GFX9-NEXT: s_lshl_b64 s[6:7], s[6:7], s12 ; GFX9-NEXT: s_cmp_lg_u32 s13, 0 -; GFX9-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 -; GFX9-NEXT: s_cselect_b64 s[6:7], s[8:9], s[6:7] +; GFX9-NEXT: s_cselect_b64 s[10:11], s[0:1], 0 +; GFX9-NEXT: s_cselect_b64 s[0:1], s[8:9], s[6:7] ; GFX9-NEXT: s_cmp_lg_u32 s14, 0 -; GFX9-NEXT: s_cselect_b64 s[6:7], s[0:1], s[6:7] +; GFX9-NEXT: s_cselect_b64 s[2:3], s[2:3], s[0:1] ; GFX9-NEXT: s_and_b32 s0, s4, 0x7f ; GFX9-NEXT: s_sub_i32 s1, s0, 64 ; GFX9-NEXT: s_sub_i32 s4, 64, s0 @@ -6181,14 +6160,14 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX9-NEXT: s_cmp_eq_u32 s0, 0 ; GFX9-NEXT: v_lshrrev_b64 v[4:5], s0, v[0:1] ; GFX9-NEXT: v_lshlrev_b64 v[6:7], s4, v[2:3] -; GFX9-NEXT: s_cselect_b32 s8, 1, 0 +; GFX9-NEXT: s_cselect_b32 s6, 1, 0 ; GFX9-NEXT: v_lshrrev_b64 v[8:9], s0, v[2:3] ; GFX9-NEXT: v_lshrrev_b64 v[2:3], s1, v[2:3] ; GFX9-NEXT: s_and_b32 s0, 1, s5 ; GFX9-NEXT: v_or_b32_e32 v4, v4, v6 ; GFX9-NEXT: v_or_b32_e32 v5, v5, v7 ; GFX9-NEXT: v_cmp_ne_u32_e64 vcc, 0, s0 -; GFX9-NEXT: s_and_b32 s0, 1, s8 +; GFX9-NEXT: s_and_b32 s0, 1, s6 ; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc ; GFX9-NEXT: v_cndmask_b32_e32 v3, v3, v5, vcc ; GFX9-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0 @@ -6196,20 +6175,19 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX9-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[0:1] ; GFX9-NEXT: v_cndmask_b32_e32 v2, 0, v8, vcc ; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v9, vcc -; GFX9-NEXT: v_or_b32_e32 v0, s2, v0 -; GFX9-NEXT: v_or_b32_e32 v1, s3, v1 -; GFX9-NEXT: v_or_b32_e32 v2, s6, v2 -; GFX9-NEXT: v_or_b32_e32 v3, s7, v3 +; GFX9-NEXT: v_or_b32_e32 v0, s10, v0 +; GFX9-NEXT: v_or_b32_e32 v1, s11, v1 +; GFX9-NEXT: v_or_b32_e32 v2, s2, v2 +; GFX9-NEXT: v_or_b32_e32 v3, s3, v3 ; GFX9-NEXT: ; return to shader part epilog ; ; GFX10-LABEL: v_fshr_i128_svs: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX10-NEXT: s_lshr_b32 s6, s1, 31 -; GFX10-NEXT: s_mov_b32 s7, 0 -; GFX10-NEXT: s_andn2_b32 s5, 0x7f, s4 +; GFX10-NEXT: s_lshr_b32 s5, s1, 31 ; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX10-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7] +; GFX10-NEXT: s_or_b32 s2, s2, s5 +; GFX10-NEXT: s_andn2_b32 s5, 0x7f, s4 ; GFX10-NEXT: s_not_b32 s10, s4 ; GFX10-NEXT: s_sub_i32 s12, s5, 64 ; GFX10-NEXT: s_sub_i32 s6, 64, s5 @@ -6259,11 +6237,10 @@ define amdgpu_ps <4 x float> @v_fshr_i128_svs(i128 inreg %lhs, i128 %rhs, i128 i ; GFX11-LABEL: v_fshr_i128_svs: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX11-NEXT: s_lshr_b32 s6, s1, 31 -; GFX11-NEXT: s_mov_b32 s7, 0 -; GFX11-NEXT: s_and_not1_b32 s5, 0x7f, s4 +; GFX11-NEXT: s_lshr_b32 s5, s1, 31 ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX11-NEXT: s_or_b64 s[2:3], s[2:3], s[6:7] +; GFX11-NEXT: s_or_b32 s2, s2, s5 +; GFX11-NEXT: s_and_not1_b32 s5, 0x7f, s4 ; GFX11-NEXT: s_not_b32 s10, s4 ; GFX11-NEXT: s_sub_i32 s12, s5, 64 ; GFX11-NEXT: s_sub_i32 s6, 64, s5 @@ -6714,81 +6691,80 @@ define i128 @v_fshr_i128_65(i128 %lhs, i128 %rhs) { define amdgpu_ps <2 x i128> @s_fshr_v2i128(<2 x i128> inreg %lhs, <2 x i128> inreg %rhs, <2 x i128> inreg %amt) { ; GFX6-LABEL: s_fshr_v2i128: ; GFX6: ; %bb.0: -; GFX6-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX6-NEXT: s_lshr_b32 s22, s1, 31 -; GFX6-NEXT: s_mov_b32 s23, 0 ; GFX6-NEXT: s_lshl_b64 s[18:19], s[0:1], 1 -; GFX6-NEXT: s_or_b64 s[0:1], s[2:3], s[22:23] -; GFX6-NEXT: s_andn2_b32 s2, 0x7f, s16 +; GFX6-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 +; GFX6-NEXT: s_lshr_b32 s0, s1, 31 +; GFX6-NEXT: s_or_b32 s2, s2, s0 +; GFX6-NEXT: s_andn2_b32 s0, 0x7f, s16 ; GFX6-NEXT: s_not_b32 s17, s16 -; GFX6-NEXT: s_sub_i32 s21, s2, 64 -; GFX6-NEXT: s_sub_i32 s22, 64, s2 -; GFX6-NEXT: s_cmp_lt_u32 s2, 64 -; GFX6-NEXT: s_cselect_b32 s28, 1, 0 -; GFX6-NEXT: s_cmp_eq_u32 s2, 0 -; GFX6-NEXT: s_cselect_b32 s29, 1, 0 -; GFX6-NEXT: s_lshr_b64 s[24:25], s[18:19], s22 -; GFX6-NEXT: s_lshl_b64 s[26:27], s[0:1], s17 -; GFX6-NEXT: s_lshl_b64 s[2:3], s[18:19], s17 -; GFX6-NEXT: s_or_b64 s[24:25], s[24:25], s[26:27] -; GFX6-NEXT: s_lshl_b64 s[18:19], s[18:19], s21 -; GFX6-NEXT: s_cmp_lg_u32 s28, 0 -; GFX6-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 -; GFX6-NEXT: s_cselect_b64 s[18:19], s[24:25], s[18:19] -; GFX6-NEXT: s_cmp_lg_u32 s29, 0 -; GFX6-NEXT: s_cselect_b64 s[18:19], s[0:1], s[18:19] -; GFX6-NEXT: s_and_b32 s0, s16, 0x7f ; GFX6-NEXT: s_sub_i32 s21, s0, 64 ; GFX6-NEXT: s_sub_i32 s22, 64, s0 ; GFX6-NEXT: s_cmp_lt_u32 s0, 64 ; GFX6-NEXT: s_cselect_b32 s26, 1, 0 ; GFX6-NEXT: s_cmp_eq_u32 s0, 0 ; GFX6-NEXT: s_cselect_b32 s27, 1, 0 -; GFX6-NEXT: s_lshr_b64 s[0:1], s[10:11], s16 +; GFX6-NEXT: s_lshr_b64 s[22:23], s[18:19], s22 +; GFX6-NEXT: s_lshl_b64 s[24:25], s[2:3], s17 +; GFX6-NEXT: s_lshl_b64 s[0:1], s[18:19], s17 +; GFX6-NEXT: s_or_b64 s[22:23], s[22:23], s[24:25] +; GFX6-NEXT: s_lshl_b64 s[18:19], s[18:19], s21 +; GFX6-NEXT: s_cmp_lg_u32 s26, 0 +; GFX6-NEXT: s_cselect_b64 s[0:1], s[0:1], 0 +; GFX6-NEXT: s_cselect_b64 s[18:19], s[22:23], s[18:19] +; GFX6-NEXT: s_cmp_lg_u32 s27, 0 +; GFX6-NEXT: s_cselect_b64 s[2:3], s[2:3], s[18:19] +; GFX6-NEXT: s_and_b32 s17, s16, 0x7f +; GFX6-NEXT: s_sub_i32 s21, s17, 64 +; GFX6-NEXT: s_sub_i32 s22, 64, s17 +; GFX6-NEXT: s_cmp_lt_u32 s17, 64 +; GFX6-NEXT: s_cselect_b32 s24, 1, 0 +; GFX6-NEXT: s_cmp_eq_u32 s17, 0 +; GFX6-NEXT: s_cselect_b32 s25, 1, 0 +; GFX6-NEXT: s_lshr_b64 s[18:19], s[10:11], s16 ; GFX6-NEXT: s_lshr_b64 s[16:17], s[8:9], s16 -; GFX6-NEXT: s_lshl_b64 s[24:25], s[10:11], s22 -; GFX6-NEXT: s_or_b64 s[16:17], s[16:17], s[24:25] +; GFX6-NEXT: s_lshl_b64 s[22:23], s[10:11], s22 +; GFX6-NEXT: s_or_b64 s[16:17], s[16:17], s[22:23] ; GFX6-NEXT: s_lshr_b64 s[10:11], s[10:11], s21 -; GFX6-NEXT: s_cmp_lg_u32 s26, 0 +; GFX6-NEXT: s_cmp_lg_u32 s24, 0 ; GFX6-NEXT: s_cselect_b64 s[10:11], s[16:17], s[10:11] -; GFX6-NEXT: s_cmp_lg_u32 s27, 0 +; GFX6-NEXT: s_cmp_lg_u32 s25, 0 ; GFX6-NEXT: s_cselect_b64 s[8:9], s[8:9], s[10:11] -; GFX6-NEXT: s_cmp_lg_u32 s26, 0 -; GFX6-NEXT: s_cselect_b64 s[10:11], s[0:1], 0 -; GFX6-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 -; GFX6-NEXT: s_lshr_b32 s22, s5, 31 -; GFX6-NEXT: s_or_b64 s[0:1], s[2:3], s[8:9] +; GFX6-NEXT: s_cmp_lg_u32 s24, 0 +; GFX6-NEXT: s_cselect_b64 s[10:11], s[18:19], 0 +; GFX6-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] ; GFX6-NEXT: s_lshl_b64 s[8:9], s[4:5], 1 -; GFX6-NEXT: s_or_b64 s[4:5], s[6:7], s[22:23] -; GFX6-NEXT: s_andn2_b32 s6, 0x7f, s20 -; GFX6-NEXT: s_or_b64 s[2:3], s[18:19], s[10:11] +; GFX6-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 +; GFX6-NEXT: s_lshr_b32 s4, s5, 31 +; GFX6-NEXT: s_or_b32 s6, s6, s4 +; GFX6-NEXT: s_andn2_b32 s4, 0x7f, s20 +; GFX6-NEXT: s_or_b64 s[2:3], s[2:3], s[10:11] ; GFX6-NEXT: s_not_b32 s16, s20 -; GFX6-NEXT: s_sub_i32 s18, s6, 64 -; GFX6-NEXT: s_sub_i32 s10, 64, s6 -; GFX6-NEXT: s_cmp_lt_u32 s6, 64 +; GFX6-NEXT: s_sub_i32 s18, s4, 64 +; GFX6-NEXT: s_sub_i32 s10, 64, s4 +; GFX6-NEXT: s_cmp_lt_u32 s4, 64 ; GFX6-NEXT: s_cselect_b32 s19, 1, 0 -; GFX6-NEXT: s_cmp_eq_u32 s6, 0 +; GFX6-NEXT: s_cmp_eq_u32 s4, 0 ; GFX6-NEXT: s_cselect_b32 s21, 1, 0 -; GFX6-NEXT: s_lshl_b64 s[6:7], s[8:9], s16 +; GFX6-NEXT: s_lshl_b64 s[4:5], s[8:9], s16 ; GFX6-NEXT: s_lshr_b64 s[10:11], s[8:9], s10 -; GFX6-NEXT: s_lshl_b64 s[16:17], s[4:5], s16 +; GFX6-NEXT: s_lshl_b64 s[16:17], s[6:7], s16 ; GFX6-NEXT: s_or_b64 s[10:11], s[10:11], s[16:17] ; GFX6-NEXT: s_lshl_b64 s[8:9], s[8:9], s18 ; GFX6-NEXT: s_cmp_lg_u32 s19, 0 -; GFX6-NEXT: s_cselect_b64 s[6:7], s[6:7], 0 +; GFX6-NEXT: s_cselect_b64 s[4:5], s[4:5], 0 ; GFX6-NEXT: s_cselect_b64 s[8:9], s[10:11], s[8:9] ; GFX6-NEXT: s_cmp_lg_u32 s21, 0 -; GFX6-NEXT: s_cselect_b64 s[8:9], s[4:5], s[8:9] -; GFX6-NEXT: s_and_b32 s4, s20, 0x7f -; GFX6-NEXT: s_sub_i32 s18, s4, 64 -; GFX6-NEXT: s_sub_i32 s16, 64, s4 -; GFX6-NEXT: s_cmp_lt_u32 s4, 64 +; GFX6-NEXT: s_cselect_b64 s[6:7], s[6:7], s[8:9] +; GFX6-NEXT: s_and_b32 s8, s20, 0x7f +; GFX6-NEXT: s_sub_i32 s18, s8, 64 +; GFX6-NEXT: s_sub_i32 s16, 64, s8 +; GFX6-NEXT: s_cmp_lt_u32 s8, 64 ; GFX6-NEXT: s_cselect_b32 s19, 1, 0 -; GFX6-NEXT: s_cmp_eq_u32 s4, 0 +; GFX6-NEXT: s_cmp_eq_u32 s8, 0 ; GFX6-NEXT: s_cselect_b32 s21, 1, 0 ; GFX6-NEXT: s_lshr_b64 s[10:11], s[12:13], s20 ; GFX6-NEXT: s_lshl_b64 s[16:17], s[14:15], s16 -; GFX6-NEXT: s_lshr_b64 s[4:5], s[14:15], s20 +; GFX6-NEXT: s_lshr_b64 s[8:9], s[14:15], s20 ; GFX6-NEXT: s_or_b64 s[10:11], s[10:11], s[16:17] ; GFX6-NEXT: s_lshr_b64 s[14:15], s[14:15], s18 ; GFX6-NEXT: s_cmp_lg_u32 s19, 0 @@ -6796,88 +6772,87 @@ define amdgpu_ps <2 x i128> @s_fshr_v2i128(<2 x i128> inreg %lhs, <2 x i128> inr ; GFX6-NEXT: s_cmp_lg_u32 s21, 0 ; GFX6-NEXT: s_cselect_b64 s[10:11], s[12:13], s[10:11] ; GFX6-NEXT: s_cmp_lg_u32 s19, 0 -; GFX6-NEXT: s_cselect_b64 s[12:13], s[4:5], 0 -; GFX6-NEXT: s_or_b64 s[4:5], s[6:7], s[10:11] -; GFX6-NEXT: s_or_b64 s[6:7], s[8:9], s[12:13] +; GFX6-NEXT: s_cselect_b64 s[8:9], s[8:9], 0 +; GFX6-NEXT: s_or_b64 s[4:5], s[4:5], s[10:11] +; GFX6-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9] ; GFX6-NEXT: ; return to shader part epilog ; ; GFX8-LABEL: s_fshr_v2i128: ; GFX8: ; %bb.0: -; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX8-NEXT: s_lshr_b32 s22, s1, 31 -; GFX8-NEXT: s_mov_b32 s23, 0 ; GFX8-NEXT: s_lshl_b64 s[18:19], s[0:1], 1 -; GFX8-NEXT: s_or_b64 s[0:1], s[2:3], s[22:23] -; GFX8-NEXT: s_andn2_b32 s2, 0x7f, s16 +; GFX8-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 +; GFX8-NEXT: s_lshr_b32 s0, s1, 31 +; GFX8-NEXT: s_or_b32 s2, s2, s0 +; GFX8-NEXT: s_andn2_b32 s0, 0x7f, s16 ; GFX8-NEXT: s_not_b32 s17, s16 -; GFX8-NEXT: s_sub_i32 s21, s2, 64 -; GFX8-NEXT: s_sub_i32 s22, 64, s2 -; GFX8-NEXT: s_cmp_lt_u32 s2, 64 -; GFX8-NEXT: s_cselect_b32 s28, 1, 0 -; GFX8-NEXT: s_cmp_eq_u32 s2, 0 -; GFX8-NEXT: s_cselect_b32 s29, 1, 0 -; GFX8-NEXT: s_lshr_b64 s[24:25], s[18:19], s22 -; GFX8-NEXT: s_lshl_b64 s[26:27], s[0:1], s17 -; GFX8-NEXT: s_lshl_b64 s[2:3], s[18:19], s17 -; GFX8-NEXT: s_or_b64 s[24:25], s[24:25], s[26:27] -; GFX8-NEXT: s_lshl_b64 s[18:19], s[18:19], s21 -; GFX8-NEXT: s_cmp_lg_u32 s28, 0 -; GFX8-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 -; GFX8-NEXT: s_cselect_b64 s[18:19], s[24:25], s[18:19] -; GFX8-NEXT: s_cmp_lg_u32 s29, 0 -; GFX8-NEXT: s_cselect_b64 s[18:19], s[0:1], s[18:19] -; GFX8-NEXT: s_and_b32 s0, s16, 0x7f ; GFX8-NEXT: s_sub_i32 s21, s0, 64 ; GFX8-NEXT: s_sub_i32 s22, 64, s0 ; GFX8-NEXT: s_cmp_lt_u32 s0, 64 ; GFX8-NEXT: s_cselect_b32 s26, 1, 0 ; GFX8-NEXT: s_cmp_eq_u32 s0, 0 ; GFX8-NEXT: s_cselect_b32 s27, 1, 0 -; GFX8-NEXT: s_lshr_b64 s[0:1], s[10:11], s16 +; GFX8-NEXT: s_lshr_b64 s[22:23], s[18:19], s22 +; GFX8-NEXT: s_lshl_b64 s[24:25], s[2:3], s17 +; GFX8-NEXT: s_lshl_b64 s[0:1], s[18:19], s17 +; GFX8-NEXT: s_or_b64 s[22:23], s[22:23], s[24:25] +; GFX8-NEXT: s_lshl_b64 s[18:19], s[18:19], s21 +; GFX8-NEXT: s_cmp_lg_u32 s26, 0 +; GFX8-NEXT: s_cselect_b64 s[0:1], s[0:1], 0 +; GFX8-NEXT: s_cselect_b64 s[18:19], s[22:23], s[18:19] +; GFX8-NEXT: s_cmp_lg_u32 s27, 0 +; GFX8-NEXT: s_cselect_b64 s[2:3], s[2:3], s[18:19] +; GFX8-NEXT: s_and_b32 s17, s16, 0x7f +; GFX8-NEXT: s_sub_i32 s21, s17, 64 +; GFX8-NEXT: s_sub_i32 s22, 64, s17 +; GFX8-NEXT: s_cmp_lt_u32 s17, 64 +; GFX8-NEXT: s_cselect_b32 s24, 1, 0 +; GFX8-NEXT: s_cmp_eq_u32 s17, 0 +; GFX8-NEXT: s_cselect_b32 s25, 1, 0 +; GFX8-NEXT: s_lshr_b64 s[18:19], s[10:11], s16 ; GFX8-NEXT: s_lshr_b64 s[16:17], s[8:9], s16 -; GFX8-NEXT: s_lshl_b64 s[24:25], s[10:11], s22 -; GFX8-NEXT: s_or_b64 s[16:17], s[16:17], s[24:25] +; GFX8-NEXT: s_lshl_b64 s[22:23], s[10:11], s22 +; GFX8-NEXT: s_or_b64 s[16:17], s[16:17], s[22:23] ; GFX8-NEXT: s_lshr_b64 s[10:11], s[10:11], s21 -; GFX8-NEXT: s_cmp_lg_u32 s26, 0 +; GFX8-NEXT: s_cmp_lg_u32 s24, 0 ; GFX8-NEXT: s_cselect_b64 s[10:11], s[16:17], s[10:11] -; GFX8-NEXT: s_cmp_lg_u32 s27, 0 +; GFX8-NEXT: s_cmp_lg_u32 s25, 0 ; GFX8-NEXT: s_cselect_b64 s[8:9], s[8:9], s[10:11] -; GFX8-NEXT: s_cmp_lg_u32 s26, 0 -; GFX8-NEXT: s_cselect_b64 s[10:11], s[0:1], 0 -; GFX8-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 -; GFX8-NEXT: s_lshr_b32 s22, s5, 31 -; GFX8-NEXT: s_or_b64 s[0:1], s[2:3], s[8:9] +; GFX8-NEXT: s_cmp_lg_u32 s24, 0 +; GFX8-NEXT: s_cselect_b64 s[10:11], s[18:19], 0 +; GFX8-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] ; GFX8-NEXT: s_lshl_b64 s[8:9], s[4:5], 1 -; GFX8-NEXT: s_or_b64 s[4:5], s[6:7], s[22:23] -; GFX8-NEXT: s_andn2_b32 s6, 0x7f, s20 -; GFX8-NEXT: s_or_b64 s[2:3], s[18:19], s[10:11] +; GFX8-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 +; GFX8-NEXT: s_lshr_b32 s4, s5, 31 +; GFX8-NEXT: s_or_b32 s6, s6, s4 +; GFX8-NEXT: s_andn2_b32 s4, 0x7f, s20 +; GFX8-NEXT: s_or_b64 s[2:3], s[2:3], s[10:11] ; GFX8-NEXT: s_not_b32 s16, s20 -; GFX8-NEXT: s_sub_i32 s18, s6, 64 -; GFX8-NEXT: s_sub_i32 s10, 64, s6 -; GFX8-NEXT: s_cmp_lt_u32 s6, 64 +; GFX8-NEXT: s_sub_i32 s18, s4, 64 +; GFX8-NEXT: s_sub_i32 s10, 64, s4 +; GFX8-NEXT: s_cmp_lt_u32 s4, 64 ; GFX8-NEXT: s_cselect_b32 s19, 1, 0 -; GFX8-NEXT: s_cmp_eq_u32 s6, 0 +; GFX8-NEXT: s_cmp_eq_u32 s4, 0 ; GFX8-NEXT: s_cselect_b32 s21, 1, 0 -; GFX8-NEXT: s_lshl_b64 s[6:7], s[8:9], s16 +; GFX8-NEXT: s_lshl_b64 s[4:5], s[8:9], s16 ; GFX8-NEXT: s_lshr_b64 s[10:11], s[8:9], s10 -; GFX8-NEXT: s_lshl_b64 s[16:17], s[4:5], s16 +; GFX8-NEXT: s_lshl_b64 s[16:17], s[6:7], s16 ; GFX8-NEXT: s_or_b64 s[10:11], s[10:11], s[16:17] ; GFX8-NEXT: s_lshl_b64 s[8:9], s[8:9], s18 ; GFX8-NEXT: s_cmp_lg_u32 s19, 0 -; GFX8-NEXT: s_cselect_b64 s[6:7], s[6:7], 0 +; GFX8-NEXT: s_cselect_b64 s[4:5], s[4:5], 0 ; GFX8-NEXT: s_cselect_b64 s[8:9], s[10:11], s[8:9] ; GFX8-NEXT: s_cmp_lg_u32 s21, 0 -; GFX8-NEXT: s_cselect_b64 s[8:9], s[4:5], s[8:9] -; GFX8-NEXT: s_and_b32 s4, s20, 0x7f -; GFX8-NEXT: s_sub_i32 s18, s4, 64 -; GFX8-NEXT: s_sub_i32 s16, 64, s4 -; GFX8-NEXT: s_cmp_lt_u32 s4, 64 +; GFX8-NEXT: s_cselect_b64 s[6:7], s[6:7], s[8:9] +; GFX8-NEXT: s_and_b32 s8, s20, 0x7f +; GFX8-NEXT: s_sub_i32 s18, s8, 64 +; GFX8-NEXT: s_sub_i32 s16, 64, s8 +; GFX8-NEXT: s_cmp_lt_u32 s8, 64 ; GFX8-NEXT: s_cselect_b32 s19, 1, 0 -; GFX8-NEXT: s_cmp_eq_u32 s4, 0 +; GFX8-NEXT: s_cmp_eq_u32 s8, 0 ; GFX8-NEXT: s_cselect_b32 s21, 1, 0 ; GFX8-NEXT: s_lshr_b64 s[10:11], s[12:13], s20 ; GFX8-NEXT: s_lshl_b64 s[16:17], s[14:15], s16 -; GFX8-NEXT: s_lshr_b64 s[4:5], s[14:15], s20 +; GFX8-NEXT: s_lshr_b64 s[8:9], s[14:15], s20 ; GFX8-NEXT: s_or_b64 s[10:11], s[10:11], s[16:17] ; GFX8-NEXT: s_lshr_b64 s[14:15], s[14:15], s18 ; GFX8-NEXT: s_cmp_lg_u32 s19, 0 @@ -6885,88 +6860,87 @@ define amdgpu_ps <2 x i128> @s_fshr_v2i128(<2 x i128> inreg %lhs, <2 x i128> inr ; GFX8-NEXT: s_cmp_lg_u32 s21, 0 ; GFX8-NEXT: s_cselect_b64 s[10:11], s[12:13], s[10:11] ; GFX8-NEXT: s_cmp_lg_u32 s19, 0 -; GFX8-NEXT: s_cselect_b64 s[12:13], s[4:5], 0 -; GFX8-NEXT: s_or_b64 s[4:5], s[6:7], s[10:11] -; GFX8-NEXT: s_or_b64 s[6:7], s[8:9], s[12:13] +; GFX8-NEXT: s_cselect_b64 s[8:9], s[8:9], 0 +; GFX8-NEXT: s_or_b64 s[4:5], s[4:5], s[10:11] +; GFX8-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9] ; GFX8-NEXT: ; return to shader part epilog ; ; GFX9-LABEL: s_fshr_v2i128: ; GFX9: ; %bb.0: -; GFX9-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX9-NEXT: s_lshr_b32 s22, s1, 31 -; GFX9-NEXT: s_mov_b32 s23, 0 ; GFX9-NEXT: s_lshl_b64 s[18:19], s[0:1], 1 -; GFX9-NEXT: s_or_b64 s[0:1], s[2:3], s[22:23] -; GFX9-NEXT: s_andn2_b32 s2, 0x7f, s16 +; GFX9-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 +; GFX9-NEXT: s_lshr_b32 s0, s1, 31 +; GFX9-NEXT: s_or_b32 s2, s2, s0 +; GFX9-NEXT: s_andn2_b32 s0, 0x7f, s16 ; GFX9-NEXT: s_not_b32 s17, s16 -; GFX9-NEXT: s_sub_i32 s21, s2, 64 -; GFX9-NEXT: s_sub_i32 s22, 64, s2 -; GFX9-NEXT: s_cmp_lt_u32 s2, 64 -; GFX9-NEXT: s_cselect_b32 s28, 1, 0 -; GFX9-NEXT: s_cmp_eq_u32 s2, 0 -; GFX9-NEXT: s_cselect_b32 s29, 1, 0 -; GFX9-NEXT: s_lshr_b64 s[24:25], s[18:19], s22 -; GFX9-NEXT: s_lshl_b64 s[26:27], s[0:1], s17 -; GFX9-NEXT: s_lshl_b64 s[2:3], s[18:19], s17 -; GFX9-NEXT: s_or_b64 s[24:25], s[24:25], s[26:27] -; GFX9-NEXT: s_lshl_b64 s[18:19], s[18:19], s21 -; GFX9-NEXT: s_cmp_lg_u32 s28, 0 -; GFX9-NEXT: s_cselect_b64 s[2:3], s[2:3], 0 -; GFX9-NEXT: s_cselect_b64 s[18:19], s[24:25], s[18:19] -; GFX9-NEXT: s_cmp_lg_u32 s29, 0 -; GFX9-NEXT: s_cselect_b64 s[18:19], s[0:1], s[18:19] -; GFX9-NEXT: s_and_b32 s0, s16, 0x7f ; GFX9-NEXT: s_sub_i32 s21, s0, 64 ; GFX9-NEXT: s_sub_i32 s22, 64, s0 ; GFX9-NEXT: s_cmp_lt_u32 s0, 64 ; GFX9-NEXT: s_cselect_b32 s26, 1, 0 ; GFX9-NEXT: s_cmp_eq_u32 s0, 0 ; GFX9-NEXT: s_cselect_b32 s27, 1, 0 -; GFX9-NEXT: s_lshr_b64 s[0:1], s[10:11], s16 +; GFX9-NEXT: s_lshr_b64 s[22:23], s[18:19], s22 +; GFX9-NEXT: s_lshl_b64 s[24:25], s[2:3], s17 +; GFX9-NEXT: s_lshl_b64 s[0:1], s[18:19], s17 +; GFX9-NEXT: s_or_b64 s[22:23], s[22:23], s[24:25] +; GFX9-NEXT: s_lshl_b64 s[18:19], s[18:19], s21 +; GFX9-NEXT: s_cmp_lg_u32 s26, 0 +; GFX9-NEXT: s_cselect_b64 s[0:1], s[0:1], 0 +; GFX9-NEXT: s_cselect_b64 s[18:19], s[22:23], s[18:19] +; GFX9-NEXT: s_cmp_lg_u32 s27, 0 +; GFX9-NEXT: s_cselect_b64 s[2:3], s[2:3], s[18:19] +; GFX9-NEXT: s_and_b32 s17, s16, 0x7f +; GFX9-NEXT: s_sub_i32 s21, s17, 64 +; GFX9-NEXT: s_sub_i32 s22, 64, s17 +; GFX9-NEXT: s_cmp_lt_u32 s17, 64 +; GFX9-NEXT: s_cselect_b32 s24, 1, 0 +; GFX9-NEXT: s_cmp_eq_u32 s17, 0 +; GFX9-NEXT: s_cselect_b32 s25, 1, 0 +; GFX9-NEXT: s_lshr_b64 s[18:19], s[10:11], s16 ; GFX9-NEXT: s_lshr_b64 s[16:17], s[8:9], s16 -; GFX9-NEXT: s_lshl_b64 s[24:25], s[10:11], s22 -; GFX9-NEXT: s_or_b64 s[16:17], s[16:17], s[24:25] +; GFX9-NEXT: s_lshl_b64 s[22:23], s[10:11], s22 +; GFX9-NEXT: s_or_b64 s[16:17], s[16:17], s[22:23] ; GFX9-NEXT: s_lshr_b64 s[10:11], s[10:11], s21 -; GFX9-NEXT: s_cmp_lg_u32 s26, 0 +; GFX9-NEXT: s_cmp_lg_u32 s24, 0 ; GFX9-NEXT: s_cselect_b64 s[10:11], s[16:17], s[10:11] -; GFX9-NEXT: s_cmp_lg_u32 s27, 0 +; GFX9-NEXT: s_cmp_lg_u32 s25, 0 ; GFX9-NEXT: s_cselect_b64 s[8:9], s[8:9], s[10:11] -; GFX9-NEXT: s_cmp_lg_u32 s26, 0 -; GFX9-NEXT: s_cselect_b64 s[10:11], s[0:1], 0 -; GFX9-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 -; GFX9-NEXT: s_lshr_b32 s22, s5, 31 -; GFX9-NEXT: s_or_b64 s[0:1], s[2:3], s[8:9] +; GFX9-NEXT: s_cmp_lg_u32 s24, 0 +; GFX9-NEXT: s_cselect_b64 s[10:11], s[18:19], 0 +; GFX9-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] ; GFX9-NEXT: s_lshl_b64 s[8:9], s[4:5], 1 -; GFX9-NEXT: s_or_b64 s[4:5], s[6:7], s[22:23] -; GFX9-NEXT: s_andn2_b32 s6, 0x7f, s20 -; GFX9-NEXT: s_or_b64 s[2:3], s[18:19], s[10:11] +; GFX9-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 +; GFX9-NEXT: s_lshr_b32 s4, s5, 31 +; GFX9-NEXT: s_or_b32 s6, s6, s4 +; GFX9-NEXT: s_andn2_b32 s4, 0x7f, s20 +; GFX9-NEXT: s_or_b64 s[2:3], s[2:3], s[10:11] ; GFX9-NEXT: s_not_b32 s16, s20 -; GFX9-NEXT: s_sub_i32 s18, s6, 64 -; GFX9-NEXT: s_sub_i32 s10, 64, s6 -; GFX9-NEXT: s_cmp_lt_u32 s6, 64 +; GFX9-NEXT: s_sub_i32 s18, s4, 64 +; GFX9-NEXT: s_sub_i32 s10, 64, s4 +; GFX9-NEXT: s_cmp_lt_u32 s4, 64 ; GFX9-NEXT: s_cselect_b32 s19, 1, 0 -; GFX9-NEXT: s_cmp_eq_u32 s6, 0 +; GFX9-NEXT: s_cmp_eq_u32 s4, 0 ; GFX9-NEXT: s_cselect_b32 s21, 1, 0 -; GFX9-NEXT: s_lshl_b64 s[6:7], s[8:9], s16 +; GFX9-NEXT: s_lshl_b64 s[4:5], s[8:9], s16 ; GFX9-NEXT: s_lshr_b64 s[10:11], s[8:9], s10 -; GFX9-NEXT: s_lshl_b64 s[16:17], s[4:5], s16 +; GFX9-NEXT: s_lshl_b64 s[16:17], s[6:7], s16 ; GFX9-NEXT: s_or_b64 s[10:11], s[10:11], s[16:17] ; GFX9-NEXT: s_lshl_b64 s[8:9], s[8:9], s18 ; GFX9-NEXT: s_cmp_lg_u32 s19, 0 -; GFX9-NEXT: s_cselect_b64 s[6:7], s[6:7], 0 +; GFX9-NEXT: s_cselect_b64 s[4:5], s[4:5], 0 ; GFX9-NEXT: s_cselect_b64 s[8:9], s[10:11], s[8:9] ; GFX9-NEXT: s_cmp_lg_u32 s21, 0 -; GFX9-NEXT: s_cselect_b64 s[8:9], s[4:5], s[8:9] -; GFX9-NEXT: s_and_b32 s4, s20, 0x7f -; GFX9-NEXT: s_sub_i32 s18, s4, 64 -; GFX9-NEXT: s_sub_i32 s16, 64, s4 -; GFX9-NEXT: s_cmp_lt_u32 s4, 64 +; GFX9-NEXT: s_cselect_b64 s[6:7], s[6:7], s[8:9] +; GFX9-NEXT: s_and_b32 s8, s20, 0x7f +; GFX9-NEXT: s_sub_i32 s18, s8, 64 +; GFX9-NEXT: s_sub_i32 s16, 64, s8 +; GFX9-NEXT: s_cmp_lt_u32 s8, 64 ; GFX9-NEXT: s_cselect_b32 s19, 1, 0 -; GFX9-NEXT: s_cmp_eq_u32 s4, 0 +; GFX9-NEXT: s_cmp_eq_u32 s8, 0 ; GFX9-NEXT: s_cselect_b32 s21, 1, 0 ; GFX9-NEXT: s_lshr_b64 s[10:11], s[12:13], s20 ; GFX9-NEXT: s_lshl_b64 s[16:17], s[14:15], s16 -; GFX9-NEXT: s_lshr_b64 s[4:5], s[14:15], s20 +; GFX9-NEXT: s_lshr_b64 s[8:9], s[14:15], s20 ; GFX9-NEXT: s_or_b64 s[10:11], s[10:11], s[16:17] ; GFX9-NEXT: s_lshr_b64 s[14:15], s[14:15], s18 ; GFX9-NEXT: s_cmp_lg_u32 s19, 0 @@ -6974,61 +6948,60 @@ define amdgpu_ps <2 x i128> @s_fshr_v2i128(<2 x i128> inreg %lhs, <2 x i128> inr ; GFX9-NEXT: s_cmp_lg_u32 s21, 0 ; GFX9-NEXT: s_cselect_b64 s[10:11], s[12:13], s[10:11] ; GFX9-NEXT: s_cmp_lg_u32 s19, 0 -; GFX9-NEXT: s_cselect_b64 s[12:13], s[4:5], 0 -; GFX9-NEXT: s_or_b64 s[4:5], s[6:7], s[10:11] -; GFX9-NEXT: s_or_b64 s[6:7], s[8:9], s[12:13] +; GFX9-NEXT: s_cselect_b64 s[8:9], s[8:9], 0 +; GFX9-NEXT: s_or_b64 s[4:5], s[4:5], s[10:11] +; GFX9-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9] ; GFX9-NEXT: ; return to shader part epilog ; ; GFX10-LABEL: s_fshr_v2i128: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX10-NEXT: s_lshr_b32 s18, s1, 31 -; GFX10-NEXT: s_mov_b32 s19, 0 -; GFX10-NEXT: s_andn2_b32 s17, 0x7f, s16 +; GFX10-NEXT: s_lshr_b32 s17, s1, 31 ; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX10-NEXT: s_or_b64 s[2:3], s[2:3], s[18:19] -; GFX10-NEXT: s_not_b32 s18, s16 -; GFX10-NEXT: s_sub_i32 s21, s17, 64 -; GFX10-NEXT: s_sub_i32 s22, 64, s17 +; GFX10-NEXT: s_or_b32 s2, s2, s17 +; GFX10-NEXT: s_andn2_b32 s17, 0x7f, s16 +; GFX10-NEXT: s_not_b32 s21, s16 +; GFX10-NEXT: s_sub_i32 s26, s17, 64 +; GFX10-NEXT: s_sub_i32 s18, 64, s17 ; GFX10-NEXT: s_cmp_lt_u32 s17, 64 -; GFX10-NEXT: s_cselect_b32 s28, 1, 0 +; GFX10-NEXT: s_cselect_b32 s27, 1, 0 ; GFX10-NEXT: s_cmp_eq_u32 s17, 0 ; GFX10-NEXT: s_cselect_b32 s17, 1, 0 -; GFX10-NEXT: s_lshr_b64 s[22:23], s[0:1], s22 -; GFX10-NEXT: s_lshl_b64 s[24:25], s[2:3], s18 -; GFX10-NEXT: s_lshl_b64 s[26:27], s[0:1], s18 -; GFX10-NEXT: s_or_b64 s[22:23], s[22:23], s[24:25] -; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], s21 -; GFX10-NEXT: s_cmp_lg_u32 s28, 0 -; GFX10-NEXT: s_cselect_b64 s[24:25], s[26:27], 0 -; GFX10-NEXT: s_cselect_b64 s[0:1], s[22:23], s[0:1] +; GFX10-NEXT: s_lshr_b64 s[18:19], s[0:1], s18 +; GFX10-NEXT: s_lshl_b64 s[22:23], s[2:3], s21 +; GFX10-NEXT: s_lshl_b64 s[24:25], s[0:1], s21 +; GFX10-NEXT: s_or_b64 s[18:19], s[18:19], s[22:23] +; GFX10-NEXT: s_lshl_b64 s[0:1], s[0:1], s26 +; GFX10-NEXT: s_cmp_lg_u32 s27, 0 +; GFX10-NEXT: s_cselect_b64 s[22:23], s[24:25], 0 +; GFX10-NEXT: s_cselect_b64 s[0:1], s[18:19], s[0:1] ; GFX10-NEXT: s_cmp_lg_u32 s17, 0 ; GFX10-NEXT: s_cselect_b64 s[2:3], s[2:3], s[0:1] ; GFX10-NEXT: s_and_b32 s0, s16, 0x7f -; GFX10-NEXT: s_sub_i32 s18, s0, 64 +; GFX10-NEXT: s_sub_i32 s21, s0, 64 ; GFX10-NEXT: s_sub_i32 s17, 64, s0 ; GFX10-NEXT: s_cmp_lt_u32 s0, 64 -; GFX10-NEXT: s_cselect_b32 s21, 1, 0 +; GFX10-NEXT: s_cselect_b32 s24, 1, 0 ; GFX10-NEXT: s_cmp_eq_u32 s0, 0 -; GFX10-NEXT: s_cselect_b32 s26, 1, 0 +; GFX10-NEXT: s_cselect_b32 s25, 1, 0 ; GFX10-NEXT: s_lshr_b64 s[0:1], s[8:9], s16 -; GFX10-NEXT: s_lshl_b64 s[22:23], s[10:11], s17 +; GFX10-NEXT: s_lshl_b64 s[18:19], s[10:11], s17 ; GFX10-NEXT: s_lshr_b64 s[16:17], s[10:11], s16 -; GFX10-NEXT: s_or_b64 s[0:1], s[0:1], s[22:23] -; GFX10-NEXT: s_lshr_b64 s[10:11], s[10:11], s18 -; GFX10-NEXT: s_cmp_lg_u32 s21, 0 +; GFX10-NEXT: s_or_b64 s[0:1], s[0:1], s[18:19] +; GFX10-NEXT: s_lshr_b64 s[10:11], s[10:11], s21 +; GFX10-NEXT: s_cmp_lg_u32 s24, 0 ; GFX10-NEXT: s_cselect_b64 s[0:1], s[0:1], s[10:11] -; GFX10-NEXT: s_cmp_lg_u32 s26, 0 +; GFX10-NEXT: s_cmp_lg_u32 s25, 0 ; GFX10-NEXT: s_cselect_b64 s[0:1], s[8:9], s[0:1] -; GFX10-NEXT: s_cmp_lg_u32 s21, 0 +; GFX10-NEXT: s_cmp_lg_u32 s24, 0 ; GFX10-NEXT: s_cselect_b64 s[8:9], s[16:17], 0 ; GFX10-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 ; GFX10-NEXT: s_or_b64 s[2:3], s[2:3], s[8:9] -; GFX10-NEXT: s_lshr_b32 s18, s5, 31 +; GFX10-NEXT: s_lshr_b32 s8, s5, 31 +; GFX10-NEXT: s_or_b64 s[0:1], s[22:23], s[0:1] +; GFX10-NEXT: s_or_b32 s6, s6, s8 ; GFX10-NEXT: s_andn2_b32 s8, 0x7f, s20 -; GFX10-NEXT: s_or_b64 s[0:1], s[24:25], s[0:1] ; GFX10-NEXT: s_lshl_b64 s[4:5], s[4:5], 1 -; GFX10-NEXT: s_or_b64 s[6:7], s[6:7], s[18:19] ; GFX10-NEXT: s_not_b32 s16, s20 ; GFX10-NEXT: s_sub_i32 s18, s8, 64 ; GFX10-NEXT: s_sub_i32 s9, 64, s8 @@ -7071,54 +7044,53 @@ define amdgpu_ps <2 x i128> @s_fshr_v2i128(<2 x i128> inreg %lhs, <2 x i128> inr ; GFX11-LABEL: s_fshr_v2i128: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 -; GFX11-NEXT: s_lshr_b32 s18, s1, 31 -; GFX11-NEXT: s_mov_b32 s19, 0 -; GFX11-NEXT: s_and_not1_b32 s17, 0x7f, s16 +; GFX11-NEXT: s_lshr_b32 s17, s1, 31 ; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], 1 -; GFX11-NEXT: s_or_b64 s[2:3], s[2:3], s[18:19] -; GFX11-NEXT: s_not_b32 s18, s16 -; GFX11-NEXT: s_sub_i32 s21, s17, 64 -; GFX11-NEXT: s_sub_i32 s22, 64, s17 +; GFX11-NEXT: s_or_b32 s2, s2, s17 +; GFX11-NEXT: s_and_not1_b32 s17, 0x7f, s16 +; GFX11-NEXT: s_not_b32 s21, s16 +; GFX11-NEXT: s_sub_i32 s26, s17, 64 +; GFX11-NEXT: s_sub_i32 s18, 64, s17 ; GFX11-NEXT: s_cmp_lt_u32 s17, 64 -; GFX11-NEXT: s_cselect_b32 s28, 1, 0 +; GFX11-NEXT: s_cselect_b32 s27, 1, 0 ; GFX11-NEXT: s_cmp_eq_u32 s17, 0 ; GFX11-NEXT: s_cselect_b32 s17, 1, 0 -; GFX11-NEXT: s_lshr_b64 s[22:23], s[0:1], s22 -; GFX11-NEXT: s_lshl_b64 s[24:25], s[2:3], s18 -; GFX11-NEXT: s_lshl_b64 s[26:27], s[0:1], s18 -; GFX11-NEXT: s_or_b64 s[22:23], s[22:23], s[24:25] -; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], s21 -; GFX11-NEXT: s_cmp_lg_u32 s28, 0 -; GFX11-NEXT: s_cselect_b64 s[24:25], s[26:27], 0 -; GFX11-NEXT: s_cselect_b64 s[0:1], s[22:23], s[0:1] +; GFX11-NEXT: s_lshr_b64 s[18:19], s[0:1], s18 +; GFX11-NEXT: s_lshl_b64 s[22:23], s[2:3], s21 +; GFX11-NEXT: s_lshl_b64 s[24:25], s[0:1], s21 +; GFX11-NEXT: s_or_b64 s[18:19], s[18:19], s[22:23] +; GFX11-NEXT: s_lshl_b64 s[0:1], s[0:1], s26 +; GFX11-NEXT: s_cmp_lg_u32 s27, 0 +; GFX11-NEXT: s_cselect_b64 s[22:23], s[24:25], 0 +; GFX11-NEXT: s_cselect_b64 s[0:1], s[18:19], s[0:1] ; GFX11-NEXT: s_cmp_lg_u32 s17, 0 ; GFX11-NEXT: s_cselect_b64 s[2:3], s[2:3], s[0:1] ; GFX11-NEXT: s_and_b32 s0, s16, 0x7f ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_sub_i32 s18, s0, 64 +; GFX11-NEXT: s_sub_i32 s21, s0, 64 ; GFX11-NEXT: s_sub_i32 s17, 64, s0 ; GFX11-NEXT: s_cmp_lt_u32 s0, 64 -; GFX11-NEXT: s_cselect_b32 s21, 1, 0 +; GFX11-NEXT: s_cselect_b32 s24, 1, 0 ; GFX11-NEXT: s_cmp_eq_u32 s0, 0 -; GFX11-NEXT: s_cselect_b32 s26, 1, 0 +; GFX11-NEXT: s_cselect_b32 s25, 1, 0 ; GFX11-NEXT: s_lshr_b64 s[0:1], s[8:9], s16 -; GFX11-NEXT: s_lshl_b64 s[22:23], s[10:11], s17 +; GFX11-NEXT: s_lshl_b64 s[18:19], s[10:11], s17 ; GFX11-NEXT: s_lshr_b64 s[16:17], s[10:11], s16 -; GFX11-NEXT: s_or_b64 s[0:1], s[0:1], s[22:23] -; GFX11-NEXT: s_lshr_b64 s[10:11], s[10:11], s18 -; GFX11-NEXT: s_cmp_lg_u32 s21, 0 +; GFX11-NEXT: s_or_b64 s[0:1], s[0:1], s[18:19] +; GFX11-NEXT: s_lshr_b64 s[10:11], s[10:11], s21 +; GFX11-NEXT: s_cmp_lg_u32 s24, 0 ; GFX11-NEXT: s_cselect_b64 s[0:1], s[0:1], s[10:11] -; GFX11-NEXT: s_cmp_lg_u32 s26, 0 +; GFX11-NEXT: s_cmp_lg_u32 s25, 0 ; GFX11-NEXT: s_cselect_b64 s[0:1], s[8:9], s[0:1] -; GFX11-NEXT: s_cmp_lg_u32 s21, 0 +; GFX11-NEXT: s_cmp_lg_u32 s24, 0 ; GFX11-NEXT: s_cselect_b64 s[8:9], s[16:17], 0 ; GFX11-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 ; GFX11-NEXT: s_or_b64 s[2:3], s[2:3], s[8:9] -; GFX11-NEXT: s_lshr_b32 s18, s5, 31 +; GFX11-NEXT: s_lshr_b32 s8, s5, 31 +; GFX11-NEXT: s_or_b64 s[0:1], s[22:23], s[0:1] +; GFX11-NEXT: s_or_b32 s6, s6, s8 ; GFX11-NEXT: s_and_not1_b32 s8, 0x7f, s20 -; GFX11-NEXT: s_or_b64 s[0:1], s[24:25], s[0:1] ; GFX11-NEXT: s_lshl_b64 s[4:5], s[4:5], 1 -; GFX11-NEXT: s_or_b64 s[6:7], s[6:7], s[18:19] ; GFX11-NEXT: s_not_b32 s16, s20 ; GFX11-NEXT: s_sub_i32 s18, s8, 64 ; GFX11-NEXT: s_sub_i32 s9, 64, s8 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir new file mode 100644 index 0000000..4b214e6 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fmaximum.mir @@ -0,0 +1,275 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py +# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9 %s +# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 -mattr=-real-true16 -run-pass=legalizer %s -o - | FileCheck -check-prefixes=GFX12 %s + +--- +name: test_fmaximum_f16 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fmaximum_f16 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s16) = G_FMAXNUM_IEEE [[TRUNC]], [[TRUNC1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s16) = COPY [[SELECT]](s16) + ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16) + ; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fmaximum_f16 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s16) = G_FMAXIMUM [[TRUNC]], [[TRUNC1]] + ; GFX12-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMAXIMUM]](s16) + ; GFX12-NEXT: $vgpr0 = COPY [[ANYEXT]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s16) = G_TRUNC %0(s32) + %2:_(s32) = COPY $vgpr1 + %3:_(s16) = G_TRUNC %2(s32) + %4:_(s16) = G_FMAXIMUM %1, %3 + %5:_(s32) = G_ANYEXT %4(s16) + $vgpr0 = COPY %5(s32) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fmaximum_f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fmaximum_f32 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fmaximum_f32 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = G_FMAXIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(s32) = G_FMAXIMUM %0, %1 + $vgpr0 = COPY %2(s32) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fmaximum_f64 +body: | + bb.0: + liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + + ; GFX9-LABEL: name: test_fmaximum_f64 + ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s64) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s64), [[COPY1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY [[SELECT]](s64) + ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[COPY2]](s64) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0_vgpr1 + ; + ; GFX12-LABEL: name: test_fmaximum_f64 + ; GFX12: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s64) = G_FMAXIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[FMAXIMUM]](s64) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0_vgpr1 + %0:_(s64) = COPY $vgpr0_vgpr1 + %1:_(s64) = COPY $vgpr2_vgpr3 + %2:_(s64) = G_FMAXIMUM %0, %1 + $vgpr0_vgpr1 = COPY %2(s64) + SI_RETURN implicit $vgpr0_vgpr1 +... +--- +name: test_fmaximum_v2f16 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fmaximum_v2f16 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(<2 x s16>) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) + ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) + ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) + ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32) + ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC2]] + ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC1]](s16), [[TRUNC3]] + ; GFX9-NEXT: [[C1:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00 + ; GFX9-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[FMAXNUM_IEEE]](<2 x s16>) + ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32) + ; GFX9-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32) + ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32) + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[TRUNC4]], [[C1]] + ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s16) = G_SELECT [[FCMP1]](s1), [[TRUNC5]], [[C1]] + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[SELECT]](s16), [[SELECT1]](s16) + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY [[BUILD_VECTOR]](<2 x s16>) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](<2 x s16>) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fmaximum_v2f16 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(<2 x s16>) = G_FMAXIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](<2 x s16>) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(<2 x s16>) = COPY $vgpr0 + %1:_(<2 x s16>) = COPY $vgpr1 + %2:_(<2 x s16>) = G_FMAXIMUM %0, %1 + $vgpr0 = COPY %2(<2 x s16>) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fmaximum_v2f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 + + ; GFX9-LABEL: name: test_fmaximum_v2f32 + ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY2]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY2]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32) + ; GFX9-NEXT: [[FMAXNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY1]], [[COPY3]] + ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY1]](s32), [[COPY3]] + ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[FCMP1]](s1), [[FMAXNUM_IEEE1]], [[C]] + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY [[SELECT1]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY4]](s32) + ; GFX9-NEXT: $vgpr1 = COPY [[COPY5]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1 + ; + ; GFX12-LABEL: name: test_fmaximum_v2f32 + ; GFX12: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX12-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = G_FMAXIMUM [[COPY]], [[COPY2]] + ; GFX12-NEXT: [[FMAXIMUM1:%[0-9]+]]:_(s32) = G_FMAXIMUM [[COPY1]], [[COPY3]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32) + ; GFX12-NEXT: $vgpr1 = COPY [[FMAXIMUM1]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(<2 x s32>) = G_BUILD_VECTOR %0(s32), %1(s32) + %3:_(s32) = COPY $vgpr2 + %4:_(s32) = COPY $vgpr3 + %5:_(<2 x s32>) = G_BUILD_VECTOR %3(s32), %4(s32) + %6:_(<2 x s32>) = G_FMAXIMUM %2, %5 + %7:_(s32), %8:_(s32) = G_UNMERGE_VALUES %6(<2 x s32>) + $vgpr0 = COPY %7(s32) + $vgpr1 = COPY %8(s32) + SI_RETURN implicit $vgpr0, implicit $vgpr1 +... +--- +name: test_fmaximum_nsz_f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fmaximum_nsz_f32 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMAXNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fmaximum_nsz_f32 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = nsz G_FMAXIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(s32) = nsz G_FMAXIMUM %0, %1 + $vgpr0 = COPY %2(s32) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fmaximum_nnan_f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fmaximum_nnan_f32 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[FMAXNUM_IEEE:%[0-9]+]]:_(s32) = G_FMAXNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[FMAXNUM_IEEE]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fmaximum_nnan_f32 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[FMAXIMUM:%[0-9]+]]:_(s32) = nnan G_FMAXIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMAXIMUM]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(s32) = nnan G_FMAXIMUM %0, %1 + $vgpr0 = COPY %2(s32) + SI_RETURN implicit $vgpr0 +... + diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir new file mode 100644 index 0000000..8ba0794 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fminimum.mir @@ -0,0 +1,275 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py +# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9 %s +# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 -mattr=-real-true16 -run-pass=legalizer %s -o - | FileCheck -check-prefixes=GFX12 %s + +--- +name: test_fminimum_f16 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fminimum_f16 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[TRUNC]], [[TRUNC1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s16) = COPY [[SELECT]](s16) + ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16) + ; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fminimum_f16 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s16) = G_FMINIMUM [[TRUNC]], [[TRUNC1]] + ; GFX12-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMINIMUM]](s16) + ; GFX12-NEXT: $vgpr0 = COPY [[ANYEXT]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s16) = G_TRUNC %0(s32) + %2:_(s32) = COPY $vgpr1 + %3:_(s16) = G_TRUNC %2(s32) + %4:_(s16) = G_FMINIMUM %1, %3 + %5:_(s32) = G_ANYEXT %4(s16) + $vgpr0 = COPY %5(s32) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fminimum_f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fminimum_f32 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fminimum_f32 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = G_FMINIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(s32) = G_FMINIMUM %0, %1 + $vgpr0 = COPY %2(s32) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fminimum_f64 +body: | + bb.0: + liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + + ; GFX9-LABEL: name: test_fminimum_f64 + ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s64) = G_FMINNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s64), [[COPY1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY [[SELECT]](s64) + ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[COPY2]](s64) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0_vgpr1 + ; + ; GFX12-LABEL: name: test_fminimum_f64 + ; GFX12: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s64) = G_FMINIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[FMINIMUM]](s64) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0_vgpr1 + %0:_(s64) = COPY $vgpr0_vgpr1 + %1:_(s64) = COPY $vgpr2_vgpr3 + %2:_(s64) = G_FMINIMUM %0, %1 + $vgpr0_vgpr1 = COPY %2(s64) + SI_RETURN implicit $vgpr0_vgpr1 +... +--- +name: test_fminimum_v2f16 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fminimum_v2f16 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(<2 x s16>) = G_FMINNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) + ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) + ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) + ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32) + ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC]](s16), [[TRUNC2]] + ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[TRUNC1]](s16), [[TRUNC3]] + ; GFX9-NEXT: [[C1:%[0-9]+]]:_(s16) = G_FCONSTANT half 0xH7E00 + ; GFX9-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[FMINNUM_IEEE]](<2 x s16>) + ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32) + ; GFX9-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32) + ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32) + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s16) = G_SELECT [[FCMP]](s1), [[TRUNC4]], [[C1]] + ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s16) = G_SELECT [[FCMP1]](s1), [[TRUNC5]], [[C1]] + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[SELECT]](s16), [[SELECT1]](s16) + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY [[BUILD_VECTOR]](<2 x s16>) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](<2 x s16>) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fminimum_v2f16 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(<2 x s16>) = G_FMINIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](<2 x s16>) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(<2 x s16>) = COPY $vgpr0 + %1:_(<2 x s16>) = COPY $vgpr1 + %2:_(<2 x s16>) = G_FMINIMUM %0, %1 + $vgpr0 = COPY %2(<2 x s16>) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fminimum_v2f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 + + ; GFX9-LABEL: name: test_fminimum_v2f32 + ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY2]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY2]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32) + ; GFX9-NEXT: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY1]], [[COPY3]] + ; GFX9-NEXT: [[FCMP1:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY1]](s32), [[COPY3]] + ; GFX9-NEXT: [[SELECT1:%[0-9]+]]:_(s32) = G_SELECT [[FCMP1]](s1), [[FMINNUM_IEEE1]], [[C]] + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY [[SELECT1]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY4]](s32) + ; GFX9-NEXT: $vgpr1 = COPY [[COPY5]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1 + ; + ; GFX12-LABEL: name: test_fminimum_v2f32 + ; GFX12: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX12-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = G_FMINIMUM [[COPY]], [[COPY2]] + ; GFX12-NEXT: [[FMINIMUM1:%[0-9]+]]:_(s32) = G_FMINIMUM [[COPY1]], [[COPY3]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32) + ; GFX12-NEXT: $vgpr1 = COPY [[FMINIMUM1]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(<2 x s32>) = G_BUILD_VECTOR %0(s32), %1(s32) + %3:_(s32) = COPY $vgpr2 + %4:_(s32) = COPY $vgpr3 + %5:_(<2 x s32>) = G_BUILD_VECTOR %3(s32), %4(s32) + %6:_(<2 x s32>) = G_FMINIMUM %2, %5 + %7:_(s32), %8:_(s32) = G_UNMERGE_VALUES %6(<2 x s32>) + $vgpr0 = COPY %7(s32) + $vgpr1 = COPY %8(s32) + SI_RETURN implicit $vgpr0, implicit $vgpr1 +... +--- +name: test_fminimum_nsz_f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fminimum_nsz_f32 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[FCMP:%[0-9]+]]:_(s1) = G_FCMP floatpred(ord), [[COPY]](s32), [[COPY1]] + ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 0x7FF8000000000000 + ; GFX9-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[FCMP]](s1), [[FMINNUM_IEEE]], [[C]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SELECT]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fminimum_nsz_f32 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = nsz G_FMINIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(s32) = nsz G_FMINIMUM %0, %1 + $vgpr0 = COPY %2(s32) + SI_RETURN implicit $vgpr0 +... +--- +name: test_fminimum_nnan_f32 +body: | + bb.0: + liveins: $vgpr0, $vgpr1 + + ; GFX9-LABEL: name: test_fminimum_nnan_f32 + ; GFX9: liveins: $vgpr0, $vgpr1 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX9-NEXT: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[COPY]], [[COPY1]] + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[FMINNUM_IEEE]](s32) + ; GFX9-NEXT: $vgpr0 = COPY [[COPY2]](s32) + ; GFX9-NEXT: SI_RETURN implicit $vgpr0 + ; + ; GFX12-LABEL: name: test_fminimum_nnan_f32 + ; GFX12: liveins: $vgpr0, $vgpr1 + ; GFX12-NEXT: {{ $}} + ; GFX12-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX12-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX12-NEXT: [[FMINIMUM:%[0-9]+]]:_(s32) = nnan G_FMINIMUM [[COPY]], [[COPY1]] + ; GFX12-NEXT: $vgpr0 = COPY [[FMINIMUM]](s32) + ; GFX12-NEXT: SI_RETURN implicit $vgpr0 + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(s32) = nnan G_FMINIMUM %0, %1 + $vgpr0 = COPY %2(s32) + SI_RETURN implicit $vgpr0 +... + diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir index c8bd8ab..d5947c8 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir @@ -18,58 +18,12 @@ body: | ; CHECK-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>) %0:_(s32) = COPY $vgpr0 %1:_(s32) = COPY $vgpr1 - %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0, 1) + %2:_(<2 x s32>) = G_BUILD_VECTOR %0, %1 $vgpr0_vgpr1 = COPY %2 ... --- -name: shufflevector_scalar_src_dst -tracksRegLiveness: true - -body: | - bb.0: - liveins: $vgpr0, $vgpr1 - - ; CHECK-LABEL: name: shufflevector_scalar_src_dst - ; CHECK: liveins: $vgpr0, $vgpr1 - ; CHECK-NEXT: {{ $}} - ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 - ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 - ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32) - ; CHECK-NEXT: $vgpr0 = COPY [[COPY2]](s32) - %0:_(s32) = COPY $vgpr0 - %1:_(s32) = COPY $vgpr1 - %2:_(s32) = G_SHUFFLE_VECTOR %0, %1, shufflemask(1) - $vgpr0 = COPY %2 - -... - ---- -name: shufflevector_scalar_dst -tracksRegLiveness: true - -body: | - bb.0: - liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 - - ; CHECK-LABEL: name: shufflevector_scalar_dst - ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 - ; CHECK-NEXT: {{ $}} - ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1 - ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3 - ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>) - ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32) - ; CHECK-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32) - ; CHECK-NEXT: $vgpr0 = COPY [[COPY3]](s32) - %0:_(<2 x s32>) = COPY $vgpr0_vgpr1 - %1:_(<2 x s32>) = COPY $vgpr2_vgpr3 - %2:_(s32) = G_SHUFFLE_VECTOR %0, %1, shufflemask(2) - $vgpr0 = COPY %2 - -... - ---- name: shufflevector_v2s32_0_1 tracksRegLiveness: true @@ -136,9 +90,7 @@ body: | ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3 ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>) ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32) - ; CHECK-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>) - ; CHECK-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32) - ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY3]](s32) + ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY2]](s32) ; CHECK-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>) %0:_(<2 x s32>) = COPY $vgpr0_vgpr1 %1:_(<2 x s32>) = COPY $vgpr2_vgpr3 @@ -374,3 +326,30 @@ body: | $vgpr0_vgpr1 = COPY %4 ... + +--- +name: shufflevector_cache_element +tracksRegLiveness: true + +body: | + bb.0: + liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + + ; CHECK-LABEL: name: shufflevector_cache_element + ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32) + ; CHECK-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32) + ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY2]](s32), [[COPY2]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY3]](s32), [[COPY3]](s32), [[COPY3]](s32) + ; CHECK-NEXT: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[BUILD_VECTOR]](<8 x s32>), [[COPY1]](s32) + ; CHECK-NEXT: $vgpr0 = COPY [[EVEC]](s32) + %0:_(s32) = COPY $vgpr0 + %1:_(s32) = COPY $vgpr1 + %2:_(<2 x s32>) = G_BUILD_VECTOR %0, %1 + %3:_(<8 x s32>) = G_SHUFFLE_VECTOR %2, %2, shufflemask(0, 0, 0, 0, 3, 3, 3, 3) + %4:_(s32) = G_EXTRACT_VECTOR_ELT %3(<8 x s32>), %1(s32) + $vgpr0 = COPY %4 + +... diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.s16.mir index 10f425c..dff8111 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.s16.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.s16.mir @@ -237,16 +237,13 @@ body: | ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 ; GFX8-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; GFX8-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) ; GFX8-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; GFX8-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[LSHR]], [[C1]] - ; GFX8-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[LSHR1]], [[C1]] ; GFX8-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 - ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32) + ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C2]](s32) ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]] - ; GFX8-NEXT: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) - ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) + ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) + ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST1]](<2 x s16>) ; ; GFX9-LABEL: name: shufflevector_v2s16_v2s16_0_0 ; GFX9: liveins: $vgpr0, $vgpr1 @@ -257,10 +254,7 @@ body: | ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) - ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) - ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16) + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC]](s16) ; GFX9-NEXT: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>) %0:_(<2 x s16>) = COPY $vgpr0 %1:_(<2 x s16>) = COPY $vgpr1 @@ -285,12 +279,10 @@ body: | ; GFX8-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; GFX8-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LSHR1]], [[C]](s32) + ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LSHR]], [[C]](s32) ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[LSHR]], [[SHL]] - ; GFX8-NEXT: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) - ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) + ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) + ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST1]](<2 x s16>) ; ; GFX9-LABEL: name: shufflevector_v2s16_v2s16_1_1 ; GFX9: liveins: $vgpr0, $vgpr1 @@ -301,10 +293,7 @@ body: | ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) - ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) - ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16) + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC]](s16) ; GFX9-NEXT: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>) %0:_(<2 x s16>) = COPY $vgpr0 %1:_(<2 x s16>) = COPY $vgpr1 @@ -329,16 +318,13 @@ body: | ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 ; GFX8-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) - ; GFX8-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) ; GFX8-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; GFX8-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[LSHR]], [[C1]] - ; GFX8-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[LSHR1]], [[C1]] ; GFX8-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 - ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32) + ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C2]](s32) ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]] - ; GFX8-NEXT: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) - ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) + ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) + ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST1]](<2 x s16>) ; ; GFX9-LABEL: name: shufflevector_v2s16_v2s16_2_2 ; GFX9: liveins: $vgpr0, $vgpr1 @@ -349,10 +335,7 @@ body: | ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) - ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) - ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) - ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16) + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC]](s16) ; GFX9-NEXT: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>) %0:_(<2 x s16>) = COPY $vgpr0 %1:_(<2 x s16>) = COPY $vgpr1 @@ -641,12 +624,10 @@ body: | ; GFX8-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) - ; GFX8-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LSHR1]], [[C]](s32) + ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LSHR]], [[C]](s32) ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[LSHR]], [[SHL]] - ; GFX8-NEXT: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) - ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) + ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) + ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST1]](<2 x s16>) ; ; GFX9-LABEL: name: shufflevector_v2s16_v2s16_3_3 ; GFX9: liveins: $vgpr0, $vgpr1 @@ -657,10 +638,7 @@ body: | ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) - ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>) - ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) - ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16) + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC]](s16) ; GFX9-NEXT: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>) %0:_(<2 x s16>) = COPY $vgpr0 %1:_(<2 x s16>) = COPY $vgpr1 @@ -733,16 +711,13 @@ body: | ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 ; GFX8-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; GFX8-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) ; GFX8-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; GFX8-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[LSHR]], [[C1]] - ; GFX8-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[LSHR1]], [[C1]] ; GFX8-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 - ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C2]](s32) + ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C2]](s32) ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]] - ; GFX8-NEXT: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) - ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) + ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) + ; GFX8-NEXT: $vgpr0 = COPY [[BITCAST1]](<2 x s16>) ; ; GFX9-LABEL: name: shufflevector_v2s16_v2s16_2_0 ; GFX9: liveins: $vgpr0, $vgpr1 @@ -753,10 +728,7 @@ body: | ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) - ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) - ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) - ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16) + ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC]](s16) ; GFX9-NEXT: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>) %0:_(<2 x s16>) = COPY $vgpr0 %1:_(<2 x s16>) = COPY $vgpr1 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/lshr.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/lshr.ll index 8533e34..518af70 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/lshr.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/lshr.ll @@ -1750,7 +1750,7 @@ define i65 @v_lshr_i65_33(i65 %value) { ; GFX6-NEXT: v_and_b32_e32 v0, 1, v2 ; GFX6-NEXT: v_lshl_b64 v[0:1], v[0:1], 31 ; GFX6-NEXT: v_lshrrev_b32_e32 v2, 1, v3 -; GFX6-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX6-NEXT: v_or_b32_e32 v0, v0, v2 ; GFX6-NEXT: v_mov_b32_e32 v2, 0 ; GFX6-NEXT: s_setpc_b64 s[30:31] ; @@ -1763,7 +1763,7 @@ define i65 @v_lshr_i65_33(i65 %value) { ; GFX8-NEXT: v_and_b32_e32 v0, 1, v2 ; GFX8-NEXT: v_lshlrev_b64 v[0:1], 31, v[0:1] ; GFX8-NEXT: v_lshrrev_b32_e32 v2, 1, v3 -; GFX8-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX8-NEXT: v_or_b32_e32 v0, v0, v2 ; GFX8-NEXT: v_mov_b32_e32 v2, 0 ; GFX8-NEXT: s_setpc_b64 s[30:31] ; @@ -1776,7 +1776,7 @@ define i65 @v_lshr_i65_33(i65 %value) { ; GFX9-NEXT: v_and_b32_e32 v0, 1, v2 ; GFX9-NEXT: v_lshlrev_b64 v[0:1], 31, v[0:1] ; GFX9-NEXT: v_lshrrev_b32_e32 v2, 1, v3 -; GFX9-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX9-NEXT: v_or_b32_e32 v0, v0, v2 ; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; @@ -1789,7 +1789,7 @@ define i65 @v_lshr_i65_33(i65 %value) { ; GFX10-NEXT: v_and_b32_e32 v0, 1, v2 ; GFX10-NEXT: v_lshrrev_b32_e32 v2, 1, v3 ; GFX10-NEXT: v_lshlrev_b64 v[0:1], 31, v[0:1] -; GFX10-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX10-NEXT: v_or_b32_e32 v0, v0, v2 ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; @@ -1800,7 +1800,7 @@ define i65 @v_lshr_i65_33(i65 %value) { ; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 1, v2 ; GFX11-NEXT: v_lshrrev_b32_e32 v2, 1, v3 ; GFX11-NEXT: v_lshlrev_b64 v[0:1], 31, v[0:1] -; GFX11-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX11-NEXT: v_or_b32_e32 v0, v0, v2 ; GFX11-NEXT: v_mov_b32_e32 v2, 0 ; GFX11-NEXT: s_setpc_b64 s[30:31] %result = lshr i65 %value, 33 @@ -1859,21 +1859,19 @@ define amdgpu_ps i65 @s_lshr_i65_33(i65 inreg %value) { ; GCN-LABEL: s_lshr_i65_33: ; GCN: ; %bb.0: ; GCN-NEXT: s_and_b64 s[2:3], s[2:3], 1 -; GCN-NEXT: s_lshr_b32 s0, s1, 1 -; GCN-NEXT: s_mov_b32 s1, 0 -; GCN-NEXT: s_lshl_b64 s[2:3], s[2:3], 31 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GCN-NEXT: s_lshr_b32 s4, s1, 1 +; GCN-NEXT: s_lshl_b64 s[0:1], s[2:3], 31 +; GCN-NEXT: s_or_b32 s0, s0, s4 ; GCN-NEXT: s_mov_b32 s2, 0 ; GCN-NEXT: ; return to shader part epilog ; ; GFX10PLUS-LABEL: s_lshr_i65_33: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_and_b64 s[2:3], s[2:3], 1 -; GFX10PLUS-NEXT: s_lshr_b32 s0, s1, 1 -; GFX10PLUS-NEXT: s_mov_b32 s1, 0 -; GFX10PLUS-NEXT: s_lshl_b64 s[2:3], s[2:3], 31 -; GFX10PLUS-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] +; GFX10PLUS-NEXT: s_lshr_b32 s4, s1, 1 +; GFX10PLUS-NEXT: s_lshl_b64 s[0:1], s[2:3], 31 ; GFX10PLUS-NEXT: s_mov_b32 s2, 0 +; GFX10PLUS-NEXT: s_or_b32 s0, s0, s4 ; GFX10PLUS-NEXT: ; return to shader part epilog %result = lshr i65 %value, 33 ret i65 %result diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/or.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/or.ll index af377b1..e0581f01 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/or.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/or.ll @@ -597,13 +597,13 @@ define amdgpu_kernel void @s_or_u64_zext_with_sregs(ptr addrspace(1) %out, ptr a ; GFX7-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GFX7-NEXT: s_mov_b32 s5, 0 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: s_load_dword s4, s[2:3], 0x0 +; GFX7-NEXT: s_load_dword s3, s[2:3], 0x0 ; GFX7-NEXT: s_mov_b32 s2, -1 -; GFX7-NEXT: s_mov_b32 s3, 0xf000 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: s_or_b64 s[4:5], s[4:5], 0x50 +; GFX7-NEXT: s_or_b32 s4, s3, 0x50 ; GFX7-NEXT: v_mov_b32_e32 v0, s4 ; GFX7-NEXT: v_mov_b32_e32 v1, s5 +; GFX7-NEXT: s_mov_b32 s3, 0xf000 ; GFX7-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX7-NEXT: s_endpgm ; @@ -616,7 +616,7 @@ define amdgpu_kernel void @s_or_u64_zext_with_sregs(ptr addrspace(1) %out, ptr a ; GFX8-NEXT: s_mov_b32 s3, 0 ; GFX8-NEXT: v_mov_b32_e32 v2, s0 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: s_or_b64 s[2:3], s[2:3], 0x50 +; GFX8-NEXT: s_or_b32 s2, s2, 0x50 ; GFX8-NEXT: v_mov_b32_e32 v0, s2 ; GFX8-NEXT: v_mov_b32_e32 v1, s3 ; GFX8-NEXT: flat_store_dwordx2 v[2:3], v[0:1] @@ -630,7 +630,7 @@ define amdgpu_kernel void @s_or_u64_zext_with_sregs(ptr addrspace(1) %out, ptr a ; GFX9-NEXT: s_load_dword s2, s[2:3], 0x0 ; GFX9-NEXT: s_mov_b32 s3, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_or_b64 s[2:3], s[2:3], 0x50 +; GFX9-NEXT: s_or_b32 s2, s2, 0x50 ; GFX9-NEXT: v_mov_b32_e32 v0, s2 ; GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] @@ -644,7 +644,7 @@ define amdgpu_kernel void @s_or_u64_zext_with_sregs(ptr addrspace(1) %out, ptr a ; GFX10-NEXT: s_load_dword s2, s[2:3], 0x0 ; GFX10-NEXT: s_mov_b32 s3, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: s_or_b64 s[2:3], s[2:3], 0x50 +; GFX10-NEXT: s_or_b32 s2, s2, 0x50 ; GFX10-NEXT: v_mov_b32_e32 v0, s2 ; GFX10-NEXT: v_mov_b32_e32 v1, s3 ; GFX10-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] @@ -658,7 +658,7 @@ define amdgpu_kernel void @s_or_u64_zext_with_sregs(ptr addrspace(1) %out, ptr a ; GFX11-NEXT: s_load_b32 s2, s[2:3], 0x0 ; GFX11-NEXT: s_mov_b32 s3, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_or_b64 s[2:3], s[2:3], 0x50 +; GFX11-NEXT: s_or_b32 s2, s2, 0x50 ; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] ; GFX11-NEXT: s_endpgm @@ -671,7 +671,7 @@ define amdgpu_kernel void @s_or_u64_zext_with_sregs(ptr addrspace(1) %out, ptr a ; GFX12-NEXT: s_load_b32 s2, s[2:3], 0x0 ; GFX12-NEXT: s_mov_b32 s3, 0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: s_or_b64 s[2:3], s[2:3], 0x50 +; GFX12-NEXT: s_or_b32 s2, s2, 0x50 ; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX12-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3 ; GFX12-NEXT: global_store_b64 v2, v[0:1], s[0:1] diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/prelegalizer-combiner-shuffle.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/prelegalizer-combiner-shuffle.mir index 6e4c6bc..31e3d97 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/prelegalizer-combiner-shuffle.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/prelegalizer-combiner-shuffle.mir @@ -147,15 +147,17 @@ body: | ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(p3) = COPY $vgpr1 - ; CHECK-NEXT: [[LOAD:%[0-9]+]]:_(<8 x s16>) = G_LOAD [[COPY]](p3) :: (load (<8 x s16>), align 8, addrspace 3) - ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[LOAD]](<8 x s16>) - ; CHECK-NEXT: G_STORE [[UV4]](s16), [[COPY1]](p3) :: (store (s16), addrspace 3) + ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 + ; CHECK-NEXT: [[PTR_ADD:%[0-9]+]]:_(p3) = G_PTR_ADD [[COPY]], [[C]](s32) + ; CHECK-NEXT: [[LOAD:%[0-9]+]]:_(s16) = G_LOAD [[PTR_ADD]](p3) :: (load (s16), addrspace 3) + ; CHECK-NEXT: G_STORE [[LOAD]](s16), [[COPY1]](p3) :: (store (s16), addrspace 3) ; CHECK-NEXT: SI_RETURN %0:_(p3) = COPY $vgpr0 %1:_(p3) = COPY $vgpr1 %2:_(<8 x s16>) = G_IMPLICIT_DEF %3:_(<8 x s16>) = G_LOAD %0(p3) :: (load (<8 x s16>), align 8, addrspace 3) - %4:_(s16) = G_SHUFFLE_VECTOR %3(<8 x s16>), %2, shufflemask(4) + %idx:_(s32) = G_CONSTANT i32 4 + %4:_(s16) = G_EXTRACT_VECTOR_ELT %3(<8 x s16>), %idx G_STORE %4(s16), %1(p3) :: (store (s16), addrspace 3) SI_RETURN ... diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sext_inreg.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sext_inreg.ll index a9b3deb..cfe655f 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sext_inreg.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sext_inreg.ll @@ -1381,7 +1381,7 @@ define i65 @v_sext_inreg_i65_33(i65 %value) { ; GFX6-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX6-NEXT: v_lshl_b64 v[0:1], v[1:2], 31 ; GFX6-NEXT: v_lshrrev_b32_e32 v3, 1, v3 -; GFX6-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX6-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX6-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX6-NEXT: s_setpc_b64 s[30:31] ; @@ -1393,7 +1393,7 @@ define i65 @v_sext_inreg_i65_33(i65 %value) { ; GFX8-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX8-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX8-NEXT: v_lshrrev_b32_e32 v3, 1, v3 -; GFX8-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX8-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX8-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX8-NEXT: s_setpc_b64 s[30:31] ; @@ -1405,7 +1405,7 @@ define i65 @v_sext_inreg_i65_33(i65 %value) { ; GFX9-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX9-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX9-NEXT: v_lshrrev_b32_e32 v3, 1, v3 -; GFX9-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX9-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX9-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; @@ -1418,7 +1418,7 @@ define i65 @v_sext_inreg_i65_33(i65 %value) { ; GFX10PLUS-NEXT: v_ashrrev_i32_e32 v2, 31, v1 ; GFX10PLUS-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX10PLUS-NEXT: v_ashrrev_i32_e32 v2, 1, v2 -; GFX10PLUS-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX10PLUS-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] %shl = shl i65 %value, 33 %ashr = ashr i65 %value, 33 @@ -1429,29 +1429,27 @@ define amdgpu_ps i65 @s_sext_inreg_i65_18(i65 inreg %value) { ; GCN-LABEL: s_sext_inreg_i65_18: ; GCN: ; %bb.0: ; GCN-NEXT: s_lshl_b64 s[2:3], s[2:3], 18 -; GCN-NEXT: s_lshr_b32 s4, s1, 14 -; GCN-NEXT: s_mov_b32 s5, 0 -; GCN-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GCN-NEXT: s_lshr_b32 s3, s1, 14 +; GCN-NEXT: s_or_b32 s2, s2, s3 ; GCN-NEXT: s_bfe_i64 s[2:3], s[2:3], 0x10000 ; GCN-NEXT: s_bfe_u64 s[0:1], s[0:1], 0x2e0000 -; GCN-NEXT: s_lshl_b32 s7, s2, 14 -; GCN-NEXT: s_mov_b32 s6, s5 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[6:7] +; GCN-NEXT: s_lshl_b32 s5, s2, 14 +; GCN-NEXT: s_mov_b32 s4, 0 +; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] ; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 18 ; GCN-NEXT: ; return to shader part epilog ; ; GFX10PLUS-LABEL: s_sext_inreg_i65_18: ; GFX10PLUS: ; %bb.0: ; GFX10PLUS-NEXT: s_lshl_b64 s[2:3], s[2:3], 18 -; GFX10PLUS-NEXT: s_lshr_b32 s4, s1, 14 -; GFX10PLUS-NEXT: s_mov_b32 s5, 0 +; GFX10PLUS-NEXT: s_lshr_b32 s3, s1, 14 ; GFX10PLUS-NEXT: s_bfe_u64 s[0:1], s[0:1], 0x2e0000 -; GFX10PLUS-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] -; GFX10PLUS-NEXT: s_mov_b32 s6, s5 +; GFX10PLUS-NEXT: s_or_b32 s2, s2, s3 +; GFX10PLUS-NEXT: s_mov_b32 s4, 0 ; GFX10PLUS-NEXT: s_bfe_i64 s[2:3], s[2:3], 0x10000 -; GFX10PLUS-NEXT: s_lshl_b32 s7, s2, 14 +; GFX10PLUS-NEXT: s_lshl_b32 s5, s2, 14 ; GFX10PLUS-NEXT: s_ashr_i64 s[2:3], s[2:3], 18 -; GFX10PLUS-NEXT: s_or_b64 s[0:1], s[0:1], s[6:7] +; GFX10PLUS-NEXT: s_or_b64 s[0:1], s[0:1], s[4:5] ; GFX10PLUS-NEXT: ; return to shader part epilog %shl = shl i65 %value, 18 %ashr = ashr i65 %shl, 18 @@ -1464,13 +1462,12 @@ define amdgpu_ps i65 @s_sext_inreg_i65_33(i65 inreg %value) { ; GCN-NEXT: s_lshl_b32 s3, s2, 1 ; GCN-NEXT: s_mov_b32 s2, 0 ; GCN-NEXT: s_lshr_b64 s[4:5], s[0:1], 31 -; GCN-NEXT: s_or_b64 s[4:5], s[2:3], s[4:5] -; GCN-NEXT: s_bfe_i64 s[4:5], s[4:5], 0x10000 -; GCN-NEXT: s_bfe_u32 s0, s0, 0x1f0000 -; GCN-NEXT: s_mov_b32 s1, s2 -; GCN-NEXT: s_lshl_b64 s[2:3], s[4:5], 31 -; GCN-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] -; GCN-NEXT: s_ashr_i32 s2, s5, 1 +; GCN-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GCN-NEXT: s_bfe_i64 s[2:3], s[2:3], 0x10000 +; GCN-NEXT: s_bfe_u32 s4, s0, 0x1f0000 +; GCN-NEXT: s_lshl_b64 s[0:1], s[2:3], 31 +; GCN-NEXT: s_or_b32 s0, s0, s4 +; GCN-NEXT: s_ashr_i32 s2, s3, 1 ; GCN-NEXT: ; return to shader part epilog ; ; GFX10PLUS-LABEL: s_sext_inreg_i65_33: @@ -1478,13 +1475,12 @@ define amdgpu_ps i65 @s_sext_inreg_i65_33(i65 inreg %value) { ; GFX10PLUS-NEXT: s_lshl_b32 s3, s2, 1 ; GFX10PLUS-NEXT: s_mov_b32 s2, 0 ; GFX10PLUS-NEXT: s_lshr_b64 s[4:5], s[0:1], 31 -; GFX10PLUS-NEXT: s_bfe_u32 s0, s0, 0x1f0000 -; GFX10PLUS-NEXT: s_or_b64 s[4:5], s[2:3], s[4:5] -; GFX10PLUS-NEXT: s_mov_b32 s1, s2 -; GFX10PLUS-NEXT: s_bfe_i64 s[4:5], s[4:5], 0x10000 -; GFX10PLUS-NEXT: s_lshl_b64 s[2:3], s[4:5], 31 -; GFX10PLUS-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3] -; GFX10PLUS-NEXT: s_ashr_i32 s2, s5, 1 +; GFX10PLUS-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GFX10PLUS-NEXT: s_bfe_u32 s4, s0, 0x1f0000 +; GFX10PLUS-NEXT: s_bfe_i64 s[2:3], s[2:3], 0x10000 +; GFX10PLUS-NEXT: s_lshl_b64 s[0:1], s[2:3], 31 +; GFX10PLUS-NEXT: s_ashr_i32 s2, s3, 1 +; GFX10PLUS-NEXT: s_or_b32 s0, s0, s4 ; GFX10PLUS-NEXT: ; return to shader part epilog %shl = shl i65 %value, 33 %ashr = ashr i65 %shl, 33 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/shufflevector-pointer-crash.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/shufflevector-pointer-crash.mir index ac903ad..e7bba9d 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/shufflevector-pointer-crash.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/shufflevector-pointer-crash.mir @@ -19,15 +19,15 @@ body: | ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:_(<2 x p0>) = G_BITCAST [[BUILD_VECTOR]](<4 x s32>) ; CHECK-NEXT: [[UV:%[0-9]+]]:_(p0), [[UV1:%[0-9]+]]:_(p0) = G_UNMERGE_VALUES [[BITCAST]](<2 x p0>) ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[UV]](p0) - ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(p0) = COPY [[COPY]](p0) - ; CHECK-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](p0) + ; CHECK-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](p0) ; CHECK-NEXT: $vgpr0 = COPY [[UV2]](s32) ; CHECK-NEXT: $vgpr1 = COPY [[UV3]](s32) ; CHECK-NEXT: SI_RETURN implicit $vgpr0, implicit $vgpr1 %0:_(p0) = G_CONSTANT i64 0 %1:_(<2 x p0>) = G_BUILD_VECTOR %0:_(p0), %0:_(p0) %2:_(<2 x p0>) = G_LOAD %0:_(p0) :: (load (<2 x p0>)) - %3:_(p0) = G_SHUFFLE_VECTOR %2:_(<2 x p0>), %1:_, shufflemask(0) + %idx:_(s32) = G_CONSTANT i32 0 + %3:_(p0) = G_EXTRACT_VECTOR_ELT %2:_(<2 x p0>), %idx %4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %3:_(p0) $vgpr0 = COPY %4:_(s32) $vgpr1 = COPY %5:_(s32) diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll index afd0f01..6831380 100644 --- a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll +++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll @@ -415,28 +415,18 @@ define amdgpu_kernel void @memcpy_known(ptr addrspace(7) %src, ptr addrspace(7) ; GISEL-GFX942-LABEL: memcpy_known: ; GISEL-GFX942: ; %bb.0: ; GISEL-GFX942-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GISEL-GFX942-NEXT: s_load_dword s7, s[4:5], 0x54 ; GISEL-GFX942-NEXT: s_load_dword s11, s[4:5], 0x34 -; GISEL-GFX942-NEXT: s_mov_b32 s7, 0 ; GISEL-GFX942-NEXT: s_load_dwordx4 s[12:15], s[4:5], 0x44 -; GISEL-GFX942-NEXT: s_mov_b32 s8, s7 +; GISEL-GFX942-NEXT: s_mov_b32 s16, 0 +; GISEL-GFX942-NEXT: v_mov_b32_e32 v0, 0x2000 ; GISEL-GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX942-NEXT: s_mov_b32 s6, s1 +; GISEL-GFX942-NEXT: s_mov_b32 s8, s1 ; GISEL-GFX942-NEXT: s_mov_b32 s9, s2 -; GISEL-GFX942-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9] -; GISEL-GFX942-NEXT: s_mov_b32 s6, s3 -; GISEL-GFX942-NEXT: s_load_dword s3, s[4:5], 0x54 -; GISEL-GFX942-NEXT: s_mov_b32 s10, s7 -; GISEL-GFX942-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] -; GISEL-GFX942-NEXT: s_mov_b32 s6, s13 -; GISEL-GFX942-NEXT: s_mov_b32 s4, s7 +; GISEL-GFX942-NEXT: s_mov_b32 s10, s3 +; GISEL-GFX942-NEXT: s_mov_b32 s4, s13 ; GISEL-GFX942-NEXT: s_mov_b32 s5, s14 -; GISEL-GFX942-NEXT: s_mov_b32 s16, 0 -; GISEL-GFX942-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] ; GISEL-GFX942-NEXT: s_mov_b32 s6, s15 -; GISEL-GFX942-NEXT: s_mov_b32 s2, s7 -; GISEL-GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX942-NEXT: s_or_b64 s[6:7], s[6:7], s[2:3] -; GISEL-GFX942-NEXT: v_mov_b32_e32 v0, 0x2000 ; GISEL-GFX942-NEXT: v_mov_b32_e32 v1, s16 ; GISEL-GFX942-NEXT: .LBB0_1: ; %load-store-loop ; GISEL-GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -491,25 +481,16 @@ define amdgpu_kernel void @memcpy_known(ptr addrspace(7) %src, ptr addrspace(7) ; GISEL-GFX1100-NEXT: s_load_b128 s[8:11], s[4:5], 0x44 ; GISEL-GFX1100-NEXT: s_load_b32 s7, s[4:5], 0x34 ; GISEL-GFX1100-NEXT: s_load_b32 s15, s[4:5], 0x54 -; GISEL-GFX1100-NEXT: s_mov_b32 s17, 0 -; GISEL-GFX1100-NEXT: s_mov_b32 s12, 0 -; GISEL-GFX1100-NEXT: s_mov_b32 s4, s17 -; GISEL-GFX1100-NEXT: s_mov_b32 s6, s17 -; GISEL-GFX1100-NEXT: v_mov_b32_e32 v0, s12 -; GISEL-GFX1100-NEXT: s_mov_b32 s14, s17 +; GISEL-GFX1100-NEXT: s_mov_b32 s4, 0 +; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GISEL-GFX1100-NEXT: v_mov_b32_e32 v0, s4 ; GISEL-GFX1100-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s1 +; GISEL-GFX1100-NEXT: s_mov_b32 s4, s1 ; GISEL-GFX1100-NEXT: s_mov_b32 s5, s2 -; GISEL-GFX1100-NEXT: s_mov_b32 s2, s17 -; GISEL-GFX1100-NEXT: s_or_b64 s[4:5], s[16:17], s[4:5] -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s3 -; GISEL-GFX1100-NEXT: s_mov_b32 s3, s10 -; GISEL-GFX1100-NEXT: s_or_b64 s[6:7], s[16:17], s[6:7] -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s9 -; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) -; GISEL-GFX1100-NEXT: s_or_b64 s[12:13], s[16:17], s[2:3] -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s11 -; GISEL-GFX1100-NEXT: s_or_b64 s[14:15], s[16:17], s[14:15] +; GISEL-GFX1100-NEXT: s_mov_b32 s6, s3 +; GISEL-GFX1100-NEXT: s_mov_b32 s12, s9 +; GISEL-GFX1100-NEXT: s_mov_b32 s13, s10 +; GISEL-GFX1100-NEXT: s_mov_b32 s14, s11 ; GISEL-GFX1100-NEXT: .LBB0_1: ; %load-store-loop ; GISEL-GFX1100-NEXT: ; =>This Inner Loop Header: Depth=1 ; GISEL-GFX1100-NEXT: v_add_nc_u32_e32 v61, s0, v0 @@ -960,28 +941,18 @@ define amdgpu_kernel void @memcpy_known_medium(ptr addrspace(7) %src, ptr addrsp ; GISEL-GFX942-LABEL: memcpy_known_medium: ; GISEL-GFX942: ; %bb.0: ; GISEL-GFX942-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GISEL-GFX942-NEXT: s_load_dword s7, s[4:5], 0x54 ; GISEL-GFX942-NEXT: s_load_dword s11, s[4:5], 0x34 -; GISEL-GFX942-NEXT: s_mov_b32 s7, 0 ; GISEL-GFX942-NEXT: s_load_dwordx4 s[12:15], s[4:5], 0x44 -; GISEL-GFX942-NEXT: s_mov_b32 s8, s7 +; GISEL-GFX942-NEXT: s_mov_b32 s16, 0 +; GISEL-GFX942-NEXT: v_mov_b32_e32 v0, 0x100 ; GISEL-GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX942-NEXT: s_mov_b32 s6, s1 +; GISEL-GFX942-NEXT: s_mov_b32 s8, s1 ; GISEL-GFX942-NEXT: s_mov_b32 s9, s2 -; GISEL-GFX942-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9] -; GISEL-GFX942-NEXT: s_mov_b32 s6, s3 -; GISEL-GFX942-NEXT: s_load_dword s3, s[4:5], 0x54 -; GISEL-GFX942-NEXT: s_mov_b32 s10, s7 -; GISEL-GFX942-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] -; GISEL-GFX942-NEXT: s_mov_b32 s6, s13 -; GISEL-GFX942-NEXT: s_mov_b32 s4, s7 +; GISEL-GFX942-NEXT: s_mov_b32 s10, s3 +; GISEL-GFX942-NEXT: s_mov_b32 s4, s13 ; GISEL-GFX942-NEXT: s_mov_b32 s5, s14 -; GISEL-GFX942-NEXT: s_mov_b32 s16, 0 -; GISEL-GFX942-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] ; GISEL-GFX942-NEXT: s_mov_b32 s6, s15 -; GISEL-GFX942-NEXT: s_mov_b32 s2, s7 -; GISEL-GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX942-NEXT: s_or_b64 s[6:7], s[6:7], s[2:3] -; GISEL-GFX942-NEXT: v_mov_b32_e32 v0, 0x100 ; GISEL-GFX942-NEXT: v_mov_b32_e32 v1, s16 ; GISEL-GFX942-NEXT: .LBB1_1: ; %load-store-loop ; GISEL-GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -1036,25 +1007,16 @@ define amdgpu_kernel void @memcpy_known_medium(ptr addrspace(7) %src, ptr addrsp ; GISEL-GFX1100-NEXT: s_load_b128 s[8:11], s[4:5], 0x44 ; GISEL-GFX1100-NEXT: s_load_b32 s7, s[4:5], 0x34 ; GISEL-GFX1100-NEXT: s_load_b32 s15, s[4:5], 0x54 -; GISEL-GFX1100-NEXT: s_mov_b32 s17, 0 -; GISEL-GFX1100-NEXT: s_mov_b32 s12, 0 -; GISEL-GFX1100-NEXT: s_mov_b32 s4, s17 -; GISEL-GFX1100-NEXT: s_mov_b32 s6, s17 -; GISEL-GFX1100-NEXT: v_mov_b32_e32 v0, s12 -; GISEL-GFX1100-NEXT: s_mov_b32 s14, s17 +; GISEL-GFX1100-NEXT: s_mov_b32 s4, 0 +; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GISEL-GFX1100-NEXT: v_mov_b32_e32 v0, s4 ; GISEL-GFX1100-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s1 +; GISEL-GFX1100-NEXT: s_mov_b32 s4, s1 ; GISEL-GFX1100-NEXT: s_mov_b32 s5, s2 -; GISEL-GFX1100-NEXT: s_mov_b32 s2, s17 -; GISEL-GFX1100-NEXT: s_or_b64 s[4:5], s[16:17], s[4:5] -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s3 -; GISEL-GFX1100-NEXT: s_mov_b32 s3, s10 -; GISEL-GFX1100-NEXT: s_or_b64 s[6:7], s[16:17], s[6:7] -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s9 -; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) -; GISEL-GFX1100-NEXT: s_or_b64 s[12:13], s[16:17], s[2:3] -; GISEL-GFX1100-NEXT: s_mov_b32 s16, s11 -; GISEL-GFX1100-NEXT: s_or_b64 s[14:15], s[16:17], s[14:15] +; GISEL-GFX1100-NEXT: s_mov_b32 s6, s3 +; GISEL-GFX1100-NEXT: s_mov_b32 s12, s9 +; GISEL-GFX1100-NEXT: s_mov_b32 s13, s10 +; GISEL-GFX1100-NEXT: s_mov_b32 s14, s11 ; GISEL-GFX1100-NEXT: .LBB1_1: ; %load-store-loop ; GISEL-GFX1100-NEXT: ; =>This Inner Loop Header: Depth=1 ; GISEL-GFX1100-NEXT: v_add_nc_u32_e32 v61, s0, v0 @@ -1228,27 +1190,18 @@ define amdgpu_kernel void @memcpy_known_small(ptr addrspace(7) %src, ptr addrspa ; GISEL-GFX942: ; %bb.0: ; GISEL-GFX942-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 ; GISEL-GFX942-NEXT: s_load_dword s11, s[4:5], 0x34 -; GISEL-GFX942-NEXT: s_mov_b32 s7, 0 -; GISEL-GFX942-NEXT: s_mov_b32 s8, s7 -; GISEL-GFX942-NEXT: s_mov_b32 s10, s7 ; GISEL-GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX942-NEXT: s_mov_b32 s6, s1 +; GISEL-GFX942-NEXT: s_mov_b32 s8, s1 ; GISEL-GFX942-NEXT: s_mov_b32 s9, s2 -; GISEL-GFX942-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9] -; GISEL-GFX942-NEXT: s_mov_b32 s6, s3 -; GISEL-GFX942-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] +; GISEL-GFX942-NEXT: s_mov_b32 s10, s3 ; GISEL-GFX942-NEXT: v_mov_b32_e32 v4, s0 ; GISEL-GFX942-NEXT: buffer_load_dwordx4 v[0:3], v4, s[8:11], 0 offen ; GISEL-GFX942-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x44 -; GISEL-GFX942-NEXT: s_load_dword s13, s[4:5], 0x54 -; GISEL-GFX942-NEXT: s_mov_b32 s4, s7 -; GISEL-GFX942-NEXT: s_mov_b32 s12, s7 +; GISEL-GFX942-NEXT: s_load_dword s7, s[4:5], 0x54 ; GISEL-GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX942-NEXT: s_mov_b32 s6, s1 +; GISEL-GFX942-NEXT: s_mov_b32 s4, s1 ; GISEL-GFX942-NEXT: s_mov_b32 s5, s2 -; GISEL-GFX942-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] ; GISEL-GFX942-NEXT: s_mov_b32 s6, s3 -; GISEL-GFX942-NEXT: s_or_b64 s[6:7], s[6:7], s[12:13] ; GISEL-GFX942-NEXT: v_mov_b32_e32 v5, s0 ; GISEL-GFX942-NEXT: s_waitcnt vmcnt(0) ; GISEL-GFX942-NEXT: buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen @@ -1261,35 +1214,24 @@ define amdgpu_kernel void @memcpy_known_small(ptr addrspace(7) %src, ptr addrspa ; GISEL-GFX1100: ; %bb.0: ; GISEL-GFX1100-NEXT: s_clause 0x1 ; GISEL-GFX1100-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GISEL-GFX1100-NEXT: s_load_b32 s7, s[4:5], 0x34 -; GISEL-GFX1100-NEXT: s_mov_b32 s13, 0 -; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GISEL-GFX1100-NEXT: s_mov_b32 s8, s13 -; GISEL-GFX1100-NEXT: s_mov_b32 s6, s13 +; GISEL-GFX1100-NEXT: s_load_b32 s11, s[4:5], 0x34 ; GISEL-GFX1100-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX1100-NEXT: s_mov_b32 s12, s1 -; GISEL-GFX1100-NEXT: s_mov_b32 s9, s2 ; GISEL-GFX1100-NEXT: v_mov_b32_e32 v4, s0 -; GISEL-GFX1100-NEXT: s_or_b64 s[0:1], s[12:13], s[8:9] -; GISEL-GFX1100-NEXT: s_mov_b32 s12, s3 -; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GISEL-GFX1100-NEXT: s_or_b64 s[2:3], s[12:13], s[6:7] -; GISEL-GFX1100-NEXT: buffer_load_b128 v[0:3], v4, s[0:3], 0 offen +; GISEL-GFX1100-NEXT: s_mov_b32 s8, s1 +; GISEL-GFX1100-NEXT: s_mov_b32 s9, s2 +; GISEL-GFX1100-NEXT: s_mov_b32 s10, s3 +; GISEL-GFX1100-NEXT: buffer_load_b128 v[0:3], v4, s[8:11], 0 offen ; GISEL-GFX1100-NEXT: s_clause 0x1 -; GISEL-GFX1100-NEXT: s_load_b128 s[8:11], s[4:5], 0x44 +; GISEL-GFX1100-NEXT: s_load_b128 s[0:3], s[4:5], 0x44 ; GISEL-GFX1100-NEXT: s_load_b32 s7, s[4:5], 0x54 -; GISEL-GFX1100-NEXT: s_mov_b32 s4, s13 ; GISEL-GFX1100-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-GFX1100-NEXT: s_mov_b32 s12, s9 -; GISEL-GFX1100-NEXT: s_mov_b32 s5, s10 -; GISEL-GFX1100-NEXT: v_mov_b32_e32 v5, s8 -; GISEL-GFX1100-NEXT: s_or_b64 s[4:5], s[12:13], s[4:5] -; GISEL-GFX1100-NEXT: s_mov_b32 s12, s11 -; GISEL-GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GISEL-GFX1100-NEXT: s_or_b64 s[6:7], s[12:13], s[6:7] +; GISEL-GFX1100-NEXT: v_mov_b32_e32 v5, s0 +; GISEL-GFX1100-NEXT: s_mov_b32 s4, s1 +; GISEL-GFX1100-NEXT: s_mov_b32 s5, s2 +; GISEL-GFX1100-NEXT: s_mov_b32 s6, s3 ; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) ; GISEL-GFX1100-NEXT: buffer_store_b128 v[0:3], v5, s[4:7], 0 offen -; GISEL-GFX1100-NEXT: buffer_load_b128 v[0:3], v4, s[0:3], 0 offen offset:16 +; GISEL-GFX1100-NEXT: buffer_load_b128 v[0:3], v4, s[8:11], 0 offen offset:16 ; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) ; GISEL-GFX1100-NEXT: buffer_store_b128 v[0:3], v5, s[4:7], 0 offen offset:16 ; GISEL-GFX1100-NEXT: s_endpgm diff --git a/llvm/test/CodeGen/AMDGPU/div_i128.ll b/llvm/test/CodeGen/AMDGPU/div_i128.ll index 6c8207a..df7f8c6 100644 --- a/llvm/test/CodeGen/AMDGPU/div_i128.ll +++ b/llvm/test/CodeGen/AMDGPU/div_i128.ll @@ -4344,7 +4344,7 @@ define i128 @v_sdiv_i128_v_pow2k(i128 %lhs) { ; GFX9-G-NEXT: v_addc_co_u32_e32 v2, vcc, 0, v3, vcc ; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2] ; GFX9-G-NEXT: v_lshrrev_b32_e32 v3, 1, v4 -; GFX9-G-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX9-G-NEXT: v_or_b32_e32 v0, v0, v3 ; GFX9-G-NEXT: v_ashrrev_i32_e32 v3, 31, v2 ; GFX9-G-NEXT: v_ashrrev_i32_e32 v2, 1, v2 ; GFX9-G-NEXT: s_setpc_b64 s[30:31] @@ -4375,14 +4375,12 @@ define i128 @v_sdiv_i128_v_pow2k(i128 %lhs) { ; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v4 ; GFX9-G-O0-NEXT: s_mov_b32 s5, 1 ; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s5 -; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v0, v0, v1 -; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, 0 -; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4 -; GFX9-G-O0-NEXT: v_lshlrev_b64 v[5:6], v2, v[5:6] -; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5 -; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v6 -; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v3 -; GFX9-G-O0-NEXT: v_or_b32_e64 v1, v1, v2 +; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v2, v0, v1 +; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-G-O0-NEXT: v_lshlrev_b64 v[5:6], v0, v[5:6] +; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v5 +; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v6 +; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v2 ; GFX9-G-O0-NEXT: s_mov_b32 s4, 31 ; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4 ; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v3, v2, v4 @@ -4437,7 +4435,7 @@ define i128 @v_udiv_i128_v_pow2k(i128 %lhs) { ; GFX9-G-NEXT: v_mov_b32_e32 v4, v1 ; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], 31, v[2:3] ; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 1, v4 -; GFX9-G-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX9-G-NEXT: v_or_b32_e32 v0, v0, v2 ; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 1, v3 ; GFX9-G-NEXT: v_mov_b32_e32 v3, 0 ; GFX9-G-NEXT: s_setpc_b64 s[30:31] @@ -4450,15 +4448,13 @@ define i128 @v_udiv_i128_v_pow2k(i128 %lhs) { ; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v3 ; GFX9-G-O0-NEXT: s_mov_b32 s4, 1 ; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4 -; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v0, v0, v1 -; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, 0 +; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v2, v0, v1 ; GFX9-G-O0-NEXT: s_mov_b32 s4, 31 -; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4 -; GFX9-G-O0-NEXT: v_lshlrev_b64 v[5:6], v2, v[4:5] -; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v5 -; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v6 -; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v4 -; GFX9-G-O0-NEXT: v_or_b32_e64 v1, v1, v2 +; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-G-O0-NEXT: v_lshlrev_b64 v[4:5], v0, v[4:5] +; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v4 +; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v5 +; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v2 ; GFX9-G-O0-NEXT: s_mov_b32 s4, 1 ; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4 ; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v2, v2, v3 diff --git a/llvm/test/CodeGen/AMDGPU/flat-saddr-load.ll b/llvm/test/CodeGen/AMDGPU/flat-saddr-load.ll index b5b2655..31344c7 100644 --- a/llvm/test/CodeGen/AMDGPU/flat-saddr-load.ll +++ b/llvm/test/CodeGen/AMDGPU/flat-saddr-load.ll @@ -2080,21 +2080,13 @@ define amdgpu_ps float @flat_load_saddr_i8_offset_or_i64_imm_offset_16(ptr addrs } define amdgpu_ps float @flat_load_saddr_i8_offset_or_i64_imm_offset_4160(ptr addrspace(6) inreg %sbase, i32 %idx) { -; GFX1250-SDAG-LABEL: flat_load_saddr_i8_offset_or_i64_imm_offset_4160: -; GFX1250-SDAG: ; %bb.0: -; GFX1250-SDAG-NEXT: v_or_b32_e32 v0, 0x1040, v0 -; GFX1250-SDAG-NEXT: v_mov_b32_e32 v1, 0 -; GFX1250-SDAG-NEXT: flat_load_u8 v0, v[0:1] -; GFX1250-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-SDAG-NEXT: ; return to shader part epilog -; -; GFX1250-GISEL-LABEL: flat_load_saddr_i8_offset_or_i64_imm_offset_4160: -; GFX1250-GISEL: ; %bb.0: -; GFX1250-GISEL-NEXT: v_mov_b32_e32 v1, 0 -; GFX1250-GISEL-NEXT: v_or_b32_e32 v0, 0x1040, v0 -; GFX1250-GISEL-NEXT: flat_load_u8 v0, v[0:1] -; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-GISEL-NEXT: ; return to shader part epilog +; GFX1250-LABEL: flat_load_saddr_i8_offset_or_i64_imm_offset_4160: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: v_or_b32_e32 v0, 0x1040, v0 +; GFX1250-NEXT: v_mov_b32_e32 v1, 0 +; GFX1250-NEXT: flat_load_u8 v0, v[0:1] +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: ; return to shader part epilog %zext.idx = zext i32 %idx to i64 %or = or i64 %zext.idx, 4160 %addr = inttoptr i64 %or to ptr diff --git a/llvm/test/CodeGen/AMDGPU/fmaximum.ll b/llvm/test/CodeGen/AMDGPU/fmaximum.ll index e59fbad..62ec010 100644 --- a/llvm/test/CodeGen/AMDGPU/fmaximum.ll +++ b/llvm/test/CodeGen/AMDGPU/fmaximum.ll @@ -1,117 +1,296 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-TRUE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-FAKE16 %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-TRUE16 %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-FAKE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s define amdgpu_ps float @test_fmaximum_f32_vv(float %a, float %b) { -; GCN-LABEL: test_fmaximum_f32_vv: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, v1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_f32_vv: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f32_vv: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.maximum.f32(float %a, float %b) ret float %val } define amdgpu_ps float @test_fmaximum_f32_ss(float inreg %a, float inreg %b) { -; GCN-LABEL: test_fmaximum_f32_ss: -; GCN: ; %bb.0: -; GCN-NEXT: s_maximum_f32 s0, s0, s1 -; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3) -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_f32_ss: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-NEXT: v_max_f32_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f32_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_maximum_f32 s0, s0, s1 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.maximum.f32(float %a, float %b) ret float %val } define amdgpu_ps float @test_fmaximum_f32_vs(float %a, float inreg %b) { -; GCN-LABEL: test_fmaximum_f32_vs: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, s0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_f32_vs: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f32_vs: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.maximum.f32(float %a, float %b) ret float %val } define amdgpu_ps float @test_fmaximum_nnan_f32(float %a, float %b) { -; GCN-LABEL: test_fmaximum_nnan_f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, v1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_nnan_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v0, v0, v1 +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_nnan_f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog %val = call nnan float @llvm.maximum.f32(float %a, float %b) ret float %val } +define amdgpu_ps float @test_fmaximum_nsz_f32(float %a, float %b) { +; GFX9-LABEL: test_fmaximum_nsz_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_nsz_f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog + %val = call nsz float @llvm.maximum.f32(float %a, float %b) + ret float %val +} + +define amdgpu_ps float @test_fmaximum_signed_zero_f32() { +; GFX9-LABEL: test_fmaximum_signed_zero_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_signed_zero_f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_mov_b32_e32 v0, 0 +; GFX12-NEXT: ; return to shader part epilog + %val = call float @llvm.maximum.f32(float -0.0, float 0.0) + ret float %val +} + define amdgpu_ps <2 x float> @test_fmaximum_v2f32(<2 x float> %a, <2 x float> %b) { -; GCN-LABEL: test_fmaximum_v2f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, v2 -; GCN-NEXT: v_maximum_f32 v1, v1, v3 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_v2f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v4, v0, v2 +; GFX9-NEXT: v_mov_b32_e32 v5, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v5, v4, vcc +; GFX9-NEXT: v_max_f32_e32 v2, v1, v3 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v2f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v2 +; GFX12-NEXT: v_maximum_f32 v1, v1, v3 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x float> @llvm.maximum.v2f32(<2 x float> %a, <2 x float> %b) ret <2 x float> %val } define amdgpu_ps <2 x float> @test_fmaximum_v2f32_ss(<2 x float> inreg %a, <2 x float> inreg %b) { -; GCN-LABEL: test_fmaximum_v2f32_ss: -; GCN: ; %bb.0: -; GCN-NEXT: s_maximum_f32 s0, s0, s2 -; GCN-NEXT: s_maximum_f32 s1, s1, s3 -; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3) -; GCN-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_v2f32_ss: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_max_f32_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-NEXT: v_max_f32_e32 v3, s1, v1 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s1, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v2f32_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_maximum_f32 s0, s0, s2 +; GFX12-NEXT: s_maximum_f32 s1, s1, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x float> @llvm.maximum.v2f32(<2 x float> %a, <2 x float> %b) ret <2 x float> %val } define amdgpu_ps <3 x float> @test_fmaximum_v3f32(<3 x float> %a, <3 x float> %b) { -; GCN-LABEL: test_fmaximum_v3f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, v3 -; GCN-NEXT: v_maximum_f32 v1, v1, v4 -; GCN-NEXT: v_maximum_f32 v2, v2, v5 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_v3f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v6, v0, v3 +; GFX9-NEXT: v_mov_b32_e32 v7, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v7, v6, vcc +; GFX9-NEXT: v_max_f32_e32 v3, v1, v4 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v7, v3, vcc +; GFX9-NEXT: v_max_f32_e32 v3, v2, v5 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v7, v3, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v3f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v3 +; GFX12-NEXT: v_maximum_f32 v1, v1, v4 +; GFX12-NEXT: v_maximum_f32 v2, v2, v5 +; GFX12-NEXT: ; return to shader part epilog %val = call <3 x float> @llvm.maximum.v3f32(<3 x float> %a, <3 x float> %b) ret <3 x float> %val } define amdgpu_ps <4 x float> @test_fmaximum_v4f32(<4 x float> %a, <4 x float> %b) { -; GCN-LABEL: test_fmaximum_v4f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, v4 -; GCN-NEXT: v_maximum_f32 v1, v1, v5 -; GCN-NEXT: v_maximum_f32 v2, v2, v6 -; GCN-NEXT: v_maximum_f32 v3, v3, v7 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_v4f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v8, v0, v4 +; GFX9-NEXT: v_mov_b32_e32 v9, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v9, v8, vcc +; GFX9-NEXT: v_max_f32_e32 v4, v1, v5 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v9, v4, vcc +; GFX9-NEXT: v_max_f32_e32 v4, v2, v6 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v9, v4, vcc +; GFX9-NEXT: v_max_f32_e32 v4, v3, v7 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v3, v7 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v9, v4, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v4f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v4 +; GFX12-NEXT: v_maximum_f32 v1, v1, v5 +; GFX12-NEXT: v_maximum_f32 v2, v2, v6 +; GFX12-NEXT: v_maximum_f32 v3, v3, v7 +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x float> @llvm.maximum.v4f32(<4 x float> %a, <4 x float> %b) ret <4 x float> %val } define amdgpu_ps <16 x float> @test_fmaximum_v16f32(<16 x float> %a, <16 x float> %b) { -; GCN-LABEL: test_fmaximum_v16f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f32 v0, v0, v16 -; GCN-NEXT: v_maximum_f32 v1, v1, v17 -; GCN-NEXT: v_maximum_f32 v2, v2, v18 -; GCN-NEXT: v_maximum_f32 v3, v3, v19 -; GCN-NEXT: v_maximum_f32 v4, v4, v20 -; GCN-NEXT: v_maximum_f32 v5, v5, v21 -; GCN-NEXT: v_maximum_f32 v6, v6, v22 -; GCN-NEXT: v_maximum_f32 v7, v7, v23 -; GCN-NEXT: v_maximum_f32 v8, v8, v24 -; GCN-NEXT: v_maximum_f32 v9, v9, v25 -; GCN-NEXT: v_maximum_f32 v10, v10, v26 -; GCN-NEXT: v_maximum_f32 v11, v11, v27 -; GCN-NEXT: v_maximum_f32 v12, v12, v28 -; GCN-NEXT: v_maximum_f32 v13, v13, v29 -; GCN-NEXT: v_maximum_f32 v14, v14, v30 -; GCN-NEXT: v_maximum_f32 v15, v15, v31 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_v16f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v32, v1, v17 +; GFX9-NEXT: v_mov_b32_e32 v33, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v17 +; GFX9-NEXT: v_max_f32_e32 v1, v0, v16 +; GFX9-NEXT: v_cmp_o_f32_e64 s[12:13], v0, v16 +; GFX9-NEXT: v_max_f32_e32 v17, v2, v18 +; GFX9-NEXT: v_cmp_o_f32_e64 s[0:1], v2, v18 +; GFX9-NEXT: v_max_f32_e32 v18, v3, v19 +; GFX9-NEXT: v_cmp_o_f32_e64 s[2:3], v3, v19 +; GFX9-NEXT: v_max_f32_e32 v19, v4, v20 +; GFX9-NEXT: v_cmp_o_f32_e64 s[4:5], v4, v20 +; GFX9-NEXT: v_max_f32_e32 v20, v5, v21 +; GFX9-NEXT: v_cmp_o_f32_e64 s[6:7], v5, v21 +; GFX9-NEXT: v_max_f32_e32 v21, v6, v22 +; GFX9-NEXT: v_cmp_o_f32_e64 s[8:9], v6, v22 +; GFX9-NEXT: v_max_f32_e32 v22, v7, v23 +; GFX9-NEXT: v_cmp_o_f32_e64 s[10:11], v7, v23 +; GFX9-NEXT: v_max_f32_e32 v23, v8, v24 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v33, v1, s[12:13] +; GFX9-NEXT: v_cndmask_b32_e32 v1, v33, v32, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v8, v24 +; GFX9-NEXT: v_max_f32_e32 v34, v9, v25 +; GFX9-NEXT: v_cndmask_b32_e32 v8, v33, v23, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v9, v25 +; GFX9-NEXT: v_max_f32_e32 v35, v10, v26 +; GFX9-NEXT: v_cndmask_b32_e32 v9, v33, v34, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v10, v26 +; GFX9-NEXT: v_max_f32_e32 v36, v11, v27 +; GFX9-NEXT: v_cndmask_b32_e32 v10, v33, v35, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v11, v27 +; GFX9-NEXT: v_max_f32_e32 v37, v12, v28 +; GFX9-NEXT: v_cndmask_b32_e32 v11, v33, v36, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v12, v28 +; GFX9-NEXT: v_max_f32_e32 v16, v13, v29 +; GFX9-NEXT: v_cndmask_b32_e32 v12, v33, v37, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v13, v29 +; GFX9-NEXT: v_cndmask_b32_e32 v13, v33, v16, vcc +; GFX9-NEXT: v_max_f32_e32 v16, v14, v30 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v14, v30 +; GFX9-NEXT: v_cndmask_b32_e32 v14, v33, v16, vcc +; GFX9-NEXT: v_max_f32_e32 v16, v15, v31 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v15, v31 +; GFX9-NEXT: v_cndmask_b32_e64 v2, v33, v17, s[0:1] +; GFX9-NEXT: v_cndmask_b32_e64 v3, v33, v18, s[2:3] +; GFX9-NEXT: v_cndmask_b32_e64 v4, v33, v19, s[4:5] +; GFX9-NEXT: v_cndmask_b32_e64 v5, v33, v20, s[6:7] +; GFX9-NEXT: v_cndmask_b32_e64 v6, v33, v21, s[8:9] +; GFX9-NEXT: v_cndmask_b32_e64 v7, v33, v22, s[10:11] +; GFX9-NEXT: v_cndmask_b32_e32 v15, v33, v16, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v16f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v16 +; GFX12-NEXT: v_maximum_f32 v1, v1, v17 +; GFX12-NEXT: v_maximum_f32 v2, v2, v18 +; GFX12-NEXT: v_maximum_f32 v3, v3, v19 +; GFX12-NEXT: v_maximum_f32 v4, v4, v20 +; GFX12-NEXT: v_maximum_f32 v5, v5, v21 +; GFX12-NEXT: v_maximum_f32 v6, v6, v22 +; GFX12-NEXT: v_maximum_f32 v7, v7, v23 +; GFX12-NEXT: v_maximum_f32 v8, v8, v24 +; GFX12-NEXT: v_maximum_f32 v9, v9, v25 +; GFX12-NEXT: v_maximum_f32 v10, v10, v26 +; GFX12-NEXT: v_maximum_f32 v11, v11, v27 +; GFX12-NEXT: v_maximum_f32 v12, v12, v28 +; GFX12-NEXT: v_maximum_f32 v13, v13, v29 +; GFX12-NEXT: v_maximum_f32 v14, v14, v30 +; GFX12-NEXT: v_maximum_f32 v15, v15, v31 +; GFX12-NEXT: ; return to shader part epilog %val = call <16 x float> @llvm.maximum.v16f32(<16 x float> %a, <16 x float> %b) ret <16 x float> %val } define amdgpu_ps half @test_fmaximum_f16_vv(half %a, half %b) { +; GFX9-LABEL: test_fmaximum_f16_vv: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f16_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; ; GFX12-SDAG-TRUE16-LABEL: test_fmaximum_f16_vv: ; GFX12-SDAG-TRUE16: ; %bb.0: ; GFX12-SDAG-TRUE16-NEXT: v_maximum_f16 v0.l, v0.l, v1.l @@ -136,35 +315,131 @@ define amdgpu_ps half @test_fmaximum_f16_vv(half %a, half %b) { } define amdgpu_ps half @test_fmaximum_f16_ss(half inreg %a, half inreg %b) { -; GCN-LABEL: test_fmaximum_f16_ss: -; GCN: ; %bb.0: -; GCN-NEXT: s_maximum_f16 s0, s0, s1 -; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3) -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fmaximum_f16_ss: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-NEXT: v_max_f16_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f16_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_maximum_f16 s0, s0, s1 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call half @llvm.maximum.f16(half %a, half %b) ret half %val } define amdgpu_ps <2 x half> @test_fmaximum_v2f16_vv(<2 x half> %a, <2 x half> %b) { -; GCN-LABEL: test_fmaximum_v2f16_vv: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_maximum_f16 v0, v0, v1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v2f16_vv: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_pk_max_f16 v2, v0, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v3, v2, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v2f16_vv: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_pk_max_f16 v2, v0, v1 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v1 +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v2f16_vv: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_maximum_f16 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b) ret <2 x half> %val } define amdgpu_ps <2 x half> @test_fmaximum_v2f16_ss(<2 x half> inreg %a, <2 x half> inreg %b) { -; GCN-LABEL: test_fmaximum_v2f16_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_maximum_f16 v0, s0, s1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v2f16_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16 +; GFX9-SDAG-NEXT: v_pk_max_f16 v1, s0, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v3 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v2f16_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-GISEL-NEXT: s_lshr_b32 s1, s1, 16 +; GFX9-GISEL-NEXT: s_lshr_b32 s2, s0, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s1 +; GFX9-GISEL-NEXT: v_pk_max_f16 v1, s0, v0 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s2, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s0, v0 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v2f16_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_maximum_f16 v0, s0, s1 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b) ret <2 x half> %val } define amdgpu_ps <3 x half> @test_fmaximum_v3f16_vv(<3 x half> %a, <3 x half> %b) { +; GFX9-SDAG-LABEL: test_fmaximum_v3f16_vv: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_pk_max_f16 v4, v1, v3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc +; GFX9-SDAG-NEXT: v_pk_max_f16 v3, v0, v2 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v3f16_vv: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_pk_max_f16 v4, v0, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v5, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v2 +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v5, v4, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_pk_max_f16 v4, v1, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v1, v3 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; ; GFX12-SDAG-LABEL: test_fmaximum_v3f16_vv: ; GFX12-SDAG: ; %bb.0: ; GFX12-SDAG-NEXT: v_pk_maximum_f16 v0, v0, v2 @@ -187,6 +462,49 @@ define amdgpu_ps <3 x half> @test_fmaximum_v3f16_vv(<3 x half> %a, <3 x half> %b } define amdgpu_ps <3 x half> @test_fmaximum_v3f16_ss(<3 x half> inreg %a, <3 x half> inreg %b) { +; GFX9-SDAG-LABEL: test_fmaximum_v3f16_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-SDAG-NEXT: v_pk_max_f16 v1, s1, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s2 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc +; GFX9-SDAG-NEXT: v_pk_max_f16 v3, s0, v3 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v4 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v3f16_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: s_lshr_b32 s5, s2, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16 +; GFX9-GISEL-NEXT: v_pk_max_f16 v1, s0, v0 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s5 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v2, v4, v3, vcc +; GFX9-GISEL-NEXT: v_pk_max_f16 v3, s1, v1 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s1, v1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; ; GFX12-SDAG-LABEL: test_fmaximum_v3f16_ss: ; GFX12-SDAG: ; %bb.0: ; GFX12-SDAG-NEXT: v_pk_maximum_f16 v0, s0, s2 @@ -206,97 +524,384 @@ define amdgpu_ps <3 x half> @test_fmaximum_v3f16_ss(<3 x half> inreg %a, <3 x ha } define amdgpu_ps <4 x half> @test_fmaximum_v4f16(<4 x half> %a, <4 x half> %b) { -; GCN-LABEL: test_fmaximum_v4f16: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_maximum_f16 v0, v0, v2 -; GCN-NEXT: v_pk_maximum_f16 v1, v1, v3 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v4f16: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_pk_max_f16 v4, v1, v3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v6, v5, v4, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_pk_max_f16 v3, v0, v2 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0 +; GFX9-SDAG-NEXT: v_perm_b32 v1, v1, v6, s0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v4f16: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_pk_max_f16 v4, v0, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v0, v2 +; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v4 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v6, v5, vcc +; GFX9-GISEL-NEXT: v_and_b32_e32 v2, 0xffff, v4 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v0, 16, v2 +; GFX9-GISEL-NEXT: v_pk_max_f16 v2, v1, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v1, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v6, v2, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v6, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v4f16: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_maximum_f16 v0, v0, v2 +; GFX12-NEXT: v_pk_maximum_f16 v1, v1, v3 +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x half> @llvm.maximum.v4f16(<4 x half> %a, <4 x half> %b) ret <4 x half> %val } define amdgpu_ps <4 x half> @test_fmaximum_v4f16_ss(<4 x half> inreg %a, <4 x half> inreg %b) { -; GCN-LABEL: test_fmaximum_v4f16_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_maximum_f16 v0, s0, s2 -; GCN-NEXT: v_pk_maximum_f16 v1, s1, s3 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v4f16_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-SDAG-NEXT: s_lshr_b32 s3, s3, 16 +; GFX9-SDAG-NEXT: v_pk_max_f16 v1, s1, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v3, v2, v1, vcc +; GFX9-SDAG-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s2 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc +; GFX9-SDAG-NEXT: v_pk_max_f16 v4, s0, v4 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, s1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v4, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v5 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-SDAG-NEXT: v_and_b32_e32 v2, 0xffff, v3 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v1, v1, 16, v2 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v4f16_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-GISEL-NEXT: s_lshr_b32 s2, s2, 16 +; GFX9-GISEL-NEXT: v_pk_max_f16 v1, s0, v0 +; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: s_lshr_b32 s2, s3, 16 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-GISEL-NEXT: s_lshr_b32 s0, s1, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, s2 +; GFX9-GISEL-NEXT: v_pk_max_f16 v2, s1, v1 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s1, v1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v4, v2, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v4, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v4f16_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_maximum_f16 v0, s0, s2 +; GFX12-NEXT: v_pk_maximum_f16 v1, s1, s3 +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x half> @llvm.maximum.v4f16(<4 x half> %a, <4 x half> %b) ret <4 x half> %val } define amdgpu_ps <2 x float> @test_fmaximum_f64_vv(double %a, double %b) { -; GCN-LABEL: test_fmaximum_f64_vv: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f64 v[0:1], v[0:1], v[2:3] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_f64_vv: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_max_f64 v[4:5], v[0:1], v[2:3] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[2:3] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v4, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_f64_vv: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_max_f64 v[4:5], v[0:1], v[2:3] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[2:3] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v5, vcc +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f64_vv: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f64 v[0:1], v[0:1], v[2:3] +; GFX12-NEXT: ; return to shader part epilog %val = call double @llvm.maximum.f64(double %a, double %b) %ret = bitcast double %val to <2 x float> ret <2 x float> %ret } define amdgpu_ps <2 x float> @test_fmaximum_f64_ss(double inreg %a, double inreg %b) { -; GCN-LABEL: test_fmaximum_f64_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f64 v[0:1], s[0:1], s[2:3] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_f64_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-SDAG-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_f64_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-GISEL-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f64_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f64 v[0:1], s[0:1], s[2:3] +; GFX12-NEXT: ; return to shader part epilog %val = call double @llvm.maximum.f64(double %a, double %b) %ret = bitcast double %val to <2 x float> ret <2 x float> %ret } define amdgpu_ps <4 x float> @test_fmaximum_v2f64_ss(<2 x double> inreg %a, <2 x double> inreg %b) { -; GCN-LABEL: test_fmaximum_v2f64_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f64 v[0:1], s[0:1], s[4:5] -; GCN-NEXT: v_maximum_f64 v[2:3], s[2:3], s[6:7] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v2f64_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s5 +; GFX9-SDAG-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s6 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s7 +; GFX9-SDAG-NEXT: v_max_f64 v[4:5], s[2:3], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v6, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v6, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v6, s[0:1] +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v2f64_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s5 +; GFX9-GISEL-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s6 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s7 +; GFX9-GISEL-NEXT: v_max_f64 v[4:5], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v6, v3, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v6, v5, s[0:1] +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v2f64_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f64 v[0:1], s[0:1], s[4:5] +; GFX12-NEXT: v_maximum_f64 v[2:3], s[2:3], s[6:7] +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x double> @llvm.maximum.v2f64(<2 x double> %a, <2 x double> %b) %ret = bitcast <2 x double> %val to <4 x float> ret <4 x float> %ret } define amdgpu_ps <8 x float> @test_fmaximum_v4f64(<4 x double> %a, <4 x double> %b) { -; GCN-LABEL: test_fmaximum_v4f64: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f64 v[0:1], v[0:1], v[8:9] -; GCN-NEXT: v_maximum_f64 v[2:3], v[2:3], v[10:11] -; GCN-NEXT: v_maximum_f64 v[4:5], v[4:5], v[12:13] -; GCN-NEXT: v_maximum_f64 v[6:7], v[6:7], v[14:15] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v4f64: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_max_f64 v[16:17], v[0:1], v[8:9] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[8:9] +; GFX9-SDAG-NEXT: v_max_f64 v[8:9], v[2:3], v[10:11] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], v[2:3], v[10:11] +; GFX9-SDAG-NEXT: v_max_f64 v[10:11], v[4:5], v[12:13] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], v[4:5], v[12:13] +; GFX9-SDAG-NEXT: v_max_f64 v[12:13], v[6:7], v[14:15] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], v[6:7], v[14:15] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v7, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v16, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v17, v7, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v8, 0, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v9, v7, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v10, 0, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v11, v7, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v12, 0, s[4:5] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v13, v7, s[4:5] +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v4f64: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_max_f64 v[16:17], v[0:1], v[8:9] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[8:9] +; GFX9-GISEL-NEXT: v_max_f64 v[8:9], v[2:3], v[10:11] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], v[2:3], v[10:11] +; GFX9-GISEL-NEXT: v_max_f64 v[10:11], v[4:5], v[12:13] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], v[4:5], v[12:13] +; GFX9-GISEL-NEXT: v_max_f64 v[12:13], v[6:7], v[14:15] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], v[6:7], v[14:15] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v18, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v16, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v18, v17, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v8, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v9, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v10, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v18, v11, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v12, s[4:5] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v18, v13, s[4:5] +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v4f64: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f64 v[0:1], v[0:1], v[8:9] +; GFX12-NEXT: v_maximum_f64 v[2:3], v[2:3], v[10:11] +; GFX12-NEXT: v_maximum_f64 v[4:5], v[4:5], v[12:13] +; GFX12-NEXT: v_maximum_f64 v[6:7], v[6:7], v[14:15] +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x double> @llvm.maximum.v4f64(<4 x double> %a, <4 x double> %b) %ret = bitcast <4 x double> %val to <8 x float> ret <8 x float> %ret } define amdgpu_ps <8 x float> @test_fmaximum_v4f64_ss(<4 x double> inreg %a, <4 x double> inreg %b) { -; GCN-LABEL: test_fmaximum_v4f64_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_maximum_f64 v[0:1], s[0:1], s[8:9] -; GCN-NEXT: v_maximum_f64 v[2:3], s[2:3], s[10:11] -; GCN-NEXT: v_maximum_f64 v[4:5], s[4:5], s[12:13] -; GCN-NEXT: v_maximum_f64 v[6:7], s[6:7], s[14:15] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fmaximum_v4f64_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s8 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s9 +; GFX9-SDAG-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v10, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s10 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s11 +; GFX9-SDAG-NEXT: v_max_f64 v[4:5], s[2:3], v[1:2] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[1:2] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s12 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s13 +; GFX9-SDAG-NEXT: v_max_f64 v[6:7], s[4:5], v[1:2] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], s[4:5], v[1:2] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s14 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s15 +; GFX9-SDAG-NEXT: v_max_f64 v[8:9], s[6:7], v[1:2] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], s[6:7], v[1:2] +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v10, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v10, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v6, 0, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v7, v10, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v8, 0, s[4:5] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v9, v10, s[4:5] +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fmaximum_v4f64_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s8 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s9 +; GFX9-GISEL-NEXT: v_max_f64 v[2:3], s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s10 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s11 +; GFX9-GISEL-NEXT: v_max_f64 v[4:5], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s12 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s13 +; GFX9-GISEL-NEXT: v_max_f64 v[6:7], s[4:5], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], s[4:5], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s14 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s15 +; GFX9-GISEL-NEXT: v_max_f64 v[8:9], s[6:7], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], s[6:7], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v10, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v3, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v10, v5, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v6, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v10, v7, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v8, s[4:5] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v10, v9, s[4:5] +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_v4f64_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f64 v[0:1], s[0:1], s[8:9] +; GFX12-NEXT: v_maximum_f64 v[2:3], s[2:3], s[10:11] +; GFX12-NEXT: v_maximum_f64 v[4:5], s[4:5], s[12:13] +; GFX12-NEXT: v_maximum_f64 v[6:7], s[6:7], s[14:15] +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x double> @llvm.maximum.v4f64(<4 x double> %a, <4 x double> %b) %ret = bitcast <4 x double> %val to <8 x float> ret <8 x float> %ret } define amdgpu_kernel void @fmaximumi_f32_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) { -; GCN-LABEL: fmaximumi_f32_move_to_valu: -; GCN: ; %bb.0: -; GCN-NEXT: s_clause 0x1 -; GCN-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GCN-NEXT: s_load_b64 s[4:5], s[4:5], 0x34 -; GCN-NEXT: v_mov_b32_e32 v0, 0 -; GCN-NEXT: s_wait_kmcnt 0x0 -; GCN-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS -; GCN-NEXT: s_wait_loadcnt 0x0 -; GCN-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS -; GCN-NEXT: s_wait_loadcnt 0x0 -; GCN-NEXT: v_maximum_f32 v1, v1, v2 -; GCN-NEXT: global_store_b32 v0, v1, s[0:1] -; GCN-NEXT: s_endpgm +; GFX9-LABEL: fmaximumi_f32_move_to_valu: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: global_load_dword v1, v0, s[2:3] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: global_load_dword v2, v0, s[6:7] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_max_f32_e32 v4, v1, v2 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX9-NEXT: global_store_dword v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +; +; GFX12-LABEL: fmaximumi_f32_move_to_valu: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_clause 0x1 +; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX12-NEXT: s_load_b64 s[4:5], s[4:5], 0x34 +; GFX12-NEXT: v_mov_b32_e32 v0, 0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS +; GFX12-NEXT: s_wait_loadcnt 0x0 +; GFX12-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS +; GFX12-NEXT: s_wait_loadcnt 0x0 +; GFX12-NEXT: v_maximum_f32 v1, v1, v2 +; GFX12-NEXT: global_store_b32 v0, v1, s[0:1] +; GFX12-NEXT: s_endpgm %a = load volatile float, ptr addrspace(1) %aptr, align 4 %b = load volatile float, ptr addrspace(1) %bptr, align 4 %v = call float @llvm.maximum.f32(float %a, float %b) @@ -305,6 +910,23 @@ define amdgpu_kernel void @fmaximumi_f32_move_to_valu(ptr addrspace(1) %out, ptr } define amdgpu_kernel void @fmaximum_f16_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) { +; GFX9-LABEL: fmaximum_f16_move_to_valu: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: global_load_ushort v1, v0, s[2:3] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_max_f16_e32 v4, v1, v2 +; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v1, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX9-NEXT: global_store_short v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +; ; GFX12-SDAG-TRUE16-LABEL: fmaximum_f16_move_to_valu: ; GFX12-SDAG-TRUE16: ; %bb.0: ; GFX12-SDAG-TRUE16-NEXT: s_clause 0x1 @@ -371,6 +993,40 @@ define amdgpu_kernel void @fmaximum_f16_move_to_valu(ptr addrspace(1) %out, ptr ret void } +define amdgpu_ps float @test_fmaximum_f32_ieee_on(float %a, float %b) #0 { +; GFX9-LABEL: test_fmaximum_f32_ieee_on: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f32_ieee_on: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog + %val = call float @llvm.maximum.f32(float %a, float %b) + ret float %val +} + +define amdgpu_ps float @test_fmaximum_f32_ieee_off(float %a, float %b) #1 { +; GFX9-LABEL: test_fmaximum_f32_ieee_off: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_max_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fmaximum_f32_ieee_off: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_maximum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog + %val = call float @llvm.maximum.f32(float %a, float %b) + ret float %val +} + declare float @llvm.maximum.f32(float, float) declare <2 x float> @llvm.maximum.v2f32(<2 x float>, <2 x float>) declare <3 x float> @llvm.maximum.v3f32(<3 x float>, <3 x float>) @@ -383,3 +1039,6 @@ declare <4 x half> @llvm.maximum.v4f16(<4 x half>, <4 x half>) declare double @llvm.maximum.f64(double, double) declare <2 x double> @llvm.maximum.v2f64(<2 x double>, <2 x double>) declare <4 x double> @llvm.maximum.v4f64(<4 x double>, <4 x double>) + +attributes #0 = { nounwind "amdgpu-ieee"="true" } +attributes #1 = { nounwind "amdgpu-ieee"="false" } diff --git a/llvm/test/CodeGen/AMDGPU/fminimum.ll b/llvm/test/CodeGen/AMDGPU/fminimum.ll index b25120f..474ac7c 100644 --- a/llvm/test/CodeGen/AMDGPU/fminimum.ll +++ b/llvm/test/CodeGen/AMDGPU/fminimum.ll @@ -1,117 +1,296 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-TRUE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-SDAG,GFX12-SDAG-FAKE16 %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-TRUE16 %s -; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX12-GISEL,GFX12-GISEL-FAKE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-SDAG %s +; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s +; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s define amdgpu_ps float @test_fminimum_f32_vv(float %a, float %b) { -; GCN-LABEL: test_fminimum_f32_vv: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, v1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_f32_vv: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f32_vv: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.minimum.f32(float %a, float %b) ret float %val } define amdgpu_ps float @test_fminimum_f32_ss(float inreg %a, float inreg %b) { -; GCN-LABEL: test_fminimum_f32_ss: -; GCN: ; %bb.0: -; GCN-NEXT: s_minimum_f32 s0, s0, s1 -; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3) -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_f32_ss: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-NEXT: v_min_f32_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f32_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_minimum_f32 s0, s0, s1 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.minimum.f32(float %a, float %b) ret float %val } define amdgpu_ps float @test_fminimum_f32_vs(float %a, float inreg %b) { -; GCN-LABEL: test_fminimum_f32_vs: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, s0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_f32_vs: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f32_vs: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call float @llvm.minimum.f32(float %a, float %b) ret float %val } define amdgpu_ps float @test_fminimum_nnan_f32(float %a, float %b) { -; GCN-LABEL: test_fminimum_nnan_f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, v1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_nnan_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v0, v0, v1 +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_nnan_f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog %val = call nnan float @llvm.minimum.f32(float %a, float %b) ret float %val } +define amdgpu_ps float @test_fminimum_nsz_f32(float %a, float %b) { +; GFX9-LABEL: test_fminimum_nsz_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_nsz_f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog + %val = call nsz float @llvm.minimum.f32(float %a, float %b) + ret float %val +} + +define amdgpu_ps float @test_fminimum_signed_zero_f32() { +; GFX9-LABEL: test_fminimum_signed_zero_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_bfrev_b32_e32 v0, 1 +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_signed_zero_f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_bfrev_b32_e32 v0, 1 +; GFX12-NEXT: ; return to shader part epilog + %val = call float @llvm.minimum.f32(float -0.0, float 0.0) + ret float %val +} + define amdgpu_ps <2 x float> @test_fminimum_v2f32(<2 x float> %a, <2 x float> %b) { -; GCN-LABEL: test_fminimum_v2f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, v2 -; GCN-NEXT: v_minimum_f32 v1, v1, v3 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_v2f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v4, v0, v2 +; GFX9-NEXT: v_mov_b32_e32 v5, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v5, v4, vcc +; GFX9-NEXT: v_min_f32_e32 v2, v1, v3 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v2f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v2 +; GFX12-NEXT: v_minimum_f32 v1, v1, v3 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x float> @llvm.minimum.v2f32(<2 x float> %a, <2 x float> %b) ret <2 x float> %val } define amdgpu_ps <2 x float> @test_fminimum_v2f32_ss(<2 x float> inreg %a, <2 x float> inreg %b) { -; GCN-LABEL: test_fminimum_v2f32_ss: -; GCN: ; %bb.0: -; GCN-NEXT: s_minimum_f32 s0, s0, s2 -; GCN-NEXT: s_minimum_f32 s1, s1, s3 -; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3) -; GCN-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_v2f32_ss: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_min_f32_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-NEXT: v_min_f32_e32 v3, s1, v1 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, s1, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v2f32_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_minimum_f32 s0, s0, s2 +; GFX12-NEXT: s_minimum_f32 s1, s1, s3 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x float> @llvm.minimum.v2f32(<2 x float> %a, <2 x float> %b) ret <2 x float> %val } define amdgpu_ps <3 x float> @test_fminimum_v3f32(<3 x float> %a, <3 x float> %b) { -; GCN-LABEL: test_fminimum_v3f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, v3 -; GCN-NEXT: v_minimum_f32 v1, v1, v4 -; GCN-NEXT: v_minimum_f32 v2, v2, v5 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_v3f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v6, v0, v3 +; GFX9-NEXT: v_mov_b32_e32 v7, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v7, v6, vcc +; GFX9-NEXT: v_min_f32_e32 v3, v1, v4 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v7, v3, vcc +; GFX9-NEXT: v_min_f32_e32 v3, v2, v5 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v7, v3, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v3f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v3 +; GFX12-NEXT: v_minimum_f32 v1, v1, v4 +; GFX12-NEXT: v_minimum_f32 v2, v2, v5 +; GFX12-NEXT: ; return to shader part epilog %val = call <3 x float> @llvm.minimum.v3f32(<3 x float> %a, <3 x float> %b) ret <3 x float> %val } define amdgpu_ps <4 x float> @test_fminimum_v4f32(<4 x float> %a, <4 x float> %b) { -; GCN-LABEL: test_fminimum_v4f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, v4 -; GCN-NEXT: v_minimum_f32 v1, v1, v5 -; GCN-NEXT: v_minimum_f32 v2, v2, v6 -; GCN-NEXT: v_minimum_f32 v3, v3, v7 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_v4f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v8, v0, v4 +; GFX9-NEXT: v_mov_b32_e32 v9, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v9, v8, vcc +; GFX9-NEXT: v_min_f32_e32 v4, v1, v5 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v9, v4, vcc +; GFX9-NEXT: v_min_f32_e32 v4, v2, v6 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v2, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v9, v4, vcc +; GFX9-NEXT: v_min_f32_e32 v4, v3, v7 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v3, v7 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v9, v4, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v4f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v4 +; GFX12-NEXT: v_minimum_f32 v1, v1, v5 +; GFX12-NEXT: v_minimum_f32 v2, v2, v6 +; GFX12-NEXT: v_minimum_f32 v3, v3, v7 +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x float> @llvm.minimum.v4f32(<4 x float> %a, <4 x float> %b) ret <4 x float> %val } define amdgpu_ps <16 x float> @test_fminimum_v16f32(<16 x float> %a, <16 x float> %b) { -; GCN-LABEL: test_fminimum_v16f32: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f32 v0, v0, v16 -; GCN-NEXT: v_minimum_f32 v1, v1, v17 -; GCN-NEXT: v_minimum_f32 v2, v2, v18 -; GCN-NEXT: v_minimum_f32 v3, v3, v19 -; GCN-NEXT: v_minimum_f32 v4, v4, v20 -; GCN-NEXT: v_minimum_f32 v5, v5, v21 -; GCN-NEXT: v_minimum_f32 v6, v6, v22 -; GCN-NEXT: v_minimum_f32 v7, v7, v23 -; GCN-NEXT: v_minimum_f32 v8, v8, v24 -; GCN-NEXT: v_minimum_f32 v9, v9, v25 -; GCN-NEXT: v_minimum_f32 v10, v10, v26 -; GCN-NEXT: v_minimum_f32 v11, v11, v27 -; GCN-NEXT: v_minimum_f32 v12, v12, v28 -; GCN-NEXT: v_minimum_f32 v13, v13, v29 -; GCN-NEXT: v_minimum_f32 v14, v14, v30 -; GCN-NEXT: v_minimum_f32 v15, v15, v31 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_v16f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v32, v1, v17 +; GFX9-NEXT: v_mov_b32_e32 v33, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v17 +; GFX9-NEXT: v_min_f32_e32 v1, v0, v16 +; GFX9-NEXT: v_cmp_o_f32_e64 s[12:13], v0, v16 +; GFX9-NEXT: v_min_f32_e32 v17, v2, v18 +; GFX9-NEXT: v_cmp_o_f32_e64 s[0:1], v2, v18 +; GFX9-NEXT: v_min_f32_e32 v18, v3, v19 +; GFX9-NEXT: v_cmp_o_f32_e64 s[2:3], v3, v19 +; GFX9-NEXT: v_min_f32_e32 v19, v4, v20 +; GFX9-NEXT: v_cmp_o_f32_e64 s[4:5], v4, v20 +; GFX9-NEXT: v_min_f32_e32 v20, v5, v21 +; GFX9-NEXT: v_cmp_o_f32_e64 s[6:7], v5, v21 +; GFX9-NEXT: v_min_f32_e32 v21, v6, v22 +; GFX9-NEXT: v_cmp_o_f32_e64 s[8:9], v6, v22 +; GFX9-NEXT: v_min_f32_e32 v22, v7, v23 +; GFX9-NEXT: v_cmp_o_f32_e64 s[10:11], v7, v23 +; GFX9-NEXT: v_min_f32_e32 v23, v8, v24 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v33, v1, s[12:13] +; GFX9-NEXT: v_cndmask_b32_e32 v1, v33, v32, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v8, v24 +; GFX9-NEXT: v_min_f32_e32 v34, v9, v25 +; GFX9-NEXT: v_cndmask_b32_e32 v8, v33, v23, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v9, v25 +; GFX9-NEXT: v_min_f32_e32 v35, v10, v26 +; GFX9-NEXT: v_cndmask_b32_e32 v9, v33, v34, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v10, v26 +; GFX9-NEXT: v_min_f32_e32 v36, v11, v27 +; GFX9-NEXT: v_cndmask_b32_e32 v10, v33, v35, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v11, v27 +; GFX9-NEXT: v_min_f32_e32 v37, v12, v28 +; GFX9-NEXT: v_cndmask_b32_e32 v11, v33, v36, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v12, v28 +; GFX9-NEXT: v_min_f32_e32 v16, v13, v29 +; GFX9-NEXT: v_cndmask_b32_e32 v12, v33, v37, vcc +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v13, v29 +; GFX9-NEXT: v_cndmask_b32_e32 v13, v33, v16, vcc +; GFX9-NEXT: v_min_f32_e32 v16, v14, v30 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v14, v30 +; GFX9-NEXT: v_cndmask_b32_e32 v14, v33, v16, vcc +; GFX9-NEXT: v_min_f32_e32 v16, v15, v31 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v15, v31 +; GFX9-NEXT: v_cndmask_b32_e64 v2, v33, v17, s[0:1] +; GFX9-NEXT: v_cndmask_b32_e64 v3, v33, v18, s[2:3] +; GFX9-NEXT: v_cndmask_b32_e64 v4, v33, v19, s[4:5] +; GFX9-NEXT: v_cndmask_b32_e64 v5, v33, v20, s[6:7] +; GFX9-NEXT: v_cndmask_b32_e64 v6, v33, v21, s[8:9] +; GFX9-NEXT: v_cndmask_b32_e64 v7, v33, v22, s[10:11] +; GFX9-NEXT: v_cndmask_b32_e32 v15, v33, v16, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v16f32: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v16 +; GFX12-NEXT: v_minimum_f32 v1, v1, v17 +; GFX12-NEXT: v_minimum_f32 v2, v2, v18 +; GFX12-NEXT: v_minimum_f32 v3, v3, v19 +; GFX12-NEXT: v_minimum_f32 v4, v4, v20 +; GFX12-NEXT: v_minimum_f32 v5, v5, v21 +; GFX12-NEXT: v_minimum_f32 v6, v6, v22 +; GFX12-NEXT: v_minimum_f32 v7, v7, v23 +; GFX12-NEXT: v_minimum_f32 v8, v8, v24 +; GFX12-NEXT: v_minimum_f32 v9, v9, v25 +; GFX12-NEXT: v_minimum_f32 v10, v10, v26 +; GFX12-NEXT: v_minimum_f32 v11, v11, v27 +; GFX12-NEXT: v_minimum_f32 v12, v12, v28 +; GFX12-NEXT: v_minimum_f32 v13, v13, v29 +; GFX12-NEXT: v_minimum_f32 v14, v14, v30 +; GFX12-NEXT: v_minimum_f32 v15, v15, v31 +; GFX12-NEXT: ; return to shader part epilog %val = call <16 x float> @llvm.minimum.v16f32(<16 x float> %a, <16 x float> %b) ret <16 x float> %val } define amdgpu_ps half @test_fminimum_f16_vv(half %a, half %b) { +; GFX9-LABEL: test_fminimum_f16_vv: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f16_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; ; GFX12-SDAG-TRUE16-LABEL: test_fminimum_f16_vv: ; GFX12-SDAG-TRUE16: ; %bb.0: ; GFX12-SDAG-TRUE16-NEXT: v_minimum_f16 v0.l, v0.l, v1.l @@ -136,35 +315,131 @@ define amdgpu_ps half @test_fminimum_f16_vv(half %a, half %b) { } define amdgpu_ps half @test_fminimum_f16_ss(half inreg %a, half inreg %b) { -; GCN-LABEL: test_fminimum_f16_ss: -; GCN: ; %bb.0: -; GCN-NEXT: s_minimum_f16 s0, s0, s1 -; GCN-NEXT: s_delay_alu instid0(SALU_CYCLE_3) -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: ; return to shader part epilog +; GFX9-LABEL: test_fminimum_f16_ss: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-NEXT: v_min_f16_e32 v1, s0, v0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f16_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_minimum_f16 s0, s0, s1 +; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX12-NEXT: v_mov_b32_e32 v0, s0 +; GFX12-NEXT: ; return to shader part epilog %val = call half @llvm.minimum.f16(half %a, half %b) ret half %val } define amdgpu_ps <2 x half> @test_fminimum_v2f16_vv(<2 x half> %a, <2 x half> %b) { -; GCN-LABEL: test_fminimum_v2f16_vv: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_minimum_f16 v0, v0, v1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v2f16_vv: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_pk_min_f16 v2, v0, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v3, v2, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v2f16_vv: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_pk_min_f16 v2, v0, v1 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v1 +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v3, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v2f16_vv: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_minimum_f16 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x half> @llvm.minimum.v2f16(<2 x half> %a, <2 x half> %b) ret <2 x half> %val } define amdgpu_ps <2 x half> @test_fminimum_v2f16_ss(<2 x half> inreg %a, <2 x half> inreg %b) { -; GCN-LABEL: test_fminimum_v2f16_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_minimum_f16 v0, s0, s1 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v2f16_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s1 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16 +; GFX9-SDAG-NEXT: v_pk_min_f16 v1, s0, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v3 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v2f16_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s1 +; GFX9-GISEL-NEXT: s_lshr_b32 s1, s1, 16 +; GFX9-GISEL-NEXT: s_lshr_b32 s2, s0, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s1 +; GFX9-GISEL-NEXT: v_pk_min_f16 v1, s0, v0 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s2, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s0, v0 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v2f16_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_minimum_f16 v0, s0, s1 +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x half> @llvm.minimum.v2f16(<2 x half> %a, <2 x half> %b) ret <2 x half> %val } define amdgpu_ps <3 x half> @test_fminimum_v3f16_vv(<3 x half> %a, <3 x half> %b) { +; GFX9-SDAG-LABEL: test_fminimum_v3f16_vv: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_pk_min_f16 v4, v1, v3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc +; GFX9-SDAG-NEXT: v_pk_min_f16 v3, v0, v2 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v3f16_vv: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_pk_min_f16 v4, v0, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v5, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v0, v2 +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, v5, v4, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_pk_min_f16 v4, v1, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v1, v3 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; ; GFX12-SDAG-LABEL: test_fminimum_v3f16_vv: ; GFX12-SDAG: ; %bb.0: ; GFX12-SDAG-NEXT: v_pk_minimum_f16 v0, v0, v2 @@ -187,6 +462,49 @@ define amdgpu_ps <3 x half> @test_fminimum_v3f16_vv(<3 x half> %a, <3 x half> %b } define amdgpu_ps <3 x half> @test_fminimum_v3f16_ss(<3 x half> inreg %a, <3 x half> inreg %b) { +; GFX9-SDAG-LABEL: test_fminimum_v3f16_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-SDAG-NEXT: v_pk_min_f16 v1, s1, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v3, s2 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc +; GFX9-SDAG-NEXT: v_pk_min_f16 v3, s0, v3 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v4 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v3f16_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: s_lshr_b32 s5, s2, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16 +; GFX9-GISEL-NEXT: v_pk_min_f16 v1, s0, v0 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s5 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v2, v4, v3, vcc +; GFX9-GISEL-NEXT: v_pk_min_f16 v3, s1, v1 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s1, v1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; ; GFX12-SDAG-LABEL: test_fminimum_v3f16_ss: ; GFX12-SDAG: ; %bb.0: ; GFX12-SDAG-NEXT: v_pk_minimum_f16 v0, s0, s2 @@ -206,97 +524,384 @@ define amdgpu_ps <3 x half> @test_fminimum_v3f16_ss(<3 x half> inreg %a, <3 x ha } define amdgpu_ps <4 x half> @test_fminimum_v4f16(<4 x half> %a, <4 x half> %b) { -; GCN-LABEL: test_fminimum_v4f16: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_minimum_f16 v0, v0, v2 -; GCN-NEXT: v_pk_minimum_f16 v1, v1, v3 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v4f16: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_pk_min_f16 v4, v1, v3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v1, v3 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v6, v5, v4, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v1, v5, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_pk_min_f16 v3, v0, v2 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, v0, v2 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v4, v5, v3, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v0, v5, v3, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: s_mov_b32 s0, 0x5040100 +; GFX9-SDAG-NEXT: v_perm_b32 v0, v0, v4, s0 +; GFX9-SDAG-NEXT: v_perm_b32 v1, v1, v6, s0 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v4f16: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_pk_min_f16 v4, v0, v2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, v0, v2 +; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v4 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v4, v6, v4, vcc +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v0, v2 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v6, v5, vcc +; GFX9-GISEL-NEXT: v_and_b32_e32 v2, 0xffff, v4 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v0, 16, v2 +; GFX9-GISEL-NEXT: v_pk_min_f16 v2, v1, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], v1, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_sdwa vcc, v1, v3 src0_sel:WORD_1 src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v6, v2, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v6, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v4f16: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_minimum_f16 v0, v0, v2 +; GFX12-NEXT: v_pk_minimum_f16 v1, v1, v3 +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x half> @llvm.minimum.v4f16(<4 x half> %a, <4 x half> %b) ret <4 x half> %val } define amdgpu_ps <4 x half> @test_fminimum_v4f16_ss(<4 x half> inreg %a, <4 x half> inreg %b) { -; GCN-LABEL: test_fminimum_v4f16_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_pk_minimum_f16 v0, s0, s2 -; GCN-NEXT: v_pk_minimum_f16 v1, s1, s3 -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v4f16_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-SDAG-NEXT: s_lshr_b32 s3, s3, 16 +; GFX9-SDAG-NEXT: v_pk_min_f16 v1, s1, v1 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x7e00 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s1, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s3 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v3, v2, v1, vcc +; GFX9-SDAG-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s1, v0 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v4, s2 +; GFX9-SDAG-NEXT: s_lshr_b32 s1, s2, 16 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v1, vcc +; GFX9-SDAG-NEXT: v_pk_min_f16 v4, s0, v4 +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-SDAG-NEXT: s_lshr_b32 s0, s0, 16 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v5, s1 +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v0, v2, v4, vcc +; GFX9-SDAG-NEXT: v_cmp_o_f16_e32 vcc, s0, v5 +; GFX9-SDAG-NEXT: v_cndmask_b32_sdwa v2, v2, v4, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-SDAG-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v0, v2, 16, v0 +; GFX9-SDAG-NEXT: v_and_b32_e32 v2, 0xffff, v3 +; GFX9-SDAG-NEXT: v_lshl_or_b32 v1, v1, 16, v2 +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v4f16_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-GISEL-NEXT: s_lshr_b32 s2, s2, 16 +; GFX9-GISEL-NEXT: v_pk_min_f16 v1, s0, v0 +; GFX9-GISEL-NEXT: s_lshr_b32 s4, s0, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v4, 0x7e00 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v0 +; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s4, v2 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc +; GFX9-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX9-GISEL-NEXT: s_lshr_b32 s2, s3, 16 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v1, 16, v0 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-GISEL-NEXT: s_lshr_b32 s0, s1, 16 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v3, s2 +; GFX9-GISEL-NEXT: v_pk_min_f16 v2, s1, v1 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e32 vcc, s0, v3 +; GFX9-GISEL-NEXT: v_cmp_o_f16_e64 s[0:1], s1, v1 +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v1, v4, v2, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_sdwa v2, v4, v2, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, v2, 16, v1 +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v4f16_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_pk_minimum_f16 v0, s0, s2 +; GFX12-NEXT: v_pk_minimum_f16 v1, s1, s3 +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x half> @llvm.minimum.v4f16(<4 x half> %a, <4 x half> %b) ret <4 x half> %val } define amdgpu_ps <2 x float> @test_fminimum_f64_vv(double %a, double %b) { -; GCN-LABEL: test_fminimum_f64_vv: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f64 v[0:1], v[0:1], v[2:3] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_f64_vv: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_min_f64 v[4:5], v[0:1], v[2:3] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[2:3] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v4, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_f64_vv: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_min_f64 v[4:5], v[0:1], v[2:3] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[2:3] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v5, vcc +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f64_vv: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f64 v[0:1], v[0:1], v[2:3] +; GFX12-NEXT: ; return to shader part epilog %val = call double @llvm.minimum.f64(double %a, double %b) %ret = bitcast double %val to <2 x float> ret <2 x float> %ret } define amdgpu_ps <2 x float> @test_fminimum_f64_ss(double inreg %a, double inreg %b) { -; GCN-LABEL: test_fminimum_f64_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f64 v[0:1], s[0:1], s[2:3] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_f64_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-SDAG-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_f64_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 +; GFX9-GISEL-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f64_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f64 v[0:1], s[0:1], s[2:3] +; GFX12-NEXT: ; return to shader part epilog %val = call double @llvm.minimum.f64(double %a, double %b) %ret = bitcast double %val to <2 x float> ret <2 x float> %ret } define amdgpu_ps <4 x float> @test_fminimum_v2f64_ss(<2 x double> inreg %a, <2 x double> inreg %b) { -; GCN-LABEL: test_fminimum_v2f64_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f64 v[0:1], s[0:1], s[4:5] -; GCN-NEXT: v_minimum_f64 v[2:3], s[2:3], s[6:7] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v2f64_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s5 +; GFX9-SDAG-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s6 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s7 +; GFX9-SDAG-NEXT: v_min_f64 v[4:5], s[2:3], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v6, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v6, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v6, s[0:1] +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v2f64_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s5 +; GFX9-GISEL-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s6 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s7 +; GFX9-GISEL-NEXT: v_min_f64 v[4:5], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v6, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v6, v3, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v6, v5, s[0:1] +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v2f64_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f64 v[0:1], s[0:1], s[4:5] +; GFX12-NEXT: v_minimum_f64 v[2:3], s[2:3], s[6:7] +; GFX12-NEXT: ; return to shader part epilog %val = call <2 x double> @llvm.minimum.v2f64(<2 x double> %a, <2 x double> %b) %ret = bitcast <2 x double> %val to <4 x float> ret <4 x float> %ret } define amdgpu_ps <8 x float> @test_fminimum_v4f64(<4 x double> %a, <4 x double> %b) { -; GCN-LABEL: test_fminimum_v4f64: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f64 v[0:1], v[0:1], v[8:9] -; GCN-NEXT: v_minimum_f64 v[2:3], v[2:3], v[10:11] -; GCN-NEXT: v_minimum_f64 v[4:5], v[4:5], v[12:13] -; GCN-NEXT: v_minimum_f64 v[6:7], v[6:7], v[14:15] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v4f64: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_min_f64 v[16:17], v[0:1], v[8:9] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, v[0:1], v[8:9] +; GFX9-SDAG-NEXT: v_min_f64 v[8:9], v[2:3], v[10:11] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], v[2:3], v[10:11] +; GFX9-SDAG-NEXT: v_min_f64 v[10:11], v[4:5], v[12:13] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], v[4:5], v[12:13] +; GFX9-SDAG-NEXT: v_min_f64 v[12:13], v[6:7], v[14:15] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], v[6:7], v[14:15] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v7, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v16, 0, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v17, v7, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v8, 0, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v9, v7, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v10, 0, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v11, v7, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v12, 0, s[4:5] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v13, v7, s[4:5] +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v4f64: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_min_f64 v[16:17], v[0:1], v[8:9] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, v[0:1], v[8:9] +; GFX9-GISEL-NEXT: v_min_f64 v[8:9], v[2:3], v[10:11] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], v[2:3], v[10:11] +; GFX9-GISEL-NEXT: v_min_f64 v[10:11], v[4:5], v[12:13] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], v[4:5], v[12:13] +; GFX9-GISEL-NEXT: v_min_f64 v[12:13], v[6:7], v[14:15] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], v[6:7], v[14:15] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v18, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v16, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v18, v17, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v8, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v9, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v10, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v18, v11, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v12, s[4:5] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v18, v13, s[4:5] +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v4f64: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f64 v[0:1], v[0:1], v[8:9] +; GFX12-NEXT: v_minimum_f64 v[2:3], v[2:3], v[10:11] +; GFX12-NEXT: v_minimum_f64 v[4:5], v[4:5], v[12:13] +; GFX12-NEXT: v_minimum_f64 v[6:7], v[6:7], v[14:15] +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x double> @llvm.minimum.v4f64(<4 x double> %a, <4 x double> %b) %ret = bitcast <4 x double> %val to <8 x float> ret <8 x float> %ret } define amdgpu_ps <8 x float> @test_fminimum_v4f64_ss(<4 x double> inreg %a, <4 x double> inreg %b) { -; GCN-LABEL: test_fminimum_v4f64_ss: -; GCN: ; %bb.0: -; GCN-NEXT: v_minimum_f64 v[0:1], s[0:1], s[8:9] -; GCN-NEXT: v_minimum_f64 v[2:3], s[2:3], s[10:11] -; GCN-NEXT: v_minimum_f64 v[4:5], s[4:5], s[12:13] -; GCN-NEXT: v_minimum_f64 v[6:7], s[6:7], s[14:15] -; GCN-NEXT: ; return to shader part epilog +; GFX9-SDAG-LABEL: test_fminimum_v4f64_ss: +; GFX9-SDAG: ; %bb.0: +; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s8 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s9 +; GFX9-SDAG-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v10, 0x7ff80000 +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s10 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s11 +; GFX9-SDAG-NEXT: v_min_f64 v[4:5], s[2:3], v[1:2] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[0:1], s[2:3], v[1:2] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s12 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s13 +; GFX9-SDAG-NEXT: v_min_f64 v[6:7], s[4:5], v[1:2] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[2:3], s[4:5], v[1:2] +; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s14 +; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s15 +; GFX9-SDAG-NEXT: v_min_f64 v[8:9], s[6:7], v[1:2] +; GFX9-SDAG-NEXT: v_cmp_u_f64_e64 s[4:5], s[6:7], v[1:2] +; GFX9-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v10, vcc +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v2, v4, 0, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v3, v5, v10, s[0:1] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v4, v6, 0, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v5, v7, v10, s[2:3] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v6, v8, 0, s[4:5] +; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v7, v9, v10, s[4:5] +; GFX9-SDAG-NEXT: ; return to shader part epilog +; +; GFX9-GISEL-LABEL: test_fminimum_v4f64_ss: +; GFX9-GISEL: ; %bb.0: +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s8 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s9 +; GFX9-GISEL-NEXT: v_min_f64 v[2:3], s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e32 vcc, s[0:1], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s10 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s11 +; GFX9-GISEL-NEXT: v_min_f64 v[4:5], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[0:1], s[2:3], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s12 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s13 +; GFX9-GISEL-NEXT: v_min_f64 v[6:7], s[4:5], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[2:3], s[4:5], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s14 +; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s15 +; GFX9-GISEL-NEXT: v_min_f64 v[8:9], s[6:7], v[0:1] +; GFX9-GISEL-NEXT: v_cmp_o_f64_e64 s[4:5], s[6:7], v[0:1] +; GFX9-GISEL-NEXT: v_mov_b32_e32 v10, 0x7ff80000 +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v3, vcc +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v2, 0, v4, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v3, v10, v5, s[0:1] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v4, 0, v6, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v5, v10, v7, s[2:3] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v6, 0, v8, s[4:5] +; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v7, v10, v9, s[4:5] +; GFX9-GISEL-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_v4f64_ss: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f64 v[0:1], s[0:1], s[8:9] +; GFX12-NEXT: v_minimum_f64 v[2:3], s[2:3], s[10:11] +; GFX12-NEXT: v_minimum_f64 v[4:5], s[4:5], s[12:13] +; GFX12-NEXT: v_minimum_f64 v[6:7], s[6:7], s[14:15] +; GFX12-NEXT: ; return to shader part epilog %val = call <4 x double> @llvm.minimum.v4f64(<4 x double> %a, <4 x double> %b) %ret = bitcast <4 x double> %val to <8 x float> ret <8 x float> %ret } define amdgpu_kernel void @fminimumi_f32_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) { -; GCN-LABEL: fminimumi_f32_move_to_valu: -; GCN: ; %bb.0: -; GCN-NEXT: s_clause 0x1 -; GCN-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GCN-NEXT: s_load_b64 s[4:5], s[4:5], 0x34 -; GCN-NEXT: v_mov_b32_e32 v0, 0 -; GCN-NEXT: s_wait_kmcnt 0x0 -; GCN-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS -; GCN-NEXT: s_wait_loadcnt 0x0 -; GCN-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS -; GCN-NEXT: s_wait_loadcnt 0x0 -; GCN-NEXT: v_minimum_f32 v1, v1, v2 -; GCN-NEXT: global_store_b32 v0, v1, s[0:1] -; GCN-NEXT: s_endpgm +; GFX9-LABEL: fminimumi_f32_move_to_valu: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: global_load_dword v1, v0, s[2:3] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: global_load_dword v2, v0, s[6:7] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_min_f32_e32 v4, v1, v2 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v1, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX9-NEXT: global_store_dword v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +; +; GFX12-LABEL: fminimumi_f32_move_to_valu: +; GFX12: ; %bb.0: +; GFX12-NEXT: s_clause 0x1 +; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX12-NEXT: s_load_b64 s[4:5], s[4:5], 0x34 +; GFX12-NEXT: v_mov_b32_e32 v0, 0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS +; GFX12-NEXT: s_wait_loadcnt 0x0 +; GFX12-NEXT: global_load_b32 v2, v0, s[4:5] scope:SCOPE_SYS +; GFX12-NEXT: s_wait_loadcnt 0x0 +; GFX12-NEXT: v_minimum_f32 v1, v1, v2 +; GFX12-NEXT: global_store_b32 v0, v1, s[0:1] +; GFX12-NEXT: s_endpgm %a = load volatile float, ptr addrspace(1) %aptr, align 4 %b = load volatile float, ptr addrspace(1) %bptr, align 4 %v = call float @llvm.minimum.f32(float %a, float %b) @@ -305,6 +910,23 @@ define amdgpu_kernel void @fminimumi_f32_move_to_valu(ptr addrspace(1) %out, ptr } define amdgpu_kernel void @fminimum_f16_move_to_valu(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) { +; GFX9-LABEL: fminimum_f16_move_to_valu: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7e00 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: global_load_ushort v1, v0, s[2:3] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] glc +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_min_f16_e32 v4, v1, v2 +; GFX9-NEXT: v_cmp_o_f16_e32 vcc, v1, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX9-NEXT: global_store_short v0, v1, s[0:1] +; GFX9-NEXT: s_endpgm +; ; GFX12-SDAG-TRUE16-LABEL: fminimum_f16_move_to_valu: ; GFX12-SDAG-TRUE16: ; %bb.0: ; GFX12-SDAG-TRUE16-NEXT: s_clause 0x1 @@ -371,6 +993,40 @@ define amdgpu_kernel void @fminimum_f16_move_to_valu(ptr addrspace(1) %out, ptr ret void } +define amdgpu_ps float @test_fminimum_f32_ieee_on(float %a, float %b) #0 { +; GFX9-LABEL: test_fminimum_f32_ieee_on: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f32_ieee_on: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog + %val = call float @llvm.minimum.f32(float %a, float %b) + ret float %val +} + +define amdgpu_ps float @test_fminimum_f32_ieee_off(float %a, float %b) #1 { +; GFX9-LABEL: test_fminimum_f32_ieee_off: +; GFX9: ; %bb.0: +; GFX9-NEXT: v_min_f32_e32 v2, v0, v1 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x7fc00000 +; GFX9-NEXT: v_cmp_o_f32_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc +; GFX9-NEXT: ; return to shader part epilog +; +; GFX12-LABEL: test_fminimum_f32_ieee_off: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_minimum_f32 v0, v0, v1 +; GFX12-NEXT: ; return to shader part epilog + %val = call float @llvm.minimum.f32(float %a, float %b) + ret float %val +} + declare float @llvm.minimum.f32(float, float) declare <2 x float> @llvm.minimum.v2f32(<2 x float>, <2 x float>) declare <3 x float> @llvm.minimum.v3f32(<3 x float>, <3 x float>) @@ -383,3 +1039,6 @@ declare <4 x half> @llvm.minimum.v4f16(<4 x half>, <4 x half>) declare double @llvm.minimum.f64(double, double) declare <2 x double> @llvm.minimum.v2f64(<2 x double>, <2 x double>) declare <4 x double> @llvm.minimum.v4f64(<4 x double>, <4 x double>) + +attributes #0 = { nounwind "amdgpu-ieee"="true" } +attributes #1 = { nounwind "amdgpu-ieee"="false" } diff --git a/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll b/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll index 94afa88..9ebf6ae 100644 --- a/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll +++ b/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll @@ -4666,21 +4666,13 @@ define amdgpu_ps float @global_load_saddr_i8_offset_or_i64_imm_offset_16(ptr add ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: ; return to shader part epilog ; -; GFX12-SDAG-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_16: -; GFX12-SDAG: ; %bb.0: -; GFX12-SDAG-NEXT: v_or_b32_e32 v0, 16, v0 -; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, 0 -; GFX12-SDAG-NEXT: global_load_u8 v0, v[0:1], off -; GFX12-SDAG-NEXT: s_wait_loadcnt 0x0 -; GFX12-SDAG-NEXT: ; return to shader part epilog -; -; GFX12-GISEL-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_16: -; GFX12-GISEL: ; %bb.0: -; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 0 -; GFX12-GISEL-NEXT: v_or_b32_e32 v0, 16, v0 -; GFX12-GISEL-NEXT: global_load_u8 v0, v[0:1], off -; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0 -; GFX12-GISEL-NEXT: ; return to shader part epilog +; GFX12-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_16: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_or_b32_e32 v0, 16, v0 +; GFX12-NEXT: v_mov_b32_e32 v1, 0 +; GFX12-NEXT: global_load_u8 v0, v[0:1], off +; GFX12-NEXT: s_wait_loadcnt 0x0 +; GFX12-NEXT: ; return to shader part epilog %zext.idx = zext i32 %idx to i64 %or = or i64 %zext.idx, 16 %addr = inttoptr i64 %or to ptr addrspace(1) @@ -4707,21 +4699,13 @@ define amdgpu_ps float @global_load_saddr_i8_offset_or_i64_imm_offset_4160(ptr a ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: ; return to shader part epilog ; -; GFX12-SDAG-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_4160: -; GFX12-SDAG: ; %bb.0: -; GFX12-SDAG-NEXT: v_or_b32_e32 v0, 0x1040, v0 -; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, 0 -; GFX12-SDAG-NEXT: global_load_u8 v0, v[0:1], off -; GFX12-SDAG-NEXT: s_wait_loadcnt 0x0 -; GFX12-SDAG-NEXT: ; return to shader part epilog -; -; GFX12-GISEL-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_4160: -; GFX12-GISEL: ; %bb.0: -; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 0 -; GFX12-GISEL-NEXT: v_or_b32_e32 v0, 0x1040, v0 -; GFX12-GISEL-NEXT: global_load_u8 v0, v[0:1], off -; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0 -; GFX12-GISEL-NEXT: ; return to shader part epilog +; GFX12-LABEL: global_load_saddr_i8_offset_or_i64_imm_offset_4160: +; GFX12: ; %bb.0: +; GFX12-NEXT: v_or_b32_e32 v0, 0x1040, v0 +; GFX12-NEXT: v_mov_b32_e32 v1, 0 +; GFX12-NEXT: global_load_u8 v0, v[0:1], off +; GFX12-NEXT: s_wait_loadcnt 0x0 +; GFX12-NEXT: ; return to shader part epilog %zext.idx = zext i32 %idx to i64 %or = or i64 %zext.idx, 4160 %addr = inttoptr i64 %or to ptr addrspace(1) diff --git a/llvm/test/CodeGen/AMDGPU/itofp.i128.ll b/llvm/test/CodeGen/AMDGPU/itofp.i128.ll index 9684712..2f9182e 100644 --- a/llvm/test/CodeGen/AMDGPU/itofp.i128.ll +++ b/llvm/test/CodeGen/AMDGPU/itofp.i128.ll @@ -1066,13 +1066,13 @@ define double @uitofp_i128_to_f64(i128 %x) { ; GISEL-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] ; GISEL-NEXT: v_lshlrev_b64 v[8:9], 30, v[2:3] ; GISEL-NEXT: v_lshrrev_b32_e32 v5, 2, v1 -; GISEL-NEXT: v_or_b32_e32 v9, v5, v8 +; GISEL-NEXT: v_or_b32_e32 v9, v8, v5 ; GISEL-NEXT: s_and_saveexec_b64 s[4:5], vcc ; GISEL-NEXT: ; %bb.11: ; %itofp-if-then20 ; GISEL-NEXT: v_lshlrev_b64 v[2:3], 29, v[2:3] ; GISEL-NEXT: v_lshrrev_b64 v[4:5], 3, v[0:1] ; GISEL-NEXT: v_lshrrev_b32_e32 v0, 3, v1 -; GISEL-NEXT: v_or_b32_e32 v9, v0, v2 +; GISEL-NEXT: v_or_b32_e32 v9, v2, v0 ; GISEL-NEXT: v_mov_b32_e32 v7, v6 ; GISEL-NEXT: ; %bb.12: ; %Flow ; GISEL-NEXT: s_or_b64 exec, exec, s[4:5] diff --git a/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll b/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll index 1e4b633..fc36ed9 100644 --- a/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll +++ b/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-nontemporal-metadata.ll @@ -45,27 +45,18 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p ; GFX9-GISEL: ; %bb.0: ; %entry ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 ; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10 -; GFX9-GISEL-NEXT: s_mov_b32 s11, 0 -; GFX9-GISEL-NEXT: s_mov_b32 s4, s11 -; GFX9-GISEL-NEXT: s_mov_b32 s6, s11 ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-GISEL-NEXT: s_mov_b32 s10, s1 +; GFX9-GISEL-NEXT: s_mov_b32 s4, s1 ; GFX9-GISEL-NEXT: s_mov_b32 s5, s2 -; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5] -; GFX9-GISEL-NEXT: s_mov_b32 s10, s3 -; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7] +; GFX9-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s0 ; GFX9-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen glc slc ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20 ; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30 -; GFX9-GISEL-NEXT: s_mov_b32 s4, s11 -; GFX9-GISEL-NEXT: s_mov_b32 s6, s11 ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-GISEL-NEXT: s_mov_b32 s10, s1 +; GFX9-GISEL-NEXT: s_mov_b32 s4, s1 ; GFX9-GISEL-NEXT: s_mov_b32 s5, s2 -; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5] -; GFX9-GISEL-NEXT: s_mov_b32 s10, s3 -; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7] +; GFX9-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s0 ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX9-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen glc slc @@ -105,27 +96,18 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p ; GFX942-GISEL: ; %bb.0: ; %entry ; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 ; GFX942-GISEL-NEXT: s_load_dword s11, s[4:5], 0x10 -; GFX942-GISEL-NEXT: s_mov_b32 s7, 0 -; GFX942-GISEL-NEXT: s_mov_b32 s8, s7 -; GFX942-GISEL-NEXT: s_mov_b32 s10, s7 ; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-GISEL-NEXT: s_mov_b32 s6, s1 +; GFX942-GISEL-NEXT: s_mov_b32 s8, s1 ; GFX942-GISEL-NEXT: s_mov_b32 s9, s2 -; GFX942-GISEL-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9] -; GFX942-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX942-GISEL-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] +; GFX942-GISEL-NEXT: s_mov_b32 s10, s3 ; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0 ; GFX942-GISEL-NEXT: buffer_load_dword v0, v0, s[8:11], 0 offen nt ; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x20 -; GFX942-GISEL-NEXT: s_load_dword s9, s[4:5], 0x30 -; GFX942-GISEL-NEXT: s_mov_b32 s4, s7 -; GFX942-GISEL-NEXT: s_mov_b32 s8, s7 +; GFX942-GISEL-NEXT: s_load_dword s7, s[4:5], 0x30 ; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-GISEL-NEXT: s_mov_b32 s6, s1 +; GFX942-GISEL-NEXT: s_mov_b32 s4, s1 ; GFX942-GISEL-NEXT: s_mov_b32 s5, s2 -; GFX942-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] ; GFX942-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX942-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9] ; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s0 ; GFX942-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX942-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen nt @@ -168,29 +150,22 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_clause 0x1 ; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; GFX10-GISEL-NEXT: s_load_dword s5, s[8:9], 0x10 -; GFX10-GISEL-NEXT: s_mov_b32 s7, 0 -; GFX10-GISEL-NEXT: s_mov_b32 s10, s7 -; GFX10-GISEL-NEXT: s_mov_b32 s4, s7 +; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10 ; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-GISEL-NEXT: s_mov_b32 s6, s1 -; GFX10-GISEL-NEXT: s_mov_b32 s11, s2 ; GFX10-GISEL-NEXT: v_mov_b32_e32 v0, s0 -; GFX10-GISEL-NEXT: s_or_b64 s[0:1], s[6:7], s[10:11] +; GFX10-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX10-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX10-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX10-GISEL-NEXT: s_or_b64 s[2:3], s[6:7], s[4:5] -; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen slc +; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen slc ; GFX10-GISEL-NEXT: s_clause 0x1 -; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20 -; GFX10-GISEL-NEXT: s_load_dword s11, s[8:9], 0x30 +; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3 +; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30 ; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-GISEL-NEXT: s_mov_b32 s6, s1 -; GFX10-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX10-GISEL-NEXT: v_mov_b32_e32 v1, s0 -; GFX10-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] +; GFX10-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX10-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX10-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX10-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[10:11] ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX10-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen glc slc ; GFX10-GISEL-NEXT: s_endpgm @@ -234,32 +209,21 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p ; GFX11-GISEL: ; %bb.0: ; %entry ; GFX11-GISEL-NEXT: s_clause 0x1 ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10 -; GFX11-GISEL-NEXT: s_mov_b32 s9, 0 -; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-GISEL-NEXT: s_mov_b32 s10, s9 -; GFX11-GISEL-NEXT: s_mov_b32 s6, s9 +; GFX11-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10 ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX11-GISEL-NEXT: s_mov_b32 s11, s2 ; GFX11-GISEL-NEXT: v_mov_b32_e32 v0, s0 -; GFX11-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11] -; GFX11-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7] -; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen slc dlc +; GFX11-GISEL-NEXT: s_mov_b32 s8, s1 +; GFX11-GISEL-NEXT: s_mov_b32 s9, s2 +; GFX11-GISEL-NEXT: s_mov_b32 s10, s3 +; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], 0 offen slc dlc ; GFX11-GISEL-NEXT: s_clause 0x1 ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20 ; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30 -; GFX11-GISEL-NEXT: s_mov_b32 s4, s9 ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX11-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX11-GISEL-NEXT: v_mov_b32_e32 v1, s0 -; GFX11-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5] -; GFX11-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GFX11-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX11-GISEL-NEXT: s_mov_b32 s5, s2 +; GFX11-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX11-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], 0 offen glc slc dlc ; GFX11-GISEL-NEXT: s_endpgm @@ -303,32 +267,21 @@ define amdgpu_kernel void @buffer_nontemporal_load_store(ptr addrspace(7) %in, p ; GFX12-GISEL: ; %bb.0: ; %entry ; GFX12-GISEL-NEXT: s_clause 0x1 ; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10 -; GFX12-GISEL-NEXT: s_mov_b32 s9, 0 -; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-GISEL-NEXT: s_mov_b32 s10, s9 -; GFX12-GISEL-NEXT: s_mov_b32 s6, s9 +; GFX12-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX12-GISEL-NEXT: s_mov_b32 s11, s2 ; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0 -; GFX12-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11] -; GFX12-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7] -; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen th:TH_LOAD_NT +; GFX12-GISEL-NEXT: s_mov_b32 s8, s1 +; GFX12-GISEL-NEXT: s_mov_b32 s9, s2 +; GFX12-GISEL-NEXT: s_mov_b32 s10, s3 +; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], null offen th:TH_LOAD_NT ; GFX12-GISEL-NEXT: s_clause 0x1 ; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20 ; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30 -; GFX12-GISEL-NEXT: s_mov_b32 s4, s9 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX12-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, s0 -; GFX12-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5] -; GFX12-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GFX12-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX12-GISEL-NEXT: s_mov_b32 s5, s2 +; GFX12-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0 ; GFX12-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], null offen th:TH_STORE_NT ; GFX12-GISEL-NEXT: s_endpgm @@ -374,28 +327,19 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp ; GFX9-GISEL: ; %bb.0: ; %entry ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 ; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10 -; GFX9-GISEL-NEXT: s_mov_b32 s11, 0 -; GFX9-GISEL-NEXT: s_mov_b32 s4, s11 -; GFX9-GISEL-NEXT: s_mov_b32 s6, s11 ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-GISEL-NEXT: s_mov_b32 s10, s1 +; GFX9-GISEL-NEXT: s_mov_b32 s4, s1 ; GFX9-GISEL-NEXT: s_mov_b32 s5, s2 -; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5] -; GFX9-GISEL-NEXT: s_mov_b32 s10, s3 -; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7] +; GFX9-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s0 ; GFX9-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen glc ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX9-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20 ; GFX9-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30 -; GFX9-GISEL-NEXT: s_mov_b32 s4, s11 -; GFX9-GISEL-NEXT: s_mov_b32 s6, s11 ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-GISEL-NEXT: s_mov_b32 s10, s1 +; GFX9-GISEL-NEXT: s_mov_b32 s4, s1 ; GFX9-GISEL-NEXT: s_mov_b32 s5, s2 -; GFX9-GISEL-NEXT: s_or_b64 s[4:5], s[10:11], s[4:5] -; GFX9-GISEL-NEXT: s_mov_b32 s10, s3 -; GFX9-GISEL-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7] +; GFX9-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s0 ; GFX9-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) @@ -436,28 +380,19 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp ; GFX942-GISEL: ; %bb.0: ; %entry ; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 ; GFX942-GISEL-NEXT: s_load_dword s11, s[4:5], 0x10 -; GFX942-GISEL-NEXT: s_mov_b32 s7, 0 -; GFX942-GISEL-NEXT: s_mov_b32 s8, s7 -; GFX942-GISEL-NEXT: s_mov_b32 s10, s7 ; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-GISEL-NEXT: s_mov_b32 s6, s1 +; GFX942-GISEL-NEXT: s_mov_b32 s8, s1 ; GFX942-GISEL-NEXT: s_mov_b32 s9, s2 -; GFX942-GISEL-NEXT: s_or_b64 s[8:9], s[6:7], s[8:9] -; GFX942-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX942-GISEL-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] +; GFX942-GISEL-NEXT: s_mov_b32 s10, s3 ; GFX942-GISEL-NEXT: v_mov_b32_e32 v0, s0 ; GFX942-GISEL-NEXT: buffer_load_dword v0, v0, s[8:11], 0 offen sc0 sc1 ; GFX942-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX942-GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x20 -; GFX942-GISEL-NEXT: s_load_dword s9, s[4:5], 0x30 -; GFX942-GISEL-NEXT: s_mov_b32 s4, s7 -; GFX942-GISEL-NEXT: s_mov_b32 s8, s7 +; GFX942-GISEL-NEXT: s_load_dword s7, s[4:5], 0x30 ; GFX942-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-GISEL-NEXT: s_mov_b32 s6, s1 +; GFX942-GISEL-NEXT: s_mov_b32 s4, s1 ; GFX942-GISEL-NEXT: s_mov_b32 s5, s2 -; GFX942-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] ; GFX942-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX942-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9] ; GFX942-GISEL-NEXT: v_mov_b32_e32 v1, s0 ; GFX942-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen sc0 sc1 ; GFX942-GISEL-NEXT: s_waitcnt vmcnt(0) @@ -501,30 +436,23 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_clause 0x1 ; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; GFX10-GISEL-NEXT: s_load_dword s5, s[8:9], 0x10 -; GFX10-GISEL-NEXT: s_mov_b32 s7, 0 -; GFX10-GISEL-NEXT: s_mov_b32 s10, s7 -; GFX10-GISEL-NEXT: s_mov_b32 s4, s7 +; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x10 ; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-GISEL-NEXT: s_mov_b32 s6, s1 -; GFX10-GISEL-NEXT: s_mov_b32 s11, s2 ; GFX10-GISEL-NEXT: v_mov_b32_e32 v0, s0 -; GFX10-GISEL-NEXT: s_or_b64 s[0:1], s[6:7], s[10:11] +; GFX10-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX10-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX10-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX10-GISEL-NEXT: s_or_b64 s[2:3], s[6:7], s[4:5] -; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen glc dlc +; GFX10-GISEL-NEXT: buffer_load_dword v0, v0, s[4:7], 0 offen glc dlc ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX10-GISEL-NEXT: s_clause 0x1 -; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-GISEL-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x20 -; GFX10-GISEL-NEXT: s_load_dword s11, s[8:9], 0x30 +; GFX10-GISEL-NEXT: s_waitcnt_depctr 0xffe3 +; GFX10-GISEL-NEXT: s_load_dword s7, s[8:9], 0x30 ; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-GISEL-NEXT: s_mov_b32 s6, s1 -; GFX10-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX10-GISEL-NEXT: v_mov_b32_e32 v1, s0 -; GFX10-GISEL-NEXT: s_or_b64 s[4:5], s[6:7], s[4:5] +; GFX10-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX10-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX10-GISEL-NEXT: s_mov_b32 s6, s3 -; GFX10-GISEL-NEXT: s_or_b64 s[6:7], s[6:7], s[10:11] ; GFX10-GISEL-NEXT: buffer_store_dword v0, v1, s[4:7], 0 offen ; GFX10-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-GISEL-NEXT: s_endpgm @@ -569,33 +497,22 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp ; GFX11-GISEL: ; %bb.0: ; %entry ; GFX11-GISEL-NEXT: s_clause 0x1 ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10 -; GFX11-GISEL-NEXT: s_mov_b32 s9, 0 -; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-GISEL-NEXT: s_mov_b32 s10, s9 -; GFX11-GISEL-NEXT: s_mov_b32 s6, s9 +; GFX11-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10 ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX11-GISEL-NEXT: s_mov_b32 s11, s2 ; GFX11-GISEL-NEXT: v_mov_b32_e32 v0, s0 -; GFX11-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11] -; GFX11-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7] -; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen glc dlc +; GFX11-GISEL-NEXT: s_mov_b32 s8, s1 +; GFX11-GISEL-NEXT: s_mov_b32 s9, s2 +; GFX11-GISEL-NEXT: s_mov_b32 s10, s3 +; GFX11-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], 0 offen glc dlc ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) ; GFX11-GISEL-NEXT: s_clause 0x1 ; GFX11-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20 ; GFX11-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30 -; GFX11-GISEL-NEXT: s_mov_b32 s4, s9 ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX11-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX11-GISEL-NEXT: v_mov_b32_e32 v1, s0 -; GFX11-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5] -; GFX11-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX11-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GFX11-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX11-GISEL-NEXT: s_mov_b32 s5, s2 +; GFX11-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX11-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], 0 offen dlc ; GFX11-GISEL-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-GISEL-NEXT: s_endpgm @@ -640,33 +557,22 @@ define amdgpu_kernel void @buffer_nontemporal_and_volatile_load_store(ptr addrsp ; GFX12-GISEL: ; %bb.0: ; %entry ; GFX12-GISEL-NEXT: s_clause 0x1 ; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x10 -; GFX12-GISEL-NEXT: s_mov_b32 s9, 0 -; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-GISEL-NEXT: s_mov_b32 s10, s9 -; GFX12-GISEL-NEXT: s_mov_b32 s6, s9 +; GFX12-GISEL-NEXT: s_load_b32 s11, s[4:5], 0x10 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX12-GISEL-NEXT: s_mov_b32 s11, s2 ; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0 -; GFX12-GISEL-NEXT: s_or_b64 s[0:1], s[8:9], s[10:11] -; GFX12-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-GISEL-NEXT: s_or_b64 s[2:3], s[8:9], s[6:7] -; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[0:3], null offen th:TH_LOAD_NT scope:SCOPE_SYS +; GFX12-GISEL-NEXT: s_mov_b32 s8, s1 +; GFX12-GISEL-NEXT: s_mov_b32 s9, s2 +; GFX12-GISEL-NEXT: s_mov_b32 s10, s3 +; GFX12-GISEL-NEXT: buffer_load_b32 v0, v0, s[8:11], null offen th:TH_LOAD_NT scope:SCOPE_SYS ; GFX12-GISEL-NEXT: s_wait_loadcnt 0x0 ; GFX12-GISEL-NEXT: s_clause 0x1 ; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x20 ; GFX12-GISEL-NEXT: s_load_b32 s7, s[4:5], 0x30 -; GFX12-GISEL-NEXT: s_mov_b32 s4, s9 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: s_mov_b32 s8, s1 -; GFX12-GISEL-NEXT: s_mov_b32 s5, s2 ; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, s0 -; GFX12-GISEL-NEXT: s_or_b64 s[4:5], s[8:9], s[4:5] -; GFX12-GISEL-NEXT: s_mov_b32 s8, s3 -; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-GISEL-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GFX12-GISEL-NEXT: s_mov_b32 s4, s1 +; GFX12-GISEL-NEXT: s_mov_b32 s5, s2 +; GFX12-GISEL-NEXT: s_mov_b32 s6, s3 ; GFX12-GISEL-NEXT: buffer_store_b32 v0, v1, s[4:7], null offen th:TH_STORE_NT scope:SCOPE_SYS ; GFX12-GISEL-NEXT: s_wait_storecnt 0x0 ; GFX12-GISEL-NEXT: s_endpgm |