https://github.com/kpyzhov updated https://github.com/llvm/llvm-project/pull/134518
>From 8643b31583b1391fb643b99e068390cb64375fdc Mon Sep 17 00:00:00 2001
From: Konstantin <konstantin.pyzhov at amd.com>
Date: Sat, 5 Apr 2025 22:40:12 -0400
Subject: [PATCH 1/2] Skip debug machine instructions in AMDGPU SIWholeQuadMode
pass.
---
llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp b/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
index 57ae7d63218dd..8ae134a7b8091 100644
--- a/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
+++ b/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
@@ -1300,6 +1300,12 @@ void SIWholeQuadMode::processBlock(MachineBasicBlock &MBB, BlockInfo &BI,
for (unsigned Idx = 0;; ++Idx) {
MachineBasicBlock::iterator Next = II;
+
+ if (II != IE && II->isDebugInstr()) {
+ II = ++Next;
+ continue;
+ }
+
char Needs = StateExact | StateWQM; // Strict mode is disabled by default.
char OutNeeds = 0;
>From 43bb2abf985c09aadf5ccb584c02442a828d1021 Mon Sep 17 00:00:00 2001
From: Konstantin <konstantin.pyzhov at amd.com>
Date: Wed, 9 Apr 2025 21:25:51 -0400
Subject: [PATCH 2/2] [AMDGPU] Skip debug machine instructions in AMDGPU
SIWholeQuadMode pass.
---
llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp | 2 +-
.../AMDGPU/si-wqm-skip-metadata-insts.mir | 109 ++++++++++++++++++
2 files changed, 110 insertions(+), 1 deletion(-)
create mode 100644 llvm/test/DebugInfo/AMDGPU/si-wqm-skip-metadata-insts.mir
diff --git a/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp b/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
index 8ae134a7b8091..56c4d1868ac05 100644
--- a/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
+++ b/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
@@ -1301,7 +1301,7 @@ void SIWholeQuadMode::processBlock(MachineBasicBlock &MBB, BlockInfo &BI,
for (unsigned Idx = 0;; ++Idx) {
MachineBasicBlock::iterator Next = II;
- if (II != IE && II->isDebugInstr()) {
+ if (II != IE && II->isMetaInstruction()) {
II = ++Next;
continue;
}
diff --git a/llvm/test/DebugInfo/AMDGPU/si-wqm-skip-metadata-insts.mir b/llvm/test/DebugInfo/AMDGPU/si-wqm-skip-metadata-insts.mir
new file mode 100644
index 0000000000000..7f25f93dbed70
--- /dev/null
+++ b/llvm/test/DebugInfo/AMDGPU/si-wqm-skip-metadata-insts.mir
@@ -0,0 +1,109 @@
+
+# RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1150 -run-pass=si-wqm -o - %s
+
+--- |
+
+
+ declare void @test() #0
+
+ !0 = !DILocalVariable(name: "rayPayload", arg: 1, scope: !661, file: !106, line: 232, type: !304)
+ !36 = !DIBasicType(name: "unsigned int", size: 32, align: 32, encoding: DW_ATE_unsigned)
+ !106 = !DIFile(filename: "./RaytracingShaderHelper.hlsli", directory: "")
+ !153 = !DIDerivedType(tag: DW_TAG_typedef, name: "UINT", file: !106, line: 20, baseType: !154)
+ !154 = !DIDerivedType(tag: DW_TAG_typedef, name: "uint", file: !106, line: 57, baseType: !36)
+ !182 = !DIDerivedType(tag: DW_TAG_typedef, name: "XMFLOAT4", file: !106, line: 17, baseType: !36)
+ !304 = !DICompositeType(tag: DW_TAG_structure_type, name: "RayPayload", file: !106, line: 60, size: 160, align: 32, elements: !305)
+ !305 = !{!306, !307}
+ !306 = !DIDerivedType(tag: DW_TAG_member, name: "color", scope: !304, file: !106, line: 62, baseType: !182, size: 128, align: 32)
+ !307 = !DIDerivedType(tag: DW_TAG_member, name: "recursionDepth", scope: !304, file: !106, line: 63, baseType: !153, size: 32, align: 32, offset: 128)
+ !325 = distinct !DISubprogram(name: "HitWorldPosition", linkageName: "\01?HitWorldPosition@@YA?AV?$vector at M$02@@XZ", scope: !106, file: !106, line: 89, type: !326, scopeLine: 90, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition)
+ !326 = !DISubroutineType(types: !327)
+ !327 = !{}
+ !661 = !DILocation(line: 0, scope: !325, inlinedAt: !662)
+ !662 = distinct !DILocation(line: 249, column: 34, scope: !325)
+ !665 = !DILocation(line: 91, column: 12, scope: !325, inlinedAt: !666)
+ !666 = distinct !DILocation(line: 239, column: 26, scope: !325)
+
+...
+---
+name: test
+body: |
+
+ bb.0:
+ successors: %bb.1(0x40000000), %bb.2(0x40000000); %bb.1(50.00%), %bb.2(50.00%)
+ liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr18, $sgpr19, $vgpr8, $vgpr9, $vgpr12, $vgpr51, $vgpr52, $vgpr53, $vgpr54, $vgpr55, $vgpr56, $vgpr57, $vgpr58, $vgpr59, $vgpr60, $vgpr61
+ %126:vgpr_32 = COPY $vgpr60, debug-instr-number 754
+ %75:vgpr_32 = COPY $vgpr9, debug-instr-number 743
+ %125:vgpr_32 = COPY $vgpr59, debug-instr-number 753
+ %124:vgpr_32 = COPY $vgpr58, debug-instr-number 752
+ %136:vgpr_32 = COPY $vgpr57, debug-instr-number 751
+ %135:vgpr_32 = COPY $vgpr56, debug-instr-number 750
+ %134:vgpr_32 = COPY $vgpr55, debug-instr-number 749
+ %133:vgpr_32 = COPY $vgpr54, debug-instr-number 748
+ %132:vgpr_32 = COPY $vgpr53, debug-instr-number 747
+ %131:vgpr_32 = COPY $vgpr52, debug-instr-number 746
+ %117:vgpr_32 = COPY $vgpr51, debug-instr-number 745
+ %78:vgpr_32 = COPY $vgpr12, debug-instr-number 744
+ %628:sreg_32 = ENTER_STRICT_WWM -1, implicit-def $exec, implicit-def $scc, implicit $exec, debug-instr-number 523
+ %74:vgpr_32 = COPY $vgpr8, debug-instr-number 742
+ $exec_lo = EXIT_STRICT_WWM %628:sreg_32, debug-instr-number 524
+ %458:sgpr_32 = COPY $sgpr19, debug-instr-number 741
+ %468:sgpr_32 = COPY $sgpr18, debug-instr-number 740
+ %466:sgpr_32 = COPY $sgpr17, debug-instr-number 739
+ %464:sgpr_32 = COPY $sgpr16, debug-instr-number 738
+ %456:sgpr_32 = COPY $sgpr15, debug-instr-number 737
+ %454:sgpr_32 = COPY $sgpr14, debug-instr-number 736
+ %452:sgpr_32 = COPY $sgpr13, debug-instr-number 735
+ %450:sgpr_32 = COPY $sgpr12, debug-instr-number 734
+ %448:sgpr_32 = COPY $sgpr11, debug-instr-number 733
+ %446:sgpr_32 = COPY $sgpr10, debug-instr-number 732
+ %444:sgpr_32 = COPY $sgpr9, debug-instr-number 731
+ %442:sgpr_32 = COPY $sgpr8, debug-instr-number 730
+ %440:sgpr_32 = COPY $sgpr7, debug-instr-number 729
+ %438:sgpr_32 = COPY $sgpr6, debug-instr-number 728
+ %436:sgpr_32 = COPY $sgpr5, debug-instr-number 727
+ %434:sgpr_32 = COPY $sgpr4, debug-instr-number 726
+ undef %291.sub0:sgpr_64 = COPY $sgpr3, debug-instr-number 725
+ %627:sgpr_32 = COPY $sgpr2, debug-instr-number 724
+ %626:sgpr_32 = COPY $sgpr1, debug-instr-number 723
+ %430:sgpr_32 = COPY $sgpr0, debug-instr-number 722
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 128, 32),debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 128), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 128, 32), debug-location !661; Raytracing.hlsl:0 @[ Raytracing.hlsl:249:34 ] line no:113
+ %0:vreg_128 = SCRATCH_LOAD_DWORDX4 %75:vgpr_32, -60, 0, implicit $exec, implicit $flat_scr, debug-instr-number 24, debug-location !665
+ %144:vreg_64 = SCRATCH_LOAD_DWORDX2 %75:vgpr_32, -44, 0, implicit $exec, implicit $flat_scr, debug-instr-number 25, debug-location !665
+ %2:vgpr_32 = SCRATCH_LOAD_DWORD %75:vgpr_32, -4, 0, implicit $exec, implicit $flat_scr, debug-instr-number 28, debug-location !665
+ %3:vgpr_32 = SCRATCH_LOAD_DWORD %75:vgpr_32, -180, 0, implicit $exec, implicit $flat_scr, debug-instr-number 29, debug-location !665
+ %4:vreg_96 = SCRATCH_LOAD_DWORDX3 %75:vgpr_32, -128, 0, implicit $exec, implicit $flat_scr, debug-instr-number 30, debug-location !665
+ %5:vreg_64 = SCRATCH_LOAD_DWORDX2 %75:vgpr_32, -92, 0, implicit $exec, implicit $flat_scr, debug-instr-number 31, debug-location !665
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 0, 32), debug-location !665; Raytracing.hlsl:249:16 line no:249
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 32, 32), debug-location !665; Raytracing.hlsl:249:16 line no:249
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 64, 32), debug-location !665; Raytracing.hlsl:249:16 line no:249
+ DBG_VALUE $noreg, $noreg, !0, !DIExpression(DW_OP_LLVM_fragment, 96, 32), debug-location !665; Raytracing.hlsl:249:16 line no:249
+ %148:vgpr_32 = V_BFE_U32_e64 %0.sub1:vreg_128, 16, 14, implicit $exec, debug-location !665; Raytracing.hlsl:251:87
+ %7:vgpr_32 = V_MUL_LO_U32_e64 %144.sub1:vreg_64, %148:vgpr_32, implicit $exec, debug-location !665; Raytracing.hlsl:251:87
+ %6:vgpr_32 = V_MUL_LO_U32_e64 %0.sub2:vreg_128, %148:vgpr_32, implicit $exec, debug-location !665; Raytracing.hlsl:251:87
+ %8:vgpr_32 = V_ADD_U32_e64 %7:vgpr_32, %144.sub0:vreg_64, 0, implicit $exec, debug-location !665; Raytracing.hlsl:251:87
+ %609:vgpr_32 = V_MOV_B32_e32 0, implicit $exec, debug-location !661; Raytracing.hlsl:251:87
+ %607:vgpr_32 = V_AND_B32_e64 65535, %0.sub1:vreg_128, implicit $exec,
+ %149:sreg_32 = V_CMP_LT_U32_e64 %8:vgpr_32, %6:vgpr_32, implicit $exec, debug-instr-number 38, debug-location !661; Raytracing.hlsl:251:87
+ %608:vgpr_32 = V_MOV_B32_e32 0, implicit $exec, debug-instr-number 612
+ %9:sreg_32 = COPY $exec_lo, implicit-def $exec_lo, debug-instr-number 566, debug-location !661; Raytracing.hlsl:251:87
+ %617:sreg_32 = S_AND_B32 %9:sreg_32, %149:sreg_32, implicit-def dead $scc, debug-instr-number 567, debug-location !661; Raytracing.hlsl:251:87
+ $exec_lo = S_MOV_B32_term %617:sreg_32, debug-instr-number 568, debug-location !661; Raytracing.hlsl:251:87
+ S_CBRANCH_EXECZ %bb.2, implicit $exec, debug-instr-number 569, debug-location !661; Raytracing.hlsl:251:87
+ S_BRANCH %bb.1, debug-instr-number 41, debug-location !661; Raytracing.hlsl:251:87
+
+ bb.1:
+
+ bb.2:
+
+
+...