[PATCH] D156893: [AMDGPU] Skip debug instruction uses while optimizing live range of a reg in SIOptimizeVGPRLiveRange

Yashwant Singh via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Aug 2 07:21:15 PDT 2023


yassingh updated this revision to Diff 546458.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D156893/new/

https://reviews.llvm.org/D156893

Files:
  llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
  llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.ll


Index: llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.ll
@@ -0,0 +1,76 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 < %s | FileCheck -check-prefix=GCN %s
+
+declare void @llvm.dbg.value(metadata, metadata, metadata) #0
+
+define void @__omp_offloading_35_36570d3__ZN6openmc31process_advance_particle_eventsEv_l252_debug___omp_outlined_debug___omp_outlined(i1 %arg) {
+; GCN-LABEL: __omp_offloading_35_36570d3__ZN6openmc31process_advance_particle_eventsEv_l252_debug___omp_outlined_debug___omp_outlined:
+; GCN:       .Lfunc_begin0:
+; GCN-NEXT:    .cfi_sections .debug_frame
+; GCN-NEXT:    .cfi_startproc
+; GCN-NEXT:  ; %bb.0: ; %bb
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_mov_b32_e32 v1, 0
+; GCN-NEXT:    v_mov_b32_e32 v2, 0
+; GCN-NEXT:    global_load_dwordx2 v[1:2], v[1:2], off
+; GCN-NEXT:    v_and_b32_e32 v0, 1, v0
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    s_xor_b64 s[4:5], vcc, -1
+; GCN-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
+; GCN-NEXT:    s_xor_b64 s[4:5], exec, s[6:7]
+; GCN-NEXT:    s_cbranch_execnz .LBB0_3
+; GCN-NEXT:  ; %bb.1: ; %Flow
+; GCN-NEXT:    s_andn2_saveexec_b64 s[4:5], s[4:5]
+; GCN-NEXT:    s_cbranch_execnz .LBB0_4
+; GCN-NEXT:  .LBB0_2: ; %bb3
+; GCN-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GCN-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+; GCN-NEXT:  .LBB0_3: ; %bb2
+; GCN-NEXT:    v_mov_b32_e32 v3, 0
+; GCN-NEXT:    v_mov_b32_e32 v4, v3
+; GCN-NEXT:    s_waitcnt vmcnt(0)
+; GCN-NEXT:    flat_store_dwordx2 v[1:2], v[3:4]
+; GCN-NEXT:    ; implicit-def: $vgpr1_vgpr2
+; GCN-NEXT:    s_andn2_saveexec_b64 s[4:5], s[4:5]
+; GCN-NEXT:    s_cbranch_execz .LBB0_2
+; GCN-NEXT:  .LBB0_4: ; %bb1
+; GCN-NEXT:    v_mov_b32_e32 v3, 0
+; GCN-NEXT:    v_mov_b32_e32 v4, v3
+; GCN-NEXT:    s_waitcnt vmcnt(0)
+; GCN-NEXT:    flat_store_dwordx2 v[1:2], v[3:4]
+; GCN-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GCN-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GCN-NEXT:    s_setpc_b64 s[30:31]
+bb:
+  %i = load ptr, ptr addrspace(1) null, align 8
+  br i1 %arg, label %bb1, label %bb2
+
+bb1:                                              ; preds = %bb
+  store double 0.000000e+00, ptr %i, align 8
+  br label %bb3
+
+bb2:                                              ; preds = %bb
+  store double 0.000000e+00, ptr %i, align 8
+  br label %bb3
+
+bb3:                                              ; preds = %bb2, %bb1
+  call void @llvm.dbg.value(metadata !DIArgList(ptr %i, i64 0), metadata !4, metadata !DIExpression(DW_OP_LLVM_arg, 0, DW_OP_LLVM_arg, 1, DW_OP_constu, 2712, DW_OP_mul, DW_OP_plus, DW_OP_plus_uconst, 2680, DW_OP_stack_value)), !dbg !9
+  ret void
+}
+
+attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!3}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 3.0 (trunk 131941)", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2)
+!1 = !DIFile(filename: "/d/j/debug-test.c", directory: "/Volumes/Data/b")
+!2 = !{}
+!3 = !{i32 1, !"Debug Info Version", i32 3}
+!4 = !DILocalVariable(name: "c", scope: !5, file: !1, line: 2, type: !8)
+!5 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 1, type: !6, scopeLine: 1, virtualIndex: 6, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0)
+!6 = !DISubroutineType(types: !7)
+!7 = !{!8}
+!8 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
+!9 = !DILocation(line: 0, scope: !5)
Index: llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
+++ llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
@@ -522,9 +522,11 @@
     auto *UseBlock = UseMI->getParent();
     // Replace uses in Endif block
     if (UseBlock == Endif) {
-      if (UseMI->isPHI()) {
+      if (UseMI->isPHI())
         O.setReg(NewReg);
-      } else {
+      else if (UseMI->isDebugInstr())
+        continue;
+      else {
         // DetectDeadLanes may mark register uses as undef without removing
         // them, in which case a non-phi instruction using the original register
         // may exist in the Endif block even though the register is not live


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D156893.546458.patch
Type: text/x-patch
Size: 4628 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230802/6b61080d/attachment.bin>


More information about the llvm-commits mailing list