[llvm] 3dc413e - [AMDGPU] Skip debug instruction uses while optimizing live range of a reg in SIOptimizeVGPRLiveRange

Yashwant Singh via llvm-commits llvm-commits at lists.llvm.org
Sun Aug 6 23:06:30 PDT 2023


Author: Yashwant Singh
Date: 2023-08-07T11:35:39+05:30
New Revision: 3dc413e25d6ccff9a9b7a3beffafef52be83e43c

URL: https://github.com/llvm/llvm-project/commit/3dc413e25d6ccff9a9b7a3beffafef52be83e43c
DIFF: https://github.com/llvm/llvm-project/commit/3dc413e25d6ccff9a9b7a3beffafef52be83e43c.diff

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

This will prevent the `assert(!O.readsReg())` from firing in
SIOptimizeVGPRLiveRange::optimizeLiveRange

Fix for #64163

Reviewed By: arsenm, #amdgpu

Differential Revision: https://reviews.llvm.org/D156893

Added: 
    llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.ll
    llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.mir

Modified: 
    llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp b/llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
index e95abae88d7a83..8204a70e72d916 100644
--- a/llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
+++ b/llvm/lib/Target/AMDGPU/SIOptimizeVGPRLiveRange.cpp
@@ -522,9 +522,11 @@ void SIOptimizeVGPRLiveRange::optimizeLiveRange(
     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

diff  --git a/llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.ll b/llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.ll
new file mode 100644
index 00000000000000..d34769ad0fcf0a
--- /dev/null
+++ b/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)

diff  --git a/llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.mir b/llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.mir
new file mode 100644
index 00000000000000..3bdcc14936fb9b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/si-optimize-vgpr-live-range-dbg-instr.mir
@@ -0,0 +1,111 @@
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -run-pass=si-opt-vgpr-liverange  %s -o - | FileCheck -check-prefix=GCN %s
+
+# SIOptimizeVGPRLiveRange shouldn't try to modify use of %5 in DBG_VALUE_LIST
+
+--- |
+  define void @dbg_instr_use(i1 %arg) #1 {
+    ret void
+  }
+
+  attributes #1 = { "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,10" "target-cpu"="gfx908" "uniform-work-group-size"="false" }
+  !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)
+...
+
+---
+name:            dbg_instr_use
+tracksRegLiveness: true
+body:             |
+  ; GCN-LABEL: name: dbg_instr_use
+  ; GCN: bb.0:
+  ; GCN-NEXT:   successors: %bb.3(0x40000000), %bb.1(0x40000000)
+  ; GCN-NEXT:   liveins: $vgpr0
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY killed $vgpr0
+  ; GCN-NEXT:   [[V_AND_B32_e32_:%[0-9]+]]:vgpr_32 = V_AND_B32_e32 1, killed [[COPY]], implicit $exec
+  ; GCN-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 1, killed [[V_AND_B32_e32_]], implicit $exec
+  ; GCN-NEXT:   [[S_XOR_B64_:%[0-9]+]]:sreg_64 = S_XOR_B64 killed [[V_CMP_EQ_U32_e64_]], -1, implicit-def dead $scc
+  ; GCN-NEXT:   [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 0, implicit $exec
+  ; GCN-NEXT:   [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 killed [[V_MOV_B]], 0, 0, implicit $exec
+  ; GCN-NEXT:   [[SI_IF:%[0-9]+]]:sreg_64 = SI_IF killed [[S_XOR_B64_]], %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+  ; GCN-NEXT:   S_BRANCH %bb.3
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT: bb.1:
+  ; GCN-NEXT:   successors: %bb.2(0x40000000), %bb.4(0x40000000)
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT:   [[PHI:%[0-9]+]]:vreg_64 = PHI [[GLOBAL_LOAD_DWORDX2_]], %bb.0, undef %13:vreg_64, %bb.3
+  ; GCN-NEXT:   [[SI_ELSE:%[0-9]+]]:sreg_64 = SI_ELSE killed [[SI_IF]], %bb.4, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+  ; GCN-NEXT:   S_BRANCH %bb.2
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT: bb.2:
+  ; GCN-NEXT:   successors: %bb.4(0x80000000)
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+  ; GCN-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_]], %subreg.sub1
+  ; GCN-NEXT:   FLAT_STORE_DWORDX2 killed [[PHI]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr
+  ; GCN-NEXT:   S_BRANCH %bb.4
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT: bb.3:
+  ; GCN-NEXT:   successors: %bb.1(0x80000000)
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT:   [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+  ; GCN-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[V_MOV_B32_e32_1]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
+  ; GCN-NEXT:   FLAT_STORE_DWORDX2 killed [[GLOBAL_LOAD_DWORDX2_]], killed [[REG_SEQUENCE1]], 0, 0, implicit $exec, implicit $flat_scr
+  ; GCN-NEXT:   S_BRANCH %bb.1
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT: bb.4:
+  ; GCN-NEXT:   SI_END_CF killed [[SI_ELSE]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+  ; GCN-NEXT:   DBG_VALUE_LIST
+  ; GCN-NEXT-SAME: %9
+  ; GCN-NEXT:   SI_RETURN
+  bb.0:
+    successors: %bb.3(0x40000000), %bb.1(0x40000000)
+    liveins: $vgpr0
+  
+    %0:vgpr_32 = COPY $vgpr0
+    %1:vgpr_32 = V_AND_B32_e32 1, %0, implicit $exec
+    %2:sreg_64 = V_CMP_EQ_U32_e64 1, killed %1, implicit $exec
+    %3:sreg_64 = S_XOR_B64 killed %2, -1, implicit-def dead $scc
+    %4:vreg_64 = V_MOV_B64_PSEUDO 0, implicit $exec
+    %5:vreg_64 = GLOBAL_LOAD_DWORDX2 killed %4, 0, 0, implicit $exec
+    %6:sreg_64 = SI_IF killed %3, %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+    S_BRANCH %bb.3
+  
+  bb.1:
+    successors: %bb.2(0x40000000), %bb.4(0x40000000)
+  
+    %7:sreg_64 = SI_ELSE %6, %bb.4, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+    S_BRANCH %bb.2
+  
+  bb.2:
+    successors: %bb.4(0x80000000)
+  
+    %8:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    %9:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %8, %subreg.sub1
+    FLAT_STORE_DWORDX2 %5, killed %9, 0, 0, implicit $exec, implicit $flat_scr
+    S_BRANCH %bb.4
+  
+  bb.3:
+    successors: %bb.1(0x80000000)
+  
+    %10:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    %11:vreg_64 = REG_SEQUENCE %10, %subreg.sub0, %10, %subreg.sub1
+    FLAT_STORE_DWORDX2 %5, killed %11, 0, 0, implicit $exec, implicit $flat_scr
+    S_BRANCH %bb.1
+  
+  bb.4:
+    SI_END_CF %7, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+    DBG_VALUE_LIST !4, !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), %5, 0, debug-location !9
+    SI_RETURN
+...


        


More information about the llvm-commits mailing list