[llvm] r301511 - AMDGPU: Fix assert in scheduler

Konstantin Zhuravlyov via llvm-commits llvm-commits at lists.llvm.org
Wed Apr 26 20:22:45 PDT 2017


Author: kzhuravl
Date: Wed Apr 26 22:22:44 2017
New Revision: 301511

URL: http://llvm.org/viewvc/llvm-project?rev=301511&view=rev
Log:
AMDGPU: Fix assert in scheduler

Assert is triggered if DBG_VALUE is first instruction in BB

Differential Revision: https://reviews.llvm.org/D32572

Added:
    llvm/trunk/test/DebugInfo/AMDGPU/dbg-value-sched-crash.ll
Modified:
    llvm/trunk/lib/Target/AMDGPU/GCNSchedStrategy.cpp

Modified: llvm/trunk/lib/Target/AMDGPU/GCNSchedStrategy.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/GCNSchedStrategy.cpp?rev=301511&r1=301510&r2=301511&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/GCNSchedStrategy.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/GCNSchedStrategy.cpp Wed Apr 26 22:22:44 2017
@@ -422,8 +422,9 @@ void GCNScheduleDAGMILive::discoverLiveI
   unsigned SGPRs = 0;
   unsigned VGPRs = 0;
 
+  auto &MI = *begin()->getParent()->getFirstNonDebugInstr();
   const SIRegisterInfo *SRI = static_cast<const SIRegisterInfo*>(TRI);
-  SlotIndex SI = LIS->getInstructionIndex(*begin()).getBaseIndex();
+  SlotIndex SI = LIS->getInstructionIndex(MI).getBaseIndex();
   assert (SI.isValid());
 
   DEBUG(dbgs() << "Region live-ins:");

Added: llvm/trunk/test/DebugInfo/AMDGPU/dbg-value-sched-crash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/DebugInfo/AMDGPU/dbg-value-sched-crash.ll?rev=301511&view=auto
==============================================================================
--- llvm/trunk/test/DebugInfo/AMDGPU/dbg-value-sched-crash.ll (added)
+++ llvm/trunk/test/DebugInfo/AMDGPU/dbg-value-sched-crash.ll Wed Apr 26 22:22:44 2017
@@ -0,0 +1,95 @@
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji < %s | FileCheck %s
+
+; Make sure we do not crash during scheduling when DBG_VALUE is the first
+; instruction in the basic block.
+
+; LLVM IR generated with the following command and OpenCL source:
+;
+; $clang -cl-std=CL2.0 -g -O2 -target amdgcn-amd-amdhsa -S -emit-llvm <path-to-file>
+;
+; kernel void kernel1(global int *A, global int *B) {
+;   if (*A == 1) {
+;     *B = 12;
+;   }
+;   if (*A == 2) {
+;     *B = 13;
+;   }
+; }
+
+declare void @llvm.dbg.value(metadata, i64, metadata, metadata)
+
+; CHECK-LABEL: {{^}}kernel1:
+define amdgpu_kernel void @kernel1(
+    i32 addrspace(1)* nocapture readonly %A,
+    i32 addrspace(1)* nocapture %B) !dbg !7  {
+entry:
+  tail call void @llvm.dbg.value(metadata i32 addrspace(1)* %A, i64 0, metadata !13, metadata !19), !dbg !20
+  tail call void @llvm.dbg.value(metadata i32 addrspace(1)* %B, i64 0, metadata !14, metadata !19), !dbg !21
+  %0 = load i32, i32 addrspace(1)* %A, align 4, !dbg !22, !tbaa !24
+  %cmp = icmp eq i32 %0, 1, !dbg !28
+  br i1 %cmp, label %if.then, label %if.end, !dbg !29
+
+if.then:                                          ; preds = %entry
+  store i32 12, i32 addrspace(1)* %B, align 4, !dbg !30, !tbaa !24
+  %.pr = load i32, i32 addrspace(1)* %A, align 4, !dbg !32, !tbaa !24
+  br label %if.end, !dbg !34
+
+if.end:                                           ; preds = %if.then, %entry
+  %1 = phi i32 [ %.pr, %if.then ], [ %0, %entry ], !dbg !32
+  %cmp1 = icmp eq i32 %1, 2, !dbg !35
+  br i1 %cmp1, label %if.then2, label %if.end3, !dbg !36
+
+if.then2:                                         ; preds = %if.end
+  store i32 13, i32 addrspace(1)* %B, align 4, !dbg !37, !tbaa !24
+  br label %if.end3, !dbg !39
+
+if.end3:                                          ; preds = %if.then2, %if.end
+  ret void, !dbg !40
+}
+
+!llvm.dbg.cu = !{!0}
+!opencl.ocl.version = !{!3}
+!llvm.module.flags = !{!4, !5}
+!llvm.ident = !{!6}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 4.0 ", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2)
+!1 = !DIFile(filename: "dbg-value-sched-crash.cl", directory: "/some/random/directory")
+!2 = !{}
+!3 = !{i32 2, i32 0}
+!4 = !{i32 2, !"Dwarf Version", i32 2}
+!5 = !{i32 2, !"Debug Info Version", i32 3}
+!6 = !{!"clang version 4.0 "}
+!7 = distinct !DISubprogram(name: "kernel1", scope: !1, file: !1, line: 1, type: !8, isLocal: false, isDefinition: true, scopeLine: 1, flags: DIFlagPrototyped, isOptimized: true, unit: !0, variables: !12)
+!8 = !DISubroutineType(types: !9)
+!9 = !{null, !10, !10}
+!10 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !11, size: 64)
+!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!12 = !{!13, !14}
+!13 = !DILocalVariable(name: "A", arg: 1, scope: !7, file: !1, line: 1, type: !10)
+!14 = !DILocalVariable(name: "B", arg: 2, scope: !7, file: !1, line: 1, type: !10)
+!15 = !{i32 1, i32 1}
+!16 = !{!"none", !"none"}
+!17 = !{!"int*", !"int*"}
+!18 = !{!"", !""}
+!19 = !DIExpression(DW_OP_constu, 1, DW_OP_swap, DW_OP_xderef)
+!20 = !DILocation(line: 1, column: 33, scope: !7)
+!21 = !DILocation(line: 1, column: 48, scope: !7)
+!22 = !DILocation(line: 2, column: 7, scope: !23)
+!23 = distinct !DILexicalBlock(scope: !7, file: !1, line: 2, column: 7)
+!24 = !{!25, !25, i64 0}
+!25 = !{!"int", !26, i64 0}
+!26 = !{!"omnipotent char", !27, i64 0}
+!27 = !{!"Simple C/C++ TBAA"}
+!28 = !DILocation(line: 2, column: 10, scope: !23)
+!29 = !DILocation(line: 2, column: 7, scope: !7)
+!30 = !DILocation(line: 3, column: 8, scope: !31)
+!31 = distinct !DILexicalBlock(scope: !23, file: !1, line: 2, column: 16)
+!32 = !DILocation(line: 5, column: 7, scope: !33)
+!33 = distinct !DILexicalBlock(scope: !7, file: !1, line: 5, column: 7)
+!34 = !DILocation(line: 4, column: 3, scope: !31)
+!35 = !DILocation(line: 5, column: 10, scope: !33)
+!36 = !DILocation(line: 5, column: 7, scope: !7)
+!37 = !DILocation(line: 6, column: 8, scope: !38)
+!38 = distinct !DILexicalBlock(scope: !33, file: !1, line: 5, column: 16)
+!39 = !DILocation(line: 7, column: 3, scope: !38)
+!40 = !DILocation(line: 8, column: 1, scope: !7)




More information about the llvm-commits mailing list