[llvm] af947d9 - [ISel] Fix crash in new FMA DAG combine

Tue Oct 4 07:19:44 PDT 2022

Author: Jay Foad
Date: 2022-10-04T15:13:18+01:00
New Revision: af947d9fcbbd71b3001ca100e15236a70c8c87b6

URL: https://github.com/llvm/llvm-project/commit/af947d9fcbbd71b3001ca100e15236a70c8c87b6
DIFF: https://github.com/llvm/llvm-project/commit/af947d9fcbbd71b3001ca100e15236a70c8c87b6.diff

LOG: [ISel] Fix crash in new FMA DAG combine

Fix a crash in the FMA combine added by D132837 and amended by D134810.
In cases where the newly created node could be folded, the combiner
would fail this assertion:

llc: DAGCombiner.cpp:268: void (anonymous namespace)::DAGCombiner::AddToWorklist(llvm::SDNode *): Assertion `N->getOpcode() != ISD::DELETED_NODE && "Deleted Node added to Worklist"' failed.

Differential Revision: https://reviews.llvm.org/D135150

Added: 
    llvm/test/CodeGen/AMDGPU/dagcombine-fma-crash.ll

Modified: 
    llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 9281b4eb92b05..db9c857f0fd9d 100644

--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -14275,7 +14275,9 @@ SDValue DAGCombiner::visitFADDForFMACombine(SDNode *N) {
         SDValue D = FMul.getOperand(1);
         SDValue CDE = DAG.getNode(PreferredFusedOpcode, SL, VT, C, D, E);
         DAG.ReplaceAllUsesOfValueWith(FMul, CDE);
-        return FMA;
+        // Replacing the inner FMul could cause the outer FMA to be simplified
+        // away.
+        return FMA.getOpcode() == ISD::DELETED_NODE ? SDValue() : FMA;
       }
 
       TmpFMA = TmpFMA->getOperand(2);

diff  --git a/llvm/test/CodeGen/AMDGPU/dagcombine-fma-crash.ll b/llvm/test/CodeGen/AMDGPU/dagcombine-fma-crash.ll
new file mode 100644
index 0000000000000..cf80833981436
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/dagcombine-fma-crash.ll
@@ -0,0 +1,80 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1030 -start-before=amdgpu-isel -stop-after=amdgpu-isel < %s | FileCheck %s
+
+define void @main(float %arg) {
+  ; CHECK-LABEL: name: main
+  ; CHECK: bb.0.bb:
+  ; CHECK-NEXT:   successors: %bb.1(0x50000000), %bb.2(0x30000000)
+  ; CHECK-NEXT:   liveins: $vgpr0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sgpr_32 = S_MOV_B32 0
+  ; CHECK-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_AND_B32_:%[0-9]+]]:sreg_32 = S_AND_B32 $exec_lo, [[S_MOV_B32_1]], implicit-def dead $scc
+  ; CHECK-NEXT:   $vcc_lo = COPY [[S_AND_B32_]]
+  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.2, implicit $vcc
+  ; CHECK-NEXT:   S_BRANCH %bb.1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1.bb2:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
+  ; CHECK-NEXT:   %20:vgpr_32 = contract reassoc nofpexcept V_FMAC_F32_e64 0, [[S_MOV_B32_]], 0, [[S_MOV_B32_]], 0, [[V_MOV_B32_e32_]], 0, 0, implicit $mode, implicit $exec
+  ; CHECK-NEXT:   %1:vgpr_32 = contract reassoc nofpexcept V_FMAC_F32_e64 0, [[COPY]], 0, [[COPY]], 0, %20, 0, 0, implicit $mode, implicit $exec
+  ; CHECK-NEXT:   %2:vgpr_32 = contract reassoc nofpexcept V_ADD_F32_e64 0, %1, 0, [[V_MOV_B32_e32_]], 0, 0, implicit $mode, implicit $exec
+  ; CHECK-NEXT:   [[S_MOV_B32_2:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2.bb11:
+  ; CHECK-NEXT:   successors: %bb.3(0x40000000), %bb.4(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[PHI:%[0-9]+]]:vgpr_32 = PHI [[DEF]], %bb.0, %1, %bb.1
+  ; CHECK-NEXT:   [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[DEF]], %bb.0, %2, %bb.1
+  ; CHECK-NEXT:   [[PHI2:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[S_MOV_B32_1]], %bb.0, [[S_MOV_B32_2]], %bb.1
+  ; CHECK-NEXT:   [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[PHI2]], implicit $exec
+  ; CHECK-NEXT:   [[S_MOV_B32_3:%[0-9]+]]:sreg_32 = S_MOV_B32 1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY [[V_CNDMASK_B32_e64_]]
+  ; CHECK-NEXT:   S_CMP_LG_U32 killed [[COPY1]], killed [[S_MOV_B32_3]], implicit-def $scc
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $scc
+  ; CHECK-NEXT:   [[S_AND_B32_1:%[0-9]+]]:sreg_32 = S_AND_B32 $exec_lo, killed [[COPY2]], implicit-def dead $scc
+  ; CHECK-NEXT:   $vcc_lo = COPY [[S_AND_B32_1]]
+  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.4, implicit $vcc
+  ; CHECK-NEXT:   S_BRANCH %bb.3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3.bb15:
+  ; CHECK-NEXT:   successors: %bb.4(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.4.bb17:
+  ; CHECK-NEXT:   SI_RETURN
+bb:
+  %i = fadd reassoc contract float 0.000000e+00, 0.000000e+00
+  %i1 = icmp ne i32 0, 0
+  br i1 %i1, label %bb2, label %bb11
+
+bb2:
+  %i3 = fmul reassoc contract float %i, %i
+  %i4 = fmul reassoc contract float %arg, %arg
+  %i5 = fadd reassoc contract float %i3, 1.000000e+00
+  %i6 = fadd reassoc contract float %i5, %i4
+  %i7 = fmul reassoc contract float %arg, %arg
+  %i8 = fmul reassoc contract float %i, %i
+  %i9 = fadd reassoc contract float %i7, %i8
+  %i10 = fadd reassoc contract float %i9, 1.000000e+00
+  br label %bb11
+
+bb11:
+  %i12 = phi float [ %i6, %bb2 ], [ undef, %bb ]
+  %i13 = phi float [ %i10, %bb2 ], [ undef, %bb ]
+  %i14 = phi i1 [ false, %bb2 ], [ true, %bb ]
+  br i1 %i14, label %bb15, label %bb17
+
+bb15:
+  %i16 = fmul reassoc contract float %i, 0.000000e+00
+  br label %bb17
+
+bb17:
+  %i18 = phi float [ %i13, %bb11 ], [ 0.000000e+00, %bb15 ]
+  %i19 = phi float [ %i12, %bb11 ], [ 0.000000e+00, %bb15 ]
+  ret void
+}


        
