Skip to content

Commit ba81d7f

Browse files
committed
[CodeGen] Always update divergence in SelectionDAG::UpdateNodeOperands
Some overloads failed to update divergence. Differential Revision: https://reviews.llvm.org/D47148 llvm-svn: 333947
1 parent 643102d commit ba81d7f

File tree

2 files changed

+32
-0
lines changed

2 files changed

+32
-0
lines changed

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6880,6 +6880,7 @@ SDNode *SelectionDAG::UpdateNodeOperands(SDNode *N, SDValue Op) {
68806880
// Now we update the operands.
68816881
N->OperandList[0].set(Op);
68826882

6883+
updateDivergence(N);
68836884
// If this gets put into a CSE map, add it.
68846885
if (InsertPos) CSEMap.InsertNode(N, InsertPos);
68856886
return N;
@@ -6959,6 +6960,7 @@ UpdateNodeOperands(SDNode *N, ArrayRef<SDValue> Ops) {
69596960
if (N->OperandList[i] != Ops[i])
69606961
N->OperandList[i].set(Ops[i]);
69616962

6963+
updateDivergence(N);
69626964
// If this gets put into a CSE map, add it.
69636965
if (InsertPos) CSEMap.InsertNode(N, InsertPos);
69646966
return N;
Lines changed: 30 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,30 @@
1+
; RUN: llc -march=amdgcn < %s
2+
3+
; Tests for a bug in SelectionDAG::UpdateNodeOperands exposed by VectorLegalizer
4+
; where divergence information is not updated.
5+
6+
declare i32 @llvm.amdgcn.workitem.id.x()
7+
8+
define amdgpu_kernel void @spam(double addrspace(1)* noalias %arg) {
9+
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
10+
%tmp1 = zext i32 %tmp to i64
11+
%tmp2 = getelementptr inbounds double, double addrspace(1)* %arg, i64 %tmp1
12+
%tmp3 = load double, double addrspace(1)* %tmp2, align 8
13+
%tmp4 = fadd double undef, 0.000000e+00
14+
%tmp5 = insertelement <2 x double> undef, double %tmp4, i64 0
15+
%tmp6 = insertelement <2 x double> %tmp5, double %tmp3, i64 1
16+
%tmp7 = insertelement <2 x double> %tmp6, double 0.000000e+00, i64 1
17+
%tmp8 = fadd <2 x double> zeroinitializer, undef
18+
%tmp9 = fadd <2 x double> %tmp7, zeroinitializer
19+
%tmp10 = extractelement <2 x double> %tmp8, i64 0
20+
%tmp11 = getelementptr inbounds double, double addrspace(1)* %tmp2, i64 2
21+
store double %tmp10, double addrspace(1)* %tmp11, align 8
22+
%tmp12 = getelementptr inbounds double, double addrspace(1)* %tmp2, i64 3
23+
store double undef, double addrspace(1)* %tmp12, align 8
24+
%tmp13 = extractelement <2 x double> %tmp9, i64 0
25+
%tmp14 = getelementptr inbounds double, double addrspace(1)* %tmp2, i64 6
26+
store double %tmp13, double addrspace(1)* %tmp14, align 8
27+
%tmp15 = getelementptr inbounds double, double addrspace(1)* %tmp2, i64 7
28+
store double 0.000000e+00, double addrspace(1)* %tmp15, align 8
29+
ret void
30+
}

0 commit comments

Comments
 (0)