[PATCH] D146683: [DAG] Conservatively check the VSELECT Operation Action in tryToFoldExtendSelectLoad
Xiang Zhang via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Mar 23 22:14:07 PDT 2023
This revision was automatically updated to reflect the committed changes.
Closed by commit rGcc86e6b0a88b: [BugFix] Fix VSELECT ISel fail (authored by xiangzhangllvm).
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D146683/new/
https://reviews.llvm.org/D146683
Files:
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
llvm/test/CodeGen/X86/vselect-post-combine.ll
Index: llvm/test/CodeGen/X86/vselect-post-combine.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/X86/vselect-post-combine.ll
@@ -0,0 +1,24 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX2
+
+define ptr @test_mul(ptr %addr) {
+; AVX2-LABEL: test_mul:
+; AVX2: # %bb.0: # %entry
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm0 = [255,0,0,0]
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX2-NEXT: vpblendvb %xmm0, (%rdi), %xmm1, %xmm0
+; AVX2-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX2-NEXT: vmovdqu %ymm0, 0
+; AVX2-NEXT: xorl %eax, %eax
+; AVX2-NEXT: vzeroupper
+; AVX2-NEXT: retq
+entry:
+ %vec0 = load <32 x i8>, ptr %addr
+ %vec1 = shufflevector <32 x i8> %vec0, <32 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, <32 x i32> <i32 0, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+ %0 = bitcast <32 x i8> %vec1 to <4 x i64>
+ %shuffle = shufflevector <4 x i64> %0, <4 x i64> zeroinitializer, <2 x i32> <i32 0, i32 1>
+ %1 = bitcast <2 x i64> %shuffle to <16 x i8>
+ %conv = zext <16 x i8> %1 to <16 x i16>
+ store <16 x i16> %conv, ptr null, align 1
+ ret ptr null
+}
Index: llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
===================================================================
--- llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -1039,6 +1039,8 @@
break;
assert(Subtarget->hasSSE41() && "Expected SSE4.1 support!");
+ assert(N->getValueType(0).getVectorElementType() != MVT::i16 &&
+ "We can't replace VSELECT with BLENDV in vXi16!");
SDValue Blendv =
CurDAG->getNode(X86ISD::BLENDV, SDLoc(N), N->getValueType(0),
N->getOperand(0), N->getOperand(1), N->getOperand(2));
Index: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -12210,7 +12210,8 @@
/// This function is called by the DAGCombiner when visiting sext/zext/aext
/// dag nodes (see for example method DAGCombiner::visitSIGN_EXTEND).
static SDValue tryToFoldExtendSelectLoad(SDNode *N, const TargetLowering &TLI,
- SelectionDAG &DAG) {
+ SelectionDAG &DAG,
+ CombineLevel Level) {
unsigned Opcode = N->getOpcode();
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
@@ -12235,10 +12236,14 @@
else if (Opcode == ISD::ZERO_EXTEND)
ExtLoadOpcode = ISD::ZEXTLOAD;
+ // Illegal VSELECT may ISel fail if happen after legalization (DAG
+ // Combine2), so we should conservatively check the OperationAction.
LoadSDNode *Load1 = cast<LoadSDNode>(Op1);
LoadSDNode *Load2 = cast<LoadSDNode>(Op2);
if (!TLI.isLoadExtLegal(ExtLoadOpcode, VT, Load1->getMemoryVT()) ||
- !TLI.isLoadExtLegal(ExtLoadOpcode, VT, Load2->getMemoryVT()))
+ !TLI.isLoadExtLegal(ExtLoadOpcode, VT, Load2->getMemoryVT()) ||
+ (N0->getOpcode() == ISD::VSELECT && Level >= AfterLegalizeTypes &&
+ TLI.getOperationAction(ISD::VSELECT, VT) != TargetLowering::Legal))
return SDValue();
SDValue Ext1 = DAG.getNode(Opcode, DL, VT, Op1);
@@ -13106,7 +13111,7 @@
return DAG.getNode(ISD::ADD, DL, VT, Zext, DAG.getAllOnesConstant(DL, VT));
}
- if (SDValue Res = tryToFoldExtendSelectLoad(N, TLI, DAG))
+ if (SDValue Res = tryToFoldExtendSelectLoad(N, TLI, DAG, Level))
return Res;
return SDValue();
@@ -13457,7 +13462,7 @@
if (SDValue V = widenAbs(N, DAG))
return V;
- if (SDValue Res = tryToFoldExtendSelectLoad(N, TLI, DAG))
+ if (SDValue Res = tryToFoldExtendSelectLoad(N, TLI, DAG, Level))
return Res;
return SDValue();
@@ -13618,7 +13623,7 @@
if (SDValue NewCtPop = widenCtPop(N, DAG))
return NewCtPop;
- if (SDValue Res = tryToFoldExtendSelectLoad(N, TLI, DAG))
+ if (SDValue Res = tryToFoldExtendSelectLoad(N, TLI, DAG, Level))
return Res;
return SDValue();
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D146683.507962.patch
Type: text/x-patch
Size: 4839 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230324/a1aaefbb/attachment.bin>
More information about the llvm-commits
mailing list