aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2024-11-20 17:46:46 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2024-11-21 10:35:56 +0000
commit600a83bf9ba2bee5ed1e9867e201f7707b1d8102 (patch)
treea62a60c06763a8f0c193cc671dbaef0ecb73e5f0
parent97ac84846a724809d28e167651b81355696d0d0b (diff)
downloadllvm-600a83bf9ba2bee5ed1e9867e201f7707b1d8102.zip
llvm-600a83bf9ba2bee5ed1e9867e201f7707b1d8102.tar.gz
llvm-600a83bf9ba2bee5ed1e9867e201f7707b1d8102.tar.bz2
[X86] IsNOT - match or(not(X),not(Y)) -> and(X,Y)
Fixes #116977
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp16
-rw-r--r--llvm/test/CodeGen/X86/vector-compare-all_of.ll42
2 files changed, 27 insertions, 31 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index fea66e9..2b497f3 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -5211,8 +5211,10 @@ static bool isConstantPowerOf2(SDValue V, unsigned EltSizeInBIts,
// Match not(pcmpgt(C, X)) -> pcmpgt(X, C - 1).
// Match not(extract_subvector(xor X, -1)) -> extract_subvector(X).
// Match not(concat_vectors(xor X, -1, xor Y, -1)) -> concat_vectors(X, Y).
+// Match or(not(X),not(Y)) -> and(X, Y).
static SDValue IsNOT(SDValue V, SelectionDAG &DAG) {
V = peekThroughBitcasts(V);
+ EVT VT = V.getValueType();
if (V.getOpcode() == ISD::XOR &&
(ISD::isBuildVectorAllOnes(V.getOperand(1).getNode()) ||
isAllOnesConstant(V.getOperand(1))))
@@ -5221,8 +5223,8 @@ static SDValue IsNOT(SDValue V, SelectionDAG &DAG) {
(isNullConstant(V.getOperand(1)) || V.getOperand(0).hasOneUse())) {
if (SDValue Not = IsNOT(V.getOperand(0), DAG)) {
Not = DAG.getBitcast(V.getOperand(0).getValueType(), Not);
- return DAG.getNode(ISD::EXTRACT_SUBVECTOR, SDLoc(Not), V.getValueType(),
- Not, V.getOperand(1));
+ return DAG.getNode(ISD::EXTRACT_SUBVECTOR, SDLoc(Not), VT, Not,
+ V.getOperand(1));
}
}
if (V.getOpcode() == X86ISD::PCMPGT &&
@@ -5255,7 +5257,15 @@ static SDValue IsNOT(SDValue V, SelectionDAG &DAG) {
if (!NotCat) return SDValue();
CatOp = DAG.getBitcast(CatOp.getValueType(), NotCat);
}
- return DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(V), V.getValueType(), CatOps);
+ return DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(V), VT, CatOps);
+ }
+ if (V.getOpcode() == ISD::OR && DAG.getTargetLoweringInfo().isTypeLegal(VT) &&
+ V.getOperand(0).hasOneUse() && V.getOperand(1).hasOneUse()) {
+ // TODO: Handle cases with single NOT operand -> ANDNP
+ if (SDValue Op1 = IsNOT(V.getOperand(1), DAG))
+ if (SDValue Op0 = IsNOT(V.getOperand(0), DAG))
+ return DAG.getNode(ISD::AND, SDLoc(V), VT, DAG.getBitcast(VT, Op0),
+ DAG.getBitcast(VT, Op1));
}
return SDValue();
}
diff --git a/llvm/test/CodeGen/X86/vector-compare-all_of.ll b/llvm/test/CodeGen/X86/vector-compare-all_of.ll
index deaf2d1..bf027a7 100644
--- a/llvm/test/CodeGen/X86/vector-compare-all_of.ll
+++ b/llvm/test/CodeGen/X86/vector-compare-all_of.ll
@@ -1599,55 +1599,41 @@ define i1 @PR116977(<32 x i8> %a, <32 x i8> %b, <32 x i8> %v) {
; SSE-LABEL: PR116977:
; SSE: # %bb.0:
; SSE-NEXT: pcmpeqb %xmm4, %xmm0
-; SSE-NEXT: pcmpeqd %xmm6, %xmm6
-; SSE-NEXT: pxor %xmm6, %xmm0
; SSE-NEXT: pcmpeqb %xmm5, %xmm1
-; SSE-NEXT: pxor %xmm6, %xmm1
; SSE-NEXT: pcmpeqb %xmm4, %xmm2
-; SSE-NEXT: pxor %xmm6, %xmm2
-; SSE-NEXT: por %xmm0, %xmm2
+; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: pcmpeqb %xmm5, %xmm3
-; SSE-NEXT: pxor %xmm6, %xmm3
-; SSE-NEXT: por %xmm1, %xmm3
-; SSE-NEXT: por %xmm2, %xmm3
+; SSE-NEXT: pand %xmm1, %xmm3
+; SSE-NEXT: pand %xmm2, %xmm3
; SSE-NEXT: pmovmskb %xmm3, %eax
-; SSE-NEXT: testl %eax, %eax
+; SSE-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE-NEXT: sete %al
; SSE-NEXT: retq
;
; AVX1-LABEL: PR116977:
; AVX1: # %bb.0:
; AVX1-NEXT: vpcmpeqb %xmm0, %xmm2, %xmm3
-; AVX1-NEXT: vpcmpeqd %xmm4, %xmm4, %xmm4
-; AVX1-NEXT: vpxor %xmm4, %xmm3, %xmm3
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
-; AVX1-NEXT: vpcmpeqb %xmm0, %xmm5, %xmm0
-; AVX1-NEXT: vpxor %xmm4, %xmm0, %xmm0
+; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4
+; AVX1-NEXT: vpcmpeqb %xmm0, %xmm4, %xmm0
; AVX1-NEXT: vpcmpeqb %xmm1, %xmm2, %xmm2
-; AVX1-NEXT: vpxor %xmm4, %xmm2, %xmm2
-; AVX1-NEXT: vpor %xmm2, %xmm3, %xmm2
+; AVX1-NEXT: vpand %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT: vpcmpeqb %xmm1, %xmm5, %xmm1
-; AVX1-NEXT: vpxor %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpor %xmm0, %xmm2, %xmm0
+; AVX1-NEXT: vpcmpeqb %xmm1, %xmm4, %xmm1
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm0, %xmm2, %xmm0
; AVX1-NEXT: vpmovmskb %xmm0, %eax
-; AVX1-NEXT: testl %eax, %eax
+; AVX1-NEXT: xorl $65535, %eax # imm = 0xFFFF
; AVX1-NEXT: sete %al
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: PR116977:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpcmpeqb %ymm0, %ymm2, %ymm0
-; AVX2-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX2-NEXT: vpxor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT: vpcmpeqb %ymm1, %ymm2, %ymm1
-; AVX2-NEXT: vpxor %ymm3, %ymm1, %ymm1
+; AVX2-NEXT: vpxor %ymm0, %ymm2, %ymm0
+; AVX2-NEXT: vpxor %ymm1, %ymm2, %ymm1
; AVX2-NEXT: vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT: vpmovmskb %ymm0, %eax
-; AVX2-NEXT: testl %eax, %eax
+; AVX2-NEXT: vptest %ymm0, %ymm0
; AVX2-NEXT: sete %al
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq