[X86] combineSelect - move vselect(cond, pshufb(x), pshufb(y)) -> or(pshufb(x), pshufb(y)) fold #145475
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…(pshufb(x), pshufb(y)) fold Move the OR(PSHUFB(),PSHUFB()) fold to reuse the createShuffleMaskFromVSELECT result and ensure it is performed before the combineX86ShufflesRecursively combine to prevent some hasOneUse failures noticed in llvm#133947 (combineX86ShufflesRecursively still unnecessarily widens vectors in several locations).
|
@llvm/pr-subscribers-backend-x86 Author: Simon Pilgrim (RKSimon) ChangesMove the OR(PSHUFB(),PSHUFB()) fold to reuse an existing createShuffleMaskFromVSELECT result and ensure it is performed before the combineX86ShufflesRecursively combine to prevent some hasOneUse failures noticed in #133947 (combineX86ShufflesRecursively still unnecessarily widens vectors in several locations). Full diff: https://github.com/llvm/llvm-project/pull/145475.diff 1 Files Affected:
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 2541182de1208..d5837ab938d4e 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -47690,12 +47690,47 @@ static SDValue combineSelect(SDNode *N, SelectionDAG &DAG,
return V;
if (N->getOpcode() == ISD::VSELECT || N->getOpcode() == X86ISD::BLENDV) {
- SmallVector<int, 64> Mask;
- if (createShuffleMaskFromVSELECT(Mask, Cond,
+ SmallVector<int, 64> CondMask;
+ if (createShuffleMaskFromVSELECT(CondMask, Cond,
N->getOpcode() == X86ISD::BLENDV)) {
// Convert vselects with constant condition into shuffles.
if (DCI.isBeforeLegalizeOps())
- return DAG.getVectorShuffle(VT, DL, LHS, RHS, Mask);
+ return DAG.getVectorShuffle(VT, DL, LHS, RHS, CondMask);
+
+ // fold vselect(cond, pshufb(x), pshufb(y)) -> or (pshufb(x), pshufb(y))
+ // by forcing the unselected elements to zero.
+ // TODO: Can we handle more shuffles with this?
+ if (LHS.hasOneUse() && RHS.hasOneUse()) {
+ SmallVector<SDValue, 1> LHSOps, RHSOps;
+ SmallVector<int, 64> LHSMask, RHSMask, ByteMask;
+ SDValue LHSShuf = peekThroughOneUseBitcasts(LHS);
+ SDValue RHSShuf = peekThroughOneUseBitcasts(RHS);
+ if (LHSShuf.getOpcode() == X86ISD::PSHUFB &&
+ RHSShuf.getOpcode() == X86ISD::PSHUFB &&
+ scaleShuffleMaskElts(VT.getSizeInBits() / 8, CondMask, ByteMask) &&
+ getTargetShuffleMask(LHSShuf, true, LHSOps, LHSMask) &&
+ getTargetShuffleMask(RHSShuf, true, RHSOps, RHSMask)) {
+ assert(ByteMask.size() == LHSMask.size() &&
+ ByteMask.size() == RHSMask.size() && "Shuffle mask mismatch");
+ for (auto [I, M] : enumerate(ByteMask)) {
+ // getConstVector sets negative shuffle mask values as undef, so
+ // ensure we hardcode SM_SentinelZero values to zero (0x80).
+ if (M < (int)ByteMask.size()) {
+ LHSMask[I] = isUndefOrZero(LHSMask[I]) ? 0x80 : LHSMask[I];
+ RHSMask[I] = 0x80;
+ } else {
+ LHSMask[I] = 0x80;
+ RHSMask[I] = isUndefOrZero(RHSMask[I]) ? 0x80 : RHSMask[I];
+ }
+ }
+ MVT ByteVT = LHSShuf.getSimpleValueType();
+ LHS = DAG.getNode(X86ISD::PSHUFB, DL, ByteVT, LHSOps[0],
+ getConstVector(LHSMask, ByteVT, DAG, DL, true));
+ RHS = DAG.getNode(X86ISD::PSHUFB, DL, ByteVT, RHSOps[0],
+ getConstVector(RHSMask, ByteVT, DAG, DL, true));
+ return DAG.getBitcast(VT, DAG.getNode(ISD::OR, DL, ByteVT, LHS, RHS));
+ }
+ }
// Attempt to combine as shuffle.
SDValue Op(N, 0);
@@ -47704,43 +47739,6 @@ static SDValue combineSelect(SDNode *N, SelectionDAG &DAG,
}
}
- // fold vselect(cond, pshufb(x), pshufb(y)) -> or (pshufb(x), pshufb(y))
- // by forcing the unselected elements to zero.
- // TODO: Can we handle more shuffles with this?
- if (N->getOpcode() == ISD::VSELECT && CondVT.isVector() && LHS.hasOneUse() &&
- RHS.hasOneUse()) {
- SmallVector<SDValue, 1> LHSOps, RHSOps;
- SmallVector<int, 64> LHSMask, RHSMask, CondMask, ByteMask;
- SDValue LHSShuf = peekThroughOneUseBitcasts(LHS);
- SDValue RHSShuf = peekThroughOneUseBitcasts(RHS);
- if (LHSShuf.getOpcode() == X86ISD::PSHUFB &&
- RHSShuf.getOpcode() == X86ISD::PSHUFB &&
- createShuffleMaskFromVSELECT(CondMask, Cond) &&
- scaleShuffleMaskElts(VT.getSizeInBits() / 8, CondMask, ByteMask) &&
- getTargetShuffleMask(LHSShuf, true, LHSOps, LHSMask) &&
- getTargetShuffleMask(RHSShuf, true, RHSOps, RHSMask)) {
- assert(ByteMask.size() == LHSMask.size() &&
- ByteMask.size() == RHSMask.size() && "Shuffle mask mismatch");
- for (auto [I, M] : enumerate(ByteMask)) {
- // getConstVector sets negative shuffle mask values as undef, so ensure
- // we hardcode SM_SentinelZero values to zero (0x80).
- if (M < (int)ByteMask.size()) {
- LHSMask[I] = isUndefOrZero(LHSMask[I]) ? 0x80 : LHSMask[I];
- RHSMask[I] = 0x80;
- } else {
- LHSMask[I] = 0x80;
- RHSMask[I] = isUndefOrZero(RHSMask[I]) ? 0x80 : RHSMask[I];
- }
- }
- MVT ByteVT = LHSShuf.getSimpleValueType();
- LHS = DAG.getNode(X86ISD::PSHUFB, DL, ByteVT, LHSOps[0],
- getConstVector(LHSMask, ByteVT, DAG, DL, true));
- RHS = DAG.getNode(X86ISD::PSHUFB, DL, ByteVT, RHSOps[0],
- getConstVector(RHSMask, ByteVT, DAG, DL, true));
- return DAG.getBitcast(VT, DAG.getNode(ISD::OR, DL, ByteVT, LHS, RHS));
- }
- }
-
// If we have SSE[12] support, try to form min/max nodes. SSE min/max
// instructions match the semantics of the common C idiom x<y?x:y but not
// x<=y?x:y, because of how they handle negative zero (which can be
|
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/123/builds/21900 Here is the relevant piece of the build log for the reference |
DrSergei
pushed a commit
to DrSergei/llvm-project
that referenced
this pull request
Jun 24, 2025
…pshufb(x), pshufb(y)) fold (llvm#145475) Move the OR(PSHUFB(),PSHUFB()) fold to reuse an existing createShuffleMaskFromVSELECT result and ensure it is performed before the combineX86ShufflesRecursively combine to prevent some hasOneUse failures noticed in llvm#133947 (combineX86ShufflesRecursively still unnecessarily widens vectors in several locations).
anthonyhatran
pushed a commit
to anthonyhatran/llvm-project
that referenced
this pull request
Jun 26, 2025
…pshufb(x), pshufb(y)) fold (llvm#145475) Move the OR(PSHUFB(),PSHUFB()) fold to reuse an existing createShuffleMaskFromVSELECT result and ensure it is performed before the combineX86ShufflesRecursively combine to prevent some hasOneUse failures noticed in llvm#133947 (combineX86ShufflesRecursively still unnecessarily widens vectors in several locations).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Move the OR(PSHUFB(),PSHUFB()) fold to reuse an existing createShuffleMaskFromVSELECT result and ensure it is performed before the combineX86ShufflesRecursively combine to prevent some hasOneUse failures noticed in #133947 (combineX86ShufflesRecursively still unnecessarily widens vectors in several locations).