Skip to content

[LLVM][CodeGen][SVE] List MVTs that are desirable for extending loads. #149153

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jul 18, 2025
Merged

[LLVM][CodeGen][SVE] List MVTs that are desirable for extending loads. #149153

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ba77472
[LLVM][CodeGen][SVE] List MVTs that are desirable for extending loads.
paulwalker-arm Jul 16, 2025
776745b
Remove nested SDValue constructor calls.
paulwalker-arm Jul 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7633,7 +7633,7 @@ SDValue DAGCombiner::visitAND(SDNode *N) {

if (SDValue(GN0, 0).hasOneUse() &&
isConstantSplatVectorMaskForType(N1.getNode(), ScalarVT) &&
TLI.isVectorLoadExtDesirable(SDValue(SDValue(GN0, 0)))) {
TLI.isVectorLoadExtDesirable(SDValue(N, 0))) {
Copy link
Contributor

@david-arm david-arm Jul 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm surprised that fixing this hasn't affected any existing tests! If I understand correctly, the argument passed to isVectorLoadExtDesirable should be the extended value - do you know why we need to wrap this in two SDValues?

Copy link
Collaborator Author

@paulwalker-arm paulwalker-arm Jul 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I needed to fix this to maintain the existing tests. Without this fix most of the sve-masked-gather-*.ll tests started to fail. I guess the nested calls to SDValues was just a typo so have removed them.

SDValue Ops[] = {GN0->getChain(), GN0->getPassThru(), GN0->getMask(),
GN0->getBasePtr(), GN0->getIndex(), GN0->getScale()};

Expand Down Expand Up @@ -15724,7 +15724,7 @@ SDValue DAGCombiner::visitSIGN_EXTEND_INREG(SDNode *N) {
// fold (sext_inreg (masked_gather x)) -> (sext_masked_gather x)
if (auto *GN0 = dyn_cast<MaskedGatherSDNode>(N0)) {
if (SDValue(GN0, 0).hasOneUse() && ExtVT == GN0->getMemoryVT() &&
TLI.isVectorLoadExtDesirable(SDValue(SDValue(GN0, 0)))) {
TLI.isVectorLoadExtDesirable(SDValue(N, 0))) {
SDValue Ops[] = {GN0->getChain(), GN0->getPassThru(), GN0->getMask(),
GN0->getBasePtr(), GN0->getIndex(), GN0->getScale()};

Expand Down
4 changes: 3 additions & 1 deletion llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6439,7 +6439,9 @@ bool AArch64TargetLowering::isVectorLoadExtDesirable(SDValue ExtVal) const {
}
}

return true;
EVT PreExtScalarVT = ExtVal->getOperand(0).getValueType().getScalarType();
return PreExtScalarVT == MVT::i8 || PreExtScalarVT == MVT::i16 ||
PreExtScalarVT == MVT::i32 || PreExtScalarVT == MVT::i64;
}

unsigned getGatherVecOpcode(bool IsScaled, bool IsSigned, bool NeedsExtend) {
Expand Down
24 changes: 24 additions & 0 deletions llvm/test/CodeGen/AArch64/sve-intrinsics-ldst-ext.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,3 +231,27 @@ define <vscale x 8 x i64> @sload_8i8_8i64(ptr %a) {
%aext = sext <vscale x 8 x i8> %aval to <vscale x 8 x i64>
ret <vscale x 8 x i64> %aext
}

; Ensure we don't try to promote a predicate load to a sign-extended load.
define <vscale x 16 x i8> @sload_16i1_16i8(ptr %addr) {
; CHECK-LABEL: sload_16i1_16i8:
; CHECK: // %bb.0:
; CHECK-NEXT: ldr p0, [x0]
; CHECK-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
; CHECK-NEXT: ret
%load = load <vscale x 16 x i1>, ptr %addr
%zext = sext <vscale x 16 x i1> %load to <vscale x 16 x i8>
ret <vscale x 16 x i8> %zext
}

; Ensure we don't try to promote a predicate load to a zero-extended load.
define <vscale x 16 x i8> @zload_16i1_16i8(ptr %addr) {
; CHECK-LABEL: zload_16i1_16i8:
; CHECK: // %bb.0:
; CHECK-NEXT: ldr p0, [x0]
; CHECK-NEXT: mov z0.b, p0/z, #1 // =0x1
; CHECK-NEXT: ret
%load = load <vscale x 16 x i1>, ptr %addr
%zext = zext <vscale x 16 x i1> %load to <vscale x 16 x i8>
ret <vscale x 16 x i8> %zext
}