Skip to content

[LoongArch] Fix broadcast load with extension. #155960

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 3 commits into from
Sep 1, 2025
Merged

[LoongArch] Fix broadcast load with extension. #155960

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
be1adea
Fix broadcast load with extension.
tangaac Aug 29, 2025
f78f7d3
address comments
tangaac Aug 29, 2025
dcaf590
address hev's comment
tangaac Sep 1, 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
9 changes: 6 additions & 3 deletions llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2516,11 +2516,14 @@ static SDValue lowerBUILD_VECTORAsBroadCastLoad(BuildVectorSDNode *BVOp,
}

// make sure that this load is valid and only has one user.
if (!IdentitySrc || !BVOp->isOnlyUserOf(IdentitySrc.getNode()))
if (!IsIdeneity || !IdentitySrc || !BVOp->isOnlyUserOf(IdentitySrc.getNode()))
return SDValue();

if (IsIdeneity) {
auto *LN = cast<LoadSDNode>(IdentitySrc);
auto *LN = cast<LoadSDNode>(IdentitySrc);
auto ExtType = LN->getExtensionType();

if ((ExtType == ISD::EXTLOAD || ExtType == ISD::NON_EXTLOAD) &&
VT.getScalarSizeInBits() == LN->getMemoryVT().getScalarSizeInBits()) {
SDVTList Tys =
LN->isIndexed()
? DAG.getVTList(VT, LN->getBasePtr().getValueType(), MVT::Other)
Expand Down
26 changes: 26 additions & 0 deletions llvm/test/CodeGen/LoongArch/lasx/broadcast-load.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,32 @@ define <4 x i64> @should_not_be_optimized(ptr %ptr, ptr %dst) {
ret <4 x i64> %tmp2
}

define <16 x i16> @should_not_be_optimized_sext_load(ptr %ptr) {
; CHECK-LABEL: should_not_be_optimized_sext_load:
; CHECK: # %bb.0:
; CHECK-NEXT: ld.b $a0, $a0, 0
; CHECK-NEXT: xvreplgr2vr.h $xr0, $a0
; CHECK-NEXT: ret
%tmp = load i8, ptr %ptr
%tmp1 = sext i8 %tmp to i16
%tmp2 = insertelement <16 x i16> zeroinitializer, i16 %tmp1, i32 0
%tmp3 = shufflevector <16 x i16> %tmp2, <16 x i16> poison, <16 x i32> zeroinitializer
ret <16 x i16> %tmp3
}

define <16 x i16> @should_not_be_optimized_zext_load(ptr %ptr) {
; CHECK-LABEL: should_not_be_optimized_zext_load:
; CHECK: # %bb.0:
; CHECK-NEXT: ld.bu $a0, $a0, 0
; CHECK-NEXT: xvreplgr2vr.h $xr0, $a0
; CHECK-NEXT: ret
%tmp = load i8, ptr %ptr
%tmp1 = zext i8 %tmp to i16
%tmp2 = insertelement <16 x i16> zeroinitializer, i16 %tmp1, i32 0
%tmp3 = shufflevector <16 x i16> %tmp2, <16 x i16> poison, <16 x i32> zeroinitializer
ret <16 x i16> %tmp3
}

define <4 x i64> @xvldrepl_d_unaligned_offset(ptr %ptr) {
; LA32-LABEL: xvldrepl_d_unaligned_offset:
; LA32: # %bb.0:
Expand Down
28 changes: 26 additions & 2 deletions llvm/test/CodeGen/LoongArch/lsx/broadcast-load.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,8 +2,6 @@
; RUN: llc --mtriple=loongarch32 -mattr=+32s,+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA32
; RUN: llc --mtriple=loongarch64 -mattr=+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA64

; TODO: Load a element and splat it to a vector could be lowerd to vldrepl

; A load has more than one user shouldn't be lowered to vldrepl
define <2 x i64> @should_not_be_optimized(ptr %ptr, ptr %dst){
; LA32-LABEL: should_not_be_optimized:
Expand Down Expand Up @@ -31,6 +29,32 @@ define <2 x i64> @should_not_be_optimized(ptr %ptr, ptr %dst){
ret <2 x i64> %tmp2
}

define <8 x i16> @should_not_be_optimized_sext_load(ptr %ptr) {
; CHECK-LABEL: should_not_be_optimized_sext_load:
; CHECK: # %bb.0:
; CHECK-NEXT: ld.b $a0, $a0, 0
; CHECK-NEXT: vreplgr2vr.h $vr0, $a0
; CHECK-NEXT: ret
%tmp = load i8, ptr %ptr
%tmp1 = sext i8 %tmp to i16
%tmp2 = insertelement <8 x i16> zeroinitializer, i16 %tmp1, i32 0
%tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> poison, <8 x i32> zeroinitializer
ret <8 x i16> %tmp3
}

define <8 x i16> @should_not_be_optimized_zext_load(ptr %ptr) {
; CHECK-LABEL: should_not_be_optimized_zext_load:
; CHECK: # %bb.0:
; CHECK-NEXT: ld.bu $a0, $a0, 0
; CHECK-NEXT: vreplgr2vr.h $vr0, $a0
; CHECK-NEXT: ret
%tmp = load i8, ptr %ptr
%tmp1 = zext i8 %tmp to i16
%tmp2 = insertelement <8 x i16> zeroinitializer, i16 %tmp1, i32 0
%tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> poison, <8 x i32> zeroinitializer
ret <8 x i16> %tmp3
}

define <2 x i64> @vldrepl_d_unaligned_offset(ptr %ptr) {
; LA32-LABEL: vldrepl_d_unaligned_offset:
; LA32: # %bb.0:
Expand Down