Skip to content

[DAGISel][ARM] Fix vector truncate combine for big-endian #118101

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 6 commits into from
Dec 4, 2024
Merged

[DAGISel][ARM] Fix vector truncate combine for big-endian #118101

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
7fb3d04
Add test showing bug
ostannard Nov 29, 2024
700a905
[DAGISel][ARM] Fix vector truncate combine for big-endian
ostannard Nov 29, 2024
804db3d
Remove unneeded target-features attribute
ostannard Nov 29, 2024
97ad28d
Merge branch 'main' into neon-big-endian
ostannard Dec 2, 2024
fc2f59a
Also test little-endian
ostannard Dec 3, 2024
3e7564b
Use existing isLE variable
ostannard Dec 4, 2024
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
5 changes: 4 additions & 1 deletion llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15495,12 +15495,15 @@ SDValue DAGCombiner::visitTRUNCATE(SDNode *N) {
unsigned BuildVecNumElts = BuildVect.getNumOperands();
unsigned TruncVecNumElts = VT.getVectorNumElements();
unsigned TruncEltOffset = BuildVecNumElts / TruncVecNumElts;
unsigned FirstElt =
DAG.getDataLayout().isBigEndian() ? (TruncEltOffset - 1) : 0;
Copy link
Collaborator

@RKSimon RKSimon Dec 3, 2024

Choose a reason for hiding this comment

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

visitTRUNCATE already has isLE - replace this with:

unsigned FirstElt =isLE ? 0 : (TruncEltOffset - 1);


assert((BuildVecNumElts % TruncVecNumElts) == 0 &&
"Invalid number of elements");

SmallVector<SDValue, 8> Opnds;
for (unsigned i = 0, e = BuildVecNumElts; i != e; i += TruncEltOffset)
for (unsigned i = FirstElt, e = BuildVecNumElts; i < e;
i += TruncEltOffset)
Opnds.push_back(BuildVect.getOperand(i));

return DAG.getBuildVector(VT, DL, Opnds);
Expand Down
31 changes: 31 additions & 0 deletions llvm/test/CodeGen/ARM/big-endian-vector-trunc.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=armebv7-unknown-none-eabihf -mattr=+neon < %s | FileCheck %s

define i32 @test(i64 %arg1) {
; CHECK-LABEL: test:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: subs r1, r1, #1
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: sbcs r0, r0, #0
; CHECK-NEXT: vldr s0, .LCPI0_0
; CHECK-NEXT: movwhs r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: vmov s1, r2
; CHECK-NEXT: vmovn.i32 d16, q0
; CHECK-NEXT: vmovn.i16 d16, q8
; CHECK-NEXT: vmov.u8 r0, d16[0]
; CHECK-NEXT: and r0, r0, #1
; CHECK-NEXT: bx lr
; CHECK-NEXT: .p2align 2
; CHECK-NEXT: @ %bb.1:
; CHECK-NEXT: .LCPI0_0:
; CHECK-NEXT: .long 0xffffffff @ float NaN
entry:
%insert_zero = insertelement <8 x i64> poison, i64 %arg1, i64 0
%splat_zero = shufflevector <8 x i64> %insert_zero, <8 x i64> poison, <8 x i32> zeroinitializer
%cmp_vec = icmp ule <8 x i64> <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>, %splat_zero
%first_cmp = extractelement <8 x i1> %cmp_vec, i32 0
%ext = zext i1 %first_cmp to i32
ret i32 %ext
}