Thanks to visit codestin.com
Credit goes to github.com

Skip to content

[VectorCombine] Shrink loads used in shufflevector rebroadcasts #128938

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 23 commits into from
Aug 4, 2025
Merged

[VectorCombine] Shrink loads used in shufflevector rebroadcasts #128938

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
23 commits
Select commit Hold shift + click to select a range
1d817c3
[AggressiveInstCombine] Shrink loads used in shufflevector rebroadcasts.
Feb 26, 2025
09ac59a
Add implementation and update tests.
Feb 26, 2025
bc8e5ce
Fix broken tests.
Feb 28, 2025
6f6fc11
Ignore non-simple loads.
Mar 4, 2025
7499a48
Move transform to VectorCombine and update tests.
Mar 12, 2025
c86d296
Address review comments and update tests.
Mar 14, 2025
16f7320
Code formatting.
Mar 14, 2025
dfe3459
Add cost analysis for shufflevector ops and update tests.
Mar 21, 2025
125c3f2
Code formatting.
Mar 21, 2025
3afabd1
Address comments.
Mar 24, 2025
875c408
Fix conflict with subvector load widening.
Mar 26, 2025
839f7bd
Fix build bot false positive for undef.
Mar 26, 2025
9518894
Address comments.
May 7, 2025
016e9a5
Address review comments.
May 19, 2025
76450c9
Address comments.
May 30, 2025
dcccc92
Formatting.
May 30, 2025
fa2ebe8
Correct renaming and remove function-style casts.
May 30, 2025
863f10f
Remove auto.
Jun 11, 2025
5f122f6
Fix cost analysis after rebase.
Jul 8, 2025
0373b90
Add debug message.
Jul 9, 2025
204d4bf
Address comments.
Jul 16, 2025
bceb357
Remove redundant code and update tests.
Jul 16, 2025
6af4449
Merge branch 'main' into shrink_loads
RKSimon Aug 4, 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
Prev Previous commit
Next Next commit
Ignore non-simple loads.
  • Loading branch information
Leon Clark committed Jul 8, 2025
commit 6f6fc11f7a4d300a8dd281826533bf5d057d8131
4 changes: 2 additions & 2 deletions clang/test/CodeGen/SystemZ/builtins-systemz-zvector-constrained.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,8 +79,8 @@ void test_core(void) {
vec_xstd2(vd, idx, ptrd);

vd = vec_splat(vd, 0);
// CHECK: shufflevector <1 x double> %{{.*}}, <1 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vlrepg
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vd = vec_splat(vd, 1);
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vrepg
Expand Down
52 changes: 26 additions & 26 deletions clang/test/CodeGen/SystemZ/builtins-systemz-zvector.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -777,80 +777,80 @@ void test_core(void) {
// CHECK: <2 x i64> splat (i64 -4503582447501313)

vsc = vec_splat(vsc, 0);
// CHECK: shufflevector <1 x i8> %{{.*}}, <1 x i8> poison, <16 x i32> zeroinitializer
// CHECK-ASM: vlrepb
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> poison, <16 x i32> zeroinitializer
// CHECK-ASM: vrepb
vsc = vec_splat(vsc, 15);
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> poison, <16 x i32> <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
// CHECK-ASM: vrepb
vuc = vec_splat(vuc, 0);
// CHECK: store volatile <16 x i8> splat (i8 {{.*}}), ptr @vuc
// CHECK-ASM: vst
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> poison, <16 x i32> zeroinitializer
// CHECK-ASM: vrepb
vuc = vec_splat(vuc, 15);
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> poison, <16 x i32> <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
// CHECK-ASM: vrepb
vbc = vec_splat(vbc, 0);
// CHECK: shufflevector <1 x i8> %{{.*}}, <1 x i8> poison, <16 x i32> zeroinitializer
// CHECK-ASM: vlrepb
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> poison, <16 x i32> zeroinitializer
// CHECK-ASM: vrepb
vbc = vec_splat(vbc, 15);
// CHECK: shufflevector <16 x i8> %{{.*}}, <16 x i8> poison, <16 x i32> <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
// CHECK-ASM: vrepb
vss = vec_splat(vss, 0);
// CHECK: shufflevector <1 x i16> %{{.*}}, <1 x i16> poison, <8 x i32> zeroinitializer
// CHECK-ASM: vlreph
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <8 x i32> zeroinitializer
// CHECK-ASM: vreph
vss = vec_splat(vss, 7);
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
// CHECK-ASM: vreph
vus = vec_splat(vus, 0);
// CHECK: store volatile <8 x i16> splat (i16 {{.*}}), ptr @vus
// CHECK-ASM: vst
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <8 x i32> zeroinitializer
// CHECK-ASM: vreph
vus = vec_splat(vus, 7);
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
// CHECK-ASM: vreph
vbs = vec_splat(vbs, 0);
// CHECK: shufflevector <1 x i16> %{{.*}}, <1 x i16> poison, <8 x i32> zeroinitializer
// CHECK-ASM: vlreph
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <8 x i32> zeroinitializer
// CHECK-ASM: vreph
vbs = vec_splat(vbs, 7);
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
// CHECK-ASM: vreph
vsi = vec_splat(vsi, 0);
// CHECK: shufflevector <1 x i32> %{{.*}}, <1 x i32> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vlrepf
// CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vrepf
vsi = vec_splat(vsi, 3);
// CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
// CHECK-ASM: vrepf
vui = vec_splat(vui, 0);
// CHECK: store volatile <4 x i32> splat (i32 {{.*}}), ptr @vui
// CHECK-ASM: vst
// CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vrepf
vui = vec_splat(vui, 3);
// CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
// CHECK-ASM: vrepf
vbi = vec_splat(vbi, 0);
// CHECK: shufflevector <1 x i32> %{{.*}}, <1 x i32> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vlrepf
// CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vrepf
vbi = vec_splat(vbi, 3);
// CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
// CHECK-ASM: vrepf
vsl = vec_splat(vsl, 0);
// CHECK: shufflevector <1 x i64> %{{.*}}, <1 x i64> poison, <2 x i32> zeroinitializer
// CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vsl = vec_splat(vsl, 1);
// CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vst
// CHECK-ASM: vrepg
vul = vec_splat(vul, 0);
// CHECK: store volatile <2 x i64> splat (i64 {{.*}}), ptr @vul
// CHECK-ASM: vst
// CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vul = vec_splat(vul, 1);
// CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vrepg
vbl = vec_splat(vbl, 0);
// CHECK: shufflevector <1 x i64> %{{.*}}, <1 x i64> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vlrepg
// CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vbl = vec_splat(vbl, 1);
// CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vrepg
vd = vec_splat(vd, 0);
// CHECK: shufflevector <1 x double> %{{.*}}, <1 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vlrepg
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vd = vec_splat(vd, 1);
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vrepg
Expand Down
12 changes: 6 additions & 6 deletions clang/test/CodeGen/SystemZ/builtins-systemz-zvector2-constrained.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,14 +130,14 @@ void test_core(void) {
// CHECK-ASM: vst

vf = vec_splat(vf, 0);
// CHECK: shufflevector <1 x float> %{{.*}}, <1 x float> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vlrepf
// CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vrepf
vf = vec_splat(vf, 1);
// CHECK: shufflevector <2 x float> %{{.*}}, <2 x float> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
// CHECK-ASM: vst
// CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
// CHECK-ASM: vrepf
vd = vec_splat(vd, 0);
// CHECK: shufflevector <1 x double> %{{.*}}, <1 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vlrepg
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vd = vec_splat(vd, 1);
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vrepg
Expand Down
12 changes: 6 additions & 6 deletions clang/test/CodeGen/SystemZ/builtins-systemz-zvector2.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -254,14 +254,14 @@ void test_core(void) {
// CHECK-ASM: vstrlr

vf = vec_splat(vf, 0);
// CHECK: shufflevector <1 x float> %{{.*}}, <1 x float> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vlrepf
// CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> poison, <4 x i32> zeroinitializer
// CHECK-ASM: vrepf
vf = vec_splat(vf, 1);
// CHECK: shufflevector <2 x float> %{{.*}}, <2 x float> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
// CHECK-ASM: vst
// CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> poison, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
// CHECK-ASM: vrepf
vd = vec_splat(vd, 0);
// CHECK: shufflevector <1 x double> %{{.*}}, <1 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vlrepg
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> zeroinitializer
// CHECK-ASM: vrepg
vd = vec_splat(vd, 1);
// CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> poison, <2 x i32> <i32 1, i32 1>
// CHECK-ASM: vrepg
Expand Down
2 changes: 1 addition & 1 deletion llvm/lib/Transforms/AggressiveInstCombine/AggressiveInstCombine.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -991,7 +991,7 @@ static bool foldPatternedLoads(Instruction &I, const DataLayout &DL) {
// poison values, attempt to shrink the load to only the lanes being used.
static bool shrinkLoadsForBroadcast(Instruction &I) {
auto *OldLoad = dyn_cast<LoadInst>(&I);
if (!OldLoad)
if (!OldLoad || !OldLoad->isSimple())
return false;

auto *VecTy = dyn_cast<FixedVectorType>(I.getType());
Expand Down
14 changes: 14 additions & 0 deletions llvm/test/Transforms/AggressiveInstCombine/load-shufflevector.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,20 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -passes=aggressive-instcombine -S < %s | FileCheck %s

define <8 x half> @shuffle_v4_v8f16_r0_1_volatile(ptr addrspace(1) nocapture readonly %arg0) local_unnamed_addr {
; CHECK-LABEL: define <8 x half> @shuffle_v4_v8f16_r0_1_volatile(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG0:%.*]]) local_unnamed_addr {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TMP0:%.*]] = load volatile <4 x half>, ptr addrspace(1) [[ARG0]], align 32
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x half> [[TMP0]], <4 x half> poison, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
; CHECK-NEXT: ret <8 x half> [[TMP1]]
;
entry:
%val0 = load volatile <4 x half>, ptr addrspace(1) %arg0, align 32
%val1 = shufflevector <4 x half> %val0, <4 x half> poison, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
ret <8 x half> %val1
}

define <8 x half> @shuffle_v4_v8f16_r0_1(ptr addrspace(1) nocapture readonly %arg0) local_unnamed_addr {
; CHECK-LABEL: define <8 x half> @shuffle_v4_v8f16_r0_1(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG0:%.*]]) local_unnamed_addr {
Expand Down