Skip to content

Commit 2afe584

Browse files
authored
Reapply "[Scalarizer] Make *_with_overflow intrinsics scalarizable (llvm#126815)" (llvm#1031)
This reverts commit 746f40f. The underlying problem was fixed by 1440f02 "[Scalarizer] Ensure valid VectorSplits for each struct element in `visitExtractValueInst` (llvm#128538)" Fixes: SC1-6029
1 parent 2b15af1 commit 2afe584

File tree

8 files changed

+150
-16
lines changed

8 files changed

+150
-16
lines changed

llvm/lib/Analysis/VectorUtils.cpp

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -128,6 +128,12 @@ bool llvm::isTriviallyScalarizable(Intrinsic::ID ID,
128128
// https://github.com/llvm/llvm-project/issues/112408
129129
switch (ID) {
130130
case Intrinsic::frexp:
131+
case Intrinsic::uadd_with_overflow:
132+
case Intrinsic::sadd_with_overflow:
133+
case Intrinsic::ssub_with_overflow:
134+
case Intrinsic::usub_with_overflow:
135+
case Intrinsic::umul_with_overflow:
136+
case Intrinsic::smul_with_overflow:
131137
return true;
132138
}
133139
return false;
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt %s -passes="function(scalarizer)" -S | FileCheck %s
3+
4+
define <3 x i32> @test_(<3 x i32> %a, <3 x i32> %b) {
5+
; CHECK-LABEL: define <3 x i32> @test_(
6+
; CHECK-SAME: <3 x i32> [[A:%.*]], <3 x i32> [[B:%.*]]) {
7+
; CHECK-NEXT: [[B_I0:%.*]] = extractelement <3 x i32> [[B]], i64 0
8+
; CHECK-NEXT: [[R_I0:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[B_I0]], i32 [[B_I0]])
9+
; CHECK-NEXT: [[B_I1:%.*]] = extractelement <3 x i32> [[B]], i64 1
10+
; CHECK-NEXT: [[R_I1:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[B_I1]], i32 [[B_I1]])
11+
; CHECK-NEXT: [[B_I2:%.*]] = extractelement <3 x i32> [[B]], i64 2
12+
; CHECK-NEXT: [[R_I2:%.*]] = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[B_I2]], i32 [[B_I2]])
13+
; CHECK-NEXT: [[EL_ELEM0:%.*]] = extractvalue { i32, i1 } [[R_I0]], 0
14+
; CHECK-NEXT: [[EL_ELEM01:%.*]] = extractvalue { i32, i1 } [[R_I1]], 0
15+
; CHECK-NEXT: [[EL_ELEM02:%.*]] = extractvalue { i32, i1 } [[R_I2]], 0
16+
; CHECK-NEXT: [[EL_UPTO0:%.*]] = insertelement <3 x i32> poison, i32 [[EL_ELEM0]], i64 0
17+
; CHECK-NEXT: [[EL_UPTO1:%.*]] = insertelement <3 x i32> [[EL_UPTO0]], i32 [[EL_ELEM01]], i64 1
18+
; CHECK-NEXT: [[EL:%.*]] = insertelement <3 x i32> [[EL_UPTO1]], i32 [[EL_ELEM02]], i64 2
19+
; CHECK-NEXT: ret <3 x i32> [[EL]]
20+
;
21+
%r = call { <3 x i32>, <3 x i1> } @llvm.sadd.with.overflow.v3i32(<3 x i32> %b, <3 x i32> %b)
22+
%el = extractvalue { <3 x i32>, <3 x i1> } %r, 0
23+
ret <3 x i32> %el
24+
}
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt %s -passes="function(scalarizer)" -S | FileCheck %s
3+
4+
define <3 x i32> @test_(<3 x i32> %a, <3 x i32> %b) {
5+
; CHECK-LABEL: define <3 x i32> @test_(
6+
; CHECK-SAME: <3 x i32> [[A:%.*]], <3 x i32> [[B:%.*]]) {
7+
; CHECK-NEXT: [[B_I0:%.*]] = extractelement <3 x i32> [[B]], i64 0
8+
; CHECK-NEXT: [[R_I0:%.*]] = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 [[B_I0]], i32 [[B_I0]])
9+
; CHECK-NEXT: [[B_I1:%.*]] = extractelement <3 x i32> [[B]], i64 1
10+
; CHECK-NEXT: [[R_I1:%.*]] = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 [[B_I1]], i32 [[B_I1]])
11+
; CHECK-NEXT: [[B_I2:%.*]] = extractelement <3 x i32> [[B]], i64 2
12+
; CHECK-NEXT: [[R_I2:%.*]] = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 [[B_I2]], i32 [[B_I2]])
13+
; CHECK-NEXT: [[EL_ELEM0:%.*]] = extractvalue { i32, i1 } [[R_I0]], 0
14+
; CHECK-NEXT: [[EL_ELEM01:%.*]] = extractvalue { i32, i1 } [[R_I1]], 0
15+
; CHECK-NEXT: [[EL_ELEM02:%.*]] = extractvalue { i32, i1 } [[R_I2]], 0
16+
; CHECK-NEXT: [[EL_UPTO0:%.*]] = insertelement <3 x i32> poison, i32 [[EL_ELEM0]], i64 0
17+
; CHECK-NEXT: [[EL_UPTO1:%.*]] = insertelement <3 x i32> [[EL_UPTO0]], i32 [[EL_ELEM01]], i64 1
18+
; CHECK-NEXT: [[EL:%.*]] = insertelement <3 x i32> [[EL_UPTO1]], i32 [[EL_ELEM02]], i64 2
19+
; CHECK-NEXT: ret <3 x i32> [[EL]]
20+
;
21+
%r = call { <3 x i32>, <3 x i1> } @llvm.smul.with.overflow.v3i32(<3 x i32> %b, <3 x i32> %b)
22+
%el = extractvalue { <3 x i32>, <3 x i1> } %r, 0
23+
ret <3 x i32> %el
24+
}
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt %s -passes="function(scalarizer)" -S | FileCheck %s
3+
4+
define <3 x i32> @test_(<3 x i32> %a, <3 x i32> %b) {
5+
; CHECK-LABEL: define <3 x i32> @test_(
6+
; CHECK-SAME: <3 x i32> [[A:%.*]], <3 x i32> [[B:%.*]]) {
7+
; CHECK-NEXT: [[B_I0:%.*]] = extractelement <3 x i32> [[B]], i64 0
8+
; CHECK-NEXT: [[R_I0:%.*]] = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 [[B_I0]], i32 [[B_I0]])
9+
; CHECK-NEXT: [[B_I1:%.*]] = extractelement <3 x i32> [[B]], i64 1
10+
; CHECK-NEXT: [[R_I1:%.*]] = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 [[B_I1]], i32 [[B_I1]])
11+
; CHECK-NEXT: [[B_I2:%.*]] = extractelement <3 x i32> [[B]], i64 2
12+
; CHECK-NEXT: [[R_I2:%.*]] = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 [[B_I2]], i32 [[B_I2]])
13+
; CHECK-NEXT: [[EL_ELEM0:%.*]] = extractvalue { i32, i1 } [[R_I0]], 0
14+
; CHECK-NEXT: [[EL_ELEM01:%.*]] = extractvalue { i32, i1 } [[R_I1]], 0
15+
; CHECK-NEXT: [[EL_ELEM02:%.*]] = extractvalue { i32, i1 } [[R_I2]], 0
16+
; CHECK-NEXT: [[EL_UPTO0:%.*]] = insertelement <3 x i32> poison, i32 [[EL_ELEM0]], i64 0
17+
; CHECK-NEXT: [[EL_UPTO1:%.*]] = insertelement <3 x i32> [[EL_UPTO0]], i32 [[EL_ELEM01]], i64 1
18+
; CHECK-NEXT: [[EL:%.*]] = insertelement <3 x i32> [[EL_UPTO1]], i32 [[EL_ELEM02]], i64 2
19+
; CHECK-NEXT: ret <3 x i32> [[EL]]
20+
;
21+
%r = call { <3 x i32>, <3 x i1> } @llvm.ssub.with.overflow.v3i32(<3 x i32> %b, <3 x i32> %b)
22+
%el = extractvalue { <3 x i32>, <3 x i1> } %r, 0
23+
ret <3 x i32> %el
24+
}

llvm/test/Transforms/Scalarizer/uadd_overflow.ll

Lines changed: 0 additions & 16 deletions
This file was deleted.
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt %s -passes="function(scalarizer)" -S | FileCheck %s
3+
4+
define <3 x i32> @test_(<3 x i32> %a, <3 x i32> %b) {
5+
; CHECK-LABEL: define <3 x i32> @test_(
6+
; CHECK-SAME: <3 x i32> [[A:%.*]], <3 x i32> [[B:%.*]]) {
7+
; CHECK-NEXT: [[B_I0:%.*]] = extractelement <3 x i32> [[B]], i64 0
8+
; CHECK-NEXT: [[R_I0:%.*]] = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 [[B_I0]], i32 [[B_I0]])
9+
; CHECK-NEXT: [[B_I1:%.*]] = extractelement <3 x i32> [[B]], i64 1
10+
; CHECK-NEXT: [[R_I1:%.*]] = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 [[B_I1]], i32 [[B_I1]])
11+
; CHECK-NEXT: [[B_I2:%.*]] = extractelement <3 x i32> [[B]], i64 2
12+
; CHECK-NEXT: [[R_I2:%.*]] = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 [[B_I2]], i32 [[B_I2]])
13+
; CHECK-NEXT: [[EL_ELEM0:%.*]] = extractvalue { i32, i1 } [[R_I0]], 0
14+
; CHECK-NEXT: [[EL_ELEM01:%.*]] = extractvalue { i32, i1 } [[R_I1]], 0
15+
; CHECK-NEXT: [[EL_ELEM02:%.*]] = extractvalue { i32, i1 } [[R_I2]], 0
16+
; CHECK-NEXT: [[EL_UPTO0:%.*]] = insertelement <3 x i32> poison, i32 [[EL_ELEM0]], i64 0
17+
; CHECK-NEXT: [[EL_UPTO1:%.*]] = insertelement <3 x i32> [[EL_UPTO0]], i32 [[EL_ELEM01]], i64 1
18+
; CHECK-NEXT: [[EL:%.*]] = insertelement <3 x i32> [[EL_UPTO1]], i32 [[EL_ELEM02]], i64 2
19+
; CHECK-NEXT: ret <3 x i32> [[EL]]
20+
;
21+
%r = call { <3 x i32>, <3 x i1> } @llvm.uadd.with.overflow.v3i32(<3 x i32> %b, <3 x i32> %b)
22+
%el = extractvalue { <3 x i32>, <3 x i1> } %r, 0
23+
ret <3 x i32> %el
24+
}
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt %s -passes="function(scalarizer)" -S | FileCheck %s
3+
4+
define <3 x i32> @test_(<3 x i32> %a, <3 x i32> %b) {
5+
; CHECK-LABEL: define <3 x i32> @test_(
6+
; CHECK-SAME: <3 x i32> [[A:%.*]], <3 x i32> [[B:%.*]]) {
7+
; CHECK-NEXT: [[B_I0:%.*]] = extractelement <3 x i32> [[B]], i64 0
8+
; CHECK-NEXT: [[R_I0:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[B_I0]], i32 [[B_I0]])
9+
; CHECK-NEXT: [[B_I1:%.*]] = extractelement <3 x i32> [[B]], i64 1
10+
; CHECK-NEXT: [[R_I1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[B_I1]], i32 [[B_I1]])
11+
; CHECK-NEXT: [[B_I2:%.*]] = extractelement <3 x i32> [[B]], i64 2
12+
; CHECK-NEXT: [[R_I2:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[B_I2]], i32 [[B_I2]])
13+
; CHECK-NEXT: [[EL_ELEM0:%.*]] = extractvalue { i32, i1 } [[R_I0]], 0
14+
; CHECK-NEXT: [[EL_ELEM01:%.*]] = extractvalue { i32, i1 } [[R_I1]], 0
15+
; CHECK-NEXT: [[EL_ELEM02:%.*]] = extractvalue { i32, i1 } [[R_I2]], 0
16+
; CHECK-NEXT: [[EL_UPTO0:%.*]] = insertelement <3 x i32> poison, i32 [[EL_ELEM0]], i64 0
17+
; CHECK-NEXT: [[EL_UPTO1:%.*]] = insertelement <3 x i32> [[EL_UPTO0]], i32 [[EL_ELEM01]], i64 1
18+
; CHECK-NEXT: [[EL:%.*]] = insertelement <3 x i32> [[EL_UPTO1]], i32 [[EL_ELEM02]], i64 2
19+
; CHECK-NEXT: ret <3 x i32> [[EL]]
20+
;
21+
%r = call { <3 x i32>, <3 x i1> } @llvm.umul.with.overflow.v3i32(<3 x i32> %b, <3 x i32> %b)
22+
%el = extractvalue { <3 x i32>, <3 x i1> } %r, 0
23+
ret <3 x i32> %el
24+
}
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt %s -passes="function(scalarizer)" -S | FileCheck %s
3+
4+
define <3 x i32> @test_(<3 x i32> %a, <3 x i32> %b) {
5+
; CHECK-LABEL: define <3 x i32> @test_(
6+
; CHECK-SAME: <3 x i32> [[A:%.*]], <3 x i32> [[B:%.*]]) {
7+
; CHECK-NEXT: [[B_I0:%.*]] = extractelement <3 x i32> [[B]], i64 0
8+
; CHECK-NEXT: [[R_I0:%.*]] = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 [[B_I0]], i32 [[B_I0]])
9+
; CHECK-NEXT: [[B_I1:%.*]] = extractelement <3 x i32> [[B]], i64 1
10+
; CHECK-NEXT: [[R_I1:%.*]] = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 [[B_I1]], i32 [[B_I1]])
11+
; CHECK-NEXT: [[B_I2:%.*]] = extractelement <3 x i32> [[B]], i64 2
12+
; CHECK-NEXT: [[R_I2:%.*]] = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 [[B_I2]], i32 [[B_I2]])
13+
; CHECK-NEXT: [[EL_ELEM0:%.*]] = extractvalue { i32, i1 } [[R_I0]], 0
14+
; CHECK-NEXT: [[EL_ELEM01:%.*]] = extractvalue { i32, i1 } [[R_I1]], 0
15+
; CHECK-NEXT: [[EL_ELEM02:%.*]] = extractvalue { i32, i1 } [[R_I2]], 0
16+
; CHECK-NEXT: [[EL_UPTO0:%.*]] = insertelement <3 x i32> poison, i32 [[EL_ELEM0]], i64 0
17+
; CHECK-NEXT: [[EL_UPTO1:%.*]] = insertelement <3 x i32> [[EL_UPTO0]], i32 [[EL_ELEM01]], i64 1
18+
; CHECK-NEXT: [[EL:%.*]] = insertelement <3 x i32> [[EL_UPTO1]], i32 [[EL_ELEM02]], i64 2
19+
; CHECK-NEXT: ret <3 x i32> [[EL]]
20+
;
21+
%r = call { <3 x i32>, <3 x i1> } @llvm.usub.with.overflow.v3i32(<3 x i32> %b, <3 x i32> %b)
22+
%el = extractvalue { <3 x i32>, <3 x i1> } %r, 0
23+
ret <3 x i32> %el
24+
}

0 commit comments

Comments
 (0)