[llvm] [InstCombine] Fold select(X >s 0, 0, -X) | smax(X, 0) to abs(X) (PR #165200)
Wenju He via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 27 19:01:36 PDT 2025
https://github.com/wenju-he updated https://github.com/llvm/llvm-project/pull/165200
>From 4629547485d5b0c4f0a0e3e2e026222bbb654c7a Mon Sep 17 00:00:00 2001
From: Wenju He <wenju.he at intel.com>
Date: Mon, 27 Oct 2025 05:41:34 +0100
Subject: [PATCH 1/6] [InstCombine] Fold select(X >s 0, 0, -X) | smax(X, 0) to
abs(X)
The IR pattern is compiled from OpenCL code:
__builtin_astype(x > (uchar2)(0) ? x : -x, uchar2);
where smax is created by foldSelectInstWithICmp + canonicalizeSPF.
smax could also come from direct elementwise max call:
int c = b > (int)(0) ? (int)(0) : -b;
int d = __builtin_elementwise_max(b, (int)(0));
*a = c | d;
---
.../InstCombine/InstCombineAndOrXor.cpp | 18 ++++++++++++
llvm/test/Transforms/InstCombine/or.ll | 28 +++++++++++++++++++
2 files changed, 46 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 3ddf182149e57..4e863ca2c6dfd 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -3997,6 +3997,20 @@ static Value *foldOrUnsignedUMulOverflowICmp(BinaryOperator &I,
return nullptr;
}
+// Fold select(X >s 0, 0, -X) | smax(X, 0) --> abs(X)
+static Value *FoldOrOfSelectSmaxToAbs(BinaryOperator &I,
+ InstCombiner::BuilderTy &Builder) {
+ CmpPredicate Pred;
+ Value *X;
+ if (match(&I, m_c_Or(m_Select(m_ICmp(Pred, m_Value(X), m_ZeroInt()),
+ m_ZeroInt(), m_Sub(m_ZeroInt(), m_Deferred(X))),
+ m_OneUse(m_Intrinsic<Intrinsic::smax>(m_Deferred(X),
+ m_ZeroInt())))) &&
+ Pred == ICmpInst::ICMP_SGT)
+ return Builder.CreateBinaryIntrinsic(Intrinsic::abs, X, Builder.getFalse());
+ return nullptr;
+}
+
// FIXME: We use commutative matchers (m_c_*) for some, but not all, matches
// here. We should standardize that construct where it is needed or choose some
// other way to ensure that commutated variants of patterns are not missed.
@@ -4545,6 +4559,10 @@ Instruction *InstCombinerImpl::visitOr(BinaryOperator &I) {
if (Value *V = SimplifyAddWithRemainder(I))
return replaceInstUsesWith(I, V);
+ // select(X >s 0, 0, -X) | smax(X, 0) -> abs(X)
+ if (Value *Res = FoldOrOfSelectSmaxToAbs(I, Builder))
+ return replaceInstUsesWith(I, Res);
+
return nullptr;
}
diff --git a/llvm/test/Transforms/InstCombine/or.ll b/llvm/test/Transforms/InstCombine/or.ll
index 6b090e982af0a..bbc79e8c16a56 100644
--- a/llvm/test/Transforms/InstCombine/or.ll
+++ b/llvm/test/Transforms/InstCombine/or.ll
@@ -2113,3 +2113,31 @@ define <4 x i32> @or_zext_nneg_minus_constant_splat(<4 x i8> %a) {
%or = or <4 x i32> %zext, splat (i32 -9)
ret <4 x i32> %or
}
+
+define i8 @or_positive_minus_non_positive_to_abs(i8 noundef %0){
+; CHECK-LABEL: @or_positive_minus_non_positive_to_abs(
+; CHECK-NEXT: [[TMP2:%.*]] = call i8 @llvm.abs.i8(i8 [[TMP0:%.*]], i1 false)
+; CHECK-NEXT: ret i8 [[TMP2]]
+;
+ %2 = icmp sgt i8 %0, zeroinitializer
+ %3 = sext i1 %2 to i8
+ %4 = sub i8 zeroinitializer, %0
+ %5 = xor i8 %3, -1
+ %6 = and i8 %4, %5
+ %7 = and i8 %0, %3
+ %8 = or i8 %6, %7
+ ret i8 %8
+}
+
+define <2 x i8> @or_select_smax_to_abs(<2 x i8> %0){
+; CHECK-LABEL: @or_select_smax_to_abs(
+; CHECK-NEXT: [[TMP2:%.*]] = call <2 x i8> @llvm.abs.v2i8(<2 x i8> [[TMP0:%.*]], i1 false)
+; CHECK-NEXT: ret <2 x i8> [[TMP2]]
+;
+ %2 = icmp sgt <2 x i8> %0, zeroinitializer
+ %3 = sub <2 x i8> zeroinitializer, %0
+ %4 = select <2 x i1> %2, <2 x i8> zeroinitializer, <2 x i8> %3
+ %5 = tail call <2 x i8> @llvm.smax.v2i8(<2 x i8> %0, <2 x i8> zeroinitializer)
+ %6 = or <2 x i8> %4, %5
+ ret <2 x i8> %6
+}
>From af62083f6139ac5c103e8342e8ed6a7c877edfb9 Mon Sep 17 00:00:00 2001
From: Wenju He <wenju.he at intel.com>
Date: Mon, 27 Oct 2025 07:32:30 +0100
Subject: [PATCH 2/6] named value, add negative test with multiple uses of
@llvm.smax
---
llvm/test/Transforms/InstCombine/or.ll | 55 ++++++++++++++++++--------
1 file changed, 39 insertions(+), 16 deletions(-)
diff --git a/llvm/test/Transforms/InstCombine/or.ll b/llvm/test/Transforms/InstCombine/or.ll
index bbc79e8c16a56..dcc8a712cd835 100644
--- a/llvm/test/Transforms/InstCombine/or.ll
+++ b/llvm/test/Transforms/InstCombine/or.ll
@@ -2114,30 +2114,53 @@ define <4 x i32> @or_zext_nneg_minus_constant_splat(<4 x i8> %a) {
ret <4 x i32> %or
}
-define i8 @or_positive_minus_non_positive_to_abs(i8 noundef %0){
+define i8 @or_positive_minus_non_positive_to_abs(i8 %a){
; CHECK-LABEL: @or_positive_minus_non_positive_to_abs(
; CHECK-NEXT: [[TMP2:%.*]] = call i8 @llvm.abs.i8(i8 [[TMP0:%.*]], i1 false)
; CHECK-NEXT: ret i8 [[TMP2]]
;
- %2 = icmp sgt i8 %0, zeroinitializer
- %3 = sext i1 %2 to i8
- %4 = sub i8 zeroinitializer, %0
- %5 = xor i8 %3, -1
- %6 = and i8 %4, %5
- %7 = and i8 %0, %3
- %8 = or i8 %6, %7
- ret i8 %8
+ %b = icmp sgt i8 %a, zeroinitializer
+ %mask = sext i1 %b to i8
+ %neg = sub i8 zeroinitializer, %a
+ %mask_inv = xor i8 %mask, -1
+ %c = and i8 %neg, %mask_inv
+ %d = and i8 %a, %mask
+ %or = or i8 %c, %d
+ ret i8 %or
}
-define <2 x i8> @or_select_smax_to_abs(<2 x i8> %0){
+define <2 x i8> @or_select_smax_to_abs(<2 x i8> %a){
; CHECK-LABEL: @or_select_smax_to_abs(
; CHECK-NEXT: [[TMP2:%.*]] = call <2 x i8> @llvm.abs.v2i8(<2 x i8> [[TMP0:%.*]], i1 false)
; CHECK-NEXT: ret <2 x i8> [[TMP2]]
;
- %2 = icmp sgt <2 x i8> %0, zeroinitializer
- %3 = sub <2 x i8> zeroinitializer, %0
- %4 = select <2 x i1> %2, <2 x i8> zeroinitializer, <2 x i8> %3
- %5 = tail call <2 x i8> @llvm.smax.v2i8(<2 x i8> %0, <2 x i8> zeroinitializer)
- %6 = or <2 x i8> %4, %5
- ret <2 x i8> %6
+ %sgt0 = icmp sgt <2 x i8> %a, zeroinitializer
+ %neg = sub <2 x i8> zeroinitializer, %a
+ %sel = select <2 x i1> %sgt0, <2 x i8> zeroinitializer, <2 x i8> %neg
+ %max = call <2 x i8> @llvm.smax.v2i8(<2 x i8> %a, <2 x i8> zeroinitializer)
+ %or = or <2 x i8> %sel, %max
+ ret <2 x i8> %or
+}
+
+declare <2 x i8> @llvm.abs.v2i8(<2 x i8>, i1)
+
+; negative test - %d has multiple uses. %or is not folded to abs.
+
+define <2 x i8> @or_select_smax_multi_uses(<2 x i8> %a){
+; CHECK-LABEL: @or_select_smax_multi_uses(
+; CHECK-NEXT: [[B:%.*]] = icmp sgt <2 x i8> [[A:%.*]], zeroinitializer
+; CHECK-NEXT: [[NEG:%.*]] = sub <2 x i8> zeroinitializer, [[A]]
+; CHECK-NEXT: [[C:%.*]] = select <2 x i1> [[B]], <2 x i8> zeroinitializer, <2 x i8> [[NEG]]
+; CHECK-NEXT: [[D:%.*]] = call <2 x i8> @llvm.smax.v2i8(<2 x i8> [[A]], <2 x i8> zeroinitializer)
+; CHECK-NEXT: [[OR1:%.*]] = or <2 x i8> [[C]], [[D]]
+; CHECK-NEXT: [[OR:%.*]] = add <2 x i8> [[OR1]], [[D]]
+; CHECK-NEXT: ret <2 x i8> [[OR]]
+;
+ %sgt0 = icmp sgt <2 x i8> %a, zeroinitializer
+ %neg = sub <2 x i8> zeroinitializer, %a
+ %sel = select <2 x i1> %sgt0, <2 x i8> zeroinitializer, <2 x i8> %neg
+ %max = call <2 x i8> @llvm.smax.v2i8(<2 x i8> %a, <2 x i8> zeroinitializer)
+ %or = or <2 x i8> %sel, %max
+ %e = add <2 x i8> %or, %max
+ ret <2 x i8> %e
}
>From 21ca09a9ba1202f19c087ebf1f0f4bf369f21527 Mon Sep 17 00:00:00 2001
From: Wenju He <wenju.he at intel.com>
Date: Mon, 27 Oct 2025 07:35:16 +0100
Subject: [PATCH 3/6] rename %e -> %add
---
llvm/test/Transforms/InstCombine/or.ll | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/llvm/test/Transforms/InstCombine/or.ll b/llvm/test/Transforms/InstCombine/or.ll
index dcc8a712cd835..61c1ac5f18a28 100644
--- a/llvm/test/Transforms/InstCombine/or.ll
+++ b/llvm/test/Transforms/InstCombine/or.ll
@@ -2161,6 +2161,6 @@ define <2 x i8> @or_select_smax_multi_uses(<2 x i8> %a){
%sel = select <2 x i1> %sgt0, <2 x i8> zeroinitializer, <2 x i8> %neg
%max = call <2 x i8> @llvm.smax.v2i8(<2 x i8> %a, <2 x i8> zeroinitializer)
%or = or <2 x i8> %sel, %max
- %e = add <2 x i8> %or, %max
- ret <2 x i8> %e
+ %add = add <2 x i8> %or, %max
+ ret <2 x i8> %add
}
>From 1e9a8cedf135cb04e02050a7e48c019674211915 Mon Sep 17 00:00:00 2001
From: Wenju He <wenju.he at intel.com>
Date: Tue, 28 Oct 2025 02:58:24 +0100
Subject: [PATCH 4/6] add TODO: fold to smax(neg, fold two smax
---
llvm/test/Transforms/InstCombine/or.ll | 39 +++++++++++++++++++++++---
1 file changed, 35 insertions(+), 4 deletions(-)
diff --git a/llvm/test/Transforms/InstCombine/or.ll b/llvm/test/Transforms/InstCombine/or.ll
index 61c1ac5f18a28..01fb9295b1e04 100644
--- a/llvm/test/Transforms/InstCombine/or.ll
+++ b/llvm/test/Transforms/InstCombine/or.ll
@@ -2119,9 +2119,9 @@ define i8 @or_positive_minus_non_positive_to_abs(i8 %a){
; CHECK-NEXT: [[TMP2:%.*]] = call i8 @llvm.abs.i8(i8 [[TMP0:%.*]], i1 false)
; CHECK-NEXT: ret i8 [[TMP2]]
;
- %b = icmp sgt i8 %a, zeroinitializer
+ %b = icmp sgt i8 %a, 0
%mask = sext i1 %b to i8
- %neg = sub i8 zeroinitializer, %a
+ %neg = sub i8 0, %a
%mask_inv = xor i8 %mask, -1
%c = and i8 %neg, %mask_inv
%d = and i8 %a, %mask
@@ -2129,6 +2129,39 @@ define i8 @or_positive_minus_non_positive_to_abs(i8 %a){
ret i8 %or
}
+; TODO Fold to smax https://alive2.llvm.org/ce/z/wDiDh2
+define i8 @or_select_smax_neg_to_abs(i8 %a){
+; CHECK-LABEL: @or_select_smax_neg_to_abs(
+; CHECK-NEXT: [[SGT0:%.*]] = icmp sgt i8 [[A:%.*]], 0
+; CHECK-NEXT: [[NEG:%.*]] = sub nsw i8 0, [[A]]
+; CHECK-NEXT: [[OR:%.*]] = select i1 [[SGT0]], i8 0, i8 [[NEG]]
+; CHECK-NEXT: ret i8 [[OR]]
+;
+ %sgt0 = icmp sgt i8 %a, 0
+ %neg = sub nsw i8 0, %a
+ %sel = select i1 %sgt0, i8 0, i8 %neg
+ ret i8 %sel
+}
+
+; TODO Fold to abs https://alive2.llvm.org/ce/z/DybfHG
+define i8 @or_select_smax_smax_to_abs(i8 %a){
+; CHECK-LABEL: @or_select_smax_smax_to_abs(
+; CHECK-NEXT: [[NEG:%.*]] = sub nsw i8 0, [[A:%.*]]
+; CHECK-NEXT: [[SEL:%.*]] = call i8 @llvm.smax.i8(i8 [[NEG]], i8 0)
+; CHECK-NEXT: [[MAX:%.*]] = call i8 @llvm.smax.i8(i8 [[A]], i8 0)
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[SEL]], [[MAX]]
+; CHECK-NEXT: ret i8 [[OR]]
+;
+ %neg = sub nsw i8 0, %a
+ %sel = call i8 @llvm.smax.i8(i8 %neg, i8 0)
+ %max = call i8 @llvm.smax.i8(i8 %a, i8 0)
+ %or = or i8 %sel, %max
+ ret i8 %or
+}
+
+declare i8 @llvm.abs.i8(i8, i1)
+declare <2 x i8> @llvm.abs.v2i8(<2 x i8>, i1)
+
define <2 x i8> @or_select_smax_to_abs(<2 x i8> %a){
; CHECK-LABEL: @or_select_smax_to_abs(
; CHECK-NEXT: [[TMP2:%.*]] = call <2 x i8> @llvm.abs.v2i8(<2 x i8> [[TMP0:%.*]], i1 false)
@@ -2142,8 +2175,6 @@ define <2 x i8> @or_select_smax_to_abs(<2 x i8> %a){
ret <2 x i8> %or
}
-declare <2 x i8> @llvm.abs.v2i8(<2 x i8>, i1)
-
; negative test - %d has multiple uses. %or is not folded to abs.
define <2 x i8> @or_select_smax_multi_uses(<2 x i8> %a){
>From e5d284b1565a2d8a43f82b35b72a52373583daee Mon Sep 17 00:00:00 2001
From: Wenju He <wenju.he at intel.com>
Date: Tue, 28 Oct 2025 10:01:14 +0800
Subject: [PATCH 5/6] Update
llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
Co-authored-by: Yingwei Zheng <dtcxzyw at qq.com>
---
llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 4e863ca2c6dfd..66fc6bbadfceb 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -3997,7 +3997,7 @@ static Value *foldOrUnsignedUMulOverflowICmp(BinaryOperator &I,
return nullptr;
}
-// Fold select(X >s 0, 0, -X) | smax(X, 0) --> abs(X)
+/// Fold select(X >s 0, 0, -X) | smax(X, 0) --> abs(X)
static Value *FoldOrOfSelectSmaxToAbs(BinaryOperator &I,
InstCombiner::BuilderTy &Builder) {
CmpPredicate Pred;
>From e19950b0ee10e926deb1afa17fd7219eb0b26a0e Mon Sep 17 00:00:00 2001
From: Wenju He <wenju.he at intel.com>
Date: Tue, 28 Oct 2025 10:01:26 +0800
Subject: [PATCH 6/6] Update
llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
Co-authored-by: Yingwei Zheng <dtcxzyw at qq.com>
---
llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 66fc6bbadfceb..a0cd19195cb9f 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -4002,11 +4002,10 @@ static Value *FoldOrOfSelectSmaxToAbs(BinaryOperator &I,
InstCombiner::BuilderTy &Builder) {
CmpPredicate Pred;
Value *X;
- if (match(&I, m_c_Or(m_Select(m_ICmp(Pred, m_Value(X), m_ZeroInt()),
+ if (match(&I, m_c_Or(m_Select(m_SpecificICmp(ICmpInst::ICMP_SGT, m_Value(X), m_ZeroInt()),
m_ZeroInt(), m_Sub(m_ZeroInt(), m_Deferred(X))),
m_OneUse(m_Intrinsic<Intrinsic::smax>(m_Deferred(X),
- m_ZeroInt())))) &&
- Pred == ICmpInst::ICMP_SGT)
+ m_ZeroInt())))))
return Builder.CreateBinaryIntrinsic(Intrinsic::abs, X, Builder.getFalse());
return nullptr;
}
More information about the llvm-commits
mailing list