[llvm] 8fa432b - [InstCombine] reduce test-for-overflow of shifted value
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 5 06:52:01 PDT 2022
Author: Tian Zhou
Date: 2022-09-05T09:51:51-04:00
New Revision: 8fa432be4f749c4ab0ee452d8f7c4337d67861a1
URL: https://github.com/llvm/llvm-project/commit/8fa432be4f749c4ab0ee452d8f7c4337d67861a1
DIFF: https://github.com/llvm/llvm-project/commit/8fa432be4f749c4ab0ee452d8f7c4337d67861a1.diff
LOG: [InstCombine] reduce test-for-overflow of shifted value
Fixes #57338.
The added code makes the following transformations:
For unsigned predicates / eq / ne:
icmp pred (x << 1), x --> icmp getSignedPredicate(pred) x, 0
icmp pred x, (x << 1) --> icmp getSignedPredicate(pred) 0, x
Some examples:
https://alive2.llvm.org/ce/z/ckn4cj
https://alive2.llvm.org/ce/z/h-4bAQ
Differential Revision: https://reviews.llvm.org/D132888
Added:
Modified:
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
llvm/test/Transforms/InstCombine/icmp-shl-1-overflow.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 834bb60a2ce67..0173f9e3d352b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4548,6 +4548,18 @@ Instruction *InstCombinerImpl::foldICmpBinOp(ICmpInst &I,
}
}
+ // For unsigned predicates / eq / ne:
+ // icmp pred (x << 1), x --> icmp getSignedPredicate(pred) x, 0
+ // icmp pred x, (x << 1) --> icmp getSignedPredicate(pred) 0, x
+ if (!ICmpInst::isSigned(Pred)) {
+ if (match(Op0, m_Shl(m_Specific(Op1), m_One())))
+ return new ICmpInst(ICmpInst::getSignedPredicate(Pred), Op1,
+ Constant::getNullValue(Op1->getType()));
+ else if (match(Op1, m_Shl(m_Specific(Op0), m_One())))
+ return new ICmpInst(ICmpInst::getSignedPredicate(Pred),
+ Constant::getNullValue(Op0->getType()), Op0);
+ }
+
if (Value *V = foldMultiplicationOverflowCheck(I))
return replaceInstUsesWith(I, V);
diff --git a/llvm/test/Transforms/InstCombine/icmp-shl-1-overflow.ll b/llvm/test/Transforms/InstCombine/icmp-shl-1-overflow.ll
index 3e34fb132a368..0803c03efdbf4 100644
--- a/llvm/test/Transforms/InstCombine/icmp-shl-1-overflow.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-shl-1-overflow.ll
@@ -6,8 +6,7 @@ declare void @usev4(<4 x i4>)
define i1 @icmp_shl_ugt_1(i8 %x) {
; CHECK-LABEL: @icmp_shl_ugt_1(
-; CHECK-NEXT: [[ADD:%.*]] = shl i8 [[X:%.*]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i8 [[ADD]], [[X]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i8 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%add = shl i8 %x, 1
@@ -18,8 +17,7 @@ define i1 @icmp_shl_ugt_1(i8 %x) {
define <2 x i1> @icmp_shl_ugt_2(<2 x i32> %_x) {
; CHECK-LABEL: @icmp_shl_ugt_2(
; CHECK-NEXT: [[X:%.*]] = add <2 x i32> [[_X:%.*]], <i32 42, i32 42>
-; CHECK-NEXT: [[ADD:%.*]] = shl <2 x i32> [[X]], <i32 1, i32 1>
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt <2 x i32> [[X]], [[ADD]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i32> [[X]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%x = add <2 x i32> <i32 42, i32 42>, %_x ; thwart complexity-based canonicalization
@@ -30,8 +28,7 @@ define <2 x i1> @icmp_shl_ugt_2(<2 x i32> %_x) {
define <3 x i1> @icmp_shl_uge_1(<3 x i7> %x) {
; CHECK-LABEL: @icmp_shl_uge_1(
-; CHECK-NEXT: [[ADD:%.*]] = shl <3 x i7> [[X:%.*]], <i7 1, i7 1, i7 1>
-; CHECK-NEXT: [[CMP:%.*]] = icmp uge <3 x i7> [[ADD]], [[X]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <3 x i7> [[X:%.*]], <i7 -1, i7 -1, i7 -1>
; CHECK-NEXT: ret <3 x i1> [[CMP]]
;
%add = shl <3 x i7> %x, <i7 1, i7 1, i7 1>
@@ -42,8 +39,7 @@ define <3 x i1> @icmp_shl_uge_1(<3 x i7> %x) {
define i1 @icmp_shl_uge_2(i5 %_x) {
; CHECK-LABEL: @icmp_shl_uge_2(
; CHECK-NEXT: [[X:%.*]] = add i5 [[_X:%.*]], 10
-; CHECK-NEXT: [[ADD:%.*]] = shl i5 [[X]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp uge i5 [[X]], [[ADD]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt i5 [[X]], 1
; CHECK-NEXT: ret i1 [[CMP]]
;
%x = add i5 42, %_x ; thwart complexity-based canonicalization
@@ -56,7 +52,7 @@ define i1 @icmp_shl_ult_1(i16 %x) {
; CHECK-LABEL: @icmp_shl_ult_1(
; CHECK-NEXT: [[ADD:%.*]] = shl i16 [[X:%.*]], 1
; CHECK-NEXT: call void @use16(i16 [[ADD]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp ult i16 [[ADD]], [[X]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt i16 [[X]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%add = shl i16 %x, 1
@@ -70,7 +66,7 @@ define <4 x i1> @icmp_shl_ult_2(<4 x i4> %_x) {
; CHECK-NEXT: [[X:%.*]] = add <4 x i4> [[_X:%.*]], <i4 -6, i4 -6, i4 -6, i4 -6>
; CHECK-NEXT: [[ADD:%.*]] = shl <4 x i4> [[X]], <i4 1, i4 1, i4 1, i4 1>
; CHECK-NEXT: call void @usev4(<4 x i4> [[ADD]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp ult <4 x i4> [[X]], [[ADD]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <4 x i4> [[X]], zeroinitializer
; CHECK-NEXT: ret <4 x i1> [[CMP]]
;
%x = add <4 x i4> <i4 42, i4 42, i4 42, i4 42>, %_x ; thwart complexity-based canonicalization
@@ -82,8 +78,7 @@ define <4 x i1> @icmp_shl_ult_2(<4 x i4> %_x) {
define <2 x i1> @icmp_shl_ule_1(<2 x i8> %x) {
; CHECK-LABEL: @icmp_shl_ule_1(
-; CHECK-NEXT: [[ADD:%.*]] = shl <2 x i8> [[X:%.*]], <i8 1, i8 poison>
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule <2 x i8> [[ADD]], [[X]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i8> [[X:%.*]], <i8 1, i8 1>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%add = shl <2 x i8> %x, <i8 1, i8 poison>
@@ -94,8 +89,7 @@ define <2 x i1> @icmp_shl_ule_1(<2 x i8> %x) {
define i1 @icmp_shl_ule_2(i8 %_x) {
; CHECK-LABEL: @icmp_shl_ule_2(
; CHECK-NEXT: [[X:%.*]] = add i8 [[_X:%.*]], 42
-; CHECK-NEXT: [[ADD:%.*]] = shl i8 [[X]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp ule i8 [[X]], [[ADD]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i8 [[X]], -1
; CHECK-NEXT: ret i1 [[CMP]]
;
%x = add i8 42, %_x ; thwart complexity-based canonicalization
@@ -106,8 +100,7 @@ define i1 @icmp_shl_ule_2(i8 %_x) {
define i1 @icmp_shl_eq_1(i8 %x) {
; CHECK-LABEL: @icmp_shl_eq_1(
-; CHECK-NEXT: [[ADD:%.*]] = shl i8 [[X:%.*]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i8 [[ADD]], [[X]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i8 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%add = shl i8 %x, 1
@@ -117,12 +110,11 @@ define i1 @icmp_shl_eq_1(i8 %x) {
define <2 x i1> @icmp_shl_eq_2(<2 x i8> %_x) {
; CHECK-LABEL: @icmp_shl_eq_2(
-; CHECK-NEXT: [[X:%.*]] = add <2 x i8> [[_X:%.*]], <i8 42, i8 42>
-; CHECK-NEXT: [[ADD:%.*]] = shl <2 x i8> [[X]], <i8 1, i8 1>
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[X]], [[ADD]]
+; CHECK-NEXT: [[X:%.*]] = sdiv <2 x i8> <i8 42, i8 42>, [[_X:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[X]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
- %x = add <2 x i8> <i8 42, i8 42>, %_x ; thwart complexity-based canonicalization
+ %x = sdiv <2 x i8> <i8 42, i8 42>, %_x ; thwart complexity-based canonicalization
%add = shl <2 x i8> %x, <i8 1, i8 1>
%cmp = icmp eq <2 x i8> %x, %add
ret <2 x i1> %cmp
@@ -130,8 +122,7 @@ define <2 x i1> @icmp_shl_eq_2(<2 x i8> %_x) {
define <2 x i1> @icmp_shl_ne_1(<2 x i8> %x) {
; CHECK-LABEL: @icmp_shl_ne_1(
-; CHECK-NEXT: [[ADD:%.*]] = shl <2 x i8> [[X:%.*]], <i8 1, i8 1>
-; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i8> [[ADD]], [[X]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i8> [[X:%.*]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%add = shl <2 x i8> %x, <i8 1, i8 1>
@@ -141,12 +132,11 @@ define <2 x i1> @icmp_shl_ne_1(<2 x i8> %x) {
define i1 @icmp_shl_ne_2(i8 %_x) {
; CHECK-LABEL: @icmp_shl_ne_2(
-; CHECK-NEXT: [[X:%.*]] = add i8 [[_X:%.*]], 42
-; CHECK-NEXT: [[ADD:%.*]] = shl i8 [[X]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp ne i8 [[X]], [[ADD]]
+; CHECK-NEXT: [[X:%.*]] = sdiv i8 42, [[_X:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne i8 [[X]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
- %x = add i8 42, %_x ; thwart complexity-based canonicalization
+ %x = sdiv i8 42, %_x ; thwart complexity-based canonicalization
%add = shl i8 %x, 1
%cmp = icmp ne i8 %x, %add
ret i1 %cmp
More information about the llvm-commits
mailing list