[llvm] 8320b97 - [InstCombine] Fold an unsigned comparison of `add nsw X, C` with a constant into a signed comparison (#103480)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Aug 14 07:31:23 PDT 2024
Author: Volodymyr Vasylkun
Date: 2024-08-14T15:31:19+01:00
New Revision: 8320b97ab9f25d726379832342f3d5d48962368d
URL: https://github.com/llvm/llvm-project/commit/8320b97ab9f25d726379832342f3d5d48962368d
DIFF: https://github.com/llvm/llvm-project/commit/8320b97ab9f25d726379832342f3d5d48962368d.diff
LOG: [InstCombine] Fold an unsigned comparison of `add nsw X, C` with a constant into a signed comparison (#103480)
Given an unsigned integer comparison of `add nsw X, C1` with some
constant `C2` we can fold it into a signed comparison of `X` and `C2 -
C1` under the following conditions:
* There's a `nsw` flag on the addition
* `C2` is non-negative
* `X + C1` is non-negative
* `C2 - C1` is non-negative
Added:
Modified:
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
llvm/test/Transforms/InstCombine/icmp-add.ll
llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 2b0347073b7d1f..10488ecb747a48 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -3077,6 +3077,12 @@ Instruction *InstCombinerImpl::foldICmpAddConstant(ICmpInst &Cmp,
return new ICmpInst(Pred, X, ConstantInt::get(Ty, NewC));
}
+ if (ICmpInst::isUnsigned(Pred) && Add->hasNoSignedWrap() &&
+ C.isNonNegative() && (C - *C2).isNonNegative() &&
+ computeConstantRange(X, /*ForSigned=*/true).add(*C2).isAllNonNegative())
+ return new ICmpInst(ICmpInst::getSignedPredicate(Pred), X,
+ ConstantInt::get(Ty, C - *C2));
+
auto CR = ConstantRange::makeExactICmpRegion(Pred, C).subtract(*C2);
const APInt &Upper = CR.getUpper();
const APInt &Lower = CR.getLower();
diff --git a/llvm/test/Transforms/InstCombine/icmp-add.ll b/llvm/test/Transforms/InstCombine/icmp-add.ll
index baa6f3d51a40ef..2ceb44b89eb9e9 100644
--- a/llvm/test/Transforms/InstCombine/icmp-add.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-add.ll
@@ -3102,3 +3102,84 @@ define i1 @uge_add_C2_pow2_C_neg(i8 %x) {
}
declare void @llvm.assume(i1)
+
+; Change an unsigned predicate to signed in icmp (add x, C1), C2
+define i1 @icmp_add_constant_with_constant_ult_to_slt(i32 range(i32 -4, 10) %x) {
+; CHECK-LABEL: @icmp_add_constant_with_constant_ult_to_slt(
+; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[X:%.*]], 8
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %add = add nsw i32 %x, 5
+ %cmp = icmp ult i32 %add, 13
+ ret i1 %cmp
+}
+
+define i1 @icmp_add_constant_with_constant_ugt_to_sgt(i32 range(i32 -4, 10) %x) {
+; CHECK-LABEL: @icmp_add_constant_with_constant_ugt_to_sgt(
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[X:%.*]], 2
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %add = add nsw i32 %x, 10
+ %cmp = icmp ugt i32 %add, 12
+ ret i1 %cmp
+}
+
+; Negative test: x + C1 may be negative
+define i1 @icmp_add_constant_with_constant_ult_to_slt_neg1(i32 range(i32 -5, 10) %x) {
+; CHECK-LABEL: @icmp_add_constant_with_constant_ult_to_slt_neg1(
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[X:%.*]], 4
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[ADD]], 20
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %add = add nsw i32 %x, 4
+ %cmp = icmp ult i32 %add, 20
+ ret i1 %cmp
+}
+
+; Negative test: missing nsw flag
+define i1 @icmp_add_constant_with_constant_ult_to_slt_neg2(i8 range(i8 -4, 120) %x) {
+; CHECK-LABEL: @icmp_add_constant_with_constant_ult_to_slt_neg2(
+; CHECK-NEXT: [[ADD:%.*]] = add i8 [[X:%.*]], 15
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i8 [[ADD]], 20
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %add = add i8 %x, 15
+ %cmp = icmp ult i8 %add, 20
+ ret i1 %cmp
+}
+
+; Negative test: C2 is negative
+define i1 @icmp_add_constant_with_constant_ult_to_slt_neg3(i32 range(i32 -4, 10) %x) {
+; CHECK-LABEL: @icmp_add_constant_with_constant_ult_to_slt_neg3(
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[X:%.*]], 4
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[ADD]], -6
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %add = add nsw i32 %x, 4
+ %cmp = icmp ult i32 %add, -6
+ ret i1 %cmp
+}
+
+; Negative test: C2 - C1 is negative
+define i1 @icmp_add_constant_with_constant_ult_to_slt_neg4(i32 range(i32 -4, 10) %x) {
+; CHECK-LABEL: @icmp_add_constant_with_constant_ult_to_slt_neg4(
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[X:%.*]], 5
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[ADD]], 2
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %add = add nsw i32 %x, 5
+ %cmp = icmp ult i32 %add, 2
+ ret i1 %cmp
+}
+
+; Same as before, but infer the range of ucmp
+define i1 @icmp_of_ucmp_plus_const_with_const(i32 %x, i32 %y) {
+; CHECK-LABEL: @icmp_of_ucmp_plus_const_with_const(
+; CHECK-NEXT: [[CMP2:%.*]] = icmp ule i32 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT: ret i1 [[CMP2]]
+;
+ %cmp1 = call i8 @llvm.ucmp(i32 %x, i32 %y)
+ %add = add i8 %cmp1, 1
+ %cmp2 = icmp ult i8 %add, 2
+ ret i1 %cmp2
+}
diff --git a/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll b/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll
index 61ed955ea13e4e..6fc52ab3f26e03 100644
--- a/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll
+++ b/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll
@@ -1359,7 +1359,7 @@ define void @PR27626_5(ptr %a, i32 %x, i32 %y, i32 %z, i64 %n) {
; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[SMAX]], -4
; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 1
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 6
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 10
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 9223372036854775804
More information about the llvm-commits
mailing list