[InstCombine] Fold mul (shr exact (X, N)), 2^N + 1 -> add (X , shr exact (X, N)) #112407
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-transforms Author: Rose (AreaZR) ChangesAlive2 Proofs: Full diff: https://github.com/llvm/llvm-project/pull/112407.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
index f4f3644acfe5ea..3378da1130ae68 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
@@ -261,6 +261,35 @@ Instruction *InstCombinerImpl::visitMul(BinaryOperator &I) {
}
}
+ // mul (lshr exact X, N), (2^N + 1) -> add (X, lshr exact (X, N))
+ {
+ Value *NewOp;
+ const APInt *ShiftC;
+ const APInt *MulAP;
+ if (match(&I, m_Mul(m_Exact(m_Shr(m_Value(NewOp), m_APInt(ShiftC))),
+ m_APInt(MulAP)))) {
+ if (BitWidth > 2 && (*MulAP - 1).isPowerOf2() &&
+ *ShiftC == MulAP->logBase2()) {
+ Value *BinOp = Op0;
+ BinaryOperator *OpBO = cast<BinaryOperator>(Op0);
+ if (!isGuaranteedNotToBeUndef(NewOp, &AC, &I, &DT))
+ NewOp = Builder.CreateFreeze(NewOp, NewOp->getName() + ".fr");
+ if (HasNUW && OpBO->getOpcode() == Instruction::AShr &&
+ OpBO->hasOneUse())
+ BinOp = Builder.CreateLShr(NewOp, ConstantInt::get(Ty, *ShiftC), "",
+ /*isExact=*/true);
+
+ auto *NewAdd = BinaryOperator::CreateAdd(NewOp, BinOp);
+ if (HasNSW && (HasNUW || OpBO->getOpcode() == Instruction::LShr ||
+ ShiftC->getZExtValue() < BitWidth - 1))
+ NewAdd->setHasNoSignedWrap(true);
+
+ NewAdd->setHasNoUnsignedWrap(HasNUW);
+ return NewAdd;
+ }
+ }
+ }
+
if (Op0->hasOneUse() && match(Op1, m_NegatedPower2())) {
// Interpret X * (-1<<C) as (-X) * (1<<C) and try to sink the negation.
// The "* (1<<C)" thus becomes a potential shifting opportunity.
diff --git a/llvm/test/Transforms/InstCombine/ashr-lshr.ll b/llvm/test/Transforms/InstCombine/ashr-lshr.ll
index 1abf1be2cbedd9..9c4611d0be3947 100644
--- a/llvm/test/Transforms/InstCombine/ashr-lshr.ll
+++ b/llvm/test/Transforms/InstCombine/ashr-lshr.ll
@@ -1077,4 +1077,172 @@ entry:
ret i32 %shr
}
+define i32 @ashr_shift_mul(i32 noundef %x) {
+; CHECK-LABEL: @ashr_shift_mul(
+; CHECK-NEXT: [[A:%.*]] = ashr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = add i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr exact i32 %x, 3
+ %res = mul i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @ashr_shift_mul_nuw(i32 noundef %x) {
+; CHECK-LABEL: @ashr_shift_mul_nuw(
+; CHECK-NEXT: [[TMP1:%.*]] = lshr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = add nuw i32 [[X]], [[TMP1]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr exact i32 %x, 3
+ %res = mul nuw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @ashr_shift_mul_nsw(i32 noundef %x) {
+; CHECK-LABEL: @ashr_shift_mul_nsw(
+; CHECK-NEXT: [[A:%.*]] = ashr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr exact i32 %x, 3
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @lshr_shift_mul_nuw(i32 noundef %x) {
+; CHECK-LABEL: @lshr_shift_mul_nuw(
+; CHECK-NEXT: [[A:%.*]] = lshr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = add nuw i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr exact i32 %x, 3
+ %res = mul nuw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @lshr_shift_mul(i32 noundef %x) {
+; CHECK-LABEL: @lshr_shift_mul(
+; CHECK-NEXT: [[A:%.*]] = lshr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = add i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr exact i32 %x, 3
+ %res = mul i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @lshr_shift_mul_nsw(i32 noundef %x) {
+; CHECK-LABEL: @lshr_shift_mul_nsw(
+; CHECK-NEXT: [[A:%.*]] = lshr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr exact i32 %x, 3
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+; Negative test
+
+define i32 @lshr_no_exact(i32 %x) {
+; CHECK-LABEL: @lshr_no_exact(
+; CHECK-NEXT: [[A:%.*]] = lshr i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = mul nuw nsw i32 [[A]], 9
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr i32 %x, 3
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+; Negative test
+
+define i32 @ashr_no_exact(i32 %x) {
+; CHECK-LABEL: @ashr_no_exact(
+; CHECK-NEXT: [[A:%.*]] = ashr i32 [[X:%.*]], 3
+; CHECK-NEXT: [[RES:%.*]] = mul nsw i32 [[A]], 9
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr i32 %x, 3
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @lshr_no_undef(i32 %x) {
+; CHECK-LABEL: @lshr_no_undef(
+; CHECK-NEXT: [[X_FR:%.*]] = freeze i32 [[X:%.*]]
+; CHECK-NEXT: [[A:%.*]] = lshr exact i32 [[X_FR]], 3
+; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[X_FR]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr exact i32 %x, 3
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @ashr_no_undef(i32 %x) {
+; CHECK-LABEL: @ashr_no_undef(
+; CHECK-NEXT: [[X_FR:%.*]] = freeze i32 [[X:%.*]]
+; CHECK-NEXT: [[A:%.*]] = ashr exact i32 [[X_FR]], 3
+; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[X_FR]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr exact i32 %x, 3
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @lshr_multiuse(i32 noundef %x) {
+; CHECK-LABEL: @lshr_multiuse(
+; CHECK-NEXT: [[A:%.*]] = lshr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: call void @use(i32 [[A]])
+; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr exact i32 %x, 3
+ call void @use(i32 %a)
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @lshr_multiuse_no_flags(i32 noundef %x) {
+; CHECK-LABEL: @lshr_multiuse_no_flags(
+; CHECK-NEXT: [[A:%.*]] = lshr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: call void @use(i32 [[A]])
+; CHECK-NEXT: [[RES:%.*]] = add i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = lshr exact i32 %x, 3
+ call void @use(i32 %a)
+ %res = mul i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @ashr_multiuse_no_flags(i32 noundef %x) {
+; CHECK-LABEL: @ashr_multiuse_no_flags(
+; CHECK-NEXT: [[A:%.*]] = ashr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: call void @use(i32 [[A]])
+; CHECK-NEXT: [[RES:%.*]] = add i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr exact i32 %x, 3
+ call void @use(i32 %a)
+ %res = mul i32 %a, 9
+ ret i32 %res
+}
+
+define i32 @ashr_multiuse(i32 noundef %x) {
+; CHECK-LABEL: @ashr_multiuse(
+; CHECK-NEXT: [[A:%.*]] = ashr exact i32 [[X:%.*]], 3
+; CHECK-NEXT: call void @use(i32 [[A]])
+; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[X]], [[A]]
+; CHECK-NEXT: ret i32 [[RES]]
+;
+ %a = ashr exact i32 %x, 3
+ call void @use(i32 %a)
+ %res = mul nsw i32 %a, 9
+ ret i32 %res
+}
+
declare void @use(i32)
|
AZero13
changed the title
AZero13
force-pushed
the
inverse-div-shift
branch
2 times, most recently
from
c6a0f7e to
35698c1
Compare
AZero13
force-pushed
the
inverse-div-shift
branch
from
35698c1 to
37b0139
Compare
llvmbot
added
the
llvm:instcombine
|
Going to ping @nikic |
|
@dtcxzyw I was wondering if you would like to merge this or not? |
dtcxzyw
reviewed
Feb 7, 2025
AZero13
force-pushed
the
inverse-div-shift
branch
2 times, most recently
from
a952c89 to
19ecf32
Compare
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
AZero13
force-pushed
the
inverse-div-shift
branch
6 times, most recently
from
708467d to
8699887
Compare
AZero13
force-pushed
the
inverse-div-shift
branch
from
8699887 to
0faada2
Compare
|
@dtcxzyw I am ready! |
AZero13
force-pushed
the
inverse-div-shift
branch
from
0faada2 to
843e173
Compare
Done! @dtcxzyw Let's go! |
AZero13
force-pushed
the
inverse-div-shift
branch
from
843e173 to
bb87faf
Compare
AZero13
force-pushed
the
inverse-div-shift
branch
from
bb87faf to
45d12f1
Compare
|
Please avoid force-pushing if possible |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/81/builds/4685 Here is the relevant piece of the build log for the reference |
flovent
pushed a commit
to flovent/llvm-project
that referenced
this pull request
Feb 13, 2025
joaosaffran
pushed a commit
to joaosaffran/llvm-project
that referenced
this pull request
Feb 14, 2025
sivan-shani
pushed a commit
to sivan-shani/llvm-project
that referenced
this pull request
Feb 24, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Alive2 Proofs:
https://alive2.llvm.org/ce/z/aJnxyp
https://alive2.llvm.org/ce/z/dyeGEv