Skip to content

[Instcombine] Avoid widening trunc+sext to trunc+shl+ashr when not profitable #160483

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
wenju-he merged 7 commits into from
Oct 28, 2025
Merged

[Instcombine] Avoid widening trunc+sext to trunc+shl+ashr when not profitable #160483

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
e9d5680
[Instcombine] Avoid widening trunc+sext to trunc+shl+ashr when not pr…
wenju-he Sep 24, 2025
1d0a139
update per copilot suggestion
wenju-he Sep 24, 2025
4160bec
refine comment
wenju-he Sep 24, 2025
4b636bf
move to separate test with native integer widths
wenju-he Sep 30, 2025
00d2e6e
NFC, rename Trunc to TruncSrc
wenju-he Sep 30, 2025
001f07d
shouldExtendExpression -> ShouldExtendExpression
wenju-he Oct 26, 2025
1c5d599
fold sext(trunc nsw) to trunc nsw
wenju-he Oct 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 18 additions & 5 deletions llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1528,7 +1528,15 @@ Instruction *InstCombinerImpl::visitSExt(SExtInst &Sext) {
}

// Try to extend the entire expression tree to the wide destination type.
if (shouldChangeType(SrcTy, DestTy) && canEvaluateSExtd(Src, DestTy)) {
bool ShouldExtendExpression = true;
Value *TruncSrc = nullptr;
// It is not desirable to extend expression in the trunc + sext pattern when
// destination type is narrower than original (pre-trunc) type.
if (match(Src, m_Trunc(m_Value(TruncSrc))))
if (TruncSrc->getType()->getScalarSizeInBits() > DestBitSize)
ShouldExtendExpression = false;
if (ShouldExtendExpression && shouldChangeType(SrcTy, DestTy) &&
canEvaluateSExtd(Src, DestTy)) {
// Okay, we can transform this! Insert the new expression now.
LLVM_DEBUG(
dbgs() << "ICE: EvaluateInDifferentType converting expression type"
Expand All @@ -1548,13 +1556,18 @@ Instruction *InstCombinerImpl::visitSExt(SExtInst &Sext) {
ShAmt);
}

Value *X;
if (match(Src, m_Trunc(m_Value(X)))) {
Value *X = TruncSrc;
if (X) {
Copy link

Copilot AI Sep 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The variable X is being reused for two different purposes. In the first block (lines 1532-1539), it's used to check the transformation condition, while later at line 1561 it's used to continue the existing trunc optimization. This creates unclear control flow since X's value depends on whether the match succeeded earlier. Consider using a separate variable name for the first check to make the code more readable.

Copilot uses AI. Check for mistakes.
// If the input has more sign bits than bits truncated, then convert
// directly to final type.
unsigned XBitSize = X->getType()->getScalarSizeInBits();
if (ComputeNumSignBits(X, &Sext) > XBitSize - SrcBitSize)
return CastInst::CreateIntegerCast(X, DestTy, /* isSigned */ true);
bool HasNSW = cast<TruncInst>(Src)->hasNoSignedWrap();
if (HasNSW || (ComputeNumSignBits(X, &Sext) > XBitSize - SrcBitSize)) {
auto *Res = CastInst::CreateIntegerCast(X, DestTy, /* isSigned */ true);
if (auto *ResTrunc = dyn_cast<TruncInst>(Res); ResTrunc && HasNSW)
ResTrunc->setHasNoSignedWrap(true);
return Res;
}

// If input is a trunc from the destination type, then convert into shifts.
if (Src->hasOneUse() && X->getType() == DestTy) {
Expand Down
11 changes: 11 additions & 0 deletions llvm/test/Transforms/InstCombine/sext-of-trunc-nsw.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,17 @@ define i24 @wide_source_matching_signbits(i32 %x) {
ret i24 %c
}

define i32 @wide_source_matching_signbits_has_nsw_flag(i64 %i) {
; CHECK-LABEL: define i32 @wide_source_matching_signbits_has_nsw_flag(
; CHECK-SAME: i64 [[I:%.*]]) {
; CHECK-NEXT: [[A:%.*]] = trunc nsw i64 [[I]] to i32
; CHECK-NEXT: ret i32 [[A]]
;
%a = trunc nsw i64 %i to i16
%b = sext i16 %a to i32
ret i32 %b
}

; negative test - not enough sign-bits

define i24 @wide_source_not_matching_signbits(i32 %x) {
Expand Down
16 changes: 16 additions & 0 deletions llvm/test/Transforms/InstCombine/trunc-sext.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt < %s -passes=instcombine -S | FileCheck %s

target datalayout = "i16:16:16-i32:32:32-i64:64:64-n16:32:64"

define i32 @test(i64 %i) {
Copy link
Member

@dtcxzyw dtcxzyw Oct 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you please add a test with trunc nsw?

define i32 @test(i64 %i) {
  %a = trunc nsw i64 %i to i16
  %b = sext i16 %a to i32
  ret i32 %b
}

It should be folded into trunc nsw i64 %i to i32.

Copy link
Contributor Author

@wenju-he wenju-he Oct 26, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you please add a test with trunc nsw?

define i32 @test(i64 %i) {
  %a = trunc nsw i64 %i to i16
  %b = sext i16 %a to i32
  ret i32 %b
}

It should be folded into trunc nsw i64 %i to i32.

Thanks for the suggestion. Done in 1c5d599. It requires an additional change to the part If the input has more sign bits than bits truncated, then convert directly to final type. Please review.

; CHECK-LABEL: define i32 @test(
; CHECK-SAME: i64 [[I:%.*]]) {
; CHECK-NEXT: [[A:%.*]] = trunc i64 [[I]] to i16
; CHECK-NEXT: [[B:%.*]] = sext i16 [[A]] to i32
; CHECK-NEXT: ret i32 [[B]]
;
%a = trunc i64 %i to i16
%b = sext i16 %a to i32
ret i32 %b
}