Skip to content

[InstCombine] Fold (trunc X) into X & Mask inside decomposeBitTestICmp #171195

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
dtcxzyw merged 4 commits into from
Jan 16, 2026
Merged

[InstCombine] Fold (trunc X) into X & Mask inside decomposeBitTestICmp #171195

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
098c069
Pre-commit tests
wermos Dec 11, 2025
749bbd1
Implement missing optimization in `decompostBitTestICmp`.
wermos Dec 11, 2025
b22fc12
Updated tests
wermos Jan 5, 2026
a98e5cc
Merge branch 'main' into trunc
wermos Jan 10, 2026
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
8 changes: 8 additions & 0 deletions llvm/lib/Analysis/CmpInstAnalysis.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,6 +162,14 @@ llvm::decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate Pred,
break;
}

// Try to convert (trunc X) eq/ne C into (X & Mask) eq/ne C
if (LookThroughTrunc && isa<TruncInst>(LHS)) {
Result.Pred = Pred;
Result.Mask = APInt::getAllOnes(C.getBitWidth());
Result.C = C;
break;
}

return std::nullopt;
}
}
Expand Down
5 changes: 3 additions & 2 deletions llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -189,8 +189,9 @@ static unsigned conjugateICmpMask(unsigned Mask) {
// Adapts the external decomposeBitTest for local use.
static bool decomposeBitTest(Value *Cond, CmpInst::Predicate &Pred, Value *&X,
Value *&Y, Value *&Z) {
auto Res = llvm::decomposeBitTest(Cond, /*LookThroughTrunc=*/true,
/*AllowNonZeroC=*/true);
auto Res =
llvm::decomposeBitTest(Cond, /*LookThroughTrunc=*/true,
/*AllowNonZeroC=*/true, /*DecomposeAnd=*/true);
if (!Res)
return false;

Expand Down
13 changes: 13 additions & 0 deletions llvm/test/Transforms/InstCombine/and-or-icmps.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3721,3 +3721,16 @@ define i1 @merge_range_check_or(i8 %a) {
%and = or i1 %cmp1, %cmp2
ret i1 %and
}

; Just a very complicated way of checking if v1 == 0.
define i1 @complicated_zero_equality_test(i64 %v1) {
; CHECK-LABEL: @complicated_zero_equality_test(
Comment thread
wermos marked this conversation as resolved.
; CHECK-NEXT: [[V5:%.*]] = icmp eq i64 [[V1:%.*]], 0
; CHECK-NEXT: ret i1 [[V5]]
;
%v2 = trunc i64 %v1 to i32
%v3 = icmp eq i32 %v2, 0
%v4 = icmp ult i64 %v1, 4294967296 ; 2 ^ 32
%v5 = and i1 %v4, %v3
ret i1 %v5
}
4 changes: 1 addition & 3 deletions llvm/test/Transforms/InstCombine/icmp-logical.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1793,9 +1793,7 @@ define <2 x i1> @masked_icmps_bmask_notmixed_or_vec(<2 x i8> %A) {
define <2 x i1> @masked_icmps_bmask_notmixed_or_vec_poison1(<2 x i8> %A) {
; CHECK-LABEL: @masked_icmps_bmask_notmixed_or_vec_poison1(
; CHECK-NEXT: [[MASK1:%.*]] = and <2 x i8> [[A:%.*]], splat (i8 15)
; CHECK-NEXT: [[TST1:%.*]] = icmp eq <2 x i8> [[MASK1]], <i8 3, i8 poison>
; CHECK-NEXT: [[TST2:%.*]] = icmp eq <2 x i8> [[A]], splat (i8 -13)
; CHECK-NEXT: [[RES:%.*]] = or <2 x i1> [[TST1]], [[TST2]]
; CHECK-NEXT: [[RES:%.*]] = icmp eq <2 x i8> [[MASK1]], splat (i8 3)
; CHECK-NEXT: ret <2 x i1> [[RES]]
;
%mask1 = and <2 x i8> %A, <i8 15, i8 15> ; 0x0f
Expand Down
18 changes: 6 additions & 12 deletions llvm/test/Transforms/InstCombine/merge-icmp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,9 +167,8 @@ define i1 @or_extra_use1(i16 %load) {
; CHECK-NEXT: [[TRUNC:%.*]] = trunc i16 [[LOAD:%.*]] to i8
; CHECK-NEXT: [[CMP1:%.*]] = icmp ne i8 [[TRUNC]], 127
; CHECK-NEXT: call void @use.i1(i1 [[CMP1]])
; CHECK-NEXT: [[AND:%.*]] = and i16 [[LOAD]], -4096
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i16 [[AND]], 20480
; CHECK-NEXT: [[OR:%.*]] = or i1 [[CMP1]], [[CMP2]]
; CHECK-NEXT: [[TMP1:%.*]] = and i16 [[LOAD]], -3841
; CHECK-NEXT: [[OR:%.*]] = icmp ne i16 [[TMP1]], 20607
; CHECK-NEXT: ret i1 [[OR]]
;
%trunc = trunc i16 %load to i8
Expand All @@ -183,12 +182,11 @@ define i1 @or_extra_use1(i16 %load) {

define i1 @or_extra_use2(i16 %load) {
; CHECK-LABEL: @or_extra_use2(
; CHECK-NEXT: [[TRUNC:%.*]] = trunc i16 [[LOAD:%.*]] to i8
; CHECK-NEXT: [[CMP1:%.*]] = icmp ne i8 [[TRUNC]], 127
; CHECK-NEXT: [[AND:%.*]] = and i16 [[LOAD]], -4096
; CHECK-NEXT: [[AND:%.*]] = and i16 [[LOAD:%.*]], -4096
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i16 [[AND]], 20480
; CHECK-NEXT: call void @use.i1(i1 [[CMP2]])
; CHECK-NEXT: [[OR:%.*]] = or i1 [[CMP1]], [[CMP2]]
; CHECK-NEXT: [[TMP1:%.*]] = and i16 [[LOAD]], -3841
; CHECK-NEXT: [[OR:%.*]] = icmp ne i16 [[TMP1]], 20607
; CHECK-NEXT: ret i1 [[OR]]
;
%trunc = trunc i16 %load to i8
Expand Down Expand Up @@ -316,11 +314,7 @@ define i1 @or_wrong_const1(i16 %load) {

define i1 @or_wrong_const2(i16 %load) {
; CHECK-LABEL: @or_wrong_const2(
; CHECK-NEXT: [[TRUNC:%.*]] = trunc i16 [[LOAD:%.*]] to i8
; CHECK-NEXT: [[CMP1:%.*]] = icmp ne i8 [[TRUNC]], 127
; CHECK-NEXT: [[AND:%.*]] = and i16 [[LOAD]], -255
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i16 [[AND]], 17665
; CHECK-NEXT: [[OR:%.*]] = or i1 [[CMP1]], [[CMP2]]
; CHECK-NEXT: [[OR:%.*]] = icmp ne i16 [[LOAD:%.*]], 17791
; CHECK-NEXT: ret i1 [[OR]]
;
%trunc = trunc i16 %load to i8
Expand Down