[msan] Fix "Add optional flag to improve instrumentation of disjoint OR (#145990)" #146799
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
…OR (llvm#145990)" The "V1" and "V2" values were already NOT'ed, hence the calculation of disjoint OR was incorrect. This patch fixes the issue by using the instruction operands directly.
thurstond
force-pushed
the
msan_fix_disjoint_or_fix
branch
from
fcb3556 to
7a7aeb0
Compare
|
@llvm/pr-subscribers-compiler-rt-sanitizer @llvm/pr-subscribers-llvm-transforms Author: Thurston Dang (thurstond) ChangesThe "V1" and "V2" values were already NOT'ed, hence the calculation of disjoint OR was incorrect. This patch fixes the issue by using the instruction operands directly. Full diff: https://github.com/llvm/llvm-project/pull/146799.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index 6511c197fcb5a..4a0eee177667e 100644
--- a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -2512,6 +2512,7 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
// S = S | (V1 & V2)
Value *S1 = getShadow(&I, 0);
Value *S2 = getShadow(&I, 1);
+ // Gotcha: V1 and V2 are NOT'ed here
Value *V1 = IRB.CreateNot(I.getOperand(0));
Value *V2 = IRB.CreateNot(I.getOperand(1));
if (V1->getType() != S1->getType()) {
@@ -2524,7 +2525,11 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
Value *S = IRB.CreateOr({S1S2, V1S2, S1V2});
if (ClPreciseDisjointOr && cast<PossiblyDisjointInst>(&I)->isDisjoint()) {
- Value *V1V2 = IRB.CreateAnd(V1, V2);
+ // "V1" and "V2" were NOT'ed above, but we still want to reuse them
+ // because they were IntCast'ed to the same type as the shadows.
+ //
+ // (V1 & V2) == ~(~V1 | ~V2) (de Morgan)
+ Value *V1V2 = IRB.CreateNot(IRB.CreateOr(V1, V2));
S = IRB.CreateOr({S, V1V2});
}
diff --git a/llvm/test/Instrumentation/MemorySanitizer/or.ll b/llvm/test/Instrumentation/MemorySanitizer/or.ll
index 2d51de13a8ebb..71beb38b466e4 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/or.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/or.ll
@@ -45,8 +45,9 @@ define i8 @test_disjoint_or(i8 %a, i8 %b) sanitize_memory {
; CHECK-IMPRECISE: [[C:%.*]] = or disjoint i8 [[A]], [[B]]
; CHECK-IMPRECISE-NEXT: store i8 [[TMP11]], ptr @__msan_retval_tls, align 8
;
-; CHECK-PRECISE: [[TMP10:%.*]] = and i8 [[TMP3]], [[TMP4]]
-; CHECK-PRECISE-NEXT: [[TMP12:%.*]] = or i8 [[TMP11]], [[TMP10]]
+; CHECK-PRECISE: [[TMP10:%.*]] = or i8 [[TMP3]], [[TMP4]]
+; CHECK-PRECISE-NEXT: [[TMP11:%.*]] = xor i8 [[TMP10]], -1
+; CHECK-PRECISE-NEXT: [[TMP12:%.*]] = or i8 [[TMP9]], [[TMP11]]
; CHECK-PRECISE-NEXT: [[C:%.*]] = or disjoint i8 [[A]], [[B]]
; CHECK-PRECISE-NEXT: store i8 [[TMP12]], ptr @__msan_retval_tls, align 8
;
|
fmayer
reviewed
Jul 2, 2025
| @@ -2512,6 +2512,7 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> { | |||
| // S = S | (V1 & V2) | |||
| Value *S1 = getShadow(&I, 0); | |||
| Value *S2 = getShadow(&I, 1); | |||
| // Gotcha: V1 and V2 are NOT'ed here | |||
There was a problem hiding this comment.
The composite names would get clunky and possibly ambiguous:
// Name is ambiguous: it's ((NotV1) & S2), as opposed to (Not (V1 & S2))
Value *NotV1S2 = IRB.CreateAnd(NotV1, S2);
Value *S1NotV2 = IRB.CreateAnd(S1, NotV2);
There was a problem hiding this comment.
maybe give them more meaningless names then, so people have to look up the definition? idk tho
There was a problem hiding this comment.
I changed the order to make it unambiguous
| @@ -2524,7 +2525,11 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> { | |||
|
|
|||
| Value *S = IRB.CreateOr({S1S2, V1S2, S1V2}); | |||
| if (ClPreciseDisjointOr && cast<PossiblyDisjointInst>(&I)->isDisjoint()) { | |||
| Value *V1V2 = IRB.CreateAnd(V1, V2); | |||
| // "V1" and "V2" were NOT'ed above, but we still want to reuse them | |||
There was a problem hiding this comment.
Is that really better than just int casting them again?
fmayer
reviewed
Jul 2, 2025
| // (V1 & V2) == ~(~V1 | ~V2) (de Morgan) | ||
| Value *V1V2 = IRB.CreateNot(IRB.CreateOr(V1, V2)); | ||
| // "V1" and "V2" were NOT'ed above | ||
| V1 = IRB.CreateIntCast(I.getOperand(0), S1->getType(), false); |
There was a problem hiding this comment.
don't reuse the variable name :(
There was a problem hiding this comment.
Renamed the earlier variables NotV1 / NotV2
fmayer
approved these changes
Jul 3, 2025
|
Failing test is flang, so YOLO |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
The "V1" and "V2" values were already NOT'ed, hence the calculation of disjoint OR in #145990 was incorrect. This patch fixes the issue by using DeMorgan's law.