[FunctionSpecialization] Fix profile count preserving logic #157939
Merged
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
The previous fix in llvm#157768 had a bug; instead of subtracting the original function's call count per call site of a specialization, we were subtracting the running total of the specialization's calls. Tracking issue: llvm#147390
|
@llvm/pr-subscribers-function-specialization @llvm/pr-subscribers-llvm-transforms Author: Alan Zhao (alanzhao1) ChangesThe previous fix in #157768 had a bug; instead of subtracting the original function's call count per call site of a specialization, we were subtracting the running total of the specialization's calls. Tracking issue: #147390 Full diff: https://github.com/llvm/llvm-project/pull/157939.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp b/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
index 30459caee1609..6af23e67f5762 100644
--- a/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
+++ b/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
@@ -802,8 +802,8 @@ bool FunctionSpecializer::run() {
if (std::optional<llvm::Function::ProfileCount> MaybeOriginalCount =
S.F->getEntryCount()) {
uint64_t OriginalCount = MaybeOriginalCount->getCount();
- if (OriginalCount >= CallCount) {
- S.F->setEntryCount(OriginalCount - CallCount);
+ if (OriginalCount >= *Count) {
+ S.F->setEntryCount(OriginalCount - *Count);
} else {
// This should generally not happen as that would mean there are
// more computed calls to the function than what was recorded.
@@ -1067,7 +1067,7 @@ Function *FunctionSpecializer::createSpecialization(Function *F,
// clone must.
Clone->setLinkage(GlobalValue::InternalLinkage);
- if (F->getEntryCount() && !ProfcheckDisableMetadataFixes)
+ if (F->getEntryCount())
Clone->setEntryCount(0);
// Initialize the lattice state of the arguments of the function clone,
diff --git a/llvm/test/Transforms/FunctionSpecialization/profile-counts.ll b/llvm/test/Transforms/FunctionSpecialization/profile-counts.ll
index d5b2e35feb118..28428927bc26c 100644
--- a/llvm/test/Transforms/FunctionSpecialization/profile-counts.ll
+++ b/llvm/test/Transforms/FunctionSpecialization/profile-counts.ll
@@ -13,23 +13,28 @@ entry:
; CHECK: if.then:
; CHECK: call i32 @foo.specialized.1(i32 %x, ptr @A)
+; CHECK: call i32 @foo.specialized.2(i32 %y, ptr @B)
if.then:
%call = call i32 @foo(i32 %x, ptr @A)
+ %call1 = call i32 @foo(i32 %y, ptr @B)
+ %call2 = call i32 @foo(i32 %y, ptr @B)
+ %add = add i32 %call, %call1
+ %add1 = add i32 %add, %call2
br label %return
; CHECK: if.else:
; CHECK: call i32 @foo.specialized.2(i32 %y, ptr @B)
if.else:
- %call1 = call i32 @foo(i32 %y, ptr @B)
+ %call3 = call i32 @foo(i32 %y, ptr @B)
br label %return
; CHECK: return:
-; CHECK: %call2 = call i32 @foo(i32 %x, ptr %z)
+; CHECK: %call3 = call i32 @foo(i32 %x, ptr %z)
return:
- %retval.0 = phi i32 [ %call, %if.then ], [ %call1, %if.else ]
- %call2 = call i32 @foo(i32 %x, ptr %z);
- %add = add i32 %retval.0, %call2
- ret i32 %add
+ %retval.0 = phi i32 [ %add1, %if.then ], [ %call3, %if.else ]
+ %call4 = call i32 @foo(i32 %x, ptr %z);
+ %add2 = add i32 %retval.0, %call4
+ ret i32 %add2
}
; CHECK: define internal i32 @foo(i32 %x, ptr %b) !prof ![[FOO_UNSPEC_PROF:[0-9]]]
@@ -44,9 +49,9 @@ entry:
; CHECK: ![[BAR_PROF]] = !{!"function_entry_count", i64 1000}
; CHECK: ![[BRANCH_PROF]] = !{!"branch_weights", i32 1, i32 3}
-; CHECK: ![[FOO_UNSPEC_PROF]] = !{!"function_entry_count", i64 234}
+; CHECK: ![[FOO_UNSPEC_PROF]] = !{!"function_entry_count", i64 500}
; CHECK: ![[FOO_SPEC_1_PROF]] = !{!"function_entry_count", i64 250}
-; CHECK: ![[FOO_SPEC_2_PROF]] = !{!"function_entry_count", i64 750}
+; CHECK: ![[FOO_SPEC_2_PROF]] = !{!"function_entry_count", i64 1250}
!0 = !{!"function_entry_count", i64 1000}
!1 = !{!"branch_weights", i32 1, i32 3}
-!2 = !{!"function_entry_count", i64 1234}
+!2 = !{!"function_entry_count", i64 2000}
|
mtrofin
approved these changes
Sep 11, 2025
mtrofin
reviewed
Sep 11, 2025
|
btw, you can remove this test from |
Done |
auto-merge was automatically disabled
September 11, 2025 05:43
Pull request was closed
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/66/builds/19111 Here is the relevant piece of the build log for the reference |
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.
The previous fix in #157768 had a bug; instead of subtracting the original function's call count per call site of a specialization, we were subtracting the running total of the specialization's calls.
Tracking issue: #147390