-
Notifications
You must be signed in to change notification settings - Fork 14.8k
[licm] don't drop MD_prof
when dropping other metadata
#152420
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.
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
Changes from all commits
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -1699,8 +1699,12 @@ static void hoist(Instruction &I, const DominatorTree *DT, const Loop *CurLoop, | |
// The check on hasMetadataOtherThanDebugLoc is to prevent us from burning | ||
// time in isGuaranteedToExecute if we don't actually have anything to | ||
// drop. It is a compile time optimization, not required for correctness. | ||
!SafetyInfo->isGuaranteedToExecute(I, DT, CurLoop)) | ||
I.dropUBImplyingAttrsAndMetadata(); | ||
!SafetyInfo->isGuaranteedToExecute(I, DT, CurLoop)) { | ||
if (ProfcheckDisableMetadataFixes) | ||
I.dropUBImplyingAttrsAndMetadata(); | ||
else | ||
I.dropUBImplyingAttrsAndMetadata({LLVMContext::MD_prof}); | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Could we directly call There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. but we'd still want to pass the other bunch of OK metadata, which would mean exposing it. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Ah. Didn't realize it wasn't exposed already. |
||
} | ||
|
||
if (isa<PHINode>(I)) | ||
// Move the new node to the end of the phi list in the destination block. | ||
|
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,45 @@ | ||
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals --version 2 | ||
; Test that hoisting conditional branches copies the debug and profiling info | ||
; metadata from the branch being hoisted. | ||
; RUN: opt -S -passes=licm %s -o - | FileCheck %s | ||
|
||
declare i32 @foo() | ||
|
||
; to_hoist should get hoisted, and that should not result | ||
; in a loss of profiling info | ||
define i32 @hoist_select(i1 %cond, i32 %a, i32 %b) nounwind { | ||
; CHECK-LABEL: define i32 @hoist_select | ||
; CHECK-SAME: (i1 [[COND:%.*]], i32 [[A:%.*]], i32 [[B:%.*]]) #[[ATTR0:[0-9]+]] { | ||
; CHECK-NEXT: entry: | ||
; CHECK-NEXT: [[TO_HOIST:%.*]] = select i1 [[COND]], i32 [[A]], i32 [[B]], !prof [[PROF0:![0-9]+]] | ||
; CHECK-NEXT: br label [[L0:%.*]] | ||
; CHECK: L0: | ||
; CHECK-NEXT: [[G:%.*]] = call i32 @foo() | ||
; CHECK-NEXT: [[SUM:%.*]] = add i32 [[G]], [[TO_HOIST]] | ||
; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[SUM]], 0 | ||
; CHECK-NEXT: br i1 [[C]], label [[L0]], label [[EXIT:%.*]], !prof [[PROF1:![0-9]+]] | ||
; CHECK: exit: | ||
; CHECK-NEXT: [[SUM_LCSSA:%.*]] = phi i32 [ [[SUM]], [[L0]] ] | ||
; CHECK-NEXT: ret i32 [[SUM_LCSSA]] | ||
; | ||
entry: | ||
br label %L0 | ||
L0: | ||
%g = call i32 @foo() | ||
%to_hoist = select i1 %cond, i32 %a, i32 %b, !prof !0 | ||
%sum = add i32 %g, %to_hoist | ||
%c = icmp eq i32 %sum, 0 | ||
br i1 %c, label %L0, label %exit, !prof !1 | ||
|
||
exit: | ||
ret i32 %sum | ||
} | ||
|
||
!0 = !{!"branch_weights", i32 2, i32 5} | ||
!1 = !{!"branch_weights", i32 101, i32 189} | ||
;. | ||
; CHECK: attributes #[[ATTR0]] = { nounwind } | ||
;. | ||
; CHECK: [[PROF0]] = !{!"branch_weights", i32 2, i32 5} | ||
; CHECK: [[PROF1]] = !{!"branch_weights", i32 101, i32 189} | ||
;. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is this useful in practice? SmallVector might already allocate enough inline storage to not require an allocation for the small number of metadata types we have.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
maybe. it's harmless though and at best a tiny efficiency.