Skip to content

[MemProf] Avoid duplicate edges between nodes #113337

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
teresajohnson merged 2 commits into from
Oct 25, 2024
Merged

[MemProf] Avoid duplicate edges between nodes #113337

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
adc7650
[MemProf] Avoid duplicate edges between nodes
teresajohnson Oct 22, 2024
64ff325
Address comment
teresajohnson Oct 25, 2024
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
16 changes: 15 additions & 1 deletion llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1352,6 +1352,12 @@ static void checkNode(const ContextNode<DerivedCCG, FuncTy, CallTy> *Node,
}
assert(NodeContextIds == CalleeEdgeContextIds);
}
// Make sure we don't end up with duplicate edges between the same caller and
// callee.
DenseSet<ContextNode<DerivedCCG, FuncTy, CallTy> *> NodeSet;
Copy link

@snehasish snehasish Oct 25, 2024

Choose a reason for hiding this comment

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

Since we only use the NodeSet to check in the assert, I think this will needlessly run in an optmized build. Should we wrap this in an #ifndef NDEBUG?

Copy link
Contributor Author

@teresajohnson teresajohnson Oct 25, 2024

Choose a reason for hiding this comment

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

Hmm, that actually goes for a bunch of stuff in this function. Note this checking is off by default and only invoked under one of the -memprof-verify-* options, so in general it won't add overhead. But in fact I think I should change this into some kind of error that doesn't rely on debug builds, so that it can be utilized for release builds as well. I am going to put this new handling under the ifndef as you suggest, but also add a TODO to make it a non-debug only checking failure.

for (const auto &E : Node->CalleeEdges)
NodeSet.insert(E->Callee);
assert(NodeSet.size() == Node->CalleeEdges.size());
}

template <typename DerivedCCG, typename FuncTy, typename CallTy>
Expand Down Expand Up @@ -3125,7 +3131,15 @@ void CallsiteContextGraph<DerivedCCG, FuncTy, CallTy>::
// from the same callers as the old node. That should be true in the current
// use case, where we will remove None-type edges after copying over all
// caller edges from the callee.
assert(IsNewNode || NewCaller->findEdgeFromCaller(OldCallerEdge->Caller));
auto *ExistingCallerEdge =
NewCaller->findEdgeFromCaller(OldCallerEdge->Caller);
assert(IsNewNode || ExistingCallerEdge);
if (ExistingCallerEdge) {
ExistingCallerEdge->getContextIds().insert(EdgeContextIdsToMove.begin(),
EdgeContextIdsToMove.end());
ExistingCallerEdge->AllocTypes |= computeAllocType(EdgeContextIdsToMove);
continue;
}
auto NewEdge = std::make_shared<ContextEdge>(
NewCaller, OldCallerEdge->Caller,
computeAllocType(EdgeContextIdsToMove), EdgeContextIdsToMove);
Expand Down
34 changes: 28 additions & 6 deletions llvm/test/ThinLTO/X86/memprof-icp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -186,9 +186,13 @@
; REMARKS-MAIN: created clone _ZN2B03barEj.memprof.1
; REMARKS-MAIN: call in clone _ZN2B03barEj marked with memprof allocation attribute notcold
; REMARKS-MAIN: call in clone _ZN2B03barEj.memprof.1 marked with memprof allocation attribute cold
; REMARKS-MAIN: call in clone _ZN2B03barEj marked with memprof allocation attribute notcold
; REMARKS-MAIN: call in clone _ZN2B03barEj.memprof.1 marked with memprof allocation attribute cold
; REMARKS-MAIN: created clone _ZN1B3barEj.memprof.1
; REMARKS-MAIN: call in clone _ZN1B3barEj marked with memprof allocation attribute notcold
; REMARKS-MAIN: call in clone _ZN1B3barEj.memprof.1 marked with memprof allocation attribute cold
; REMARKS-MAIN: call in clone _ZN1B3barEj marked with memprof allocation attribute notcold
; REMARKS-MAIN: call in clone _ZN1B3barEj.memprof.1 marked with memprof allocation attribute cold
; REMARKS-FOO: created clone _Z3fooR2B0j.memprof.1
;; In each version of foo we should have promoted the indirect call to two conditional
;; direct calls, one to B::bar and one to B0::bar. The cloned version of foo should call
Expand All @@ -208,10 +212,10 @@
; REMARKS-FOO: call in clone _ZN1B3barEj marked with memprof allocation attribute notcold
; REMARKS-FOO: call in clone _ZN1B3barEj.memprof.1 marked with memprof allocation attribute cold

; STATS: 2 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) during whole program analysis
; STATS-BE: 4 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) during ThinLTO backend
; STATS: 2 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) during whole program analysis
; STATS-BE: 4 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) during ThinLTO backend
; STATS: 4 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) during whole program analysis
; STATS-BE: 8 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) during ThinLTO backend
; STATS: 4 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) during whole program analysis
; STATS-BE: 8 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) during ThinLTO backend
; STATS: 3 memprof-context-disambiguation - Number of function clones created during whole program analysis
; STATS-BE: 5 memprof-context-disambiguation - Number of function clones created during ThinLTO backend

Expand Down Expand Up @@ -247,8 +251,8 @@
; IR: attributes #[[NOTCOLD]] = {{.*}} "memprof"="notcold"
; IR: attributes #[[COLD]] = {{.*}} "memprof"="cold"

; STATS-BE-DISTRIB: 2 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) during ThinLTO backend
; STATS-BE-DISTRIB: 2 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) during ThinLTO backend
; STATS-BE-DISTRIB: 4 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) during ThinLTO backend
; STATS-BE-DISTRIB: 4 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) during ThinLTO backend
; STATS-BE-DISTRIB: 3 memprof-context-disambiguation - Number of function clones created during ThinLTO backend

;--- foo.ll
Expand Down Expand Up @@ -298,6 +302,9 @@ declare i32 @_Z3fooR2B0j(ptr, i32)
define i32 @_ZN2B03barEj(ptr %this, i32 %s) {
entry:
%call = tail call ptr @_Znwm(i64 noundef 4) #0, !memprof !33, !callsite !38
;; Second allocation in this function, to ensure that indirect edges to the
;; same callee are partitioned correctly.
%call2 = tail call ptr @_Znwm(i64 noundef 4) #0, !memprof !45, !callsite !50
store volatile i32 0, ptr %call, align 4
ret i32 0
}
Expand All @@ -311,6 +318,9 @@ declare void @_ZdlPvm()
define i32 @_ZN1B3barEj(ptr %this, i32 %s) {
entry:
%call = tail call ptr @_Znwm(i64 noundef 4) #0, !memprof !39, !callsite !44
;; Second allocation in this function, to ensure that indirect edges to the
;; same callee are partitioned correctly.
%call2 = tail call ptr @_Znwm(i64 noundef 4) #0, !memprof !51, !callsite !56
store volatile i32 0, ptr %call, align 4
ret i32 0
}
Expand Down Expand Up @@ -367,3 +377,15 @@ attributes #0 = { builtin allocsize(0) }
!42 = !{!43, !"cold"}
!43 = !{i64 4457553070050523782, i64 -2101080423462424381, i64 -6490791336773930154}
!44 = !{i64 4457553070050523782}
!45 = !{!46, !48}
!46 = !{!47, !"notcold"}
!47 = !{i64 456, i64 -2101080423462424381, i64 5188446645037944434}
!48 = !{!49, !"cold"}
!49 = !{i64 456, i64 -2101080423462424381, i64 5583420417449503557}
!50 = !{i64 456}
!51 = !{!52, !54}
!52 = !{!53, !"notcold"}
!53 = !{i64 789, i64 -2101080423462424381, i64 132626519179914298}
!54 = !{!55, !"cold"}
!55 = !{i64 789, i64 -2101080423462424381, i64 -6490791336773930154}
!56 = !{i64 789}
Loading