[StandardInstrumentation] Annotate loops with the function name

[annamthomas]

When analyzing pass debug output it is helpful to have the function name along with the loop name.

[annamthomas]

I've not updated all the failing tests check lines, want to know this is okay before going forward.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: None (annamthomas)

Changes

When analyzing pass debug output it is helpful to have the function name along with the loop name.

---

Full diff: https://github.com/llvm/llvm-project/pull/90756.diff

3 Files Affected:

• (modified) llvm/lib/Passes/StandardInstrumentations.cpp (+2-1)
• (modified) llvm/test/Transforms/LoopPredication/invalidate-analyses.ll (+2-2)
• (modified) llvm/test/Transforms/SimpleLoopUnswitch/nontrivial-unswitch-markloopasdeleted.ll (+2-2)

diff --git a/llvm/lib/Passes/StandardInstrumentations.cpp b/llvm/lib/Passes/StandardInstrumentations.cpp
index 63490c83e85f05..990ace8b04f87b 100644
--- a/llvm/lib/Passes/StandardInstrumentations.cpp
+++ b/llvm/lib/Passes/StandardInstrumentations.cpp
@@ -245,7 +245,8 @@ std::string getIRName(Any IR) {
     return C->getName();
 
   if (const auto *L = unwrapIR<Loop>(IR))
-    return L->getName().str();
+    return "Loop: [ " + L->getName().str() + " ] In Function: [ " +
+           L->getHeader()->getParent()->getName().str() + " ]";
 
   if (const auto *MF = unwrapIR<MachineFunction>(IR))
     return MF->getName().str();
diff --git a/llvm/test/Transforms/LoopPredication/invalidate-analyses.ll b/llvm/test/Transforms/LoopPredication/invalidate-analyses.ll
index 7afacd564939b5..b3eb345e1ba6cd 100644
--- a/llvm/test/Transforms/LoopPredication/invalidate-analyses.ll
+++ b/llvm/test/Transforms/LoopPredication/invalidate-analyses.ll
@@ -5,10 +5,10 @@
 ;       please update this test some other analysis that isn't preserved.
 
 ; CHECK: Running analysis: LazyValueAnalysis on drop_a_wc_and_leave_early
-; CHECK: Running pass: LoopPredicationPass on loop
+; CHECK: Running pass: LoopPredicationPass on Loop: [ loop ] In Function: [ drop_a_wc_and_leave_early ] 
 ; CHECK: Invalidating analysis: LazyValueAnalysis on drop_a_wc_and_leave_early
 ; CHECK: Running analysis: LazyValueAnalysis on drop_a_wc_and_leave
-; CHECK: Running pass: LoopPredicationPass on loop
+; CHECK: Running pass: LoopPredicationPass on Loop: [ loop ] In Function: [ drop_a_wc_and_leave ] 
 ; CHECK: Invalidating analysis: LazyValueAnalysis on drop_a_wc_and_leave
 
 
diff --git a/llvm/test/Transforms/SimpleLoopUnswitch/nontrivial-unswitch-markloopasdeleted.ll b/llvm/test/Transforms/SimpleLoopUnswitch/nontrivial-unswitch-markloopasdeleted.ll
index c8e1291b9cd55c..e47b76f58833b3 100644
--- a/llvm/test/Transforms/SimpleLoopUnswitch/nontrivial-unswitch-markloopasdeleted.ll
+++ b/llvm/test/Transforms/SimpleLoopUnswitch/nontrivial-unswitch-markloopasdeleted.ll
@@ -17,9 +17,9 @@
 ; SimpleLoopUnswitch not marking the Loop as removed, so we missed clearing
 ; the analysis caches.
 ;
-; CHECK: Running pass: SimpleLoopUnswitchPass on loop_begin
+; CHECK: Running pass: SimpleLoopUnswitchPass on Loop: [ loop_begin ] In Function: [ test6 ]
 ; CHECK-NEXT: Running analysis: OuterAnalysisManagerProxy
-; CHECK-NEXT: Clearing all analysis results for: loop_a_inner
+; CHECK-NEXT: Clearing all analysis results for: Loop [ loop_a_inner ] In Function: [ test6 ]
 
 
 ; When running loop-distribute the second time we can see that loop_a_inner

[aeubanks]

makes sense to me, although bikeshedding a bit, I'd prefer
Running pass: SimpleLoopUnswitchPass on loop %loop_begin in function test6

happy to hear any other tweaks to what actually gets printed

[annamthomas]

makes sense to me, although bikeshedding a bit, I'd prefer
Running pass: SimpleLoopUnswitchPass on loop %loop_begin in function test6

Yes, I can change to that. I'll wait a day to see if anyone has other preferences.

[jamieschmeiser]

I like the idea of making the loop easier to identify. Have you considered demangling the function name also, or is that already handled in getName()? However, that may make filtering difficult... what about printing both the mangled and demangled name if it is different?

This is a suggestion; I wouldn't block this change for demangling.

[nikic]

makes sense to me, although bikeshedding a bit, I'd prefer Running pass: SimpleLoopUnswitchPass on loop %loop_begin in function test6

+1

I like the idea of making the loop easier to identify. Have you considered demangling the function name also, or is that already handled in getName()? However, that may make filtering difficult... what about printing both the mangled and demangled name if it is different?

I would not find demangling helpful here -- if I'm looking at debug logs, I want to line them up with IR, not with source.

[jamieschmeiser]

I would not find demangling helpful here -- if I'm looking at debug logs, I want to line them up with IR, not with source.

I guess it depends on the situation as to whether demangling would be useful. I find demangled names easier to understand but the filtering mechanisms work with mangled names. This is why I suggested both but that may be too much info on the line. At the risk of over-designing something simple, perhaps demangling could be controlled by a hidden option?

Demangled or not, adding the function name is a good step.

[annamthomas]

Suggestion about formatting noted.
For now, I'm just going to be adding the function name to progress ahead with this (and general helps with the use cases I'm after).

[aeubanks]

lgtm with comment addressed

[jamieschmeiser]

I had lost track of #87626 but the memory of it was haunting me during this review. It hasn't landed yet but it may be something you want to look at.