[llvm] [memprof] Dump call site matching information (PR #125130)

[Teresa Johnson via llvm-commits]

================
@@ -0,0 +1,114 @@
+; Tests that the compiler dumps call site matches upon request.
+;
+; The test case is generated from:
+;
+; // main
+; // |
+; // f1 (noinline)
+; // |
+; // f2
+; // |
+; // f3 (noinline)
+; // |
+; // new
+;
+; __attribute__((noinline)) char *f3() { return ::new char[4]; }
+;
+; static char *f2() { return f3(); }
+;
+; __attribute__((noinline)) static char *f1() { return f2(); }
+;
+; int main() {
+;   f1();
+;   return 0;
+; }
+;
+; Here we expect to match two inline call stacks:
+;
+; - [main]
+; - [f1, f2]
+;
+; Note that f3 is considered to be an allocation site, not a call site, because
+; it directly calls new after inlining.
+
+; REQUIRES: x86_64-linux
+; RUN: split-file %s %t
+; RUN: llvm-profdata merge %t/memprof-dump-matched-call-site.yaml -o %t/memprof-dump-matched-call-site.memprofdata
+; RUN: opt < %t/memprof-dump-matched-call-site.ll -passes='memprof-use<profile-filename=%t/memprof-dump-matched-call-site.memprofdata>' -memprof-print-match-info -S 2>&1 | FileCheck %s
+
+;--- memprof-dump-matched-call-site.yaml
+---
+HeapProfileRecords:
+  - GUID:            main
+    AllocSites:      []
+    CallSites:
+      - - { Function: main, LineOffset: 1, Column: 3, IsInlineFrame: false }
+  - GUID:            _ZL2f1v
+    AllocSites:      []
+    CallSites:
+      - - { Function: _ZL2f2v, LineOffset: 0, Column: 28, IsInlineFrame: true }
+        - { Function: _ZL2f1v, LineOffset: 0, Column: 54, IsInlineFrame: false }
+  - GUID:            _ZL2f2v
+    AllocSites:      []
+    CallSites:
+      - - { Function: _ZL2f2v, LineOffset: 0, Column: 28, IsInlineFrame: true }
+        - { Function: _ZL2f1v, LineOffset: 0, Column: 54, IsInlineFrame: false }
+  - GUID:            _Z2f3v
+    AllocSites:
+      - Callstack:
+          - { Function: _Z2f3v, LineOffset: 0, Column: 47, IsInlineFrame: false }
+          - { Function: _ZL2f2v, LineOffset: 0, Column: 28, IsInlineFrame: true }
+          - { Function: _ZL2f1v, LineOffset: 0, Column: 54, IsInlineFrame: false }
+          - { Function: main, LineOffset: 1, Column: 3, IsInlineFrame: false }
+        MemInfoBlock:
+          AllocCount:      1
+          TotalSize:       4
+          TotalLifetime:   0
+          TotalLifetimeAccessDensity: 0
+    CallSites:       []
+...
+;--- memprof-dump-matched-call-site.ll
+; CHECK: MemProf notcold context with id 3894143216621363392 has total profiled size 4 is matched
+; CHECK: MemProf callsite match for inline call stack 4745611964195289084 10616861955219347331
----------------
teresajohnson wrote:

The other alternative is to have the compiler optionally dump out this correspondence for each matched allocation, however, there is already a need to deduplicate these across the compilations and that would increase the volume quite a bit. The upside is that we wouldn't need another tool and tool invocation. The other upside is that the compiler trims unneeded parts of the context during matching, which would reduce the volume and the need for redoing this analysis externally. But that can be considered for a follow on change if needed.

https://github.com/llvm/llvm-project/pull/125130