[llvm] Reapply "[MemProf] Streamline and avoid unnecessary context id duplication (#107918)" (PR #110036)

[Teresa Johnson via llvm-commits]

================
@@ -0,0 +1,102 @@
+;; This test ensures that the logic which assigns calls to stack nodes
+;; correctly handles a case where multiple nodes have stack ids that
+;; overlap with each other but have different last nodes (can happen with
+;; inlining into various levels of a call chain). Specifically, when we
+;; have one that is duplicated (e.g. unrolling), we need to correctly
+;; handle the case where the context id has already been assigned to
+;; a different call with a different last node.
+
+;; -stats requires asserts
+; REQUIRES: asserts
+
+; RUN: opt -passes=memprof-context-disambiguation -supports-hot-cold-new \
+; RUN:	-memprof-verify-ccg -memprof-verify-nodes \
+; RUN:  -stats -pass-remarks=memprof-context-disambiguation \
+; RUN:	%s -S 2>&1 | FileCheck %s --check-prefix=IR \
+; RUN:  --check-prefix=STATS --check-prefix=REMARKS
+
+; REMARKS: created clone _Z1Ab.memprof.1
+; REMARKS: created clone _Z3XZNv.memprof.1
+; REMARKS: call in clone main assigned to call function clone _Z3XZNv.memprof.1
+;; Make sure the inlined context in _Z3XZNv, which partially overlaps the stack
+;; ids in the shorter inlined context of Z2XZv, correctly calls a cloned
+;; version of Z1Ab, which will call the cold annotated allocation.
+; REMARKS: call in clone _Z3XZNv.memprof.1 assigned to call function clone _Z1Ab.memprof.1
+; REMARKS: call in clone _Z1Ab.memprof.1 marked with memprof allocation attribute cold
+; REMARKS: call in clone main assigned to call function clone _Z3XZNv
+; REMARKS: call in clone _Z3XZNv assigned to call function clone _Z1Ab
+; REMARKS: call in clone _Z1Ab marked with memprof allocation attribute notcold
+
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define dso_local void @_Z1Ab() {
+entry:
+  %call = call noalias noundef nonnull ptr @_Znam(i64 noundef 10) #1, !memprof !0, !callsite !5
+  ret void
+}
+
+; Function Attrs: nobuiltin
+declare ptr @_Znam(i64) #0
+
+;; Inlining of stack id 2 into 3. Assume this is called from somewhere else.
+define dso_local void @_Z2XZv() local_unnamed_addr #0 {
+entry:
+  ;; Simulate duplication of the callsite (e.g. unrolling).
+  tail call void @_Z1Ab(), !callsite !6
+  tail call void @_Z1Ab(), !callsite !6
----------------
teresajohnson wrote:

done

https://github.com/llvm/llvm-project/pull/110036