[llvm] [Metadata] Handle memprof, callsite merging when one is missing. (PR #132106)
Snehasish Kumar via llvm-commits
llvm-commits at lists.llvm.org
Wed Mar 19 14:46:09 PDT 2025
https://github.com/snehasish created https://github.com/llvm/llvm-project/pull/132106
For memprof and callsite metadata we want to pick one deterministically
and keep that even if one of them may be missing.
>From 14e634e3a036c02b25c2fbb7caefe157ff7bfa01 Mon Sep 17 00:00:00 2001
From: Snehasish Kumar <snehasishk at google.com>
Date: Wed, 19 Mar 2025 21:41:19 +0000
Subject: [PATCH] [Metadata] Handle memprof, callsite merging when one is
missing.
For memprof and callsite metadata we want to pick one deterministically
and keep that even if one of them may be missing.
---
llvm/lib/Transforms/Utils/Local.cpp | 36 +++++++++---
.../SimplifyCFG/merge-calls-memprof.ll | 55 +++++++++++++++++--
2 files changed, 78 insertions(+), 13 deletions(-)
diff --git a/llvm/lib/Transforms/Utils/Local.cpp b/llvm/lib/Transforms/Utils/Local.cpp
index 95f0d099aacb5..161c7c875e0eb 100644
--- a/llvm/lib/Transforms/Utils/Local.cpp
+++ b/llvm/lib/Transforms/Utils/Local.cpp
@@ -3355,8 +3355,14 @@ static void combineMetadata(Instruction *K, const Instruction *J,
case LLVMContext::MD_invariant_group:
// Preserve !invariant.group in K.
break;
+ // Keep empty cases for mmra, memprof, and callsite to prevent them from
+ // being removed as unknown metadata. The actual merging is handled
+ // separately below.
case LLVMContext::MD_mmra:
- // Combine MMRAs
+ [[fallthrough]];
+ case LLVMContext::MD_memprof:
+ [[fallthrough]];
+ case LLVMContext::MD_callsite:
break;
case LLVMContext::MD_align:
if (!AAOnly && (DoesKMove || !K->hasMetadata(LLVMContext::MD_noundef)))
@@ -3369,14 +3375,6 @@ static void combineMetadata(Instruction *K, const Instruction *J,
K->setMetadata(Kind,
MDNode::getMostGenericAlignmentOrDereferenceable(JMD, KMD));
break;
- case LLVMContext::MD_memprof:
- if (!AAOnly)
- K->setMetadata(Kind, MDNode::getMergedMemProfMetadata(KMD, JMD));
- break;
- case LLVMContext::MD_callsite:
- if (!AAOnly)
- K->setMetadata(Kind, MDNode::getMergedCallsiteMetadata(KMD, JMD));
- break;
case LLVMContext::MD_preserve_access_index:
// Preserve !preserve.access.index in K.
break;
@@ -3420,6 +3418,26 @@ static void combineMetadata(Instruction *K, const Instruction *J,
K->setMetadata(LLVMContext::MD_mmra,
MMRAMetadata::combine(K->getContext(), JMMRA, KMMRA));
}
+
+ // Merge memprof metadata.
+ // Handle separately to support cases where only one instruction has the
+ // metadata.
+ auto JMemProf = J->getMetadata(LLVMContext::MD_memprof);
+ auto KMemProf = K->getMetadata(LLVMContext::MD_memprof);
+ if (!AAOnly && (JMemProf || KMemProf)) {
+ K->setMetadata(LLVMContext::MD_memprof,
+ MDNode::getMergedMemProfMetadata(KMemProf, JMemProf));
+ }
+
+ // Merge callsite metadata.
+ // Handle separately to support cases where only one instruction has the
+ // metadata.
+ auto JCallSite = J->getMetadata(LLVMContext::MD_callsite);
+ auto KCallSite = K->getMetadata(LLVMContext::MD_callsite);
+ if (!AAOnly && (JCallSite || KCallSite)) {
+ K->setMetadata(LLVMContext::MD_callsite,
+ MDNode::getMergedCallsiteMetadata(KCallSite, JCallSite));
+ }
}
void llvm::combineMetadataForCSE(Instruction *K, const Instruction *J,
diff --git a/llvm/test/Transforms/SimplifyCFG/merge-calls-memprof.ll b/llvm/test/Transforms/SimplifyCFG/merge-calls-memprof.ll
index 10c6aeb26ba76..d15eeb7b69fee 100644
--- a/llvm/test/Transforms/SimplifyCFG/merge-calls-memprof.ll
+++ b/llvm/test/Transforms/SimplifyCFG/merge-calls-memprof.ll
@@ -1,5 +1,3 @@
-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
-
;; Test to ensure that memprof related metadata is not dropped when
;; instructions are combined. Currently the metadata from the first instruction
;; is kept, which prevents full loss of profile context information.
@@ -32,6 +30,51 @@ if.end: ; preds = %if.else, %if.then
ret ptr %x.0
}
+define dso_local noundef nonnull ptr @_Z9test_leftb(i1 noundef zeroext %b) local_unnamed_addr #0 {
+; CHECK-LABEL: define dso_local noundef nonnull ptr @_Z9test_leftb(
+; CHECK-SAME: i1 noundef zeroext [[B:%.*]]) local_unnamed_addr {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CALL:%.*]] = call noalias noundef nonnull dereferenceable(4) ptr @_Znwm(i64 noundef 4), !memprof [[META0:![0-9]+]], !callsite [[META3:![0-9]+]]
+; CHECK-NEXT: ret ptr [[CALL]]
+;
+entry:
+ br i1 %b, label %if.then, label %if.else
+
+if.then: ; preds = %entry
+ %call = call noalias noundef nonnull dereferenceable(4) ptr @_Znwm(i64 noundef 4), !memprof !0, !callsite !3
+ br label %if.end
+
+if.else: ; preds = %entry
+ %call1 = call noalias noundef nonnull dereferenceable(4) ptr @_Znwm(i64 noundef 4)
+ br label %if.end
+
+if.end: ; preds = %if.else, %if.then
+ %x.0 = phi ptr [ %call, %if.then ], [ %call1, %if.else ]
+ ret ptr %x.0
+}
+
+define dso_local noundef nonnull ptr @_Z10test_rightb(i1 noundef zeroext %b) local_unnamed_addr #0 {
+; CHECK-LABEL: define dso_local noundef nonnull ptr @_Z10test_rightb(
+; CHECK-SAME: i1 noundef zeroext [[B:%.*]]) local_unnamed_addr {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CALL:%.*]] = call noalias noundef nonnull dereferenceable(4) ptr @_Znwm(i64 noundef 4), !memprof [[META4:![0-9]+]], !callsite [[META7:![0-9]+]]
+; CHECK-NEXT: ret ptr [[CALL]]
+;
+entry:
+ br i1 %b, label %if.then, label %if.else
+
+if.then: ; preds = %entry
+ %call = call noalias noundef nonnull dereferenceable(4) ptr @_Znwm(i64 noundef 4)
+ br label %if.end
+
+if.else: ; preds = %entry
+ %call1 = call noalias noundef nonnull dereferenceable(4) ptr @_Znwm(i64 noundef 4), !memprof !4, !callsite !7
+ br label %if.end
+
+if.end: ; preds = %if.else, %if.then
+ %x.0 = phi ptr [ %call, %if.then ], [ %call1, %if.else ]
+ ret ptr %x.0
+}
declare ptr @_Znwm(i64) nounwind readonly
@@ -43,9 +86,13 @@ declare ptr @_Znwm(i64) nounwind readonly
!5 = !{!6, !"cold"}
!6 = !{i64 123, i64 -2101080423462424381, i64 5188446645037944434}
!7 = !{i64 123}
-;.
+
; CHECK: [[META0]] = !{[[META1:![0-9]+]]}
; CHECK: [[META1]] = !{[[META2:![0-9]+]], !"notcold"}
; CHECK: [[META2]] = !{i64 -852997907418798798, i64 -2101080423462424381, i64 5188446645037944434}
; CHECK: [[META3]] = !{i64 -852997907418798798}
-;.
+; CHECK: [[META4]] = !{[[META5:![0-9]+]]}
+; CHECK: [[META5]] = !{[[META6:![0-9]+]], !"cold"}
+; CHECK: [[META6]] = !{i64 123, i64 -2101080423462424381, i64 5188446645037944434}
+; CHECK: [[META7]] = !{i64 123}
+
More information about the llvm-commits
mailing list