[licm] clone `MD_prof` when hoisting conditional branch #152232

mtrofin · 2025-08-06T00:51:16Z

The profiling - related metadata information for the hoisted conditional branch should be copied from the original branch, not from the current terminator of the block it's hoisted to.

The patch adds a way to disable the fix just so we can do an ablation test, after which the flag will be removed. The same flag will be reused for other similar fixes.

(This was identified through profcheck (see Issue #147390), and this PR addresses most of the test failures (when running under profcheck) under Transforms/LICM.)

mtrofin · 2025-08-06T00:51:31Z

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-08-06T17:27:25Z

@llvm/pr-subscribers-debuginfo
@llvm/pr-subscribers-llvm-ir

@llvm/pr-subscribers-llvm-analysis

Author: Mircea Trofin (mtrofin)

Changes

The metadata information for the hoisted conditional branch should be copied from the original branch, not from the current terminator of the block it's hoisted to. Note that debug info was previously copied by ReplaceInstWithInst as described in the previous statement. MD_prof was dropped outright.

(This was identified through profcheck (see Issue #147390), and this PR addresses most of the test failures (when running under profcheck) under Transforms/LICM.)

Full diff: https://github.com/llvm/llvm-project/pull/152232.diff

3 Files Affected:

(modified) llvm/lib/Analysis/BranchProbabilityInfo.cpp (+7)
(modified) llvm/lib/Transforms/Scalar/LICM.cpp (+14-3)
(added) llvm/test/Transforms/LICM/hoist-phi-metadata.ll (+46)

diff --git a/llvm/lib/Analysis/BranchProbabilityInfo.cpp b/llvm/lib/Analysis/BranchProbabilityInfo.cpp
index 671e023415a6e..1467bcd61ec8b 100644
--- a/llvm/lib/Analysis/BranchProbabilityInfo.cpp
+++ b/llvm/lib/Analysis/BranchProbabilityInfo.cpp
@@ -59,6 +59,13 @@ static cl::opt<std::string> PrintBranchProbFuncName(
     cl::desc("The option to specify the name of the function "
              "whose branch probability info is printed."));
 
+// Flag used for an ablation performance test, Issue #147390. Placing it here
+// because referencing Analysis should be feasible from anywhere. Will be
+// removed after the ablation test.
+cl::opt<bool> DisableProfilingInfoCorrectPropagation(
+    "profcheck-disable-fixes", cl::Hidden, cl::init(false),
+    cl::desc("Temporary flag, will be used for an ablation test"));
+
 INITIALIZE_PASS_BEGIN(BranchProbabilityInfoWrapperPass, "branch-prob",
                       "Branch Probability Analysis", false, true)
 INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
diff --git a/llvm/lib/Transforms/Scalar/LICM.cpp b/llvm/lib/Transforms/Scalar/LICM.cpp
index c3f80f901a120..3a17e34573cc4 100644
--- a/llvm/lib/Transforms/Scalar/LICM.cpp
+++ b/llvm/lib/Transforms/Scalar/LICM.cpp
@@ -218,6 +218,8 @@ using PointersAndHasReadsOutsideSet =
 static SmallVector<PointersAndHasReadsOutsideSet, 0>
 collectPromotionCandidates(MemorySSA *MSSA, AliasAnalysis *AA, Loop *L);
 
+extern cl::opt<bool> DisableProfilingInfoCorrectPropagation;
+
 namespace {
 struct LoopInvariantCodeMotion {
   bool runOnLoop(Loop *L, AAResults *AA, LoopInfo *LI, DominatorTree *DT,
@@ -857,9 +859,18 @@ class ControlFlowHoister {
     }
 
     // Now finally clone BI.
-    ReplaceInstWithInst(
-        HoistTarget->getTerminator(),
-        BranchInst::Create(HoistTrueDest, HoistFalseDest, BI->getCondition()));
+    auto *NewBI =
+        BranchInst::Create(HoistTrueDest, HoistFalseDest, BI->getCondition());
+    ReplaceInstWithInst(HoistTarget->getTerminator(), NewBI);
+    // Copy all the metadata. In particular:
+    // - debug info (critical to Sample-based profiling) should be the same as
+    // the original branch, not that of HoistTarget->getTerminator(), which is
+    // what ReplaceInstWithInst would use.
+    // - md_prof should also come from the original branch - since the condition
+    // was hoisted, the branch probabilities shouldn't change.
+    if (!DisableProfilingInfoCorrectPropagation)
+      NewBI->copyMetadata(*BI);
+
     ++NumClonedBranches;
 
     assert(CurLoop->getLoopPreheader() &&
diff --git a/llvm/test/Transforms/LICM/hoist-phi-metadata.ll b/llvm/test/Transforms/LICM/hoist-phi-metadata.ll
new file mode 100644
index 0000000000000..031a00d11bbf3
--- /dev/null
+++ b/llvm/test/Transforms/LICM/hoist-phi-metadata.ll
@@ -0,0 +1,46 @@
+; Test that hoising conditional branches copies over profiling metadata
+; RUN: opt -S -passes=licm -licm-control-flow-hoisting=1 %s -o - | FileCheck %s
+
+; CHECK-LABEL: @triangle_phi
+define void @triangle_phi(i32 %x, ptr %p) {
+; CHECK-LABEL: entry:
+; CHECK: %cmp1 = icmp sgt i32 %x, 0
+; CHECK: br i1 %cmp1, label %[[IF_LICM:.*]], label %[[THEN_LICM:.*]], !prof !0
+entry:
+  br label %loop
+
+; CHECK: [[IF_LICM]]:
+; CHECK: %add = add i32 %x, 1
+; CHECK: br label %[[THEN_LICM]]
+
+; CHECK: [[THEN_LICM]]:
+; CHECK: phi i32 [ %add, %[[IF_LICM]] ], [ %x, %entry ]
+; CHECK: store i32 %phi, ptr %p
+; CHECK: %cmp2 = icmp ne i32 %phi, 0
+; CHECK: br label %loop
+
+loop:
+  %cmp1 = icmp sgt i32 %x, 0
+  br i1 %cmp1, label %if, label %then, !prof !0
+
+if:
+  %add = add i32 %x, 1
+  br label %then
+
+; CHECK-LABEL: then:
+; CHECK: br i1 %cmp2, label %loop, label %end, !prof !1
+then:
+  %phi = phi i32 [ %add, %if ], [ %x, %loop ]
+  store i32 %phi, ptr %p
+  %cmp2 = icmp ne i32 %phi, 0
+  br i1 %cmp2, label %loop, label %end, !prof !1
+
+; CHECK-LABEL: end:
+end:
+  ret void
+}
+
+; CHECK: !0 = !{!"branch_weights", i32 5, i32 7}
+; CHECK: !1 = !{!"branch_weights", i32 13, i32 11}
+!0 = !{!"branch_weights", i32 5, i32 7}
+!1 = !{!"branch_weights", i32 13, i32 11}

llvmbot · 2025-08-06T17:27:26Z

@llvm/pr-subscribers-llvm-transforms

Author: Mircea Trofin (mtrofin)

Changes

The metadata information for the hoisted conditional branch should be copied from the original branch, not from the current terminator of the block it's hoisted to. Note that debug info was previously copied by ReplaceInstWithInst as described in the previous statement. MD_prof was dropped outright.

(This was identified through profcheck (see Issue #147390), and this PR addresses most of the test failures (when running under profcheck) under Transforms/LICM.)

Full diff: https://github.com/llvm/llvm-project/pull/152232.diff

3 Files Affected:

(modified) llvm/lib/Analysis/BranchProbabilityInfo.cpp (+7)
(modified) llvm/lib/Transforms/Scalar/LICM.cpp (+14-3)
(added) llvm/test/Transforms/LICM/hoist-phi-metadata.ll (+46)

diff --git a/llvm/lib/Analysis/BranchProbabilityInfo.cpp b/llvm/lib/Analysis/BranchProbabilityInfo.cpp
index 671e023415a6e..1467bcd61ec8b 100644
--- a/llvm/lib/Analysis/BranchProbabilityInfo.cpp
+++ b/llvm/lib/Analysis/BranchProbabilityInfo.cpp
@@ -59,6 +59,13 @@ static cl::opt<std::string> PrintBranchProbFuncName(
     cl::desc("The option to specify the name of the function "
              "whose branch probability info is printed."));
 
+// Flag used for an ablation performance test, Issue #147390. Placing it here
+// because referencing Analysis should be feasible from anywhere. Will be
+// removed after the ablation test.
+cl::opt<bool> DisableProfilingInfoCorrectPropagation(
+    "profcheck-disable-fixes", cl::Hidden, cl::init(false),
+    cl::desc("Temporary flag, will be used for an ablation test"));
+
 INITIALIZE_PASS_BEGIN(BranchProbabilityInfoWrapperPass, "branch-prob",
                       "Branch Probability Analysis", false, true)
 INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
diff --git a/llvm/lib/Transforms/Scalar/LICM.cpp b/llvm/lib/Transforms/Scalar/LICM.cpp
index c3f80f901a120..3a17e34573cc4 100644
--- a/llvm/lib/Transforms/Scalar/LICM.cpp
+++ b/llvm/lib/Transforms/Scalar/LICM.cpp
@@ -218,6 +218,8 @@ using PointersAndHasReadsOutsideSet =
 static SmallVector<PointersAndHasReadsOutsideSet, 0>
 collectPromotionCandidates(MemorySSA *MSSA, AliasAnalysis *AA, Loop *L);
 
+extern cl::opt<bool> DisableProfilingInfoCorrectPropagation;
+
 namespace {
 struct LoopInvariantCodeMotion {
   bool runOnLoop(Loop *L, AAResults *AA, LoopInfo *LI, DominatorTree *DT,
@@ -857,9 +859,18 @@ class ControlFlowHoister {
     }
 
     // Now finally clone BI.
-    ReplaceInstWithInst(
-        HoistTarget->getTerminator(),
-        BranchInst::Create(HoistTrueDest, HoistFalseDest, BI->getCondition()));
+    auto *NewBI =
+        BranchInst::Create(HoistTrueDest, HoistFalseDest, BI->getCondition());
+    ReplaceInstWithInst(HoistTarget->getTerminator(), NewBI);
+    // Copy all the metadata. In particular:
+    // - debug info (critical to Sample-based profiling) should be the same as
+    // the original branch, not that of HoistTarget->getTerminator(), which is
+    // what ReplaceInstWithInst would use.
+    // - md_prof should also come from the original branch - since the condition
+    // was hoisted, the branch probabilities shouldn't change.
+    if (!DisableProfilingInfoCorrectPropagation)
+      NewBI->copyMetadata(*BI);
+
     ++NumClonedBranches;
 
     assert(CurLoop->getLoopPreheader() &&
diff --git a/llvm/test/Transforms/LICM/hoist-phi-metadata.ll b/llvm/test/Transforms/LICM/hoist-phi-metadata.ll
new file mode 100644
index 0000000000000..031a00d11bbf3
--- /dev/null
+++ b/llvm/test/Transforms/LICM/hoist-phi-metadata.ll
@@ -0,0 +1,46 @@
+; Test that hoising conditional branches copies over profiling metadata
+; RUN: opt -S -passes=licm -licm-control-flow-hoisting=1 %s -o - | FileCheck %s
+
+; CHECK-LABEL: @triangle_phi
+define void @triangle_phi(i32 %x, ptr %p) {
+; CHECK-LABEL: entry:
+; CHECK: %cmp1 = icmp sgt i32 %x, 0
+; CHECK: br i1 %cmp1, label %[[IF_LICM:.*]], label %[[THEN_LICM:.*]], !prof !0
+entry:
+  br label %loop
+
+; CHECK: [[IF_LICM]]:
+; CHECK: %add = add i32 %x, 1
+; CHECK: br label %[[THEN_LICM]]
+
+; CHECK: [[THEN_LICM]]:
+; CHECK: phi i32 [ %add, %[[IF_LICM]] ], [ %x, %entry ]
+; CHECK: store i32 %phi, ptr %p
+; CHECK: %cmp2 = icmp ne i32 %phi, 0
+; CHECK: br label %loop
+
+loop:
+  %cmp1 = icmp sgt i32 %x, 0
+  br i1 %cmp1, label %if, label %then, !prof !0
+
+if:
+  %add = add i32 %x, 1
+  br label %then
+
+; CHECK-LABEL: then:
+; CHECK: br i1 %cmp2, label %loop, label %end, !prof !1
+then:
+  %phi = phi i32 [ %add, %if ], [ %x, %loop ]
+  store i32 %phi, ptr %p
+  %cmp2 = icmp ne i32 %phi, 0
+  br i1 %cmp2, label %loop, label %end, !prof !1
+
+; CHECK-LABEL: end:
+end:
+  ret void
+}
+
+; CHECK: !0 = !{!"branch_weights", i32 5, i32 7}
+; CHECK: !1 = !{!"branch_weights", i32 13, i32 11}
+!0 = !{!"branch_weights", i32 5, i32 7}
+!1 = !{!"branch_weights", i32 13, i32 11}

llvm/test/Transforms/LICM/hoist-phi-metadata.ll

llvm/lib/Analysis/BranchProbabilityInfo.cpp

snehasish · 2025-08-06T18:20:00Z

llvm/lib/Transforms/Scalar/LICM.cpp

+    // - md_prof should also come from the original branch - since the condition
+    // was hoisted, the branch probabilities shouldn't change.
+    if (!DisableProfilingInfoCorrectPropagation)
+      NewBI->copyMetadata(*BI);


For debug information, it looks like ReplaceInstWithInst first copies over the wrong debug info, then we update it to the right one as part of this copyMetadata call? The hidden manipulation of some types of metadata seems like a design that could be improved.

This code probably shouldn't be using ReplaceInstWithInst at all. The things it does are largely useless in this context.

Should probably do something along these lines?

auto *NewBI = BranchInst::Create(..., HoistTarget->getTerminator()); NewBI->copyMetadata(*BI); HoistTarget->getTerminator()->eraseFromParent();

nikic · 2025-08-06T19:40:59Z

llvm/lib/Transforms/Scalar/LICM.cpp

+    // what ReplaceInstWithInst would use.
+    // - md_prof should also come from the original branch - since the condition
+    // was hoisted, the branch probabilities shouldn't change.
+    if (!ProfcheckDisableMetadataFixes)


Control flow hoisting in LICM is not enabled by default, so I don't think this option is useful.

For its temporary purpose, it's easier if I just control all these kinds of changes this way. I'll remove it after (basically I want to lump all fixes I can detect via profcheck, and it's OK if some aren't enabled anyway - just want to approximate the benefit)

llvm/test/Transforms/LICM/hoist-phi-metadata.ll

nikic · 2025-08-06T20:41:53Z

llvm/lib/Transforms/Scalar/LICM.cpp

+    // Copy all the metadata. In particular:
+    // - debug info (critical to Sample-based profiling) should be the same
+    // as the original branch, not that of HoistTarget->getTerminator(),
+    // which is what ReplaceInstWithInst would use.


Hm, looking at https://llvm.org/docs/HowToUpdateDebugInfo.html#when-to-preserve-an-instruction-location again, it specifically calls out LICM as a case where the debug location should be dropped, not preserved.

Surprising / counter-intuitive, since the article does talk about SamplePGO. @vedantk who edited that last.

We had a similar discussion about merging source locations and it's intended usage with AutoFDO. We find that retaining some location (though potentially misleading) is better for profile quality. See #129960 for discussion and updates to the section just below the linked one above where the guidance has been updated. We haven't updated the guidance for dropping source locations though this is probably a good time as any to have the conversation and make the change.

Edit: updated PR link to the correct one.

Besides, in this case it wouldn't be misleading, for AutoFDO purposes either (or someone correct me if I'm missing something)

Debug info usage of debuggers and SamplePGO is at odds here. Debuggers want this dropped, while SamplePGO does not. Our debug info policy is generally targeted towards debuggers.

Filed #152767 to track the addition of a new flag and to update the documentation.

I don't think we should enable it by default for O2/O3 though. I received pushback even when suggesting we add it under -fdebug-info-for-profiling. It should be purely opt-in especially for the cases where the location may be non-representable.

To keep this PR moving along can we leave the behaviour for debug information unchanged (add a TODO with the issue number) and only address the profile metadata being dropped?

Oh, if there's precedent re. auto-enabling, sgtm.

Adding the flag and change in behavior is trivial, especially since (my earlier comment) we actually shouldn't just move all the metadata over. The documentation part of Issue #152767 can be addressed separately, imho.

Agree the change here is small but we should do this as a separate change to solicit feedback from other debug info stakeholders.

You'll still want to explicitly set the location to dropped here to not run afoul of verification (https://discourse.llvm.org/t/rfc-require-real-or-annotated-source-locations-on-all-instructions/86816).

Ack. Kept the behavior we had previously for debuginfo

mtrofin · 2025-08-07T01:16:39Z

llvm/lib/IR/Instruction.cpp

+// FIXME: Flag used for an ablation performance test, Issue #147390. Placing it
+// here because referencing IR should be feasible from anywhere. Will be
+// removed after the ablation test.
+cl::opt<bool> ProfcheckDisableMetadataFixes(


moved it to IR because... well, see the stacked PR (still draft because I am thinking where to actually put the test)

This comment doesn't make sense anymore, but I'd leave it in IR to avoid later churn.

llvm/lib/Transforms/Scalar/LICM.cpp

snehasish

lgtm.

llvm/lib/Transforms/Scalar/LICM.cpp

nikic

The code change LGTM. For the option, I think I was wrong in saying that this needs to be in a header with LLVM_ABI. I see lots of other places that do cross-imports of options with just extern cl::opt<>. @andrurogerz Is it okay to just have cl::opt<bool> ProfcheckDisableMetadataFixes in Instructions.cpp and extern cl::opt<bool> ProfcheckDisableMetadataFixes in LICM.cpp?

andrurogerz · 2025-08-12T18:58:14Z

Is it okay to just have cl::opt ProfcheckDisableMetadataFixes in Instructions.cpp and extern cl::opt ProfcheckDisableMetadataFixes in LICM.cpp?

You will need LLVM_ABI on the extern declaration otherwise the client won't know it is mean to be DLL-imported on Windows. It is only required on the definition in LICM.cpp if that file doesn't #include the header where it is declared (looks like it doesn't?)

nikic · 2025-08-12T19:17:00Z

@andrurogerz This only matters if we want to export the option from the DLL though, right? If it's just purely internal (but used across multiple files) it should be fine without annotations? We seems to have lots of options like this, MaxDevirtIterations to pick a random example (defined in llvm/lib/Analysis/CallGraphSCCPass.cpp, imported in llvm/lib/Passes/PassBuilderPipelines.cpp, neither using LLVM_ABI).

andrurogerz · 2025-08-12T19:22:39Z

This only matters if we want to export the option from the DLL though, right?

@nikic yes, good point. I missed that it was internal only. You only need to annotate it for export if a test or other executable needs to reference the symbol. I see that it is declared in a public header, though, can it move to a private header? I realize it is referenced cross-library so perhaps not.

mtrofin · 2025-08-12T20:30:44Z

I'm assuming in the Windows case, there's a LLVM.dll that we're talking about here? i.e. the component boundary isn't IR vs Transforms, it's more coarse-grained.

mtrofin · 2025-08-12T23:59:17Z

Merge activity

Aug 12, 11:59 PM UTC: A user started a stack merge that includes this pull request via Graphite.
Aug 13, 12:01 AM UTC: @mtrofin merged this pull request with Graphite.

nikic · 2025-08-13T09:24:53Z

I'm assuming in the Windows case, there's a LLVM.dll that we're talking about here? i.e. the component boundary isn't IR vs Transforms, it's more coarse-grained.

Yeah, exactly.

mtrofin marked this pull request as ready for review August 6, 2025 17:26

llvmbot added llvm:analysis Includes value tracking, cost tables and constant folding llvm:transforms labels Aug 6, 2025

mtrofin commented Aug 6, 2025

View reviewed changes

llvm/test/Transforms/LICM/hoist-phi-metadata.ll Outdated Show resolved Hide resolved

mtrofin requested review from jdenny-ornl, john-brawn-arm and snehasish August 6, 2025 17:27

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from 667dcb0 to 14815d4 Compare August 6, 2025 17:30

snehasish reviewed Aug 6, 2025

View reviewed changes

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch 2 times, most recently from 96a8b81 to 12346fa Compare August 6, 2025 19:27

nikic reviewed Aug 6, 2025

View reviewed changes

llvm/test/Transforms/LICM/hoist-phi-metadata.ll Show resolved Hide resolved

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from 12346fa to 0494ce7 Compare August 6, 2025 19:49

nikic reviewed Aug 6, 2025

View reviewed changes

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch 2 times, most recently from 37ea48f to 82b2df1 Compare August 7, 2025 00:55

llvmbot added the llvm:ir label Aug 7, 2025

mtrofin mentioned this pull request Aug 7, 2025

[licm] don't drop MD_prof when dropping other metadata #152420

Open

mtrofin commented Aug 7, 2025

View reviewed changes

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from 82b2df1 to 9bea77f Compare August 8, 2025 18:22

llvmbot added the debuginfo label Aug 8, 2025

mtrofin requested a review from nikic August 8, 2025 18:26

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from 9bea77f to ac134fb Compare August 9, 2025 05:25

nikic reviewed Aug 9, 2025

View reviewed changes

llvm/lib/Transforms/Scalar/LICM.cpp Outdated Show resolved Hide resolved

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch 2 times, most recently from cac2eba to b8aa221 Compare August 11, 2025 14:47

snehasish approved these changes Aug 11, 2025

View reviewed changes

llvm/lib/Transforms/Scalar/LICM.cpp Outdated Show resolved Hide resolved

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from b8aa221 to 223404b Compare August 11, 2025 22:27

mtrofin changed the title ~~[licm] clone metadata when hoisting conditional branch~~ [licm] clone MD_prof when hoisting conditional branch Aug 12, 2025

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from 223404b to 871de1e Compare August 12, 2025 16:59

mtrofin mentioned this pull request Aug 12, 2025

[nfc][pgo] Expose traversal of VP metadata elements #153237

Closed

nikic approved these changes Aug 12, 2025

View reviewed changes

mtrofin mentioned this pull request Aug 12, 2025

[nfc][pgo] const-ify some APIs in InstrProfSymtab #153284

Merged

[licm] clone metadata when hoisting conditional branch

8ed35ac

mtrofin force-pushed the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch from 871de1e to 8ed35ac Compare August 12, 2025 21:40

mtrofin mentioned this pull request Aug 12, 2025

[JTS] Propagate profile info #153305

Draft

mtrofin merged commit 374cbfd into main Aug 13, 2025
11 of 12 checks passed

mtrofin deleted the users/mtrofin/08-05-_licm_clone_metadata_when_hoisting_conditional_branch branch August 13, 2025 00:01

[licm] clone MD_prof when hoisting conditional branch #152232

[licm] clone MD_prof when hoisting conditional branch #152232

Conversation

mtrofin commented Aug 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mtrofin commented Aug 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Aug 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Aug 6, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

snehasish Aug 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mtrofin Aug 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

snehasish left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

nikic left a comment

Choose a reason for hiding this comment

Uh oh!

andrurogerz commented Aug 12, 2025

Uh oh!

nikic commented Aug 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

andrurogerz commented Aug 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mtrofin commented Aug 12, 2025

Uh oh!

mtrofin commented Aug 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merge activity

Uh oh!

Uh oh!

nikic commented Aug 13, 2025

Uh oh!

Uh oh!

[licm] clone `MD_prof` when hoisting conditional branch #152232

[licm] clone `MD_prof` when hoisting conditional branch #152232

mtrofin commented Aug 6, 2025 •

edited

Loading

mtrofin commented Aug 6, 2025 •

edited

Loading

llvmbot commented Aug 6, 2025 •

edited

Loading

snehasish Aug 6, 2025 •

edited

Loading

mtrofin Aug 7, 2025 •

edited

Loading

nikic commented Aug 12, 2025 •

edited

Loading

andrurogerz commented Aug 12, 2025 •

edited

Loading

mtrofin commented Aug 12, 2025 •

edited

Loading