[InstCombinePHI] Enhance PHI CSE to remove redundant phis #163453

CongzheUalberta · 2025-10-14T21:07:17Z

Enhanced PHI CSE to eliminate redundant PHIs, which could clean up the IR and open up opportunities for other passes such as loop vectorization.

Motivation:

Given the following range() function,

void range(float *q, float *c, float &vl, float &vr)
{
    vl = +1e20;
    vr = -1e20;    
    for (int i = 0; i < 128; i++) {
        float tmp = (*q) - (*c);
        if (tmp < vl)
            vl = tmp;
        if (tmp > vr)
            vr = tmp;
        q++;
        c++;
    }
    return;
}

The IR that is right before loop vectorization is shown below. Here range() is inlined into its caller and becomes the single BB for.body.

for.body:                                    ; preds = %entry, %for.body
  %v0 = phi float [ 0x4415AF1D80000000, %entry ], [ %v0.1, %for.body ]
  %v1 = phi float [ 0xC415AF1D80000000, %entry ], [ %v1.1, %for.body ]
  %phi.to.remove = phi float [ 0xC415AF1D80000000, %entry ], [ %phi.to.remove.next, %for.body ]  (<= redundant, needs clean-up)
  %i = phi i32 [ 0, %entry ], [ %inc.i, %for.body ]
  %q = phi ptr [ %m, %entry ], [ %q.next, %for.body ]
  %c = phi ptr [ %n, %entry ], [ %c.next, %for.body ]
  %q.load = load float, ptr %q
  %c.load = load float, ptr %c
  %sub = fsub float %q.load, %c.load
  %cmp1 = fcmp olt float %sub, %v0
  %v0.1 = select i1 %cmp1, float %sub, float %v0
  %same.as.v1 = select i1 %cmp1, float %v1, float %phi.to.remove  (<= redundant, needs clean-up)
  %cmp2 = fcmp ogt float  %sub, %same.as.v1
  %v1.1 = select i1 %cmp2, float %sub, float %v1
  %phi.to.remove.next = select i1 %cmp2, float %sub, float %same.as.v1  (<= redundant, needs clean-up)
  %inc.i = add nuw nsw i32 %i, 1
  %q.next = getelementptr inbounds float, ptr %q, i64 1
  %c.next = getelementptr inbounds float, ptr %c, i64 1
  %exitcond = icmp eq i32 %inc.i, %count
  br i1 %exitcond, label %exit, label %for.body

llvm trunk is not able to vectorize it because there are redundant phi (%phi.to.remove) and redundant select instructions (%phi.to.remove.next, %same.as.v1).

Those redundant instructions just act exactly the same as %v1, hence they are purely redundant and should be eliminated.
This patch identifies the redundant phi and eliminates it, as a result the loop could get vectorized and it could improve one of our internal workloads by 10%.

How the redundant phi was generated:

It was initially introduced by GVN that did load-in-loop-pre, which partially eliminated the load of %v1 and introduced in one of its predecessors this load %.pre = load float, ptr %v1. %.pre eventually became the redundant phi that was not cleaned up.

Compiler explorer: https://godbolt.org/z/f4ncn3Kjo
Please refer to the IR before vectorization on main(), and IR before and after GVN on range().

llvmbot · 2025-10-14T22:11:33Z

@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Congzhe (CongzheUalberta)

Changes

Enhanced PHI CSE to eliminate redundant PHIs, which could clean up the IR and open up opportunities for other passes such as loop vectorization.

Motivation:

Given the following range() function,

void range(float *q, float *c, float &amp;vl, float &amp;vr)
{
    vl = +1e20;
    vr = -1e20;    
    for (int i = 0; i &lt; 128; i++) {
        float tmp = (*q) - (*c);
        if (tmp &lt; vl)
            vl = tmp;
        if (tmp &gt; vr)
            vr = tmp;
        q++;
        c++;
    }
    return;
}

The IR that is right before loop vectorization is shown below. Here range() is inlined into its caller and becomes the single BB for.body.

for.body:                                    ; preds = %entry, %for.body
  %v0 = phi float [ 0x4415AF1D80000000, %entry ], [ %v0.1, %for.body ]
  %v1 = phi float [ 0xC415AF1D80000000, %entry ], [ %v1.1, %for.body ]
  %phi.to.remove = phi float [ 0xC415AF1D80000000, %entry ], [ %phi.to.remove.next, %for.body ]  (&lt;= redundant, needs clean-up)
  %i = phi i32 [ 0, %entry ], [ %inc.i, %for.body ]
  %q = phi ptr [ %m, %entry ], [ %q.next, %for.body ]
  %c = phi ptr [ %n, %entry ], [ %c.next, %for.body ]
  %q.load = load float, ptr %q
  %c.load = load float, ptr %c
  %sub = fsub float %q.load, %c.load
  %cmp1 = fcmp olt float %sub, %v0
  %v0.1 = select i1 %cmp1, float %sub, float %v0
  %same.as.v1 = select i1 %cmp1, float %v1, float %phi.to.remove  (&lt;= redundant, needs clean-up)
  %cmp2 = fcmp ogt float  %sub, %same.as.v1
  %v1.1 = select i1 %cmp2, float %sub, float %v1
  %phi.to.remove.next = select i1 %cmp2, float %sub, float %same.as.v1  (&lt;= redundant, needs clean-up)
  %inc.i = add nuw nsw i32 %i, 1
  %q.next = getelementptr inbounds float, ptr %q, i64 1
  %c.next = getelementptr inbounds float, ptr %c, i64 1
  %exitcond = icmp eq i32 %inc.i, %count
  br i1 %exitcond, label %exit, label %for.body

llvm trunk is not able to vectorize it because there are redundant phi (%phi.to.remove) and redundant select instructions (%phi.to.remove.next, %same.as.v1).

Those redundant instructions just act exactly the same as %v1, hence they are purely redundant and should be eliminated.
This patch identifies the redundant phi and eliminates it, as a result the loop could get vectorized and performance could get improved.

How the redundant phi was generated:

It was initially introduced by GVN that did load-in-loop-pre, which partially eliminated the load of %v1 and introduced in one of its predecessors this load %.pre = load float, ptr %v1. %.pre eventually became the redundant phi that was not cleaned up.

Compiler explorer: https://godbolt.org/z/f4ncn3Kjo
Please refer to the IR before vectorization on main(), and IR before and after GVN on range().

Full diff: https://github.com/llvm/llvm-project/pull/163453.diff

2 Files Affected:

(modified) llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp (+84-5)
(added) llvm/test/Transforms/InstCombine/enhanced-phi-cse.ll (+61)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp b/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp
index 9815644f5f43d..e736e89a3a146 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp
@@ -1621,11 +1621,90 @@ Instruction *InstCombinerImpl::visitPHINode(PHINode &PN) {
     // Note that even though we've just canonicalized this PHI, due to the
     // worklist visitation order, there are no guarantess that *every* PHI
     // has been canonicalized, so we can't just compare operands ranges.
-    if (!PN.isIdenticalToWhenDefined(&IdenticalPN))
-      continue;
-    // Just use that PHI instead then.
-    ++NumPHICSEs;
-    return replaceInstUsesWith(PN, &IdenticalPN);
+    if (PN.isIdenticalToWhenDefined(&IdenticalPN)) {
+      // Just use that PHI instead then.
+      ++NumPHICSEs;
+      return replaceInstUsesWith(PN, &IdenticalPN);
+    }
+
+    // Look for the following pattern and do PHI CSE to clean up the
+    // redundant %phi. Here %phi, %1 and %phi.next perform the same
+    // functionality as %identicalPhi and hence %phi can be eliminated.
+    //
+    // BB1:
+    //   %identicalPhi = phi [ X, %BB0 ], [ %identicalPhi.next, %BB1 ]
+    //   %phi = phi [ X, %BB0 ], [ %phi.next, %BB1 ]
+    //   ...
+    //   %identicalPhi.next = select %cmp, %val, %identicalPhi
+    //   %1 = select %cmp2, %identicalPhi, float %phi
+    //   %phi.next = select %cmp, %val, %1
+    //
+    // Prove that %phi and %identicalPhi are the same by induction:
+    //
+    // Base case: Both %phi and %identicalPhi are equal on entry to the loop.
+    // Inductive case:
+    // Suppose %phi and %identicalPhi are equal at iteration i.
+    // We look at their values at iteration i+1 which are %phi.next and
+    // %identicalPhi.next. They would have become different only when %cmp is
+    // false and the corresponding values %1 and %identicalPhi differ.
+    //
+    // The only condition when %1 and %identicalPh could differ is when %cmp2
+    // is false and %1 is %phi, which contradicts our inductive hypothesis
+    // that %phi and %identicalPhi are equal. Thus %phi and %identicalPhi are
+    // always equal at iteration i+1.
+
+    if (PN.getNumIncomingValues() == 2 && PN.getNumUses() == 1) {
+      unsigned diffVals = 0;
+      unsigned diffValIdx = 0;
+      // Check that only the backedge incoming value is different.
+      for (unsigned i = 0; i < 2; i++) {
+        if (PN.getIncomingValue(i) != IdenticalPN.getIncomingValue(i)) {
+          diffVals++;
+          diffValIdx = i;
+        }
+      }
+      BasicBlock *CurBB = PN.getParent();
+      if (diffVals == 2 || PN.getIncomingBlock(diffValIdx) != CurBB)
+        continue;
+      // Now check that the backedge incoming values are two select
+      // instructions that are in the same BB, and have the same condition,
+      // true value.
+      auto *Val = PN.getIncomingValue(diffValIdx);
+      auto *IdenticalVal = IdenticalPN.getIncomingValue(diffValIdx);
+      if (!isa<SelectInst>(Val) || !isa<SelectInst>(IdenticalVal))
+        continue;
+
+      auto *SI = cast<SelectInst>(Val);
+      auto *IdenticalSI = cast<SelectInst>(IdenticalVal);
+      if (SI->getParent() != CurBB || IdenticalSI->getParent() != CurBB)
+        continue;
+      if (SI->getCondition() != IdenticalSI->getCondition() ||
+          SI->getTrueValue() != IdenticalSI->getTrueValue())
+        continue;
+
+      // Now check that the false values, i.e., %1 and %identicalPhi,
+      // are essentially the same value within the same BB.
+      auto SameSelAndPhi = [&](SelectInst *SI, PHINode *IdenticalPN,
+                               PHINode *PN) {
+        if (SI->getTrueValue() == IdenticalPN) {
+          return SI->getFalseValue() == PN;
+        }
+        return false;
+      };
+      auto *FalseVal = SI->getFalseValue();
+      auto *IdenticalSIFalseVal =
+          dyn_cast<PHINode>(IdenticalSI->getFalseValue());
+      if (!isa<SelectInst>(FalseVal) || !IdenticalSIFalseVal ||
+          IdenticalSIFalseVal != &IdenticalPN)
+        continue;
+      auto *FalseValSI = cast<SelectInst>(FalseVal);
+      if (FalseValSI->getParent() != CurBB ||
+          !SameSelAndPhi(FalseValSI, &IdenticalPN, &PN))
+        continue;
+
+      ++NumPHICSEs;
+      return replaceInstUsesWith(PN, &IdenticalPN);
+    }
   }
 
   // If this is an integer PHI and we know that it has an illegal type, see if
diff --git a/llvm/test/Transforms/InstCombine/enhanced-phi-cse.ll b/llvm/test/Transforms/InstCombine/enhanced-phi-cse.ll
new file mode 100644
index 0000000000000..ae589b7450465
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/enhanced-phi-cse.ll
@@ -0,0 +1,61 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -S -passes=instcombine -instcombine-enhanced-phi-cse=true | FileCheck %s
+@A = extern_weak global float, align 4
+
+; %phi.to.remove acts the same as %v1, and can be eliminated with PHI CSE.
+define void @enhanced_phi_cse(ptr %m, ptr %n, i32 %count) {
+; CHECK-LABEL: @enhanced_phi_cse(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[V0:%.*]] = phi float [ 0x4415AF1D80000000, [[ENTRY:%.*]] ], [ [[V0_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[V1:%.*]] = phi float [ 0xC415AF1D80000000, [[ENTRY]] ], [ [[V1_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INC_I:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[Q:%.*]] = phi ptr [ [[M:%.*]], [[ENTRY]] ], [ [[Q_NEXT:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[C:%.*]] = phi ptr [ [[N:%.*]], [[ENTRY]] ], [ [[C_NEXT:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[Q_LOAD:%.*]] = load float, ptr [[Q]], align 4
+; CHECK-NEXT:    [[C_LOAD:%.*]] = load float, ptr [[C]], align 4
+; CHECK-NEXT:    [[SUB:%.*]] = fsub float [[Q_LOAD]], [[C_LOAD]]
+; CHECK-NEXT:    [[CMP1:%.*]] = fcmp olt float [[SUB]], [[V0]]
+; CHECK-NEXT:    [[V0_1]] = select i1 [[CMP1]], float [[SUB]], float [[V0]]
+; CHECK-NEXT:    [[CMP2:%.*]] = fcmp ogt float [[SUB]], [[V1]]
+; CHECK-NEXT:    [[V1_1]] = select i1 [[CMP2]], float [[SUB]], float [[V1]]
+; CHECK-NEXT:    [[INC_I]] = add nuw nsw i32 [[I]], 1
+; CHECK-NEXT:    [[Q_NEXT]] = getelementptr inbounds float, ptr [[Q]], i64 1
+; CHECK-NEXT:    [[C_NEXT]] = getelementptr inbounds float, ptr [[C]], i64 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INC_I]], [[COUNT:%.*]]
+; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT:%.*]], label [[FOR_BODY]]
+; CHECK:       exit:
+; CHECK-NEXT:    store float [[V1_1]], ptr @A, align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                    ; preds = %entry, %for.body
+  %v0 = phi float [ 0x4415AF1D80000000, %entry ], [ %v0.1, %for.body ]
+  %v1 = phi float [ 0xC415AF1D80000000, %entry ], [ %v1.1, %for.body ]
+  %phi.to.remove = phi float [ 0xC415AF1D80000000, %entry ], [ %phi.to.remove.next, %for.body ]
+  %i = phi i32 [ 0, %entry ], [ %inc.i, %for.body ]
+  %q = phi ptr [ %m, %entry ], [ %q.next, %for.body ]
+  %c = phi ptr [ %n, %entry ], [ %c.next, %for.body ]
+  %q.load = load float, ptr %q
+  %c.load = load float, ptr %c
+  %sub = fsub float %q.load, %c.load
+  %cmp1 = fcmp olt float %sub, %v0
+  %v0.1 = select i1 %cmp1, float %sub, float %v0
+  %same.as.v1 = select i1 %cmp1, float %v1, float %phi.to.remove
+  %cmp2 = fcmp ogt float  %sub, %same.as.v1
+  %v1.1 = select i1 %cmp2, float %sub, float %v1
+  %phi.to.remove.next = select i1 %cmp2, float %sub, float %same.as.v1
+  %inc.i = add nuw nsw i32 %i, 1
+  %q.next = getelementptr inbounds float, ptr %q, i64 1
+  %c.next = getelementptr inbounds float, ptr %c, i64 1
+  %exitcond = icmp eq i32 %inc.i, %count
+  br i1 %exitcond, label %exit, label %for.body
+
+exit:
+  %vl.1.lcssa = phi float [ %v1.1, %for.body ]
+  store float %vl.1.lcssa, ptr @A
+  ret void
+}

Enhanced PHI CSE to eliminate redundant PHIs, which could clean up the IR and open up opportunities for other passes such as loop vectorization.

dtcxzyw

I'd expect that this fold should be implemented in simplifySelectInst and simplify %1 = select %cmp2, %identicalPhi, %phi -> %identicalPhi. It is not a strong requirement, so it's up to you.

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

dtcxzyw · 2025-10-15T16:49:31Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+    // that %phi and %identicalPhi are equal. Thus %phi and %identicalPhi are
+    // always equal at iteration i+1.
+
+    if (PN.getNumIncomingValues() == 2 && PN.getNumUses() == 1) {


What is the reason for the one-use checking here?

Here it is supposed to check for the target pattern that is the redundant cycle of phi (%phi.to.remove) and two select instructions (%phi.to.remove.next, %same.as.v1). Here and for the rest of my replies I'm using the notion of those variables (%phi.to.remove, %phi.to.remove.next, %same.as.v1) from the motivative IR in the description which is the same as enhanced_phi_cse() in the test file.

I've now improved the code and I'm checking SI->getNumUses() == 1 on line 1866 instead, to make sure that %phi.to.remove.next (which had been checked to be a select) has only one use that is the backedge incoming value of %phi.to.remove. Hope it is more readable now.

As I said before, we can do the same fold by simplifying %1 = select %cmp2, %identicalPhi, %phi -> %identicalPhi. Since we never check the uses of instructions in InstSimplify, I don't think it is necessary in this case.

I've thought about it more and I do agree with you that the one-use checking is not necessary in this case. I've now deleted the one-use checking. Thanks for pointing it out!

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

dtcxzyw · 2025-10-15T16:51:59Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+        }
+      }
+      BasicBlock *CurBB = PN.getParent();
+      if (diffVals == 2 || PN.getIncomingBlock(diffValIdx) != CurBB)


The header and latch block can be different.

In this patch I limited it to the case where the loop is a single BB, hence redundant cycle of phi and selects are in the same BB and control flow equivalent. With multiple BB and control flow divergence, it would become more complex and not very straightforward to prove that they are purely redundant. Therefore I'd like to make it work for the single BB case, and then I could possibly extend it to multi-BB scenarios. Does it sound reasonable to you?

Can you please provide a counterexample to demonstrate that removing this constraint may lead to a crash or miscompilation?

I thought more about it and I think you are right. I could not come up with a counterexample that would break my code when this constraint was removed - it would work correctly without the single BB constraint. I've now removed the constraint and simplified the code. Thanks for this comment.

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

dtcxzyw · 2025-10-17T17:57:03Z

Crash reproducer:

; bin/opt -passes=instcombine test.ll -S
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"

define fastcc i32 @vduse_queue_check_inflights() {
._crit_edge:
  br label %.lr.ph54

.lr.ph54:
  %0 = phi ptr [ null, %._crit_edge ], [ %3, %.lr.ph54 ]
  %1 = phi ptr [ %3, %.lr.ph54 ], [ null, %._crit_edge ]
  %2 = load i8, ptr %1, align 8
  %.not51 = icmp eq i8 %2, 0
  %3 = select i1 %.not51, ptr %0, ptr null
  br label %.lr.ph54
}

opt: /home/dtcxzyw/WorkSpace/Projects/compilers/llvm-project/llvm/include/llvm/IR/Instructions.h:2817: llvm::Value* llvm::PHINode::getIncomingValueForBlock(const llvm::BasicBlock*) const: Assertion `Idx >= 0 && "Invalid basic block argument!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace and instructions to reproduce the bug.
Stack dump:
0.      Program arguments: bin/opt -passes=instcombine reduced.ll -S
1.      Running pass "function(instcombine<max-iterations=1;verify-fixpoint>)" on module "reduced.ll"
2.      Running pass "instcombine<max-iterations=1;verify-fixpoint>" on function "vduse_queue_check_inflights"
 #0 0x000071702d854192 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMSupport.so.22.0git+0x254192)
 #1 0x000071702d85075f llvm::sys::RunSignalHandlers() (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMSupport.so.22.0git+0x25075f)
 #2 0x000071702d8508ac SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0
 #3 0x000071702d245330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)
 #4 0x000071702d29eb2c __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
 #5 0x000071702d29eb2c __pthread_kill_internal ./nptl/pthread_kill.c:78:10
 #6 0x000071702d29eb2c pthread_kill ./nptl/pthread_kill.c:89:10
 #7 0x000071702d24527e raise ./signal/../sysdeps/posix/raise.c:27:6
 #8 0x000071702d2288ff abort ./stdlib/abort.c:81:7
 #9 0x000071702d22881b _nl_load_domain ./intl/loadmsgcat.c:1177:9
#10 0x000071702d23b517 (/lib/x86_64-linux-gnu/libc.so.6+0x3b517)
#11 0x0000717027c38c4a llvm::PHINode::getIncomingValueForBlock(llvm::BasicBlock const*) const (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x38c4a)
#12 0x0000717027d5bfa9 llvm::InstCombinerImpl::visitPHINode(llvm::PHINode&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x15bfa9)
#13 0x0000717027c68ff8 llvm::InstCombinerImpl::run() (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x68ff8)
#14 0x0000717027c6a1c2 combineInstructionsOverFunction(llvm::Function&, llvm::InstructionWorklist&, llvm::AAResults*, llvm::AssumptionCache&, llvm::TargetLibraryInfo&, llvm::TargetTransformInfo&, llvm::DominatorTree&, llvm::OptimizationRemarkEmitter&, llvm::BlockFrequencyInfo*, llvm::BranchProbabilityInfo*, llvm::ProfileSummaryInfo*, llvm::InstCombineOptions const&) InstructionCombining.cpp:0:0
#15 0x0000717027c6b204 llvm::InstCombinePass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x6b204)
#16 0x000071702a3a8ee5 llvm::detail::PassModel<llvm::Function, llvm::InstCombinePass, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libPolly.so.22.0git+0x1a8ee5)
#17 0x0000717026b2c299 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMCore.so.22.0git+0x32c299)
#18 0x000071702cadd335 llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMX86CodeGen.so.22.0git+0xdd335)
#19 0x0000717026b2a722 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMCore.so.22.0git+0x32a722)
#20 0x000071702dad5285 llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMOptDriver.so.22.0git+0x20285)
#21 0x0000717026b2af4d llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMCore.so.22.0git+0x32af4d)
#22 0x000071702dae2261 llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::ArrayRef<std::function<void (llvm::PassBuilder&)>>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool, bool) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMOptDriver.so.22.0git+0x2d261)
#23 0x000071702daed436 optMain (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMOptDriver.so.22.0git+0x38436)
#24 0x000071702d22a1ca __libc_start_call_main ./csu/../sysdeps/nptl/libc_start_call_main.h:74:3
#25 0x000071702d22a28b call_init ./csu/../csu/libc-start.c:128:20
#26 0x000071702d22a28b __libc_start_main ./csu/../csu/libc-start.c:347:5
#27 0x0000622cb7c80095 _start (bin/opt+0x1095)
Aborted (core dumped)

CongzheUalberta · 2025-10-17T18:52:05Z

Crash reproducer:

; bin/opt -passes=instcombine test.ll -S
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"

define fastcc i32 @vduse_queue_check_inflights() {
._crit_edge:
  br label %.lr.ph54

.lr.ph54:
  %0 = phi ptr [ null, %._crit_edge ], [ %3, %.lr.ph54 ]
  %1 = phi ptr [ %3, %.lr.ph54 ], [ null, %._crit_edge ]
  %2 = load i8, ptr %1, align 8
  %.not51 = icmp eq i8 %2, 0
  %3 = select i1 %.not51, ptr %0, ptr null
  br label %.lr.ph54
}

opt: /home/dtcxzyw/WorkSpace/Projects/compilers/llvm-project/llvm/include/llvm/IR/Instructions.h:2817: llvm::Value* llvm::PHINode::getIncomingValueForBlock(const llvm::BasicBlock*) const: Assertion `Idx >= 0 && "Invalid basic block argument!"' failed.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace and instructions to reproduce the bug.
Stack dump:
0.      Program arguments: bin/opt -passes=instcombine reduced.ll -S
1.      Running pass "function(instcombine<max-iterations=1;verify-fixpoint>)" on module "reduced.ll"
2.      Running pass "instcombine<max-iterations=1;verify-fixpoint>" on function "vduse_queue_check_inflights"
 #0 0x000071702d854192 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMSupport.so.22.0git+0x254192)
 #1 0x000071702d85075f llvm::sys::RunSignalHandlers() (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMSupport.so.22.0git+0x25075f)
 #2 0x000071702d8508ac SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0
 #3 0x000071702d245330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)
 #4 0x000071702d29eb2c __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
 #5 0x000071702d29eb2c __pthread_kill_internal ./nptl/pthread_kill.c:78:10
 #6 0x000071702d29eb2c pthread_kill ./nptl/pthread_kill.c:89:10
 #7 0x000071702d24527e raise ./signal/../sysdeps/posix/raise.c:27:6
 #8 0x000071702d2288ff abort ./stdlib/abort.c:81:7
 #9 0x000071702d22881b _nl_load_domain ./intl/loadmsgcat.c:1177:9
#10 0x000071702d23b517 (/lib/x86_64-linux-gnu/libc.so.6+0x3b517)
#11 0x0000717027c38c4a llvm::PHINode::getIncomingValueForBlock(llvm::BasicBlock const*) const (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x38c4a)
#12 0x0000717027d5bfa9 llvm::InstCombinerImpl::visitPHINode(llvm::PHINode&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x15bfa9)
#13 0x0000717027c68ff8 llvm::InstCombinerImpl::run() (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x68ff8)
#14 0x0000717027c6a1c2 combineInstructionsOverFunction(llvm::Function&, llvm::InstructionWorklist&, llvm::AAResults*, llvm::AssumptionCache&, llvm::TargetLibraryInfo&, llvm::TargetTransformInfo&, llvm::DominatorTree&, llvm::OptimizationRemarkEmitter&, llvm::BlockFrequencyInfo*, llvm::BranchProbabilityInfo*, llvm::ProfileSummaryInfo*, llvm::InstCombineOptions const&) InstructionCombining.cpp:0:0
#15 0x0000717027c6b204 llvm::InstCombinePass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMInstCombine.so.22.0git+0x6b204)
#16 0x000071702a3a8ee5 llvm::detail::PassModel<llvm::Function, llvm::InstCombinePass, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libPolly.so.22.0git+0x1a8ee5)
#17 0x0000717026b2c299 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMCore.so.22.0git+0x32c299)
#18 0x000071702cadd335 llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMX86CodeGen.so.22.0git+0xdd335)
#19 0x0000717026b2a722 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMCore.so.22.0git+0x32a722)
#20 0x000071702dad5285 llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMOptDriver.so.22.0git+0x20285)
#21 0x0000717026b2af4d llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/../lib/libLLVMCore.so.22.0git+0x32af4d)
#22 0x000071702dae2261 llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::ArrayRef<std::function<void (llvm::PassBuilder&)>>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool, bool) (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMOptDriver.so.22.0git+0x2d261)
#23 0x000071702daed436 optMain (/home/dtcxzyw/WorkSpace/Projects/compilers/LLVM/llvm-build/bin/../lib/libLLVMOptDriver.so.22.0git+0x38436)
#24 0x000071702d22a1ca __libc_start_call_main ./csu/../sysdeps/nptl/libc_start_call_main.h:74:3
#25 0x000071702d22a28b call_init ./csu/../csu/libc-start.c:128:20
#26 0x000071702d22a28b __libc_start_main ./csu/../csu/libc-start.c:347:5
#27 0x0000622cb7c80095 _start (bin/opt+0x1095)
Aborted (core dumped)

Thanks for the case.

I've also checked other failures in the last CI, they occurred because with %0 = phi ptr [ null, %._crit_edge ], [ %3, %.lr.ph54 ] and %1 = phi ptr [ %3, %.lr.ph54 ], [ null, %._crit_edge ], DiffVals actually is 0 and this case slipped into my code since I did not explicitly have if (DiffVals == 0) continue.

I've fixed it by using if (DiffVals != 1) continue because the target pattern of this patch is explicitly that one of the two incoming values of phis differ. CI has passed, and I've also checked that your example IR works fine.

dtcxzyw

I find that it is more natural to handle this pattern in simplifySelectInst:

bool isSimplifierIdenticalPHI(PHINode &PHI, PHINode &IdenticalPHI) {
  if (PHI.getParent() != IdenticalPHI.getParent())
    return false;
  // Check incoming values
  ...
  // Check next values
  ...
}
Value *simplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal,
                                 const SimplifyQuery &Q, unsigned MaxRecurse) {
  ...

  if (auto *TruePHI = dyn_cast<PHINode>(TrueVal)) {
    if (auto *FalsePHI = dyn_cast<PHINode>(FalseVal)) {
    if (isSimplifierIdenticalPHI(*TruePHI, *FalsePHI))
      return FalseVal;
    if (isSimplifierIdenticalPHI(*TruePHI, *FalsePHI))
      return FalseVal;
    if (isSimplifierIdenticalPHI(*FalsePHI, *TruePHI))
      return TrueVal;
  }
  }
  return nullptr;
}

dtcxzyw · 2025-10-18T09:27:58Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+      if (!isa<SelectInst>(Val) || !isa<SelectInst>(IdenticalVal))
+        continue;
+
+      auto *SI = cast<SelectInst>(Val);


Use dyn_cast instead of isa + cast.

Thanks, I've used dyn_cast now.

dtcxzyw · 2025-10-18T09:28:16Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+        continue;
+      if (cast<PHINode>(IdenticalSIOtherVal) != &IdenticalPN)
+        continue;
+      auto *SIOtherValAsSel = cast<SelectInst>(SIOtherVal);


Use dyn_cast instead of isa + cast.

Thanks, I've used dyn_cast now.

dtcxzyw · 2025-10-18T09:29:01Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+      };
+      if (!isa<SelectInst>(SIOtherVal) || !isa<PHINode>(IdenticalSIOtherVal))
+        continue;
+      if (cast<PHINode>(IdenticalSIOtherVal) != &IdenticalPN)


Suggested change

if (cast<PHINode>(IdenticalSIOtherVal) != &IdenticalPN)

if (IdenticalSIOtherVal != &IdenticalPN)

Cast doesn't change the pointer.

Addressed accordingly.

dtcxzyw · 2025-10-18T09:35:00Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+        if (SI->getTrueValue() == IdenticalPN) {
+          return SI->getFalseValue() == PN;
+        }
+        return false;


Suggested change

if (SI->getTrueValue() == IdenticalPN) {

return SI->getFalseValue() == PN;

}

return false;

return SI->getTrueValue() == IdenticalPN && SI->getFalseValue() == PN;

We don't need a lambda function.

Removed the lambda function.

dtcxzyw · 2025-10-18T09:37:04Z

llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp

+    //   ...
+    //   %identicalPhi.next = select %cmp, %val, %identicalPhi
+    //                      (or select %cmp, %identicalPhi, %val)
+    //   %1 = select %cmp2, %identicalPhi, %phi


Can you please also add a commuted test with %1 = select %cmp2, %phi, %identicalPhi?

Twe more test cases added that reflect the commuted test, i.e., select_with_identical_phi_3() and select_with_identical_phi_4().

CongzheUalberta · 2025-10-20T23:33:57Z

I find that it is more natural to handle this pattern in simplifySelectInst:

bool isSimplifierIdenticalPHI(PHINode &PHI, PHINode &IdenticalPHI) {
  if (PHI.getParent() != IdenticalPHI.getParent())
    return false;
  // Check incoming values
  ...
  // Check next values
  ...
}
Value *simplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal,
                                 const SimplifyQuery &Q, unsigned MaxRecurse) {
  ...

  if (auto *TruePHI = dyn_cast<PHINode>(TrueVal)) {
    if (auto *FalsePHI = dyn_cast<PHINode>(FalseVal)) {
    if (isSimplifierIdenticalPHI(*TruePHI, *FalsePHI))
      return FalseVal;
    if (isSimplifierIdenticalPHI(*TruePHI, *FalsePHI))
      return FalseVal;
    if (isSimplifierIdenticalPHI(*FalsePHI, *TruePHI))
      return TrueVal;
  }
  }
  return nullptr;
}

I have updated the patch so that the feature is implemented in simplifySelectInst now. Aslo added two more test cases.

dtcxzyw · 2025-10-21T14:48:04Z

llvm/lib/Analysis/InstructionSimplify.cpp

+// The only condition when %1 and %identicalPh could differ is when %cmp2
+// is false and %1 is %phi, which contradicts our inductive hypothesis
+// that %phi and %identicalPhi are equal. Thus %phi and %identicalPhi are
+// always equal at iteration i+1.


Use /// for header comments.

Thanks, I've used /// now.

dtcxzyw · 2025-10-21T14:50:58Z

llvm/lib/Analysis/InstructionSimplify.cpp

+  if (SI->getCondition() != IdenticalSI->getCondition() ||
+      (SI->getTrueValue() != IdenticalSI->getTrueValue() &&
+       SI->getFalseValue() != IdenticalSI->getFalseValue()))


Suggested change

if (SI->getCondition() != IdenticalSI->getCondition() ||

(SI->getTrueValue() != IdenticalSI->getTrueValue() &&

SI->getFalseValue() != IdenticalSI->getFalseValue()))

if (SI->getCondition() != IdenticalSI->getCondition())

Thanks, updated accordingly.

dtcxzyw · 2025-10-21T14:51:13Z

llvm/lib/Analysis/InstructionSimplify.cpp

+  if (SI->getTrueValue() == IdenticalSI->getTrueValue()) {
+    SIOtherVal = dyn_cast<SelectInst>(SI->getFalseValue());
+    IdenticalSIOtherVal = IdenticalSI->getFalseValue();
+  } else {


Suggested change

} else {

} else if (SI->getFalseValue() == IdenticalSI->getFalseValue()) {

Updated accordingly.

dtcxzyw · 2025-10-21T14:51:30Z

llvm/lib/Analysis/InstructionSimplify.cpp

+  } else {
+    SIOtherVal = dyn_cast<SelectInst>(SI->getTrueValue());
+    IdenticalSIOtherVal = IdenticalSI->getTrueValue();
+  }


Suggested change

}

} else {

return false;

}

Updated accordingly.

dtcxzyw

LG

dtcxzyw · 2025-10-21T16:40:25Z

llvm/lib/Analysis/InstructionSimplify.cpp

 }

+/// Look for the following pattern and simplify %1 to %identicalPhi.
+/// Here %phi, %1 and %phi.next perform the same functionality as


It would be better to give %1 a meaningful name.

I've changed %1 to %to_fold.

CongzheUalberta · 2025-10-21T17:12:57Z

LG

Thanks a lot for the review!

CongzheUalberta force-pushed the enhanced-phi-cse branch from 6432743 to 306b429 Compare October 14, 2025 21:39

CongzheUalberta marked this pull request as ready for review October 14, 2025 22:10

CongzheUalberta requested a review from nikic as a code owner October 14, 2025 22:10

llvmbot added llvm:instcombine Covers the InstCombine, InstSimplify and AggressiveInstCombine passes llvm:transforms labels Oct 14, 2025

CongzheUalberta requested review from Chengjunp, aengelke, antoniofrighetto, davemgreen and fhahn October 14, 2025 22:21

[InstCombinePHI] Enhance PHI CSE to remove redundant phis

4715789

Enhanced PHI CSE to eliminate redundant PHIs, which could clean up the IR and open up opportunities for other passes such as loop vectorization.

CongzheUalberta force-pushed the enhanced-phi-cse branch from 5843a60 to 4715789 Compare October 14, 2025 22:47

antoniofrighetto requested a review from dtcxzyw October 15, 2025 12:33

dtcxzyw reviewed Oct 15, 2025

View reviewed changes

CongzheUalberta force-pushed the enhanced-phi-cse branch from ff369b4 to 78bfbef Compare October 15, 2025 23:05

address reviewer's comments

fbc4a5f

CongzheUalberta force-pushed the enhanced-phi-cse branch from 78bfbef to fbc4a5f Compare October 15, 2025 23:16

dtcxzyw mentioned this pull request Oct 17, 2025

Task submission dtcxzyw/llvm-opt-benchmark#1312

Open

zyw-bot mentioned this pull request Oct 17, 2025

pre-commit: PR163453 dtcxzyw/llvm-opt-benchmark#2943

Closed

Address reviewer's comments and simplify the code.

4cfc7bf

CongzheUalberta force-pushed the enhanced-phi-cse branch from 1e9166c to 4cfc7bf Compare October 17, 2025 18:06

zyw-bot mentioned this pull request Oct 18, 2025

pre-commit: PR163453 dtcxzyw/llvm-opt-benchmark#2944

Closed

dtcxzyw mentioned this pull request Oct 18, 2025

Fuzz PR163453 dtcxzyw/llvm-fuzz-service#147

Closed

dtcxzyw reviewed Oct 18, 2025

View reviewed changes

llvmbot added the llvm:analysis Includes value tracking, cost tables and constant folding label Oct 20, 2025

Re-implemented in InstSimplify.

ee823bc

CongzheUalberta force-pushed the enhanced-phi-cse branch from fc743e2 to ee823bc Compare October 20, 2025 23:40

dtcxzyw reviewed Oct 21, 2025

View reviewed changes

zyw-bot mentioned this pull request Oct 21, 2025

pre-commit: PR163453 dtcxzyw/llvm-opt-benchmark#2963

Closed

Address reviewer's comments made on Oct 21st.

60777d5

dtcxzyw approved these changes Oct 21, 2025

View reviewed changes

Address reviewer's comments made on Oct 21st, 2nd part.

4d8a564

CongzheUalberta merged commit 9a9fbbb into llvm:main Oct 21, 2025
10 checks passed

	if (cast<PHINode>(IdenticalSIOtherVal) != &IdenticalPN)
	if (IdenticalSIOtherVal != &IdenticalPN)

	} else {
	} else if (SI->getFalseValue() == IdenticalSI->getFalseValue()) {

[InstCombinePHI] Enhance PHI CSE to remove redundant phis #163453

[InstCombinePHI] Enhance PHI CSE to remove redundant phis #163453

Conversation

CongzheUalberta commented Oct 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation:

Uh oh!

llvmbot commented Oct 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation:

Uh oh!

dtcxzyw left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

CongzheUalberta Oct 15, 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!

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!

Uh oh!

Uh oh!

dtcxzyw commented Oct 17, 2025

Uh oh!

CongzheUalberta commented Oct 17, 2025

Uh oh!

dtcxzyw left a comment

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!

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!

CongzheUalberta commented Oct 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

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

CongzheUalberta commented Oct 14, 2025 •

edited

Loading

llvmbot commented Oct 14, 2025 •

edited

Loading

CongzheUalberta Oct 15, 2025 •

edited

Loading

CongzheUalberta commented Oct 20, 2025 •

edited

Loading