[llvm] 23db972 - [InstCombine] Fix infinite loop in min/max load/store bitcast combine (PR44835)

[Nikita Popov via llvm-commits]

Author: Nikita Popov
Date: 2020-02-08T16:55:22+01:00
New Revision: 23db9724d0e5490fa5a2a726acf015f84e2c87cf

URL: https://github.com/llvm/llvm-project/commit/23db9724d0e5490fa5a2a726acf015f84e2c87cf
DIFF: https://github.com/llvm/llvm-project/commit/23db9724d0e5490fa5a2a726acf015f84e2c87cf.diff

LOG: [InstCombine] Fix infinite loop in min/max load/store bitcast combine (PR44835)

Fixes https://bugs.llvm.org/show_bug.cgi?id=44835. Skip the transform
if it wouldn't actually do anything (apart from removing and reinserting
the same instructions).

Note that the test case doesn't loop on current master anymore, only
on the LLVM 10 release branch. The issue is already mitigated on master
due to worklist order fixes, but we should fix the root cause there as well.

As a side note, we should probably assert in combineLoadToNewType()
that it does not combine to the same type. Not doing this here, because
this assertion would also be triggered in another place right now.

Differential Revision: https://reviews.llvm.org/D74278

Added: 
    llvm/test/Transforms/InstCombine/pr44835.ll

Modified: 
    llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index a2be01585b05..4db482646ab2 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -1320,6 +1320,11 @@ static bool removeBitcastsFromLoadStoreOnMinMax(InstCombiner &IC,
   if (!isMinMaxWithLoads(LoadAddr, CmpLoadTy))
     return false;
 
+  // Make sure the type would actually change.
+  // This condition can be hit with chains of bitcasts.
+  if (LI->getType() == CmpLoadTy)
+    return false;
+
   // Make sure we're not changing the size of the load/store.
   const auto &DL = IC.getDataLayout();
   if (DL.getTypeStoreSizeInBits(LI->getType()) !=

diff  --git a/llvm/test/Transforms/InstCombine/pr44835.ll b/llvm/test/Transforms/InstCombine/pr44835.ll
new file mode 100644
index 000000000000..46c9fca6bb7a
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/pr44835.ll
@@ -0,0 +1,29 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S -instcombine < %s | FileCheck %s
+
+; This test used to cause an infinite loop in the load/store min/max bitcast
+; transform.
+
+define void @test(i32* %p, i32* %p2) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:    [[V:%.*]] = load i32, i32* [[P:%.*]], align 4
+; CHECK-NEXT:    [[V2:%.*]] = load i32, i32* [[P2:%.*]], align 4
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[V2]], [[V]]
+; CHECK-NEXT:    [[TMP1:%.*]] = select i1 [[CMP]], i32 [[V2]], i32 [[V]]
+; CHECK-NEXT:    store i32 [[TMP1]], i32* [[P]], align 4
+; CHECK-NEXT:    ret void
+;
+  %v = load i32, i32* %p, align 4
+  %v2 = load i32, i32* %p2, align 4
+  %cmp = icmp ult i32 %v2, %v
+  %sel = select i1 %cmp, i32* %p2, i32* %p
+  %p8 = bitcast i32* %p to i8*
+  %sel8 = bitcast i32* %sel to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %p8, i8* align 4 %sel8, i64 4, i1 false)
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8* noalias nocapture readonly, i64, i1 immarg) #0
+
+attributes #0 = { argmemonly nounwind willreturn }