[llvm] 5556660 - [MemCpyOpt] Handle read from lifetime.start with offset

Nikita Popov via llvm-commits llvm-commits at lists.llvm.org
Sat Mar 13 11:38:23 PST 2021


Author: Nikita Popov
Date: 2021-03-13T20:38:09+01:00
New Revision: 55566609711f284581d822576705354444e6a7f0

URL: https://github.com/llvm/llvm-project/commit/55566609711f284581d822576705354444e6a7f0
DIFF: https://github.com/llvm/llvm-project/commit/55566609711f284581d822576705354444e6a7f0.diff

LOG: [MemCpyOpt] Handle read from lifetime.start with offset

This fixes a regression from the MemDep-based implementation:
MemDep completely ignores lifetime.start intrinsics that aren't
MustAlias -- this is probably unsound, but it does mean that the
MemDep based implementation successfully eliminated memcpy's from
lifetime.start if the memcpy happens at an offset, rather than
the base address of the alloca.

Add a special case for the case where the lifetime.start spans the
whole alloca (which is pretty much the only kind of lifetime.start
that frontends ever emit), as we don't need to figure out our exact
aliasing relationship in that case, the whole alloca is dead prior
to the call.

If this doesn't cover all practically relevant cases, then it
would be possible to make use of the recently added PartialAlias
clobber offsets to make this more precise.

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
    llvm/test/Transforms/MemCpyOpt/memcpy-undef.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index f2f11c2d142c..5c1c6b2a2d8f 100644
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -1253,6 +1253,18 @@ static bool hasUndefContentsMSSA(MemorySSA *MSSA, AliasAnalysis *AA, Value *V,
       if (AA->isMustAlias(V, II->getArgOperand(1)) &&
           LTSize->getZExtValue() >= Size->getZExtValue())
         return true;
+
+      // If the lifetime.start covers a whole alloca (as it almost always does)
+      // and we're querying a pointer based on that alloca, then we know the
+      // memory is definitely undef, regardless of how exactly we alias. The
+      // size also doesn't matter, as an out-of-bounds access would be UB.
+      AllocaInst *Alloca = dyn_cast<AllocaInst>(getUnderlyingObject(V));
+      if (getUnderlyingObject(II->getArgOperand(1)) == Alloca) {
+        DataLayout DL = Alloca->getModule()->getDataLayout();
+        if (Optional<TypeSize> AllocaSize = Alloca->getAllocationSizeInBits(DL))
+          if (*AllocaSize == LTSize->getValue() * 8)
+            return true;
+      }
     }
   }
 

diff  --git a/llvm/test/Transforms/MemCpyOpt/memcpy-undef.ll b/llvm/test/Transforms/MemCpyOpt/memcpy-undef.ll
index 561c4cc3d645..983aba204af4 100644
--- a/llvm/test/Transforms/MemCpyOpt/memcpy-undef.ll
+++ b/llvm/test/Transforms/MemCpyOpt/memcpy-undef.ll
@@ -67,20 +67,12 @@ define void @test_lifetime_may_alias(i8* %lifetime, i8* %src, i8* %dst) {
 
 ; lifetime.start on full alloca size, copy in range.
 define void @test_lifetime_partial_alias_1(i8* noalias %dst) {
-; NO-MSSA-LABEL: @test_lifetime_partial_alias_1(
-; NO-MSSA-NEXT:    [[A:%.*]] = alloca [16 x i8], align 1
-; NO-MSSA-NEXT:    [[A_I8:%.*]] = bitcast [16 x i8]* [[A]] to i8*
-; NO-MSSA-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[A_I8]])
-; NO-MSSA-NEXT:    [[GEP:%.*]] = getelementptr i8, i8* [[A_I8]], i64 8
-; NO-MSSA-NEXT:    ret void
-;
-; MSSA-LABEL: @test_lifetime_partial_alias_1(
-; MSSA-NEXT:    [[A:%.*]] = alloca [16 x i8], align 1
-; MSSA-NEXT:    [[A_I8:%.*]] = bitcast [16 x i8]* [[A]] to i8*
-; MSSA-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[A_I8]])
-; MSSA-NEXT:    [[GEP:%.*]] = getelementptr i8, i8* [[A_I8]], i64 8
-; MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[DST:%.*]], i8* [[GEP]], i64 8, i1 false)
-; MSSA-NEXT:    ret void
+; CHECK-LABEL: @test_lifetime_partial_alias_1(
+; CHECK-NEXT:    [[A:%.*]] = alloca [16 x i8], align 1
+; CHECK-NEXT:    [[A_I8:%.*]] = bitcast [16 x i8]* [[A]] to i8*
+; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[A_I8]])
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, i8* [[A_I8]], i64 8
+; CHECK-NEXT:    ret void
 ;
   %a = alloca [16 x i8]
   %a.i8 = bitcast [16 x i8]* %a to i8*
@@ -92,20 +84,12 @@ define void @test_lifetime_partial_alias_1(i8* noalias %dst) {
 
 ; lifetime.start on full alloca size, copy out of range.
 define void @test_lifetime_partial_alias_2(i8* noalias %dst) {
-; NO-MSSA-LABEL: @test_lifetime_partial_alias_2(
-; NO-MSSA-NEXT:    [[A:%.*]] = alloca [16 x i8], align 1
-; NO-MSSA-NEXT:    [[A_I8:%.*]] = bitcast [16 x i8]* [[A]] to i8*
-; NO-MSSA-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[A_I8]])
-; NO-MSSA-NEXT:    [[GEP:%.*]] = getelementptr i8, i8* [[A_I8]], i64 8
-; NO-MSSA-NEXT:    ret void
-;
-; MSSA-LABEL: @test_lifetime_partial_alias_2(
-; MSSA-NEXT:    [[A:%.*]] = alloca [16 x i8], align 1
-; MSSA-NEXT:    [[A_I8:%.*]] = bitcast [16 x i8]* [[A]] to i8*
-; MSSA-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[A_I8]])
-; MSSA-NEXT:    [[GEP:%.*]] = getelementptr i8, i8* [[A_I8]], i64 8
-; MSSA-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[DST:%.*]], i8* [[GEP]], i64 16, i1 false)
-; MSSA-NEXT:    ret void
+; CHECK-LABEL: @test_lifetime_partial_alias_2(
+; CHECK-NEXT:    [[A:%.*]] = alloca [16 x i8], align 1
+; CHECK-NEXT:    [[A_I8:%.*]] = bitcast [16 x i8]* [[A]] to i8*
+; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[A_I8]])
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, i8* [[A_I8]], i64 8
+; CHECK-NEXT:    ret void
 ;
   %a = alloca [16 x i8]
   %a.i8 = bitcast [16 x i8]* %a to i8*


        


More information about the llvm-commits mailing list