[llvm] 54067c5 - [SROA] Use memcpy if type size does not match store size

[Nikita Popov via llvm-commits]

Author: Nikita Popov
Date: 2023-12-22T10:19:22+01:00
New Revision: 54067c5fbe9fc13ab195cdddb8f17e18d72b5fe4

URL: https://github.com/llvm/llvm-project/commit/54067c5fbe9fc13ab195cdddb8f17e18d72b5fe4
DIFF: https://github.com/llvm/llvm-project/commit/54067c5fbe9fc13ab195cdddb8f17e18d72b5fe4.diff

LOG: [SROA] Use memcpy if type size does not match store size

The original memcpy also copies the padding, so make sure that
this is still the case after splitting.

Fixes https://github.com/llvm/llvm-project/issues/64081.

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/SROA.cpp
    llvm/test/Transforms/SROA/pr64081.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/SROA.cpp b/llvm/lib/Transforms/Scalar/SROA.cpp
index 24da26c9f0f253..656abdb0abbffa 100644
--- a/llvm/lib/Transforms/Scalar/SROA.cpp
+++ b/llvm/lib/Transforms/Scalar/SROA.cpp
@@ -3285,6 +3285,7 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
         (BeginOffset > NewAllocaBeginOffset || EndOffset < NewAllocaEndOffset ||
          SliceSize !=
              DL.getTypeStoreSize(NewAI.getAllocatedType()).getFixedValue() ||
+         !DL.typeSizeEqualsStoreSize(NewAI.getAllocatedType()) ||
          !NewAI.getAllocatedType()->isSingleValueType());
 
     // If we're just going to emit a memcpy, the alloca hasn't changed, and the

diff  --git a/llvm/test/Transforms/SROA/pr64081.ll b/llvm/test/Transforms/SROA/pr64081.ll
index e03e5c497b5fb5..4b893842138263 100644
--- a/llvm/test/Transforms/SROA/pr64081.ll
+++ b/llvm/test/Transforms/SROA/pr64081.ll
@@ -3,18 +3,17 @@
 
 %B = type { i1, i3 }
 
-; FIXME: This is a miscompile.
 define void @test(i7 %x) {
 ; CHECK-LABEL: define void @test(
 ; CHECK-SAME: i7 [[X:%.*]]) {
 ; CHECK-NEXT:  bb:
 ; CHECK-NEXT:    [[RES:%.*]] = alloca [2 x i8], align 1
+; CHECK-NEXT:    [[TMP_SROA_0:%.*]] = alloca i1, align 8
 ; CHECK-NEXT:    [[TMP_SROA_1:%.*]] = alloca i3, align 1
 ; CHECK-NEXT:    store i7 [[X]], ptr [[TMP_SROA_1]], align 1
-; CHECK-NEXT:    store i1 undef, ptr [[RES]], align 1
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 1 [[RES]], ptr align 8 [[TMP_SROA_0]], i64 1, i1 false)
 ; CHECK-NEXT:    [[TMP_SROA_1_0_RES_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[RES]], i64 1
-; CHECK-NEXT:    [[TMP_SROA_1_0_TMP_SROA_1_0_COPYLOAD:%.*]] = load i3, ptr [[TMP_SROA_1]], align 1
-; CHECK-NEXT:    store i3 [[TMP_SROA_1_0_TMP_SROA_1_0_COPYLOAD]], ptr [[TMP_SROA_1_0_RES_SROA_IDX]], align 1
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 1 [[TMP_SROA_1_0_RES_SROA_IDX]], ptr align 1 [[TMP_SROA_1]], i64 1, i1 false)
 ; CHECK-NEXT:    [[TMP0:%.*]] = call i8 @use(ptr [[RES]])
 ; CHECK-NEXT:    ret void
 ;