2 files changed, 41 insertions, 3 deletions
diff --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 72127bb0d6e..550757a09f5 100644
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -876,6 +876,12 @@ bool MemCpyOpt::processMemSetMemCpyDependence(MemCpyInst *MemCpy,
 
   IRBuilder<> Builder(MemCpy->getNextNode());
 
+  // If the sizes have different types (i32 vs i64), promote both to i64.
+  if (DestSize->getType() != SrcSize->getType()) {
+    DestSize = Builder.CreateZExt(DestSize, Builder.getInt64Ty());
+    SrcSize = Builder.CreateZExt(SrcSize, Builder.getInt64Ty());
+  }
+
   Value *MemsetLen =
       Builder.CreateSelect(Builder.CreateICmpULE(DestSize, SrcSize),
                            ConstantInt::getNullValue(DestSize->getType()),
diff --git a/llvm/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll b/llvm/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll
index 5fa5a915a5a..cc94f58dbfb 100644
--- a/llvm/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll
+++ b/llvm/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll
@@ -8,14 +8,44 @@ target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 ; CHECK-DAG: [[ULE:%[0-9]+]] = icmp ule i64 %dst_size, %src_size
 ; CHECK-DAG: [[SIZEDIFF:%[0-9]+]] = sub i64 %dst_size, %src_size
 ; CHECK-DAG: [[SIZE:%[0-9]+]] = select i1 [[ULE]], i64 0, i64 [[SIZEDIFF]]
-; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[DST]], i8 0, i64 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[DST]], i8 %c, i64 [[SIZE]], i32 1, i1 false)
 ; CHECK-NEXT: ret void
-define void @test(i8* %src, i64 %src_size, i8* %dst, i64 %dst_size) {
-  call void @llvm.memset.p0i8.i64(i8* %dst, i8 0, i64 %dst_size, i32 1, i1 false)
+define void @test(i8* %src, i64 %src_size, i8* %dst, i64 %dst_size, i8 %c) {
+  call void @llvm.memset.p0i8.i64(i8* %dst, i8 %c, i64 %dst_size, i32 1, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
+  ret void
+}
+
+; CHECK-LABEL: define void @test_different_types_i32_i64
+; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
+; CHECK-DAG: [[DSTSIZE:%[0-9]+]] = zext i32 %dst_size to i64
+; CHECK-DAG: [[DST:%[0-9]+]] = getelementptr i8, i8* %dst, i64 %src_size
+; CHECK-DAG: [[ULE:%[0-9]+]] = icmp ule i64 [[DSTSIZE]], %src_size
+; CHECK-DAG: [[SIZEDIFF:%[0-9]+]] = sub i64 [[DSTSIZE]], %src_size
+; CHECK-DAG: [[SIZE:%[0-9]+]] = select i1 [[ULE]], i64 0, i64 [[SIZEDIFF]]
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[DST]], i8 %c, i64 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: ret void
+define void @test_different_types_i32_i64(i8* %dst, i8* %src, i32 %dst_size, i64 %src_size, i8 %c) {
+  call void @llvm.memset.p0i8.i32(i8* %dst, i8 %c, i32 %dst_size, i32 1, i1 false)
   call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
   ret void
 }
 
+; CHECK-LABEL: define void @test_different_types_i64_i32
+; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %src_size, i32 1, i1 false)
+; CHECK-DAG: [[SRCSIZE:%[0-9]+]] = zext i32 %src_size to i64
+; CHECK-DAG: [[DST:%[0-9]+]] = getelementptr i8, i8* %dst, i64 [[SRCSIZE]]
+; CHECK-DAG: [[ULE:%[0-9]+]] = icmp ule i64 %dst_size, [[SRCSIZE]]
+; CHECK-DAG: [[SIZEDIFF:%[0-9]+]] = sub i64 %dst_size, [[SRCSIZE]]
+; CHECK-DAG: [[SIZE:%[0-9]+]] = select i1 [[ULE]], i64 0, i64 [[SIZEDIFF]]
+; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[DST]], i8 %c, i64 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: ret void
+define void @test_different_types_i64_i32(i8* %dst, i8* %src, i64 %dst_size, i32 %src_size, i8 %c) {
+  call void @llvm.memset.p0i8.i64(i8* %dst, i8 %c, i64 %dst_size, i32 1, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %src_size, i32 1, i1 false)
+  ret void
+}
+
 ; CHECK-LABEL: define void @test_align_same
 ; CHECK: call void @llvm.memset.p0i8.i64(i8* {{.*}}, i8 0, i64 {{.*}}, i32 8, i1 false)
 define void @test_align_same(i8* %src, i8* %dst, i64 %dst_size) {
@@ -52,3 +82,5 @@ define void @test_different_dst(i8* %dst2, i8* %src, i64 %src_size, i8* %dst, i6
 
 declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1)
 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1)
+declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i32, i1)
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i32, i1)