[MemCpyOpt] Don't move the memset when optimizing memset+memcpy.

Fixes PR23599, another miscompile introduced by r235232: when there is
another dependency on the destination of the created memset (i.e., the
part of the original destination that the memcpy doesn't depend on)
between the memcpy and the original memset, we would insert the created
memset after the memcpy, and thus after the other dependency.

Instead, insert the created memset right after the old one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237858 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 7ac968159c5a4f4ba633480519d567661f30c8dc..ae5501e588477d4839b4dc1eeee42fe677a36000 100644 (file)
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -874,7 +874,7 @@ bool MemCpyOpt::processMemSetMemCpyDependence(MemCpyInst *MemCpy,
     if (ConstantInt *SrcSizeC = dyn_cast<ConstantInt>(SrcSize))
       Align = MinAlign(SrcSizeC->getZExtValue(), DestAlign);
 
-  IRBuilder<> Builder(MemCpy->getNextNode());
+  IRBuilder<> Builder(MemSet->getNextNode());
 
   // If the sizes have different types, zext the smaller one.
   if (DestSize->getType() != SrcSize->getType()) {

diff --git a/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll b/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll
index 983ec44335f2da210aac4501ecc9e98505cd9e7c..25411f2698278684bbee8b2c2ba0e155a95a5c1d 100644 (file)
--- a/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll
+++ b/test/Transforms/MemCpyOpt/memset-memcpy-redundant-memset.ll
@@ -3,12 +3,12 @@
 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 
 ; CHECK-LABEL: define void @test
-; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
 ; CHECK-DAG: [[DST:%[0-9]+]] = getelementptr i8, i8* %dst, i64 %src_size
 ; 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 %c, i64 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
 ; CHECK-NEXT: ret void
 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)
@@ -17,13 +17,13 @@ define void @test(i8* %src, i64 %src_size, i8* %dst, i64 %dst_size, i8 %c) {
 }
 
 ; 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: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_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)
@@ -32,13 +32,13 @@ define void @test_different_types_i32_i64(i8* %dst, i8* %src, i32 %dst_size, i64
 }
 
 ; CHECK-LABEL: define void @test_different_types_i128_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 i128
 ; CHECK-DAG: [[DST:%[0-9]+]] = getelementptr i8, i8* %dst, i128 [[SRCSIZE]]
 ; CHECK-DAG: [[ULE:%[0-9]+]] = icmp ule i128 %dst_size, [[SRCSIZE]]
 ; CHECK-DAG: [[SIZEDIFF:%[0-9]+]] = sub i128 %dst_size, [[SRCSIZE]]
 ; CHECK-DAG: [[SIZE:%[0-9]+]] = select i1 [[ULE]], i128 0, i128 [[SIZEDIFF]]
 ; CHECK-NEXT: call void @llvm.memset.p0i8.i128(i8* [[DST]], i8 %c, i128 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %src_size, i32 1, i1 false)
 ; CHECK-NEXT: ret void
 define void @test_different_types_i128_i32(i8* %dst, i8* %src, i128 %dst_size, i32 %src_size, i8 %c) {
   call void @llvm.memset.p0i8.i128(i8* %dst, i8 %c, i128 %dst_size, i32 1, i1 false)
@@ -47,13 +47,13 @@ define void @test_different_types_i128_i32(i8* %dst, i8* %src, i128 %dst_size, i
 }
 
 ; CHECK-LABEL: define void @test_different_types_i32_i128
-; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i128(i8* %dst, i8* %src, i128 %src_size, i32 1, i1 false)
 ; CHECK-DAG: [[DSTSIZE:%[0-9]+]] = zext i32 %dst_size to i128
 ; CHECK-DAG: [[DST:%[0-9]+]] = getelementptr i8, i8* %dst, i128 %src_size
 ; CHECK-DAG: [[ULE:%[0-9]+]] = icmp ule i128 [[DSTSIZE]], %src_size
 ; CHECK-DAG: [[SIZEDIFF:%[0-9]+]] = sub i128 [[DSTSIZE]], %src_size
 ; CHECK-DAG: [[SIZE:%[0-9]+]] = select i1 [[ULE]], i128 0, i128 [[SIZEDIFF]]
 ; CHECK-NEXT: call void @llvm.memset.p0i8.i128(i8* [[DST]], i8 %c, i128 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i128(i8* %dst, i8* %src, i128 %src_size, i32 1, i1 false)
 ; CHECK-NEXT: ret void
 define void @test_different_types_i32_i128(i8* %dst, i8* %src, i32 %dst_size, i128 %src_size, i8 %c) {
   call void @llvm.memset.p0i8.i32(i8* %dst, i8 %c, i32 %dst_size, i32 1, i1 false)
@@ -62,13 +62,13 @@ define void @test_different_types_i32_i128(i8* %dst, i8* %src, i32 %dst_size, i1
 }
 
 ; 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: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %src_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)
@@ -102,12 +102,12 @@ define void @test_align_memcpy(i8* %src, i8* %dst, i64 %dst_size) {
 
 ; CHECK-LABEL: define void @test_non_i8_dst_type
 ; CHECK-NEXT: %dst = bitcast i64* %dst_pi64 to i8*
-; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
 ; CHECK-DAG: [[DST:%[0-9]+]] = getelementptr i8, i8* %dst, i64 %src_size
 ; 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 %c, i64 [[SIZE]], i32 1, i1 false)
+; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %src_size, i32 1, i1 false)
 ; CHECK-NEXT: ret void
 define void @test_non_i8_dst_type(i8* %src, i64 %src_size, i64* %dst_pi64, i64 %dst_size, i8 %c) {
   %dst = bitcast i64* %dst_pi64 to i8*