MemCpyOptimizer: don't create new addrspace casts

This isn't safe on all targets, and since we don't have a way
to know it's safe, avoid doing it for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297788 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index c5e69cd0f9915d34600d68af8985ea3c6cfb2c13..f161d6710abf3701cf83ba3993381b19dc935004 100644 (file)
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -932,6 +932,17 @@ bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpy, Value *cpyDest,
   if (MR != MRI_NoModRef)
     return false;
 
+  // We can't create address space casts here because we don't know if they're
+  // safe for the target.
+  if (cpySrc->getType()->getPointerAddressSpace() !=
+      cpyDest->getType()->getPointerAddressSpace())
+    return false;
+  for (unsigned i = 0; i < CS.arg_size(); ++i)
+    if (CS.getArgument(i)->stripPointerCasts() == cpySrc &&
+        cpySrc->getType()->getPointerAddressSpace() !=
+        CS.getArgument(i)->getType()->getPointerAddressSpace())
+      return false;
+
   // All the checks have passed, so do the transformation.
   bool changedArgument = false;
   for (unsigned i = 0; i < CS.arg_size(); ++i)

diff --git a/test/Transforms/MemCpyOpt/memcpy.ll b/test/Transforms/MemCpyOpt/memcpy.ll
index 6181543cfc63a7c2b6b121638ebe86e5942adda4..f638d8d57018501d47c297cafce7db083d16983f 100644 (file)
--- a/test/Transforms/MemCpyOpt/memcpy.ll
+++ b/test/Transforms/MemCpyOpt/memcpy.ll
@@ -202,6 +202,21 @@ define void @test10(%opaque* noalias nocapture sret %x, i32 %y) {
   ret void
 }
 
+; don't create new addressspacecasts when we don't know they're safe for the target
+define void @test11([20 x i32] addrspace(1)* nocapture dereferenceable(80) %P) {
+  %A = alloca [20 x i32], align 4
+  %a = bitcast [20 x i32]* %A to i8*
+  %b = bitcast [20 x i32] addrspace(1)* %P to i8 addrspace(1)*
+  call void @llvm.memset.p0i8.i64(i8* %a, i8 0, i64 80, i32 4, i1 false)
+  call void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* %b, i8* %a, i64 80, i32 4, i1 false)
+  ret void
+; CHECK-LABEL: @test11(
+; CHECK-NOT: addrspacecast
+}
+
+declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) nounwind
+declare void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* nocapture, i8* nocapture, i64, i32, i1) nounwind
+
 declare void @f1(%struct.big* nocapture sret)
 declare void @f2(%struct.big*)