Currently clang emits file-scope asm during *both* host and device
compilation modes which is usually a wrong thing to do.
There's no way to attach any attribute to an __asm statement, so
there's no way to differentiate between host-side and device-side
file-scope asm. This patch makes clang to match nvcc behavior and
emit file-scope-asm only during host-side compilation.
Differential Revision: http://reviews.llvm.org/D9270
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@235905
91177308-0d34-0410-b5e6-
96231b3b80d8
break;
case Decl::FileScopeAsm: {
+ // File-scope asm is ignored during device-side CUDA compilation.
+ if (LangOpts.CUDA && LangOpts.CUDAIsDevice)
+ break;
auto *AD = cast<FileScopeAsmDecl>(D);
getModule().appendModuleInlineAsm(AD->getAsmString()->getString());
break;
#include "Inputs/cuda.h"
+// This has to be at the top of the file as that's where file-scope
+// asm ends up.
+// CHECK-HOST: module asm "file scope asm is host only"
+// CHECK-DEVICE-NOT: module asm "file scope asm is host only"
+__asm__("file scope asm is host only");
+
// CHECK-HOST-NOT: constantdata = global
// CHECK-DEVICE: constantdata = global
__constant__ char constantdata[256];