[CUDA] Only allow __global__ on free functions and static member functions.

Summary:
Warn for NVCC compatibility if you declare a static member function or
inline function as __global__.

Reviewers: tra

Subscribers: jhen, echristo, cfe-commits

Differential Revision: http://reviews.llvm.org/D16261

llvm-svn: 258263

4 files changed, 36 insertions, 9 deletions

diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index d031b53bef3..6e652d782ff 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -6403,6 +6403,8 @@ def err_deleted_function_use : Error<"attempt to use a deleted function">;
 
 def err_kern_type_not_void_return : Error<
   "kernel function type %0 must have void return type">;
+def err_kern_is_nonstatic_method : Error<
+  "kernel function %0 must be a free function or static member function">;
 def err_config_scalar_return : Error<
   "CUDA special function 'cudaConfigureCall' must have scalar return type">;
 def err_kern_call_not_global_function : Error<
@@ -6415,6 +6417,12 @@ def err_ref_bad_target : Error<
 def warn_host_calls_from_host_device : Warning<
   "calling __host__ function %0 from __host__ __device__ function %1 can lead to runtime errors">,
   InGroup<CudaCompat>;
+def warn_kern_is_method : Extension<
+  "kernel function %0 is a member function; this may not be accepted by nvcc">,
+  InGroup<CudaCompat>;
+def warn_kern_is_inline : Warning<
+  "ignored 'inline' attribute on kernel function %0">,
+  InGroup<CudaCompat>;
 
 def warn_non_pod_vararg_with_format_string : Warning<
   "cannot pass %select{non-POD|non-trivial}0 object of type %1 to variadic "
diff --git a/clang/lib/Sema/SemaCUDA.cpp b/clang/lib/Sema/SemaCUDA.cpp
index 61dfdd3f720..568c765984f 100644
--- a/clang/lib/Sema/SemaCUDA.cpp
+++ b/clang/lib/Sema/SemaCUDA.cpp
@@ -273,12 +273,9 @@ static bool
 resolveCalleeCUDATargetConflict(Sema::CUDAFunctionTarget Target1,
                                 Sema::CUDAFunctionTarget Target2,
                                 Sema::CUDAFunctionTarget *ResolvedTarget) {
-  if (Target1 == Sema::CFT_Global && Target2 == Sema::CFT_Global) {
-    // TODO: this shouldn't happen, really. Methods cannot be marked __global__.
-    // Clang should detect this earlier and produce an error. Then this
-    // condition can be changed to an assertion.
-    return true;
-  }
+  // Only free functions and static member functions may be global.
+  assert(Target1 != Sema::CFT_Global);
+  assert(Target2 != Sema::CFT_Global);
 
   if (Target1 == Sema::CFT_HostDevice) {
     *ResolvedTarget = Target2;
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index 31741be618a..e0ce3adce42 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -3617,11 +3617,21 @@ static void handleGlobalAttr(Sema &S, Decl *D, const AttributeList &Attr) {
                               : FixItHint());
     return;
   }
+  if (const auto *Method = dyn_cast<CXXMethodDecl>(FD)) {
+    if (Method->isInstance()) {
+      S.Diag(Method->getLocStart(), diag::err_kern_is_nonstatic_method)
+          << Method;
+      return;
+    }
+    S.Diag(Method->getLocStart(), diag::warn_kern_is_method) << Method;
+  }
+  // Only warn for "inline" when compiling for host, to cut down on noise.
+  if (FD->isInlineSpecified() && !S.getLangOpts().CUDAIsDevice)
+    S.Diag(FD->getLocStart(), diag::warn_kern_is_inline) << FD;
 
   D->addAttr(::new (S.Context)
               CUDAGlobalAttr(Attr.getRange(), S.Context,
                              Attr.getAttributeSpellingListIndex()));
-
 }
 
 static void handleGNUInlineAttr(Sema &S, Decl *D, const AttributeList &Attr) {
diff --git a/clang/test/SemaCUDA/bad-attributes.cu b/clang/test/SemaCUDA/bad-attributes.cu
index 7e01e141de1..4cb43e240b0 100644
--- a/clang/test/SemaCUDA/bad-attributes.cu
+++ b/clang/test/SemaCUDA/bad-attributes.cu
@@ -4,8 +4,8 @@
 //
 // You should be able to run this file through nvcc for compatibility testing.
 //
-// RUN: %clang_cc1 -fsyntax-only -verify %s
-// RUN: %clang_cc1 -fcuda-is-device -fsyntax-only -verify %s
+// RUN: %clang_cc1 -fsyntax-only -Wcuda-compat -verify -DEXPECT_INLINE_WARNING %s
+// RUN: %clang_cc1 -fcuda-is-device -fsyntax-only -Wcuda-compat -verify %s
 
 #include "Inputs/cuda.h"
 
@@ -47,3 +47,15 @@ __global__ __device__ void z11();  // expected-error {{attributes are not compat
 // expected-note@-1 {{conflicting attribute is here}}
 __global__ __host__ void z12();  // expected-error {{attributes are not compatible}}
 // expected-note@-1 {{conflicting attribute is here}}
+
+struct S {
+  __global__ void foo() {};  // expected-error {{must be a free function or static member function}}
+  __global__ static void bar(); // expected-warning {{kernel function 'bar' is a member function}}
+  // Although this is implicitly inline, we shouldn't warn.
+  __global__ static void baz() {}; // expected-warning {{kernel function 'baz' is a member function}}
+};
+
+__global__ static inline void foobar() {};
+#ifdef EXPECT_INLINE_WARNING
+// expected-warning@-2 {{ignored 'inline' attribute on kernel function 'foobar'}}
+#endif