[PATCH] D16047: [OpenCL] Add Sema checks for OpenCL 2.0

[Xiuli PAN via cfe-commits]

pxli168 added inline comments.

================
Comment at: include/clang/Basic/DiagnosticSemaKinds.td:593
@@ -592,2 +592,3 @@
   InGroup<MissingDeclarations>;
+def err_no_declarators : Error<"declaration does not declare anything">;
 def ext_typedef_without_a_name : ExtWarn<"typedef requires a name">,
----------------
I think this no declarator is just for bit-field in C. And OpenCL C does not support the bit-field I think we need not to support that.
If you think it is useless I think we can remove it then.

================
Comment at: include/clang/Basic/DiagnosticSemaKinds.td:7666
@@ +7665,3 @@
+def err_opencl_atomic_init_addressspace : Error<
+  "initialization of atomic variables is restricted to variables in global address space in opencl">;
+def err_opencl_block_proto_variadic : Error<
----------------
It seems that this two are very different and need a big select. Then it seems two maybe OK.

================
Comment at: lib/Sema/SemaDecl.cpp:5724
@@ +5723,3 @@
+  // Pipes can only be passed as arguments to a function.
+  if (getLangOpts().OpenCL && getLangOpts().OpenCLVersion >= 200 &&
+      R->isPipeType()) {
----------------
It seems all other check for OpenCL goes this way. I don't know if we just check OpenCL version will cause any error.

================
Comment at: lib/Sema/SemaDecl.cpp:6718
@@ +6717,3 @@
+    const BlockPointerType *BlkTy = T->getAs<BlockPointerType>();
+    assert(BlkTy && "Not a block pointer.");
+
----------------
agree. I will remove it.

================
Comment at: lib/Sema/SemaExpr.cpp:6299
@@ +6298,3 @@
+    // should output error for both LHS and RHS, use | instead ||
+    if (checkBlockType(*this, LHS.get()) | checkBlockType(*this, RHS.get()))
+      return QualType();
----------------
But how can we know if LHS or RHS is Block, or they both are?

================
Comment at: lib/Sema/SemaExpr.cpp:10095
@@ -10060,1 +10094,3 @@
     Result = PT->getPointeeType();
+    // OpenCL v2.0 s6.12.5 --The unary operators (* and &) cannot be used with a
+    // Block.
----------------
Fixed

================
Comment at: lib/Sema/SemaInit.cpp:6138
@@ -6137,1 +6137,3 @@
 
+  // OpenCL v2.0 s6.13.1.1 - This macro can only be used to initialize atomic
+  // objects that are declared in program scope in the global address space.
----------------
yes, thank you!

================
Comment at: lib/Sema/SemaInit.cpp:6147
@@ +6146,3 @@
+        Args.size() > 0) {
+      const Expr *Init = Args[0];
+      S.Diag(Init->getLocStart(), diag::err_opencl_atomic_init_addressspace)
----------------
I just follow the surrounding code style. 

================
Comment at: test/CodeGenOpenCL/opencl_types.cl:39
@@ -38,3 @@
-
-void __attribute__((overloadable)) bad1(image1d_t *b, image2d_t *c, image2d_t *d) {}
-// CHECK-LABEL: @{{_Z4bad1P11ocl_image1dP11ocl_image2dS2_|"\\01\?bad1@@\$\$J0YAXPE?APAUocl_image1d@@PE?APAUocl_image2d@@1 at Z"}}
----------------
As you can see, it used pointer of image type which I just added a semacheck for it.

================
Comment at: test/SemaOpenCL/invalid-decl.cl:11
@@ +10,3 @@
+void test2(image1d_t *i){} // expected-error {{pointer to image is invalid in OpenCL}}
+
+void test3() {
----------------
I will find some place to put these test in.

================
Comment at: test/SemaOpenCL/invalid-decl.cl:19
@@ +18,2 @@
+  atomic_int guide = ATOMIC_VAR_INIT(42); // expected-error {{initialization of atomic variables is restricted to variables in global address space in opencl}}
+}
----------------
It seems strange to put sema test cases in Parser, just let them be together then.


http://reviews.llvm.org/D16047