[SYCL][Clang] Add __sycl_detail__::add_ir_annotations_member attribute #5879
Conversation
These changes introduce the new `__sycl_detail__::add_ir_annotations_member`. This attribute is similar in behaviour to `__sycl_detail__::add_ir_attributes_*` attributes but will generate annotation intrinsic calls rather than LLVM IR attributes. Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
|
Note: This is based of https://github.com/intel/llvm/blob/sycl/sycl/doc/design/CompileTimeProperties.md design document, with the change from the generated intrinsic call carrying metadata to it referencing a constant global variable with the string literals, similar to what I will make a patch for the design document shortly. Ping @gmlueck for awareness. |
clang/lib/CodeGen/CodeGenModule.cpp
Outdated
| auto *Bitcasted = llvm::ConstantExpr::getBitCast(GV, GlobalsInt8PtrTy); | ||
|
|
||
| Lookup = Bitcasted; | ||
| return Bitcasted; |
There was a problem hiding this comment.
| auto *Bitcasted = llvm::ConstantExpr::getBitCast(GV, GlobalsInt8PtrTy); | |
| Lookup = Bitcasted; | |
| return Bitcasted; | |
| return llvm::ConstantExpr::getBitCast(GV, GlobalsInt8PtrTy); |
There was a problem hiding this comment.
Lookup is a reference to an entry in the map to existing values, so setting also sets the entry for future reuse of the created value. I have added comments to the function and renamed it to LookupRef to make it a little more clear what the intention here is.
Additionally, I realized I didn't add a test for using the optional filter in codegen. It has been added, but it exposed a bug in this function. It has been fixed, but it does mean the function has been changed.
| @@ -2873,6 +2873,43 @@ void CodeGenModule::AddGlobalAnnotations(const ValueDecl *D, | |||
| Annotations.push_back(EmitAnnotateAttr(GV, I, D->getLocation())); | |||
| } | |||
|
|
|||
| llvm::Constant *CodeGenModule::EmitSYCLAnnotationArgs( | |||
| const SYCLAddIRAnnotationsMemberAttr *Attr) { | |||
There was a problem hiding this comment.
I would appreciate some comments in this code.
There was a problem hiding this comment.
Absolutely! I have added comments. Let me know if you think it is sufficient.
|
|
||
| // Tests the AST produced from instantiating templates using | ||
| // __sycl_detail__::add_ir_annotations_member attributes with pack expansion | ||
| // arguments. |
There was a problem hiding this comment.
I will need to come back to review this file.
…lue reuse tests Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
Signed-off-by: Steffen Larsen <steffen.larsen@intel.com>
…annotations_member (#5884) * [SYCL][Doc] Adjust design for compile-time properties through add_ir_annotations_member During implementation of the attribute and translation of annotations on fields, the design was conflicting with existing features. This commit makes the following design changes: * Change the `llvm.ptr.annotation` intrinsic call produced by `[[__sycl_detail__::add_ir_annotations_member()]]` to use a pointer to a constant global variable instead of metadata. This is done to adhere to the signature of the intrinsic. * Change the representation consumed by the translator from a new SPIR-V builtin to an extended version of existing decoration parsing using `llvm.ptr.annotation`. Implementation of these are in review #5879 and KhronosGroup/SPIRV-LLVM-Translator#1446 Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
These changes introduce the new
__sycl_detail__::add_ir_annotations_member. This attribute is similar in behaviour to__sycl_detail__::add_ir_attributes_*attributes but will generate annotation intrinsic calls rather than LLVM IR attributes.