[Mlir-commits] [mlir] [mlir][gpu] Add metadata attributes for storing kernel metadata in GPU objects (PR #95292)

[Oleksandr Alex Zinenko]

================
@@ -25,15 +25,19 @@ bool mlirAttributeIsAGPUObjectAttr(MlirAttribute attr) {
 
 MlirAttribute mlirGPUObjectAttrGet(MlirContext mlirCtx, MlirAttribute target,
                                    uint32_t format, MlirStringRef objectStrRef,
-                                   MlirAttribute mlirObjectProps) {
+                                   MlirAttribute mlirObjectProps,
+                                   MlirAttribute mlirKernelsAttr) {
   MLIRContext *ctx = unwrap(mlirCtx);
   llvm::StringRef object = unwrap(objectStrRef);
   DictionaryAttr objectProps;
   if (mlirObjectProps.ptr != nullptr)
     objectProps = llvm::cast<DictionaryAttr>(unwrap(mlirObjectProps));
-  return wrap(gpu::ObjectAttr::get(ctx, unwrap(target),
-                                   static_cast<gpu::CompilationTarget>(format),
-                                   StringAttr::get(ctx, object), objectProps));
+  gpu::KernelTableAttr kernels;
+  if (mlirKernelsAttr.ptr != nullptr)
+    kernels = llvm::cast<gpu::KernelTableAttr>(unwrap(mlirKernelsAttr));
----------------
ftynse wrote:

Since kernels attribute may be null anyway, how about introducing a `mlirGPUObjectAttrGetWithKernels` in addition to the existing, asserting non-null, and thus avoiding breakage in C API.

https://github.com/llvm/llvm-project/pull/95292