gpu: update nested_hit to gpu refactor

Author: syurkevi

src/common/concat_pd.hpp

@@ -280,7 +280,8 @@ struct concat_pd_t : public primitive_desc_t {
         return safe_ptr_assign(*concat_pd, _pd.release()); \
     } \
     status_t create_primitive( \
-            std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> &primitive, \
+            std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> \
+                    &primitive, \
             dnnl::impl::engine_t *engine, const cache_blob_t &cache_blob) \
             const override { \
         return primitive_t::create_primitive_common<__VA_ARGS__, pd_t>( \

src/common/primitive_desc.hpp

@@ -487,7 +487,8 @@ struct primitive_desc_t : public c_compatible {
         return new_pd.release(); \
     } \
     status_t create_primitive( \
-            std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> &primitive, \
+            std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> \
+                    &primitive, \
             dnnl::impl::engine_t *engine, const cache_blob_t &cache_blob) \
             const override { \
         return primitive_t::create_primitive_common<impl_type, pd_t>( \

src/common/sum_pd.hpp

@@ -208,7 +208,8 @@ struct sum_pd_t : public primitive_desc_t {
         return safe_ptr_assign(*sum_pd, _pd.release()); \
     } \
     status_t create_primitive( \
-            std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> &primitive, \
+            std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> \
+                    &primitive, \
             dnnl::impl::engine_t *engine, const cache_blob_t &cache_blob) \
             const override { \
         return primitive_t::create_primitive_common<__VA_ARGS__, pd_t>( \

src/gpu/gpu_primitive.hpp

@@ -73,7 +73,13 @@ struct primitive_t : public impl::primitive_t {
             std::shared_ptr<impl::primitive_t> &primitive,
             const std::shared_ptr<primitive_desc_t> &pd,
             impl::engine_t *engine) {
-        CHECK(pd->create_primitive(primitive, engine, cache_blob()));
+        std::pair<std::shared_ptr<impl::primitive_t>, cache_state_t> p;
+        CHECK(pd->create_primitive_nested(p, engine, cache_blob()));
+
+        if (p.second == cache_state_t::kernel_hit) {
+            creation_cached_state_ = cache_state_t::nested_primitive_hit;
+        }
+        primitive = p.first;
         register_compute_block(new compute_block_t(primitive.get()));
         return status::success;
     }

src/gpu/intel/gpu_primitive.hpp

@@ -155,18 +155,6 @@ struct gpu_primitive_t : public gpu::primitive_t {
         return status::success;
     }
 
-    status_t create_nested_primitive(std::shared_ptr<primitive_t> &primitive,
-            const std::shared_ptr<primitive_desc_t> &pd, engine_t *engine) {
-        std::pair<std::shared_ptr<primitive_t>, cache_state_t> p;
-        CHECK(pd->create_primitive_nested(p, engine, cache_blob()));
-        if (p.second == cache_state_t::kernel_hit) {
-            creation_cached_state_ = cache_state_t::nested_primitive_hit;
-        }
-        primitive = p.first;
-        register_primitive(primitive.get());
-        return status::success;
-    }
-
     // TODO: use inheritance for exec_ctx_t to get rid of such places...
     static status_t parallel_for(const gemm_exec_ctx_t &ctx,
             const compute::nd_range_t &range, const compute::kernel_t &kernel,