changed a way to use kernel_impl_params in save and load

openvinotoolkit · Nov 7, 2022 · e72ffa4 · e72ffa4
1 parent b2033a4
commit e72ffa4
Show file tree

Hide file tree

Showing 23 changed files with 89 additions and 76 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/impls/cpu/detection_output.cpp b/src/plugins/intel_gpu/src/graph/impls/cpu/detection_output.cpp
@@ -68,11 +68,11 @@ struct detection_output_impl : typed_primitive_impl<detection_output> {
         nms_type = (node.get_primitive()->decrease_label_id ? NMSType::MXNET : NMSType::CAFFE);
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob << make_data(&nms_type, sizeof(NMSType));
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         ib >> make_data(&nms_type, sizeof(NMSType));
     }
 

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp
@@ -48,13 +48,13 @@ struct activation_impl : typed_primitive_impl_ocl<activation> {
         return args;
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _is_parameterized;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _is_parameterized;
     }
 

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/binary_convolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/binary_convolution.cpp
@@ -75,13 +75,13 @@ struct binary_convolution_impl : typed_primitive_impl_ocl<binary_convolution> {
     int32_t get_split() const override { return _split; }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
     }
 

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp
@@ -85,13 +85,13 @@ struct concatenation_impl : typed_primitive_impl_ocl<concatenation> {
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _can_be_optimized;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _can_be_optimized;
     }
 

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp
@@ -80,15 +80,15 @@ struct convolution_impl : typed_primitive_impl_ocl<convolution> {
     bool get_depthwise_sep_opt() const override { return _depthwise_sep_opt; }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
         ob << _groups;
         ob << _depthwise_sep_opt;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
         ib >> _groups;
         ib >> _depthwise_sep_opt;

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp
@@ -44,13 +44,13 @@ struct crop_impl : typed_primitive_impl_ocl<crop> {
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _can_be_optimized;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _can_be_optimized;
     }
 

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/custom_primitive.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/custom_primitive.cpp
@@ -85,12 +85,12 @@ struct custom_gpu_primitive_impl : typed_primitive_impl<custom_gpu_primitive> {
         return {_kernel_id};
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob << *cl_kernel;
         ob << _kernel_id;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         cl_kernel = std::make_shared<kernel_selector::cl_kernel_data>();
         ib >> *cl_kernel;
         ib >> _kernel_id;

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp
@@ -41,14 +41,14 @@ struct deconvolution_impl : typed_primitive_impl_ocl<deconvolution> {
         _groups = node.get_groups();
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
         ob << _groups;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
         ib >> _groups;
     }

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/deformable_convolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/deformable_convolution.cpp
@@ -42,14 +42,14 @@ struct deformable_conv_impl : typed_primitive_impl_ocl<deformable_conv> {
         _groups = node.get_groups();
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
         ob << _groups;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
         ib >> _groups;
     }

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/generic_layer.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/generic_layer.cpp
@@ -44,12 +44,12 @@ struct generic_layer_impl : typed_primitive_impl<generic_layer> {
         _kernel_id = arg.get_program().add_kernel(arg.get_primitive()->generic_params.clKernel->code.kernelString);
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob <<_cl_kernel_data;
         ob << _kernel_id;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         ib >> _cl_kernel_data;
         ib >> _kernel_id;
     }

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp b/src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp
@@ -73,15 +73,15 @@ struct typed_primitive_impl_ocl : public typed_primitive_impl<PType> {
 
     bool is_cpu() const override { return false; }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob << make_data(&_kernel_data.internalBufferDataType, sizeof(kernel_selector::Datatype));
         ob << _kernel_data.internalBufferSizes;
         ob << _kernel_data.kernels;
         ob << _kernel_ids;
         ob << _kernel_args;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         ib >> make_data(&_kernel_data.internalBufferDataType, sizeof(kernel_selector::Datatype));
         ib >> _kernel_data.internalBufferSizes;
         ib >> _kernel_data.kernels;

diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/reorder.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/reorder.cpp
@@ -40,14 +40,14 @@ struct reorder_impl : typed_primitive_impl_ocl<reorder> {
         _has_mean = node.has_mean();
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _can_be_optimized;
         ob << _has_mean;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _can_be_optimized;
         ib >> _has_mean;
     }

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/concatenation_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/concatenation_onednn.cpp
@@ -63,16 +63,17 @@ struct concatenation_onednn : typed_primitive_onednn_impl<concatenation, void, d
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         if (_prim.get(true) == nullptr) {
             ob << false;
             return;
         } else {
             ob << true;
         }
 
-        parent::save(ob, impl_params);
+        parent::save(ob);
 
+        const kernel_impl_params* impl_params = reinterpret_cast<kernel_impl_params*>(ob.getKernlImplParams());
         auto prim = impl_params->typed_desc<concatenation>();
         ob << prim->axis;
 
@@ -81,18 +82,19 @@ struct concatenation_onednn : typed_primitive_onednn_impl<concatenation, void, d
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         bool has_prim;
         ib >> has_prim;
 
         if (!has_prim)
             return;
 
-        parent::load(ib, impl_params);
+        parent::load(ib);
 
         int64_t prim_axis;
         ib >> prim_axis;
 
+        const kernel_impl_params* impl_params = reinterpret_cast<kernel_impl_params*>(ib.getKernlImplParams());
         auto desc = get_concatenation_descriptor(*impl_params, prim_axis, ib.get_engine());
         _pd = *desc;
 

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp
@@ -185,8 +185,8 @@ struct convolution_onednn : typed_primitive_onednn_impl<convolution, dnnl::convo
 
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_convolution_desc_t));
 
@@ -195,8 +195,8 @@ struct convolution_onednn : typed_primitive_onednn_impl<convolution, dnnl::convo
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::convolution_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_convolution_desc_t));

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/deconvolution_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/deconvolution_onednn.cpp
@@ -104,8 +104,8 @@ struct deconvolution_onednn : typed_primitive_onednn_impl<deconvolution, dnnl::d
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_deconvolution_desc_t));
 
@@ -114,8 +114,8 @@ struct deconvolution_onednn : typed_primitive_onednn_impl<deconvolution, dnnl::d
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::deconvolution_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_deconvolution_desc_t));

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/fully_connected_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/fully_connected_onednn.cpp
@@ -166,8 +166,8 @@ struct fully_connected_onednn : typed_primitive_onednn_impl<fully_connected, dnn
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_inner_product_desc_t));
 
@@ -176,8 +176,8 @@ struct fully_connected_onednn : typed_primitive_onednn_impl<fully_connected, dnn
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::inner_product_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_inner_product_desc_t));

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp
@@ -209,8 +209,8 @@ struct gemm_onednn : typed_primitive_onednn_impl<gemm, dnnl::matmul::desc> {
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_matmul_desc_t));
 
@@ -219,8 +219,8 @@ struct gemm_onednn : typed_primitive_onednn_impl<gemm, dnnl::matmul::desc> {
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::matmul::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_matmul_desc_t));

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/pooling_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/pooling_onednn.cpp
@@ -64,8 +64,8 @@ struct pooling_onednn : typed_primitive_onednn_impl<pooling, dnnl::pooling_forwa
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_pooling_desc_t));
 
@@ -74,8 +74,8 @@ struct pooling_onednn : typed_primitive_onednn_impl<pooling, dnnl::pooling_forwa
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::pooling_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_pooling_desc_t));

diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/reduction_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/reduction_onednn.cpp
@@ -93,8 +93,8 @@ struct reduction_onednn : typed_primitive_onednn_impl<reduce, dnnl::reduction::d
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_reduction_desc_t));
 
@@ -103,8 +103,8 @@ struct reduction_onednn : typed_primitive_onednn_impl<reduce, dnnl::reduction::d
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::reduction::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_reduction_desc_t));