Added DEX support for (MaxPool + AvgPool) Backprop op for CPU backend (#1302)

* - Added DEX support for MaxPoolBackprop op for CPU backend

* Added DEX execution support for AvgPoolBackprop

src/ngraph/runtime/cpu/builder/avg_pool.cpp

@@ -108,7 +108,85 @@ namespace ngraph
                 }
             }
 
+            template <>
+            void Builder::BUILDER_DECL(ngraph::op::AvgPoolBackprop)
+            {
+                auto apb = static_cast<const ngraph::op::AvgPoolBackprop*>(node);
+
+                auto& functors = external_function->get_functors();
+                auto& tensor_data = external_function->get_tensor_data();
+
+                auto delta_shape = args[0].get_shape();
+                auto out_shape = out[0].get_shape();
+
+                auto& delta_tensor = tensor_data[args[0].get_name()];
+                auto& out_tensor = tensor_data[out[0].get_name()];
+
+                auto window_shape = apb->get_window_shape();
+                auto window_movement_strides = apb->get_window_movement_strides();
+                auto padding_below = apb->get_padding_below();
+                auto padding_above = apb->get_padding_above();
+                auto include_padding_in_avg_computation =
+                    apb->get_include_padding_in_avg_computation();
+
+                if (runtime::cpu::mkldnn_utils::use_mkldnn_kernel(node))
+                {
+                    auto& mkldnn_emitter = external_function->get_mkldnn_emitter();
+                    auto diff_dst_desc = mkldnn_emitter->build_memory_descriptor(
+                        args[0], runtime::cpu::mkldnn_utils::get_input_mkldnn_format(node, 0));
+                    auto diff_src_desc = mkldnn_emitter->build_memory_descriptor(
+                        out[0], runtime::cpu::mkldnn_utils::get_output_mkldnn_format(node, 0));
+
+                    size_t avg_pool_index = mkldnn_emitter->build_pooling_backward(
+                        (apb->get_include_padding_in_avg_computation()
+                             ? mkldnn::algorithm::pooling_avg_include_padding
+                             : mkldnn::algorithm::pooling_avg_exclude_padding),
+                        diff_dst_desc,
+                        diff_src_desc,
+                        apb->get_window_movement_strides(),
+                        apb->get_window_shape(),
+                        apb->get_padding_below(),
+                        apb->get_padding_above());
+
+                    auto& deps = mkldnn_emitter->get_primitive_deps(avg_pool_index);
+                    auto functor = [&, avg_pool_index](CPURuntimeContext* ctx) {
+                        cpu::mkldnn_utils::set_memory_ptr(ctx, deps[0], delta_tensor);
+                        cpu::mkldnn_utils::set_memory_ptr(ctx, deps[1], out_tensor);
+                        cpu::mkldnn_utils::mkldnn_invoke_primitive(ctx, avg_pool_index);
+                    };
+                    functors.emplace_back(functor);
+                }
+                else
+                {
+                    std::function<decltype(runtime::cpu::kernel::avg_pool_backprop<float>)> kernel;
+
+                    SELECT_KERNEL(
+                        kernel, out[0].get_element_type(), runtime::cpu::kernel::avg_pool_backprop);
+
+                    auto functor = [&,
+                                    kernel,
+                                    delta_shape,
+                                    out_shape,
+                                    window_shape,
+                                    window_movement_strides,
+                                    padding_below,
+                                    padding_above,
+                                    include_padding_in_avg_computation](CPURuntimeContext* ctx) {
+                        kernel(delta_tensor,
+                               out_tensor,
+                               delta_shape,
+                               out_shape,
+                               window_shape,
+                               window_movement_strides,
+                               padding_below,
+                               padding_above,
+                               include_padding_in_avg_computation);
+                    };
+                    functors.emplace_back(functor);
+                }
+            }
             REGISTER_OP_BUILDER(AvgPool);
+            REGISTER_OP_BUILDER(AvgPoolBackprop);
         }
     }
 }

src/ngraph/runtime/cpu/builder/max_pool.cpp

@@ -102,6 +102,95 @@ namespace ngraph
                     functors.emplace_back(functor);
                 }
             }
+            template <>
+            void Builder::BUILDER_DECL(ngraph::op::MaxPoolBackprop)
+            {
+                auto mpb = static_cast<const ngraph::op::MaxPoolBackprop*>(node);
+
+                auto& functors = external_function->get_functors();
+                auto& tensor_data = external_function->get_tensor_data();
+
+                auto arg_fwd_shape = args[0].get_shape();
+                auto delta_shape = args[1].get_shape();
+                auto out_shape = out[0].get_shape();
+
+                auto& arg_fwd_tensor = tensor_data[args[0].get_name()];
+                auto& delta_tensor = tensor_data[args[1].get_name()];
+                auto& out_tensor = tensor_data[out[0].get_name()];
+
+                auto window_shape = mpb->get_window_shape();
+                auto window_movement_strides = mpb->get_window_movement_strides();
+                auto padding_below = mpb->get_padding_below();
+                auto padding_above = mpb->get_padding_above();
+
+                if (runtime::cpu::mkldnn_utils::use_mkldnn_kernel(node))
+                {
+                    auto& mkldnn_emitter = external_function->get_mkldnn_emitter();
+                    auto fprop_src_desc = mkldnn_emitter->build_memory_descriptor(
+                        args[0], runtime::cpu::mkldnn_utils::get_input_mkldnn_format(node, 0));
+                    auto diff_dst_desc = mkldnn_emitter->build_memory_descriptor(
+                        args[1], runtime::cpu::mkldnn_utils::get_input_mkldnn_format(node, 1));
+                    auto diff_src_desc = mkldnn_emitter->build_memory_descriptor(
+                        out[0], runtime::cpu::mkldnn_utils::get_output_mkldnn_format(node, 0));
+
+                    size_t max_pool_index = mkldnn_emitter->build_max_pooling_backward(
+                        mkldnn::algorithm::pooling_max,
+                        fprop_src_desc,
+                        diff_dst_desc,
+                        diff_src_desc,
+                        mpb->get_window_movement_strides(),
+                        mpb->get_window_shape(),
+                        mpb->get_padding_below(),
+                        mpb->get_padding_above());
+
+                    auto& fdeps = mkldnn_emitter->get_primitive_deps(max_pool_index - 1);
+                    auto functor_fprop = [&, max_pool_index](CPURuntimeContext* ctx) {
+                        cpu::mkldnn_utils::set_memory_ptr(ctx, fdeps[0], arg_fwd_tensor);
+                        cpu::mkldnn_utils::set_memory_ptr(ctx, fdeps[1], out_tensor);
+                        cpu::mkldnn_utils::set_memory_ptr(
+                            ctx, fdeps[2], ctx->mkldnn_workspaces[fdeps[3]]);
+                        cpu::mkldnn_utils::mkldnn_invoke_primitive(ctx, max_pool_index - 1);
+                    };
+                    functors.emplace_back(functor_fprop);
+                    auto& bdeps = mkldnn_emitter->get_primitive_deps(max_pool_index);
+                    auto functor_bprop = [&, max_pool_index](CPURuntimeContext* ctx) {
+                        cpu::mkldnn_utils::set_memory_ptr(ctx, bdeps[0], delta_tensor);
+                        cpu::mkldnn_utils::set_memory_ptr(
+                            ctx, bdeps[1], ctx->mkldnn_workspaces[bdeps[3]]);
+                        cpu::mkldnn_utils::set_memory_ptr(ctx, bdeps[2], out_tensor);
+                        cpu::mkldnn_utils::mkldnn_invoke_primitive(ctx, max_pool_index);
+                    };
+                    functors.emplace_back(functor_bprop);
+                }
+                else
+                {
+                    std::function<decltype(runtime::cpu::kernel::max_pool_backprop<float>)> kernel;
+
+                    SELECT_KERNEL(
+                        kernel, out[0].get_element_type(), runtime::cpu::kernel::max_pool_backprop);
+
+                    auto functor = [&,
+                                    kernel,
+                                    arg_fwd_shape,
+                                    delta_shape,
+                                    out_shape,
+                                    window_shape,
+                                    window_movement_strides,
+                                    padding_below,
+                                    padding_above](CPURuntimeContext* ctx) {
+                        kernel(arg_fwd_tensor,
+                               delta_tensor,
+                               out_tensor,
+                               delta_shape,
+                               arg_fwd_shape,
+                               window_shape,
+                               window_movement_strides,
+                               padding_below,
+                               padding_above);
+                    };
+                    functors.emplace_back(functor);
+                }
+            }
 
             template <>
             void Builder::BUILDER_DECL(ngraph::op::MaxPoolWithIndices)
@@ -190,6 +279,7 @@ namespace ngraph
             }
 
             REGISTER_OP_BUILDER(MaxPool);
+            REGISTER_OP_BUILDER(MaxPoolBackprop);
             REGISTER_OP_BUILDER(MaxPoolWithIndices);
             REGISTER_OP_BUILDER(MaxPoolWithIndicesBackprop);
         }

src/ngraph/runtime/cpu/kernel/avg_pool.hpp

@@ -48,6 +48,29 @@ namespace ngraph
                                                      padding_above,
                                                      include_padding_in_avg_computation);
                 }
+
+                template <typename ElementType>
+                void avg_pool_backprop(void* delta,
+                                       void* out,
+                                       const Shape& delta_shape,
+                                       const Shape& out_shape,
+                                       const Shape& window_shape,
+                                       const Strides& window_movement_strides,
+                                       const Shape& padding_below,
+                                       const Shape& padding_above,
+                                       bool include_padding_in_avg_computation)
+                {
+                    reference::avg_pool_backprop<ElementType>(
+                        static_cast<const ElementType*>(delta),
+                        static_cast<ElementType*>(out),
+                        delta_shape,
+                        out_shape,
+                        window_shape,
+                        window_movement_strides,
+                        padding_below,
+                        padding_above,
+                        include_padding_in_avg_computation);
+                }
             }
         }
     }

src/ngraph/runtime/cpu/kernel/max_pool.hpp

@@ -46,6 +46,29 @@ namespace ngraph
                                                      padding_below,
                                                      padding_above);
                 }
+
+                template <typename ElementType>
+                void max_pool_backprop(void* arg_forward,
+                                       void* delta,
+                                       void* out,
+                                       const Shape& delta_shape,
+                                       const Shape& out_shape, // same as arg_forward_shape
+                                       const Shape& window_shape,
+                                       const Strides& window_movement_strides,
+                                       const Shape& padding_below,
+                                       const Shape& padding_above)
+                {
+                    reference::max_pool_backprop<ElementType>(
+                        static_cast<const ElementType*>(arg_forward),
+                        static_cast<const ElementType*>(delta),
+                        static_cast<ElementType*>(out),
+                        delta_shape,
+                        out_shape,
+                        window_shape,
+                        window_movement_strides,
+                        padding_below,
+                        padding_above);
+                }
             }
         }
     }