From 1cf5ebc4c5e3f02a9e094a6f1521b3d481a2b94b Mon Sep 17 00:00:00 2001
From: Ke Zhang <kezhan@microsoft.com>
Date: Thu, 1 Aug 2019 13:22:15 -0700
Subject: [PATCH] copyfromhost/copytohost are not needed for mkldnn ep (#1532)

* memcpy is not necessary for mkldnn ep to copy from/to host.

* update
---
 .../mkldnn/mkldnn_execution_provider.cc       | 31 ++-----------------
 1 file changed, 3 insertions(+), 28 deletions(-)

diff --git a/onnxruntime/core/providers/mkldnn/mkldnn_execution_provider.cc b/onnxruntime/core/providers/mkldnn/mkldnn_execution_provider.cc
index d58b507a96..93cb36116f 100644
--- a/onnxruntime/core/providers/mkldnn/mkldnn_execution_provider.cc
+++ b/onnxruntime/core/providers/mkldnn/mkldnn_execution_provider.cc
@@ -5,39 +5,18 @@
 #pragma warning(disable : 4996)
 #endif
 
-#include "mkldnn_execution_provider.h"
 #include "core/framework/allocator.h"
-#include "core/framework/memcpy.h"
-#include "core/framework/kernel_registry.h"
-#include "mkldnn_fwd.h"
 #include "core/framework/compute_capability.h"
+#include "core/framework/kernel_registry.h"
 #include "core/providers/mkldnn/subgraph/mkldnn_func_kernel.h"
+#include "mkldnn_execution_provider.h"
+#include "mkldnn_fwd.h"
 
 namespace onnxruntime {
 
 constexpr const char* MKLDNN = "MklDnn";
 constexpr const char* MKLDNN_CPU = "MklDnnCpu";
 
-namespace mkl_dnn {
-
-ONNX_OPERATOR_KERNEL_EX(
-    MemcpyFromHost,
-    kOnnxDomain,
-    1,
-    kMklDnnExecutionProvider,
-    KernelDefBuilder().InputMemoryType<OrtMemTypeCPUInput>(0).TypeConstraint("T", DataTypeImpl::AllTensorTypes()),
-    Memcpy);
-
-ONNX_OPERATOR_KERNEL_EX(
-    MemcpyToHost,
-    kOnnxDomain,
-    1,
-    kMklDnnExecutionProvider,
-    KernelDefBuilder().OutputMemoryType<OrtMemTypeCPUOutput>(0).TypeConstraint("T", DataTypeImpl::AllTensorTypes()),
-    Memcpy);
-
-}  // namespace mkl_dnn
-
 MKLDNNExecutionProvider::MKLDNNExecutionProvider(const MKLDNNExecutionProviderInfo& info)
     : IExecutionProvider{onnxruntime::kMklDnnExecutionProvider} {
   DeviceAllocatorRegistrationInfo default_allocator_info({OrtMemTypeDefault,
@@ -65,8 +44,6 @@ MKLDNNExecutionProvider::~MKLDNNExecutionProvider() {
 namespace mkl_dnn {
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 1, Conv);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 7, Gemm);
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 1, MemcpyFromHost);
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 1, MemcpyToHost);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 6, Relu);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 6, Sum);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 7, BatchNormalization);
@@ -81,8 +58,6 @@ void RegisterMKLDNNKernels(KernelRegistry& kernel_registry) {
   static const BuildKernelCreateInfoFn function_table[] = {
       BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 1, Conv)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 7, Gemm)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 1, MemcpyFromHost)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 1, MemcpyToHost)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 6, Relu)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 6, Sum)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kMklDnnExecutionProvider, kOnnxDomain, 7, BatchNormalization)>,