Add caching mechanizm to requantize_mkldnn_op (#22223)

paddle/fluid/operators/mkldnn/requantize_mkldnn_op.cc

@@ -21,14 +21,10 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
-using mkldnn::memory;
+using dnnl::memory;
-using mkldnn::primitive;
+using dnnl::reorder;
-using mkldnn::reorder;
 using platform::to_void_cast;
 using Tensor = framework::Tensor;
-using framework::DataLayout;
-using mkldnn::stream;
-using platform::GetMKLDNNFormat;
 
 template <typename T>
 class ReQuantOpKernel : public framework::OpKernel<T> {
@@ -42,42 +38,66 @@ class ReQuantOpKernel : public framework::OpKernel<T> {
         ctx.template device_context<platform::MKLDNNDeviceContext>();
     const auto& engine = dev_ctx.GetEngine();
 
-    std::vector<primitive> pipeline;
+    auto src_tz = paddle::framework::vectorize(input->dims());
-    auto src_tz = paddle::framework::vectorize<int64_t>(input->dims());
+
-    auto dst_tz = paddle::framework::vectorize<int64_t>(output->dims());
+    std::string key = platform::CreateKey(src_tz, scale_in, scale_out,
-    mkldnn::memory::data_type src_dt =
+                                          ctx.OutputName("Output"));
-        paddle::framework::ToMKLDNNDataType(input->type());
+    const std::string key_prim = key + "@reorder_p";
-    mkldnn::memory::data_type dst_dt = src_dt;
+    const std::string key_src_mem = key + "@src_mem";
-    MKLDNNMemoryFormat src_fmt = MKLDNNMemoryFormat::nhwc;
+    const std::string key_dst_mem = key + "@dst_mem";
-    MKLDNNMemoryFormat dst_fmt = MKLDNNMemoryFormat::nhwc;
+
+    std::shared_ptr<dnnl::memory> src_memory;
+    std::shared_ptr<dnnl::memory> dst_memory;
+    std::shared_ptr<reorder> reorder_p;
+    reorder_p = std::static_pointer_cast<reorder>(dev_ctx.GetBlob(key_prim));
 
     const T* input_data = input->data<T>();
     T* output_data = output->mutable_data<T>(ctx.GetPlace());
-    float scale_shift = scale_out / scale_in;
 
-    mkldnn::primitive_attr attri;
+    if (reorder_p == nullptr) {
+      dnnl::primitive_attr attri;
       int mask = 0;
+      float scale_shift = scale_out / scale_in;
       attri.set_output_scales(mask, {scale_shift});
 
-    auto src_md = platform::MKLDNNMemDesc({src_tz}, src_dt, src_fmt);
+      auto dst_tz = paddle::framework::vectorize(output->dims());
-    auto src_memory = std::make_shared<mkldnn::memory>(
+      dnnl::memory::data_type src_dt =
-        src_md, engine, to_void_cast<T>(input_data));
+          paddle::framework::ToMKLDNNDataType(input->type());
-
+      dnnl::memory::data_type dst_dt = src_dt;
-    auto dst_md = platform::MKLDNNMemDesc({dst_tz}, dst_dt, dst_fmt);
+
-    auto dst_memory =
+      auto src_md =
-        mkldnn::memory(dst_md, engine, to_void_cast<T>(output_data));
+          platform::MKLDNNMemDesc({src_tz}, src_dt, MKLDNNMemoryFormat::nhwc);
-
+      src_memory = std::make_shared<dnnl::memory>(src_md, engine,
-    auto reorder_pd = std::shared_ptr<reorder::primitive_desc>(
+                                                  to_void_cast<T>(input_data));
-        new reorder::primitive_desc(*src_memory, dst_memory, attri));
+
-
+      auto dst_md =
-    auto reorder_p = std::shared_ptr<reorder>(new reorder(*reorder_pd));
+          platform::MKLDNNMemDesc({dst_tz}, dst_dt, MKLDNNMemoryFormat::nhwc);
+      dst_memory = std::make_shared<dnnl::memory>(dst_md, engine,
+                                                  to_void_cast<T>(output_data));
+
+      auto reorder_pd =
+          reorder::primitive_desc(*src_memory, *dst_memory, attri);
+      reorder_p = std::make_shared<reorder>(reorder_pd);
+
+      dev_ctx.SetBlob(key_prim, reorder_p);
+      dev_ctx.SetBlob(key_src_mem, src_memory);
+      dev_ctx.SetBlob(key_dst_mem, dst_memory);
+    } else {
+      src_memory =
+          std::static_pointer_cast<dnnl::memory>(dev_ctx.GetBlob(key_src_mem));
+      src_memory->set_data_handle(to_void_cast<T>(input_data));
+
+      dst_memory =
+          std::static_pointer_cast<dnnl::memory>(dev_ctx.GetBlob(key_dst_mem));
+      dst_memory->set_data_handle(output_data);
+    }
 
-    mkldnn::stream astream(engine);
+    dnnl::stream astream(engine);
-    reorder_p->execute(astream, *src_memory, dst_memory);
+    reorder_p->execute(astream, *src_memory, *dst_memory);
     astream.wait();
 
-    output->set_layout(DataLayout::kMKLDNN);
+    output->set_layout(framework::DataLayout::kMKLDNN);
-    output->set_format(GetMKLDNNFormat(dst_memory));
+    output->set_format(platform::GetMKLDNNFormat(*dst_memory));
   }
 };