[OneDNN] Fix Convolution Performance when IC <=4 (PaddlePaddle#58435)

danleifeng · Oct 30, 2023 · ae10e7c · ae10e7c
1 parent 6ccca1d
commit ae10e7c
Show file tree

Hide file tree

Showing 2 changed files with 90 additions and 1 deletion.
diff --git a/paddle/phi/kernels/onednn/conv_handler.h b/paddle/phi/kernels/onednn/conv_handler.h
@@ -180,7 +180,7 @@ class ConvOneDNNHandlerT
         weights_md = funcs::OneDNNMemDesc(
             weights_tz, data_type, funcs::OneDNNMemoryFormat::any);
       }
-      if (input->dims().size() == 4 && input->dims()[1] == 3) {
+      if (input->dims().size() == 4 && input->dims()[1] <= 4) {
         chosen_memory_format = funcs::OneDNNMemoryFormat::nhwc;
       }
       const auto dst_md = funcs::OneDNNMemDesc(

diff --git a/test/cpp/fluid/mkldnn/test_conv_mkldnn_nhwc.cc b/test/cpp/fluid/mkldnn/test_conv_mkldnn_nhwc.cc
@@ -108,3 +108,92 @@ TEST(test_conv2d_output, int8) {
 
   op->Run(scope, cpu_place);
 }
+TEST(test_conv2d_output, ic1) {
+  paddle::framework::Scope scope;
+  paddle::platform::CPUPlace cpu_place;
+
+  paddle::framework::OpDesc conv2d_op(nullptr);
+  conv2d_op.SetType("conv2d");
+  conv2d_op.SetInput("Input", {"conv2d-X"});
+  conv2d_op.SetInput("Filter", {"conv2d-Y"});
+  conv2d_op.SetOutput("Output", {"conv2d-Out"});
+
+  AddVarToScope<float>("conv2d-X", &scope, {1, 1, 224, 224});
+  AddVarToScope<float>("conv2d-Y", &scope, {64, 1, 7, 7});
+  AddVarToScope<float>("conv2d-Out", &scope, {1, 64, 218, 218});
+
+  const std::vector<int> strides({1, 1});
+  const std::vector<int> paddings({1, 1});
+  const std::vector<int> dilations({1, 1});
+  const int groups = 1;
+
+  conv2d_op.SetAttr("strides", strides);
+  conv2d_op.SetAttr("paddings", paddings);
+  conv2d_op.SetAttr("dilations", dilations);
+  conv2d_op.SetAttr("groups", groups);
+  conv2d_op.SetAttr("use_mkldnn", true);
+
+  auto op = paddle::framework::OpRegistry::CreateOp(conv2d_op);
+
+  op->Run(scope, cpu_place);
+}
+
+TEST(test_conv2d_output, ic2) {
+  paddle::framework::Scope scope;
+  paddle::platform::CPUPlace cpu_place;
+
+  paddle::framework::OpDesc conv2d_op(nullptr);
+  conv2d_op.SetType("conv2d");
+  conv2d_op.SetInput("Input", {"conv2d-X"});
+  conv2d_op.SetInput("Filter", {"conv2d-Y"});
+  conv2d_op.SetOutput("Output", {"conv2d-Out"});
+
+  AddVarToScope<float>("conv2d-X", &scope, {1, 2, 224, 224});
+  AddVarToScope<float>("conv2d-Y", &scope, {64, 2, 7, 7});
+  AddVarToScope<float>("conv2d-Out", &scope, {1, 64, 218, 218});
+
+  const std::vector<int> strides({1, 1});
+  const std::vector<int> paddings({1, 1});
+  const std::vector<int> dilations({1, 1});
+  const int groups = 1;
+
+  conv2d_op.SetAttr("strides", strides);
+  conv2d_op.SetAttr("paddings", paddings);
+  conv2d_op.SetAttr("dilations", dilations);
+  conv2d_op.SetAttr("groups", groups);
+  conv2d_op.SetAttr("use_mkldnn", true);
+
+  auto op = paddle::framework::OpRegistry::CreateOp(conv2d_op);
+
+  op->Run(scope, cpu_place);
+}
+
+TEST(test_conv2d_output, ic4) {
+  paddle::framework::Scope scope;
+  paddle::platform::CPUPlace cpu_place;
+
+  paddle::framework::OpDesc conv2d_op(nullptr);
+  conv2d_op.SetType("conv2d");
+  conv2d_op.SetInput("Input", {"conv2d-X"});
+  conv2d_op.SetInput("Filter", {"conv2d-Y"});
+  conv2d_op.SetOutput("Output", {"conv2d-Out"});
+
+  AddVarToScope<float>("conv2d-X", &scope, {1, 4, 224, 224});
+  AddVarToScope<float>("conv2d-Y", &scope, {64, 4, 7, 7});
+  AddVarToScope<float>("conv2d-Out", &scope, {1, 64, 218, 218});
+
+  const std::vector<int> strides({1, 1});
+  const std::vector<int> paddings({1, 1});
+  const std::vector<int> dilations({1, 1});
+  const int groups = 1;
+
+  conv2d_op.SetAttr("strides", strides);
+  conv2d_op.SetAttr("paddings", paddings);
+  conv2d_op.SetAttr("dilations", dilations);
+  conv2d_op.SetAttr("groups", groups);
+  conv2d_op.SetAttr("use_mkldnn", true);
+
+  auto op = paddle::framework::OpRegistry::CreateOp(conv2d_op);
+
+  op->Run(scope, cpu_place);
+}