Add Conv Transpose BF16 (#30877)

* Add conv transpose BF16

* Share function GetWeightsTz

* Adjust to review and fix op compatibility

* Add bias to unique handler name

* Remove errors related to paddle enforce

* Add conv2d_transpose to bf16 list and kernel refator

paddle/fluid/framework/ir/graph_pattern_detector.cc

@@ -2192,9 +2192,9 @@ PDNode *patterns::Bfloat16Placement::operator()(
     const std::unordered_set<std::string> &bfloat16_enabled_op_types) {
   std::unordered_set<std::string> supported_op_types =
       std::unordered_set<std::string>(
-          {"concat", "conv2d", "elementwise_add", "elementwise_mul", "fc",
-           "fusion_gru", "gelu", "layer_norm", "matmul", "pool2d", "reshape2",
-           "softmax", "sum", "transpose2"});
+          {"concat", "conv2d", "conv2d_transpose", "elementwise_add",
+           "elementwise_mul", "fc", "fusion_gru", "gelu", "layer_norm",
+           "matmul", "pool2d", "reshape2", "softmax", "sum", "transpose2"});
   if (!bfloat16_enabled_op_types.empty()) {
     supported_op_types = bfloat16_enabled_op_types;
   }

paddle/fluid/framework/ir/mkldnn/conv_bias_mkldnn_fuse_pass.cc

@@ -160,7 +160,7 @@ REGISTER_PASS(conv_transpose_bias_mkldnn_fuse_pass,
 REGISTER_PASS_CAPABILITY(conv_transpose_bias_mkldnn_fuse_pass)
     .AddCombination(
         paddle::framework::compatible::OpVersionComparatorCombination()
-            .LE("conv2d_transpose", 1)
+            .LE("conv2d_transpose", 2)
             .LE("elementwise_add", 1));
 
 REGISTER_PASS(conv3d_bias_mkldnn_fuse_pass,

paddle/fluid/framework/ir/quant_conv2d_dequant_fuse_pass.cc

@@ -329,7 +329,7 @@ REGISTER_PASS_CAPABILITY(quant_conv2d_dequant_fuse_pass)
         paddle::framework::compatible::OpVersionComparatorCombination()
             .LE("conv2d", 1)
             .EQ("fc", 0)
-            .LE("conv2d_transpose", 1)
+            .LE("conv2d_transpose", 2)
             .EQ("fake_quantize_abs_max", 0)
             .EQ("fake_quantize_range_abs_max", 0)
             .EQ("fake_quantize_moving_average_abs_max", 0)

paddle/fluid/inference/analysis/ir_passes/tensorrt_subgraph_pass.cc

@@ -390,7 +390,7 @@ REGISTER_PASS_CAPABILITY(tensorrt_subgraph_pass)
             .LE("elementwise_add", 1)
             .LE("elementwise_mul", 1)
             .EQ("prelu", 0)
-            .LE("conv2d_transpose", 1)
+            .LE("conv2d_transpose", 2)
             .LE("leaky_relu", 1)
             .EQ("fc", 0)
             .EQ("shuffle_channel", 0)

paddle/fluid/operators/conv_transpose_op.cc

@@ -290,6 +290,15 @@ void Conv2DTransposeOpMaker::Make() {
   AddAttr<bool>("use_mkldnn",
                 "(bool, default false) Only used in mkldnn kernel")
       .SetDefault(false);
+  AddAttr<bool>("force_fp32_output",
+                "(bool, default false) Force BF16 kernel output FP32, only "
+                "used in MKL-DNN BF16")
+      .SetDefault(false);
+  AddAttr<std::string>(
+      "mkldnn_data_type",
+      "(string, default \"float32\"). Data type of mkldnn kernel")
+      .SetDefault("float32")
+      .InEnum({"float32", "bfloat16"});
   AddAttr<bool>("fuse_relu", "(bool, default false) Only used in mkldnn kernel")
       .SetDefault(false);
   AddAttr<std::string>("fuse_activation",
@@ -671,7 +680,17 @@ REGISTER_OP_VERSION(conv2d_transpose)
             "output_padding",
             "In order to add additional size to one side of each dimension "
             "in the output",
-            std::vector<int>{}));
+            std::vector<int>{}))
+    .AddCheckpoint(
+        R"ROC(
+      Upgrade conv2d transpose to add a new attributes [force_fp32_output, mkldnn_data_type].
+    )ROC",
+        paddle::framework::compatible::OpVersionDesc()
+            .NewAttr("force_fp32_output",
+                     "Force BF16 kernel output FP32, only used in MKL-DNN BF16",
+                     false)
+            .NewAttr("mkldnn_data_type", "Data type of mkldnn kernel",
+                     "float32"));
 
 REGISTER_OP_VERSION(conv3d_transpose)
     .AddCheckpoint(

paddle/fluid/operators/mkldnn/conv_mkldnn_op.cc

@@ -33,18 +33,6 @@ using mkldnn::stream;
 using platform::GetMKLDNNFormat;
 using platform::to_void_cast;
 
-inline void GetWeightsTz(std::vector<int64_t>& weights_tz,  // NOLINT
-                         const int groups) {
-  if (groups > 1) {
-    // if (is_conv3d) [o, i, d, h, w]->[g, o/g, i, d, h, w]
-    // else [o, i, h, w] -> [g, o/g, i, h, w]
-    weights_tz.push_back(0);
-    std::rotate(weights_tz.begin(), weights_tz.end() - 1, weights_tz.end());
-    weights_tz[0] = groups;
-    weights_tz[1] = weights_tz[1] / groups;
-  }
-}
-
 inline MKLDNNMemoryFormat GetWeightsFormat(const MKLDNNMemoryFormat format,
                                            const int groups,
                                            const bool is_conv3d) {
@@ -198,7 +186,7 @@ class ConvMKLDNNHandlerT
       const auto src_tz = paddle::framework::vectorize(input->dims());
 
       auto weights_tz = paddle::framework::vectorize(filter->dims());
-      GetWeightsTz(weights_tz, groups);
+      platform::GetGroupConvWeightsTz(weights_tz, groups);
 
       const auto dst_tz = paddle::framework::vectorize(output->dims());
 
@@ -322,7 +310,7 @@ class ConvMKLDNNHandlerT
     } else {
       const K* filter_data = filter->data<K>();
       auto weights_tz = framework::vectorize(filter->dims());
-      GetWeightsTz(weights_tz, groups);
+      platform::GetGroupConvWeightsTz(weights_tz, groups);
 
       auto user_src_md = platform::MKLDNNMemDesc(
           weights_tz, platform::MKLDNNGetDataType<K>(),
@@ -640,7 +628,7 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
       auto weights_tz = paddle::framework::vectorize(filter->dims());
       int g = std::max(groups, 1);
 
-      GetWeightsTz(weights_tz, g);
+      platform::GetGroupConvWeightsTz(weights_tz, g);
       auto dst_tz = paddle::framework::vectorize(output->dims());
 
       std::transform(dilations.begin(), dilations.end(), dilations.begin(),
@@ -959,7 +947,7 @@ class ConvMKLDNNGradOpKernel : public paddle::framework::OpKernel<T> {
     auto weights_tz = paddle::framework::vectorize(filter->dims());
 
     int g = std::max(groups, 1);
-    GetWeightsTz(weights_tz, g);
+    platform::GetGroupConvWeightsTz(weights_tz, g);
     auto dst_tz = paddle::framework::vectorize(output_grad->dims());
 
     auto src_format = input->format();

paddle/fluid/operators/mkldnn/conv_transpose_mkldnn_op.cc

@@ -25,245 +25,339 @@ namespace operators {
 using Tensor = framework::Tensor;
 using framework::DataLayout;
 
-template <typename T>
-class ConvTransposeMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
+inline mkldnn::memory::dims GetWeightsTz(const Tensor* filter,
+                                         const int groups) {
+  auto iohw_weights_tz = framework::vectorize(filter->dims());
+  auto weights_tz = iohw_weights_tz;
+
+  // IOHW -> OIHW
+  weights_tz[0] = iohw_weights_tz[1];
+  weights_tz[1] = iohw_weights_tz[0];
+  int g = std::max(groups, 1);
+  platform::GetGroupConvWeightsTz(weights_tz, g);
+  return weights_tz;
+}
+
+template <typename T, typename K, typename T_out>
+class ConvTransposeMKLDNNHandlerT
+    : public platform::MKLDNNHandlerT<T, mkldnn::deconvolution_forward> {
  public:
-  void Compute(const paddle::framework::ExecutionContext& ctx) const override {
-    PADDLE_ENFORCE_EQ(platform::is_cpu_place(ctx.GetPlace()), true,
-                      paddle::platform::errors::PreconditionNotMet(
-                          "Operator DNNL ConvTranspose must use CPUPlace"));
-    const bool is_test = ctx.Attr<bool>("is_test");
-    PADDLE_ENFORCE_EQ(is_test, true,
-                      platform::errors::InvalidArgument(
-                          "ConvTransposeMKLDNN works only for inference. "
-                          "Set is_test = True. but got is_test=False ."));
-
-    auto& dev_ctx =
-        ctx.template device_context<paddle::platform::MKLDNNDeviceContext>();
-    const auto& mkldnn_engine = dev_ctx.GetEngine();
-
-    auto* input = ctx.Input<Tensor>("Input");
-    auto* filter = ctx.Input<Tensor>("Filter");
-    auto* bias = ctx.HasInput("Bias") ? ctx.Input<Tensor>("Bias") : nullptr;
-    auto* output = ctx.Output<Tensor>("Output");
-
-    PADDLE_ENFORCE_EQ(
-        input->layout(), DataLayout::kMKLDNN,
-        platform::errors::InvalidArgument(
-            "Got wrong layout = %d for Input tensor.", input->layout()));
-    PADDLE_ENFORCE_NE(input->format(), MKLDNNMemoryFormat::undef,
-                      platform::errors::InvalidArgument(
-                          "Got wrong format for Input tensor."));
-
-    PADDLE_ENFORCE_EQ(
-        filter->layout(), DataLayout::kMKLDNN,
-        platform::errors::InvalidArgument(
-            "The filter tensor's laytout should be %d, but got %d.",
-            DataLayout::kMKLDNN, filter->layout()));
-    PADDLE_ENFORCE_NE(filter->format(), MKLDNNMemoryFormat::undef,
-                      platform::errors::InvalidArgument(
-                          "Got wrong formats for Filter tensor."));
-
-    PADDLE_ENFORCE_EQ(
-        input->dims().size(), 4,
-        platform::errors::InvalidArgument(
-            "Input must be with 4 dimensions, i.e. NCHW. but got dimension =%d",
-            input->dims().size()));
-    PADDLE_ENFORCE_EQ(
-        filter->dims().size(), 4,
-        platform::errors::InvalidArgument("Filter must be with 4 dimensions, "
-                                          "i.e. OIHW, but got dimension =%d",
-                                          filter->dims().size()));
+  ConvTransposeMKLDNNHandlerT(const framework::ExecutionContext& ctx,
+                              const platform::MKLDNNDeviceContext& dev_ctx,
+                              const mkldnn::engine mkldnn_engine,
+                              platform::Place cpu_place, const Tensor* input,
+                              const Tensor* filter, const Tensor* bias,
+                              Tensor* output, const std::string& unique_name)
+      : platform::MKLDNNHandlerT<T, mkldnn::deconvolution_forward>(
+            dev_ctx, mkldnn_engine, cpu_place,
+            platform::CreateKey(dev_ctx, framework::vectorize(input->dims()),
+                                unique_name)) {
+    if (!this->isCached()) {
+      const bool is_test = ctx.Attr<bool>("is_test");
+      PADDLE_ENFORCE_EQ(is_test, true,
+                        platform::errors::InvalidArgument(
+                            "ConvTransposeMKLDNN works only for inference. "
+                            "The attribute \'is_test\' value should be set to "
+                            "True, but got is_test=False."));
 
-    if (bias) {
       PADDLE_ENFORCE_EQ(
-          bias->layout(), DataLayout::kMKLDNN,
+          input->layout(), DataLayout::kMKLDNN,
           platform::errors::InvalidArgument(
-              "The bias tensor's laytout should be %d, but got %d.",
-              DataLayout::kMKLDNN, bias->layout()));
-      PADDLE_ENFORCE_NE(bias->format(), MKLDNNMemoryFormat::undef,
+              "Got wrong layout = %d for Input tensor.", input->layout()));
+      PADDLE_ENFORCE_NE(input->format(), MKLDNNMemoryFormat::undef,
                         platform::errors::InvalidArgument(
-                            "Got wrong format for Bias tensor."));
+                            "Got wrong format for Input tensor. The input "
+                            "format is undefined."));
 
       PADDLE_ENFORCE_EQ(
-          bias->dims().size(), 1,
-          platform::errors::InvalidArgument("Bias must only have 1 dimension, "
-                                            "i.e. X, but got dimension = %d .",
-                                            bias->dims().size()));
-    }
-
-    std::vector<int> strides_temp = ctx.Attr<std::vector<int>>("strides");
-    std::vector<int64_t> strides(begin(strides_temp), end(strides_temp));
-
-    std::vector<int> paddings_temp = ctx.Attr<std::vector<int>>("paddings");
-    std::vector<int64_t> paddings(begin(paddings_temp), end(paddings_temp));
+          filter->layout(), DataLayout::kMKLDNN,
+          platform::errors::InvalidArgument(
+              "The filter tensor's laytout should be %d, but got %d.",
+              DataLayout::kMKLDNN, filter->layout()));
+      PADDLE_ENFORCE_NE(filter->format(), MKLDNNMemoryFormat::undef,
+                        platform::errors::InvalidArgument(
+                            "Got wrong formats for Filter tensor."));
 
-    std::vector<int> dilations_temp = ctx.Attr<std::vector<int>>("dilations");
-    std::vector<int64_t> dilations(begin(dilations_temp), end(dilations_temp));
+      PADDLE_ENFORCE_EQ(
+          input->dims().size(), 4,
+          platform::errors::InvalidArgument("Input must be with 4 dimensions, "
+                                            "i.e. NCHW. but got dimension =%d",
+                                            input->dims().size()));
+      PADDLE_ENFORCE_EQ(
+          filter->dims().size(), 4,
+          platform::errors::InvalidArgument("Filter must be with 4 dimensions, "
+                                            "i.e. OIHW, but got dimension =%d",
+                                            filter->dims().size()));
+
+      if (bias) {
+        PADDLE_ENFORCE_EQ(
+            bias->layout(), DataLayout::kMKLDNN,
+            platform::errors::InvalidArgument(
+                "The bias tensor's laytout should be %d, but got %d.",
+                DataLayout::kMKLDNN, bias->layout()));
+        PADDLE_ENFORCE_NE(bias->format(), MKLDNNMemoryFormat::undef,
+                          platform::errors::InvalidArgument(
+                              "Got wrong format for Bias tensor."));
+
+        PADDLE_ENFORCE_EQ(bias->dims().size(), 1,
+                          platform::errors::InvalidArgument(
+                              "Bias must only have 1 dimension, "
+                              "i.e. X, but got dimension = %d .",
+                              bias->dims().size()));
+      }
 
-    int groups = ctx.Attr<int>("groups");
-    std::string padding_algorithm = ctx.Attr<std::string>("padding_algorithm");
+      std::vector<int> strides_temp = ctx.Attr<std::vector<int>>("strides");
+      mkldnn::memory::dims strides(begin(strides_temp), end(strides_temp));
 
-    PADDLE_ENFORCE_EQ(
-        strides.size(), 2,
-        platform::errors::Unimplemented(
-            "Now we only support 2d oneDNN convolution transpose op"));
+      std::vector<int> paddings_temp = ctx.Attr<std::vector<int>>("paddings");
+      mkldnn::memory::dims paddings(begin(paddings_temp), end(paddings_temp));
 
-    auto input_dims = input->dims();
-    auto data_dims = framework::slice_ddim(input_dims, 2, input_dims.size());
-    auto filter_dims = filter->dims();
-    auto filter_data_dims =
-        framework::slice_ddim(filter_dims, 2, filter_dims.size());
+      std::vector<int> dilations_temp = ctx.Attr<std::vector<int>>("dilations");
+      mkldnn::memory::dims dilations(begin(dilations_temp),
+                                     end(dilations_temp));
 
-    auto ksize = framework::vectorize(filter_data_dims);
+      int groups = ctx.Attr<int>("groups");
+      std::string padding_algorithm =
+          ctx.Attr<std::string>("padding_algorithm");
 
-    UpdatePaddingAndDilation(&paddings, &dilations, padding_algorithm,
-                             data_dims, strides, ksize);
+      PADDLE_ENFORCE_EQ(
+          strides.size(), 2,
+          platform::errors::Unimplemented(
+              "Now we only support 2d oneDNN convolution transpose op"));
+
+      const auto& input_dims = input->dims();
+      const auto data_dims =
+          framework::slice_ddim(input_dims, 2, input_dims.size());
+      const auto& filter_dims = filter->dims();
+      const auto filter_data_dims =
+          framework::slice_ddim(filter_dims, 2, filter_dims.size());
+
+      const auto ksize = framework::vectorize(filter_data_dims);
+
+      UpdatePaddingAndDilation(&paddings, &dilations, padding_algorithm,
+                               data_dims, strides, ksize);
+
+      std::transform(dilations.begin(), dilations.end(), dilations.begin(),
+                     [](int64_t i) { return i - 1; });
+
+      const auto src_tz = framework::vectorize(input->dims());
+      const auto weights_tz = GetWeightsTz(filter, groups);
+      const auto dst_tz = framework::vectorize(output->dims());
+      const auto mkldnn_paddings = platform::ToMkldnnPadding(paddings);
+
+      /* create memory descriptor for convolution without specified format
+       * ('any') which lets a primitive (convolution in this case) choose
+       * the memory format preferred for best performance
+       */
+      const auto chosen_memory_format = MKLDNNMemoryFormat::any;
+      const std::string fuse_activation =
+          ctx.Attr<std::string>("fuse_activation");
+      const float fuse_alpha = ctx.Attr<float>("fuse_alpha");
+      const float fuse_beta = ctx.Attr<float>("fuse_beta");
+
+      auto data_type = mkldnn::memory::data_type::f32;
+      if (ctx.Attr<std::string>("mkldnn_data_type") == "bfloat16" ||
+          std::is_same<T_out, platform::bfloat16>::value)
+        data_type = mkldnn::memory::data_type::bf16;
+
+      const auto src_md =
+          platform::MKLDNNMemDesc(src_tz, data_type, chosen_memory_format);
+      const auto weights_md =
+          platform::MKLDNNMemDesc(weights_tz, data_type, chosen_memory_format);
+      const auto dst_md = platform::MKLDNNMemDesc(
+          dst_tz, platform::MKLDNNGetDataType<T_out>(), chosen_memory_format);
+
+      const mkldnn::primitive_attr conv_trans_attr =
+          CreatePostOps(fuse_activation, fuse_alpha, fuse_beta);
+      auto fwd_prop_kind = is_test ? mkldnn::prop_kind::forward_inference
+                                   : mkldnn::prop_kind::forward_training;
+      if (bias) {
+        std::vector<int64_t> bias_tz = framework::vectorize(bias->dims());
+        const auto bias_md =
+            platform::MKLDNNMemDesc(bias_tz, data_type, MKLDNNMemoryFormat::x);
+        this->AcquireForwardPrimitiveDescriptor(
+            conv_trans_attr, fwd_prop_kind,
+            dnnl::algorithm::deconvolution_direct, src_md, weights_md, bias_md,
+            dst_md, strides, dilations, mkldnn_paddings[0], mkldnn_paddings[1]);
+      } else {
+        this->AcquireForwardPrimitiveDescriptor(
+            conv_trans_attr, fwd_prop_kind,
+            dnnl::algorithm::deconvolution_direct, src_md, weights_md, dst_md,
+            strides, dilations, mkldnn_paddings[0], mkldnn_paddings[1]);
+      }
+    }
+  }
 
-    std::transform(dilations.begin(), dilations.end(), dilations.begin(),
-                   [](int64_t i) { return i - 1; });
+  mkldnn::primitive_attr CreatePostOps(const std::string& fuse_activation,
+                                       const float& fuse_alpha,
+                                       const float& fuse_beta) {
+    mkldnn::primitive_attr conv_attr;
+    mkldnn::post_ops post_operations;
+
+    // Fusion with ReLU layer is executed through the PostOps feature. Create a
+    // PostOps object and configure it to execute an eltwise relu operation.
+    if (fuse_activation == "relu" || fuse_activation == "leaky_relu") {
+      constexpr float scale = 1.0f;
+      post_operations.append_eltwise(scale, mkldnn::algorithm::eltwise_relu,
+                                     fuse_alpha, fuse_beta);
+    } else if (fuse_activation == "relu6") {
+      constexpr float scale = 1.0f;
+      post_operations.append_eltwise(scale,
+                                     mkldnn::algorithm::eltwise_bounded_relu,
+                                     fuse_alpha, fuse_beta);
+    } else if (fuse_activation == "swish") {
+      constexpr float scale = 1.0f;
+      post_operations.append_eltwise(scale, mkldnn::algorithm::eltwise_swish,
+                                     fuse_alpha, fuse_beta);
+    }
+    conv_attr.set_post_ops(post_operations);
+    return conv_attr;
+  }
 
+  std::shared_ptr<mkldnn::memory> AcquireSrcMemoryWithReorder(
+      const framework::Tensor* input) {
     const T* input_data = input->data<T>();
-    const T* filter_data = filter->data<T>();
-
-    auto src_tz = paddle::framework::vectorize<int64_t>(input->dims());
-    auto iohw_weights_tz =
-        paddle::framework::vectorize<int64_t>(filter->dims());
-    auto weights_tz = iohw_weights_tz;
-
-    // IOHW -> OIHW
-    weights_tz[0] = iohw_weights_tz[1];
-    weights_tz[1] = iohw_weights_tz[0];
-
-    // Custom Reorder from IOHW to OIHW
-    auto iohw2oihw_reorder =
-        [&iohw_weights_tz](const T* filter_data) -> std::shared_ptr<T> {
-      int o = iohw_weights_tz[1];
-      int c = iohw_weights_tz[0];
-      int h = iohw_weights_tz[2];
-      int w = iohw_weights_tz[3];
-      std::shared_ptr<T> reordered_filter_data(new T[o * c * h * w](),
-                                               std::default_delete<T[]>());
-      for (int i = 0; i < c; ++i) {
-        for (int j = 0; j < o; ++j) {
-          int in_offset = j * h * w + i * o * h * w;
-          int out_offset = j * c * h * w + i * h * w;
-          std::memcpy(&(reordered_filter_data.get())[out_offset],
-                      &filter_data[in_offset], h * w * sizeof(T));
-        }
+    const std::string user_key_suffix{"@src_mem_p_user"};
+    auto user_src_mem_p = this->AcquireMemory(user_key_suffix);
+    if (!user_src_mem_p) {
+      auto user_src_md = platform::MKLDNNMemDesc(
+          framework::vectorize(input->dims()), platform::MKLDNNGetDataType<T>(),
+          input->format());
+      return this->AcquireMemoryWithReorder(
+          user_src_md, this->fwd_pd_->src_desc(),
+          platform::to_void_cast<T>(input_data), "@src_mem_p");
+    } else {
+      const std::string target_key_suffix{"@src_mem_p_target"};
+      const auto target_src_mem_p = this->AcquireMemory(target_key_suffix);
+      user_src_mem_p->set_data_handle(platform::to_void_cast<T>(input_data));
+      if (user_src_mem_p != target_src_mem_p) {
+        this->AcquireReorder(user_src_mem_p, target_src_mem_p, "@src_mem_p");
       }
+      return target_src_mem_p;
+    }
+  }
+
+  std::shared_ptr<mkldnn::memory> AcquireWeightsMemoryWithReorder(
+      const framework::Tensor* filter, const int& groups, const bool& is_test) {
+    // This is workaround to make execution faster, delete
+    // if statement after including md inside Tensor
+    auto weights_mem_p = this->AcquireMemory("@weights_mem_p_target");
+    if (is_test && weights_mem_p) {
+      return weights_mem_p;
+    } else {
+      const K* filter_data = filter->data<K>();
+      auto weights_tz = GetWeightsTz(filter, groups);
+      int g = std::max(groups, 1);
+
+      auto user_src_md = platform::MKLDNNMemDesc(
+          weights_tz, platform::MKLDNNGetDataType<K>(),
+          (g == 1) ? filter->format() : MKLDNNMemoryFormat::goihw);
+
+      auto iohw_weights_tz = framework::vectorize(filter->dims());
+      // Custom Reorder from IOHW to OIHW
+      auto iohw2oihw_reorder =
+          [&iohw_weights_tz](const K* filter_data) -> std::shared_ptr<K> {
+        int o = iohw_weights_tz[1];
+        int c = iohw_weights_tz[0];
+        int h = iohw_weights_tz[2];
+        int w = iohw_weights_tz[3];
+        std::shared_ptr<K> reordered_filter_data(new K[o * c * h * w](),
+                                                 std::default_delete<K[]>());
+        for (int i = 0; i < c; ++i) {
+          for (int j = 0; j < o; ++j) {
+            int in_offset = j * h * w + i * o * h * w;
+            int out_offset = j * c * h * w + i * h * w;
+            std::memcpy(&(reordered_filter_data.get())[out_offset],
+                        &filter_data[in_offset], h * w * sizeof(K));
+          }
+        }
 
-      return reordered_filter_data;
-    };
-
-    int g = std::max(groups, 1);
-    if (g > 1) {
-      int o = weights_tz[0];
-      int i = weights_tz[1];
-      int h = weights_tz[2];
-      int w = weights_tz[3];
-      weights_tz.resize(5);
-      weights_tz[0] = g;
-      weights_tz[1] = o / g;
-      weights_tz[2] = i;
-      weights_tz[3] = h;
-      weights_tz[4] = w;
+        return reordered_filter_data;
+      };
+
+      return this->template AcquireMemoryWithReorder<K>(
+          user_src_md, this->fwd_pd_->weights_desc(),
+          platform::to_void_cast<K>(filter_data), "@weights_mem_p", is_test,
+          iohw2oihw_reorder);
     }
-    auto dst_tz = paddle::framework::vectorize<int64_t>(output->dims());
-
-    // Get unique name for storing MKLDNN primitives
-    const std::string key =
-        platform::CreateKey(dev_ctx, src_tz, ctx.OutputName("Output"));
-
-    std::vector<mkldnn::primitive> pipeline;
-
-    auto user_src_md = platform::MKLDNNMemDesc(
-        {src_tz}, platform::MKLDNNGetDataType<T>(), input->format());
-    auto user_weights_md = platform::MKLDNNMemDesc(
-        {weights_tz}, platform::MKLDNNGetDataType<T>(),
-        (g == 1) ? MKLDNNMemoryFormat::oihw : MKLDNNMemoryFormat::goihw);
-
-    /* create memory descriptor for convolution without specified format
-     * ('any') which lets a primitive (convolution in this case) choose
-     * the memory format preferred for best performance
-     */
-    auto chosen_memory_format = MKLDNNMemoryFormat::any;
-    std::string fuse_activation = ctx.Attr<std::string>("fuse_activation");
-    float fuse_alpha = ctx.Attr<float>("fuse_alpha");
-    float fuse_beta = ctx.Attr<float>("fuse_beta");
-
-    auto src_md = platform::MKLDNNMemDesc(
-        src_tz, platform::MKLDNNGetDataType<T>(), chosen_memory_format);
-    auto weights_md = platform::MKLDNNMemDesc(
-        weights_tz, platform::MKLDNNGetDataType<T>(), chosen_memory_format);
-    std::vector<int64_t> bias_tz;
-    auto dst_md = platform::MKLDNNMemDesc(
-        dst_tz, platform::MKLDNNGetDataType<T>(), chosen_memory_format);
-
-    platform::ConvTransposeMKLDNNHandler handler(dev_ctx, mkldnn_engine, key);
-    // create a deconv(conv transpose) primitive descriptor and save it for
-    // usage in backward
-    std::shared_ptr<mkldnn::deconvolution_forward::primitive_desc>
-        conv_transpose_pd;
-    auto fwd_prop_kind = is_test ? mkldnn::prop_kind::forward_inference
-                                 : mkldnn::prop_kind::forward_training;
-    if (bias) {
-      bias_tz = paddle::framework::vectorize<int64_t>(bias->dims());
-      auto bias_md = platform::MKLDNNMemDesc(
-          bias_tz, platform::MKLDNNGetDataType<T>(), MKLDNNMemoryFormat::x);
-      conv_transpose_pd = handler.AcquireConvolutionPrimitiveDescriptor(
-          src_md, weights_md, bias_md, dst_md, strides, dilations, paddings,
-          mkldnn_engine, fuse_activation, fuse_alpha, fuse_beta, false,
-          fwd_prop_kind);
+  }
+
+  std::shared_ptr<mkldnn::memory> AcquireBiasMemoryWithReorder(
+      const framework::Tensor* bias, const bool& is_test) {
+    auto bias_mem_p = this->AcquireMemory("@bias_mem_p_target");
+    if (is_test && bias_mem_p) {
+      return bias_mem_p;
     } else {
-      conv_transpose_pd = handler.AcquireConvolutionPrimitiveDescriptor(
-          src_md, weights_md, boost::none, dst_md, strides, dilations, paddings,
-          mkldnn_engine, fuse_activation, fuse_alpha, fuse_beta, false,
-          fwd_prop_kind);
+      const K* bias_data = bias->data<K>();
+      auto user_bias_md = platform::MKLDNNMemDesc(
+          framework::vectorize(bias->dims()), platform::MKLDNNGetDataType<K>(),
+          MKLDNNMemoryFormat::x);
+      return this->AcquireMemoryWithReorder(
+          user_bias_md, this->fwd_pd_->bias_desc(),
+          platform::to_void_cast<K>(bias_data), "@bias_mem_p", is_test);
     }
+  }
+};
 
-    // create mkldnn memory from input tensors (data/weights)
-    auto user_src_memory_p = handler.AcquireSrcMemory(
-        user_src_md, platform::to_void_cast<T>(input_data));
-    auto user_weights_memory_p = handler.AcquireWeightsMemory(
-        user_weights_md, platform::to_void_cast<T>(filter_data),
-        is_test ? iohw2oihw_reorder : platform::user_function());
+template <typename T, typename K>
+class ConvTransposeMKLDNNOpKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    PADDLE_ENFORCE_EQ(platform::is_cpu_place(ctx.GetPlace()), true,
+                      platform::errors::PreconditionNotMet(
+                          "Operator DNNL ConvTranspose must use CPUPlace"));
+    const bool is_bfloat16 =
+        ctx.Attr<std::string>("mkldnn_data_type") == "bfloat16";
+    const bool force_fp32_output = ctx.Attr<bool>("force_fp32_output");
+    if (is_bfloat16) {
+      if (force_fp32_output)
+        Execute<float>(ctx);
+      else
+        Execute<platform::bfloat16>(ctx);
+    } else {
+      Execute<float>(ctx);
+    }
+  }
 
-    // create reorder primitive if the input format is not the preferred one
-    auto src_memory_p =
-        handler.AcquireSrcMemoryFromPrimitive(user_src_memory_p, pipeline);
-    auto weights_memory_p = handler.AcquireWeightsMemoryFromPrimitive(
-        user_weights_memory_p, pipeline, is_test);
+  template <typename T_out>
+  void Execute(const framework::ExecutionContext& ctx) const {
+    auto& dev_ctx =
+        ctx.template device_context<platform::MKLDNNDeviceContext>();
+    const auto& mkldnn_engine = dev_ctx.GetEngine();
 
-    auto output_data =
-        output->mutable_data<T>(ctx.GetPlace(), handler.GetDstMemorySize());
-    auto dst_memory_p = handler.AcquireDstMemoryFromPrimitive(
-        platform::to_void_cast<T>(output_data));
+    const bool is_test = ctx.Attr<bool>("is_test");
 
-    auto conv_p = handler.AcquireConvolution();
+    const auto* input = ctx.Input<Tensor>("Input");
+    const auto* filter = ctx.Input<Tensor>("Filter");
+    const auto* bias =
+        ctx.HasInput("Bias") ? ctx.Input<Tensor>("Bias") : nullptr;
+    auto* output = ctx.Output<Tensor>("Output");
+    const std::string unique_name = ctx.InputName("Input") +
+                                    ctx.InputName("Filter") +
+                                    (bias ? ctx.InputName("Bias") : "");
+    ConvTransposeMKLDNNHandlerT<T, K, T_out> handler(
+        ctx, dev_ctx, mkldnn_engine, ctx.GetPlace(), input, filter, bias,
+        output, unique_name);
+    auto src_memory_p = handler.AcquireSrcMemoryWithReorder(input);
+    auto weights_memory_p = handler.AcquireWeightsMemoryWithReorder(
+        filter, ctx.Attr<int>("groups"), is_test);
+
+    std::shared_ptr<dnnl::memory> dst_memory_p =
+        handler.template AcquireDstMemory<T_out>(output);
+    auto conv_p = handler.AcquireForwardPrimitive();
+
+    std::unordered_map<int, dnnl::memory> args = {
+        {MKLDNN_ARG_SRC, *src_memory_p},
+        {MKLDNN_ARG_WEIGHTS, *weights_memory_p},
+        {MKLDNN_ARG_DST, *dst_memory_p}};
 
-    auto& astream = platform::MKLDNNDeviceContext::tls().get_stream();
     if (bias) {
-      const T* bias_data = bias->data<T>();
-      auto user_bias_md = platform::MKLDNNMemDesc(
-          {bias_tz}, platform::MKLDNNGetDataType<T>(), MKLDNNMemoryFormat::x);
-      auto user_bias_memory_p = handler.AcquireBiasMemory(
-          user_bias_md, platform::to_void_cast<T>(bias_data));
-
-      auto bias_memory_p =
-          handler.AcquireBiasMemoryFromPrimitive(user_bias_memory_p, pipeline);
-
-      conv_p->execute(astream, {{MKLDNN_ARG_SRC, *src_memory_p},
-                                {MKLDNN_ARG_WEIGHTS, *weights_memory_p},
-                                {MKLDNN_ARG_BIAS, *bias_memory_p},
-                                {MKLDNN_ARG_DST, *dst_memory_p}});
-    } else {
-      conv_p->execute(astream, {{MKLDNN_ARG_SRC, *src_memory_p},
-                                {MKLDNN_ARG_WEIGHTS, *weights_memory_p},
-                                {MKLDNN_ARG_DST, *dst_memory_p}});
+      auto bias_memory_p = handler.AcquireBiasMemoryWithReorder(bias, is_test);
+      args.insert({MKLDNN_ARG_BIAS, *bias_memory_p});
     }
+    auto& astream = platform::MKLDNNDeviceContext::tls().get_stream();
+    conv_p->execute(astream, args);
     astream.wait();
-
     output->set_layout(DataLayout::kMKLDNN);
     output->set_format(platform::GetMKLDNNFormat(*dst_memory_p));
   }
@@ -274,5 +368,7 @@ class ConvTransposeMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 
-REGISTER_OP_KERNEL(conv2d_transpose, MKLDNN, ::paddle::platform::CPUPlace,
-                   ops::ConvTransposeMKLDNNOpKernel<float>);
+REGISTER_OP_KERNEL(
+    conv2d_transpose, MKLDNN, ::paddle::platform::CPUPlace,
+    ops::ConvTransposeMKLDNNOpKernel<float, float>,
+    ops::ConvTransposeMKLDNNOpKernel<paddle::platform::bfloat16, float>);

paddle/fluid/platform/mkldnn_helper.h

@@ -492,6 +492,19 @@ inline std::vector<std::vector<int64_t>> ToMkldnnPadding(
   }
 }
 
+// The function adjusts the vector of weight dimensions for group convolutions
+inline void GetGroupConvWeightsTz(std::vector<int64_t>& weights_tz,  // NOLINT
+                                  const int groups) {
+  if (groups > 1) {
+    // if (is_conv3d) [o, i, d, h, w]->[g, o/g, i, d, h, w]
+    // else [o, i, h, w] -> [g, o/g, i, h, w]
+    weights_tz.push_back(0);
+    std::rotate(weights_tz.begin(), weights_tz.end() - 1, weights_tz.end());
+    weights_tz[0] = groups;
+    weights_tz[1] = weights_tz[1] / groups;
+  }
+}
+
 inline bool HasOpINT8DataType(const paddle::framework::OpDesc* op) {
   return (op->GetAttrIfExists<std::string>("mkldnn_data_type") == "int8" ||
           op->GetAttrIfExists<bool>("use_quantizer"));

paddle/fluid/platform/mkldnn_reuse.h

@@ -250,10 +250,12 @@ class MKLDNNHandlerT {
     astream.wait();
   }
 
+  template <typename F = T>
   std::shared_ptr<mkldnn::memory> AcquireMemoryWithReorder(
       const mkldnn::memory::desc& user_md,
       const mkldnn::memory::desc& target_md, void* ptr,
-      const std::string& suffix, bool is_persistent = false) {
+      const std::string& suffix, bool is_persistent = false,
+      std::function<std::shared_ptr<F>(const F*)> custom_reorder_func = {}) {
     const auto target_key = key_ + suffix + "_target";
     const auto key_reorder_p = key_ + suffix + "reorder_p";
     const auto user_key = key_ + suffix + "_user";
@@ -262,6 +264,12 @@ class MKLDNNHandlerT {
         std::static_pointer_cast<dnnl::memory>(dev_ctx_.GetBlob(target_key));
 
     if (target_memory_p == nullptr) {
+      if (custom_reorder_func) {
+        auto reordered_data =
+            custom_reorder_func(reinterpret_cast<const F*>(ptr));
+        dev_ctx_.SetBlob(key_reorder_p + "-custom_reorder", reordered_data);
+        ptr = reinterpret_cast<void*>(reordered_data.get());
+      }
       auto user_memory_p =
           std::make_shared<dnnl::memory>(user_md, engine_, ptr);
       if (user_md != target_md) {
@@ -1288,6 +1296,5 @@ static void SetDstMemoryQuantized(
   dst_memory.reset(
       new mkldnn::memory(*dst_md, engine, to_void_cast<T>(output_data)));
 }
-
 }  // namespace platform
 }  // namespace paddle

python/paddle/fluid/tests/unittests/mkldnn/test_conv2d_transpose_bf16_mkldnn_op.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import unittest
+import numpy as np
+import paddle.fluid.core as core
+from paddle.fluid.tests.unittests.op_test import OpTest, convert_float_to_uint16
+
+from paddle.fluid.tests.unittests.test_conv2d_transpose_op import conv2dtranspose_forward_naive
+from paddle import enable_static
+
+
+def conv2d_bias_naive(out, bias):
+    _, out_c, _, _ = out.shape
+
+    for l in range(out_c):
+        out[:, l, :, :] = out[:, l, :, :] + bias[l]
+    return out
+
+
+@unittest.skipIf(not core.supports_bfloat16(),
+                 "place does not support BF16 evaluation")
+class TestConv2DTransposeBF16MKLDNNOp(OpTest):
+    def test_check_output(self):
+        self.check_output_with_place(core.CPUPlace())
+
+    def test_check_grad(self):
+        pass
+
+    def test_check_grad_no_input(self):
+        pass
+
+    def test_check_grad_no_filter(self):
+        pass
+
+    def init_op_type(self):
+        self.data_format = "NCHW"
+        self.op_type = 'conv2d_transpose'
+        self._cpu_only = True
+
+    def init_test_case(self):
+        self.pad = [0, 0]
+        self.fuse_bias = False
+        self.use_mkldnn = True
+        self.is_test = True
+        self.bias_size = None
+        self.fuse_activation = ""
+        self.fuse_alpha = 0.0
+        self.fuse_beta = 0.0
+        self.stride = [1, 1]
+        self.dilations = [1, 1]
+        self.input_size = [2, 3, 5, 5]  # NCHW
+        f_c = self.input_size[1]
+        self.filter_size = [f_c, 6, 3, 3]
+        self.groups = 1
+        self.output_size = None
+        self.output_padding = []
+        self.data_format = "NCHW"
+        self.pad = [0, 0]
+        self.padding_algorithm = "EXPLICIT"
+        self.force_fp32_output = False
+
+    def setUp(self):
+        self.input_type = np.uint16
+        self.dtype = np.uint16
+        self.mkldnn_data_type = "bfloat16"
+        self.init_op_type()
+        self.init_test_case()
+
+        input = np.random.random(self.input_size).astype(np.float32)
+        filter = np.random.random(self.filter_size).astype(np.float32)
+
+        self.attrs = {
+            'strides': self.stride,
+            'paddings': self.pad,
+            'padding_algorithm': self.padding_algorithm,
+            'groups': self.groups,
+            'dilations': self.dilations,
+            'is_test': self.is_test,
+            'use_mkldnn': self.use_mkldnn,
+            'mkldnn_data_type': self.mkldnn_data_type,
+            'force_fp32_output': self.force_fp32_output,
+            'data_format': self.data_format,
+            'fuse_activation': self.fuse_activation,
+            'fuse_alpha': self.fuse_alpha,
+            'fuse_beta': self.fuse_beta
+        }
+        if self.output_size is not None:
+            self.attrs['output_size'] = self.output_size
+
+        if len(self.output_padding) > 0:
+            self.attrs['output_padding'] = self.output_padding
+
+        output = conv2dtranspose_forward_naive(input, filter,
+                                               self.attrs).astype(np.float32)
+
+        if self.input_type is not np.float32:
+            input = convert_float_to_uint16(input)
+
+        self.inputs = {
+            'Input': input.view(self.input_type),
+            'Filter': OpTest.np_dtype_to_fluid_dtype(filter)
+        }
+
+        if self.fuse_bias and self.bias_size is not None:
+            bias = np.random.random(self.bias_size).astype(np.float32)
+            output = conv2d_bias_naive(output, bias)
+            output = output.astype(np.float32)
+            self.attrs['fuse_bias'] = self.fuse_bias
+            self.inputs['Bias'] = OpTest.np_dtype_to_fluid_dtype(bias)
+
+        if self.fuse_activation == "relu":
+            output = np.maximum(output, 0).astype(np.float32)
+        output = output.astype(np.float32)
+
+        if not self.force_fp32_output:
+            output = convert_float_to_uint16(output, self.attrs['data_format'])
+
+        self.outputs['Output'] = output
+
+
+class TestMKLDNNFuseBias(TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestMKLDNNFuseBias, self).init_test_case()
+        self.pad = [1, 1]
+        self.fuse_bias = True
+        self.bias_size = [6]
+
+
+class TestMKLDNNWithPad(TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestMKLDNNWithPad, self).init_test_case()
+        self.pad = [1, 1]
+        self.input_size = [2, 3, 10, 10]
+
+
+class TestMKLDNNWithStride(TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestMKLDNNWithStride, self).init_test_case()
+        self.pad = [1, 1]
+        self.stride = [2, 2]
+        self.input_size = [2, 3, 6, 6]  # NCHW
+
+
+class TestMKLDNNWithAsymPad(TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestMKLDNNWithAsymPad, self).init_test_case()
+        self.pad = [0, 0, 1, 2]
+        self.padding_algorithm = "EXPLICIT"
+
+
+class TestMKLDNNWithSamePad(TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestMKLDNNWithSamePad, self).init_test_case()
+        self.pad = [0, 0]
+        self.padding_algorithm = "SAME"
+
+
+class TestMKLDNNWithValidPad(TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestMKLDNNWithValidPad, self).init_test_case()
+        self.pad = [1, 1]
+        self.padding_algorithm = "VALID"
+
+
+class TestMKLDNNWithValidPad_NHWC(TestMKLDNNWithValidPad):
+    def init_test_case(self):
+        super(TestMKLDNNWithValidPad_NHWC, self).init_test_case()
+        self.data_format = 'NHWC'
+        N, C, H, W = self.input_size
+        self.input_size = [N, H, W, C]
+
+
+class TestConv2DTransposeMKLDNNWithDilationsExplicitPad(
+        TestConv2DTransposeBF16MKLDNNOp):
+    def init_test_case(self):
+        super(TestConv2DTransposeMKLDNNWithDilationsExplicitPad,
+              self).init_test_case()
+        self.stride = [2, 1]
+        self.dilations = [1, 2]
+        self.groups = 1
+        self.input_size = [4, 3, 8, 7]  # NCHW
+        f_c = self.input_size[1]
+        self.filter_size = [f_c, 6, 4, 3]
+        self.pad = [1, 3, 2, 1]
+        self.padding_algorithm = "EXPLICIT"
+
+
+if __name__ == '__main__':
+    enable_static()
+    unittest.main()

python/paddle/fluid/tests/unittests/mkldnn/test_conv2d_transpose_mkldnn_op.py

@@ -82,6 +82,8 @@ class TestConv2DTransposeMKLDNNOp(TestConv2DTransposeOp):
         self.attrs['fuse_activation'] = self.fuse_activation
         self.attrs['fuse_alpha'] = self.fuse_alpha
         self.attrs['fuse_beta'] = self.fuse_beta
+        self.attrs['mkldnn_data_type'] = 'float32'
+        self.attrs['force_fp32_output'] = False
 
         self.outputs['Output'] = output
 
@@ -150,3 +152,8 @@ class TestConv2DTransposeMKLDNNWithDilationsExplicitPad(
         self.filter_size = [f_c, 6, 4, 3]
         self.pad = [1, 3, 2, 1]
         self.padding_algorithm = "EXPLICIT"
+
+
+if __name__ == '__main__':
+    enable_static()
+    unittest.main()

python/paddle/fluid/tests/unittests/op_test.py

@@ -221,12 +221,18 @@ def copy_bits_from_float_to_uint16(f):
     return struct.unpack('<I', struct.pack('<f', f))[0] >> 16
 
 
-def convert_float_to_uint16(float_list):
+def convert_float_to_uint16(float_list, data_format="NCHW"):
+    if data_format == "NHWC":
+        float_list = np.transpose(float_list, [0, 3, 1, 2])
+
     new_output = []
     for x in np.nditer(float_list):
         new_output.append(np.uint16(copy_bits_from_float_to_uint16(x)))
+    new_output = np.reshape(new_output, float_list.shape).view(np.uint16)
 
-    return np.reshape(new_output, float_list.shape).view(np.uint16)
+    if data_format == "NHWC":
+        new_output = np.transpose(new_output, [0, 2, 3, 1])
+    return new_output
 
 
 class OpTest(unittest.TestCase):

tools/static_mode_white_list.py

@@ -590,6 +590,7 @@ STATIC_MODE_TESTING_LIST = [
     'test_conv2d_int8_mkldnn_op',
     'test_conv2d_mkldnn_op',
     'test_conv2d_transpose_mkldnn_op',
+    'test_conv2d_transpose_bf16_mkldnn_op',
     'test_conv3d_mkldnn_op',
     'test_dequantize_mkldnn_op',
     'test_elementwise_add_mkldnn_op',