add FusedBatchNoramEx gpu kernel

mindspore/ccsrc/backend/kernel_compiler/gpu/nn/fused_batch_norm_ex_gpu_kernel.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#include "backend/kernel_compiler/gpu/nn/fused_batch_norm_ex_gpu_kernel.h"
+
+namespace mindspore {
+namespace kernel {
+MS_REG_GPU_KERNEL_ONE(FusedBatchNormEx,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      FusedBatchNormExGpuKernel, float)
+MS_REG_GPU_KERNEL_ONE(FusedBatchNormEx,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat16)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat16)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      FusedBatchNormExGpuKernel, half)
+
+MS_REG_GPU_KERNEL_ONE(FusedBatchNormExWithActivation,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      FusedBatchNormExGpuKernel, float)
+MS_REG_GPU_KERNEL_ONE(FusedBatchNormExWithActivation,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat16)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat16)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      FusedBatchNormExGpuKernel, half)
+
+MS_REG_GPU_KERNEL_ONE(FusedBatchNormExWithAddAndActivation,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      FusedBatchNormExGpuKernel, float)
+MS_REG_GPU_KERNEL_ONE(FusedBatchNormExWithAddAndActivation,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat16)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat16)
+                        .AddOutputAttr(kNumberTypeFloat16)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      FusedBatchNormExGpuKernel, half)
+}  // namespace kernel
+}  // namespace mindspore

mindspore/ccsrc/backend/kernel_compiler/gpu/nn/fused_batch_norm_ex_gpu_kernel.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * 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.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_NN_FUSED_BATCH_NORM_EX_GPU_KERNEL_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_NN_FUSED_BATCH_NORM_EX_GPU_KERNEL_H_
+
+#include <vector>
+#include <string>
+#include "backend/kernel_compiler/gpu/gpu_kernel.h"
+#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
+#include "backend/kernel_compiler/gpu/kernel_constants.h"
+#include "utils/utils.h"
+
+namespace mindspore {
+namespace kernel {
+template <typename T>
+class FusedBatchNormExGpuKernel : public GpuKernel {
+ public:
+  FusedBatchNormExGpuKernel()
+      : input_x_size_(0),
+        input_z_size_(0),
+        para_size_(0),
+        output_size_(0),
+        workspace_size_(0),
+        reserve_size_(0),
+        mode_(CUDNN_BATCHNORM_SPATIAL),
+        bn_ops_(CUDNN_BATCHNORM_OPS_BN),
+        epsilon_(10e-5),
+        exp_avg_factor_(0.1),
+        is_null_input_(false),
+        x_desc_(nullptr),
+        y_desc_(nullptr),
+        z_desc_(nullptr),
+        scale_bias_mean_var_desc_(nullptr),
+        activation_desc_(nullptr),
+        handle_(nullptr),
+        cudnn_data_type_(CUDNN_DATA_FLOAT) {}
+  ~FusedBatchNormExGpuKernel() override { DestroyResource(); }
+
+  const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
+  const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
+  const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
+
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+              const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
+    VARIABLE_NOT_USED(workspace);
+    VARIABLE_NOT_USED(stream_ptr);
+    if (is_null_input_) {
+      return true;
+    }
+    auto x = GetDeviceAddress<T>(inputs, 0);
+    auto scale = GetDeviceAddress<float>(inputs, 1);
+    auto bias = GetDeviceAddress<float>(inputs, 2);
+    auto runing_mean = GetDeviceAddress<float>(inputs, 3);
+    auto runnig_variance = GetDeviceAddress<float>(inputs, 4);
+    T *z = nullptr;
+    if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+      z = GetDeviceAddress<T>(inputs, 5);
+    }
+
+    auto y = GetDeviceAddress<T>(outputs, 0);
+    auto save_mean = GetDeviceAddress<float>(outputs, 3);
+    auto save_variance = GetDeviceAddress<float>(outputs, 4);
+    auto reserve_addr = GetDeviceAddress<float>(outputs, 5);
+    T *workspace_addr = nullptr;
+    if (workspace_size_ != 0) {
+      workspace_addr = GetDeviceAddress<T>(workspace, 0);
+    }
+    const float alpha = 1;
+    const float beta = 0;
+    CHECK_CUDNN_RET_WITH_EXCEPT(
+      cudnnBatchNormalizationForwardTrainingEx(handle_, mode_, bn_ops_, &alpha, &beta, x_desc_, x, z_desc_, z, y_desc_,
+                                               y, scale_bias_mean_var_desc_, scale, bias, exp_avg_factor_, runing_mean,
+                                               runnig_variance, epsilon_, save_mean, save_variance, activation_desc_,
+                                               workspace_addr, workspace_size_, reserve_addr, reserve_size_),
+      "Kernel launch failed");
+    return true;
+  }
+
+  bool Init(const CNodePtr &kernel_node) override {
+    MS_EXCEPTION_IF_NULL(kernel_node);
+    std::string kernel_name = AnfAlgo::GetCNodeName(kernel_node);
+    if (kernel_name == kFusedBatchNormEx) {
+      bn_ops_ = CUDNN_BATCHNORM_OPS_BN;
+    } else if (kernel_name == kFusedBatchNormExWithActivation) {
+      bn_ops_ = CUDNN_BATCHNORM_OPS_BN_ACTIVATION;
+    } else if (kernel_name == kFusedBatchNormExWithAddAndActivation) {
+      bn_ops_ = CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION;
+    } else {
+      MS_LOG(EXCEPTION) << "Invalid kernel name: " << kernel_name;
+    }
+
+    InitResource();
+    mode_ = CUDNN_BATCHNORM_SPATIAL_PERSISTENT;
+    epsilon_ = GetAttr<float>(kernel_node, "epsilon");
+    exp_avg_factor_ = GetAttr<float>(kernel_node, "momentum");
+
+    cudnn_data_type_ = GetCudnnDataType(TypeIdLabel(AnfAlgo::GetInputDeviceDataType(kernel_node, 0)));
+    size_t input_num = AnfAlgo::GetInputTensorNum(kernel_node);
+    if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+      if (input_num != 6) {
+        MS_LOG(EXCEPTION) << "input tensor size is " << input_num << ", " << kernel_name << " should be 6";
+      }
+    } else {
+      if (input_num != 5) {
+        MS_LOG(EXCEPTION) << "input tensor size is " << input_num << ", " << kernel_name << "  should be 5";
+      }
+    }
+
+    auto shape = AnfAlgo::GetInputDeviceShape(kernel_node, 0);
+    if (shape.size() != 4) {
+      MS_LOG(EXCEPTION) << "tensor shape is " << shape.size() << ", FusedBatchNormExGpuKernel should be 4";
+    }
+    is_null_input_ = CHECK_NULL_INPUT(shape);
+    if (is_null_input_) {
+      MS_LOG(WARNING) << "FusedBatchNormExGpuKernel input is null";
+      InitSizeLists();
+      return true;
+    }
+    auto format = AnfAlgo::GetInputFormat(kernel_node, 0);
+    SetTensorDescriptor(format, shape);
+    InitSizeLists();
+    return true;
+  }
+
+ protected:
+  void InitResource() override {
+    handle_ = device::gpu::GPUDeviceManager::GetInstance().GetCudnnHandle();
+    CHECK_CUDNN_RET_WITH_EXCEPT(cudnnCreateTensorDescriptor(&x_desc_), "Create x desc failed");
+    CHECK_CUDNN_RET_WITH_EXCEPT(cudnnCreateTensorDescriptor(&y_desc_), "Create y desc failed");
+    if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnCreateTensorDescriptor(&z_desc_), "Create z desc failed");
+    }
+    CHECK_CUDNN_RET_WITH_EXCEPT(cudnnCreateTensorDescriptor(&scale_bias_mean_var_desc_), "Create para desc failed");
+
+    if (bn_ops_ != CUDNN_BATCHNORM_OPS_BN) {
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnCreateActivationDescriptor(&activation_desc_),
+                                  "Create activation descriptor failed");
+    }
+  }
+
+  void InitSizeLists() override {
+    if (!is_null_input_) {
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnGetTensorSizeInBytes(x_desc_, &input_x_size_), "Get input x size failed");
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnGetTensorSizeInBytes(scale_bias_mean_var_desc_, &para_size_),
+                                  "Get para size failed");
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnGetTensorSizeInBytes(y_desc_, &output_size_), "Get output size failed");
+      if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+        CHECK_CUDNN_RET_WITH_EXCEPT(cudnnGetTensorSizeInBytes(z_desc_, &input_z_size_), "Get input z size failed");
+      }
+
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize(
+                                    handle_, mode_, bn_ops_, x_desc_, z_desc_, y_desc_, scale_bias_mean_var_desc_,
+                                    activation_desc_, &workspace_size_),
+                                  "cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize failed");
+
+      CHECK_CUDNN_RET_WITH_EXCEPT(cudnnGetBatchNormalizationTrainingExReserveSpaceSize(
+                                    handle_, mode_, bn_ops_, activation_desc_, x_desc_, &reserve_size_),
+                                  "Get reserve size failed");
+    }
+
+    input_size_list_.push_back(input_x_size_);  // input x
+    input_size_list_.push_back(para_size_);     // scale
+    input_size_list_.push_back(para_size_);     // bias
+    input_size_list_.push_back(para_size_);     // mean
+    input_size_list_.push_back(para_size_);     // variance
+    if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+      input_size_list_.push_back(input_z_size_);  // input z
+    }
+
+    output_size_list_.push_back(output_size_);   // output
+    output_size_list_.push_back(para_size_);     // save scale
+    output_size_list_.push_back(para_size_);     // save bias
+    output_size_list_.push_back(para_size_);     // save mean
+    output_size_list_.push_back(para_size_);     // save variance
+    output_size_list_.push_back(reserve_size_);  // reserve space
+
+    workspace_size_list_.push_back(workspace_size_);
+  }
+
+ private:
+  void SetTensorDescriptor(const std::string &format, const std::vector<size_t> &shape) {
+    cudnnTensorFormat_t cudnn_format;
+    int batch, channel, height, width;
+    if (format == kOpFormat_NHWC) {
+      batch = SizeToInt(shape[0]);
+      height = SizeToInt(shape[1]);
+      width = SizeToInt(shape[2]);
+      channel = SizeToInt(shape[3]);
+      cudnn_format = CUDNN_TENSOR_NHWC;
+    } else {
+      batch = SizeToInt(shape[0]);
+      channel = SizeToInt(shape[1]);
+      height = SizeToInt(shape[2]);
+      width = SizeToInt(shape[3]);
+      cudnn_format = CUDNN_TENSOR_NCHW;
+    }
+
+    CHECK_CUDNN_RET_WITH_EXCEPT(
+      cudnnSetTensor4dDescriptor(x_desc_, cudnn_format, cudnn_data_type_, batch, channel, height, width),
+      "Set x desc failed");
+
+    CHECK_CUDNN_RET_WITH_EXCEPT(
+      cudnnSetTensor4dDescriptor(y_desc_, cudnn_format, cudnn_data_type_, batch, channel, height, width),
+      "Set y desc failed");
+
+    if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+      CHECK_CUDNN_RET_WITH_EXCEPT(
+        cudnnSetTensor4dDescriptor(z_desc_, cudnn_format, cudnn_data_type_, batch, channel, height, width),
+        "Set z desc failed");
+    }
+
+    CHECK_CUDNN_RET_WITH_EXCEPT(
+      cudnnSetTensor4dDescriptor(scale_bias_mean_var_desc_, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, channel, 1, 1),
+      "Set para desc failed");
+
+    if (bn_ops_ != CUDNN_BATCHNORM_OPS_BN) {
+      CHECK_CUDNN_RET_WITH_EXCEPT(
+        cudnnSetActivationDescriptor(activation_desc_, CUDNN_ACTIVATION_RELU, CUDNN_NOT_PROPAGATE_NAN, 0.0),
+        "cudnnSetActivationDescriptor failed");
+    }
+  }
+
+  void DestroyResource() noexcept {
+    CHECK_CUDNN_RET_WITH_ERROR(cudnnDestroyTensorDescriptor(x_desc_), "Destroy x desc failed");
+    CHECK_CUDNN_RET_WITH_ERROR(cudnnDestroyTensorDescriptor(y_desc_), "Destroy y desc failed");
+    CHECK_CUDNN_RET_WITH_ERROR(cudnnDestroyTensorDescriptor(scale_bias_mean_var_desc_), "Destroy para desc failed");
+    if (bn_ops_ == CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION) {
+      CHECK_CUDNN_RET_WITH_ERROR(cudnnDestroyTensorDescriptor(z_desc_), "Destroy z desc failed");
+    }
+
+    if (bn_ops_ != CUDNN_BATCHNORM_OPS_BN) {
+      CHECK_CUDNN_RET_WITH_ERROR(cudnnDestroyActivationDescriptor(activation_desc_),
+                                 "Destroy activation descriptor failed");
+    }
+  }
+
+  size_t input_x_size_;
+  size_t input_z_size_;
+  size_t para_size_;
+  size_t output_size_;
+  size_t workspace_size_;
+  size_t reserve_size_;
+  cudnnBatchNormMode_t mode_;
+  cudnnBatchNormOps_t bn_ops_;
+  double epsilon_;
+  double exp_avg_factor_;
+  bool is_null_input_;
+  cudnnTensorDescriptor_t x_desc_;
+  cudnnTensorDescriptor_t y_desc_;
+  cudnnTensorDescriptor_t z_desc_;
+  cudnnTensorDescriptor_t scale_bias_mean_var_desc_;
+  cudnnActivationDescriptor_t activation_desc_;
+
+  cudnnHandle_t handle_;
+  cudnnDataType_t cudnn_data_type_;
+  std::vector<size_t> input_size_list_;
+  std::vector<size_t> output_size_list_;
+  std::vector<size_t> workspace_size_list_;
+};
+}  // namespace kernel
+}  // namespace mindspore
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_NN_FUSED_BATCH_NORM_EX_GPU_KERNEL_H_

mindspore/ccsrc/utils/utils.h

@@ -42,6 +42,12 @@ constexpr auto kFusedBN3OpName = "FusedBN3";
 constexpr auto kBNGrad1OpName = "BNGrad1";
 constexpr auto kBNGrad2OpName = "BNGrad2";
 constexpr auto kBNGrad3OpName = "BNGrad3";
+constexpr auto kFusedBatchNormEx = "FusedBatchNormEx";
+constexpr auto kFusedBatchNormExWithActivation = "FusedBatchNormExWithActivation";
+constexpr auto kFusedBatchNormExWithAddAndActivation = "FusedBatchNormExWithAddAndActivation";
+constexpr auto kFusedBatchNormGradEx = "FusedBatchNormGradEx";
+constexpr auto kFusedBatchNormGradExWithActivation = "FusedBatchNormGradExWithActivation";
+constexpr auto kFusedBatchNormGradExWithAddAndActivation = "FusedBatchNormGradExWithAddAndActivation";
 constexpr auto kClearZeroOpName = "ClearZero";
 constexpr auto kAtomicAddrCleanOpName = "AtomicAddrClean";
 constexpr auto kGetNextOpName = "GetNext";