Gpu support LayerNorm kernel

mindspore/ccsrc/kernel/gpu/cuda_impl/layer_norm_grad_impl.cu

+/**
+ * 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 <stdio.h>
+#include <stdint.h>
+#include <cuda_runtime.h>
+#include "kernel/gpu/cuda_impl/layer_norm_grad_impl.cuh"
+
+constexpr int NUM_PER_THREAD_REDUCE = 4;
+constexpr int WARP_SIZE = 32;
+
+template <typename T>
+inline __device__ void GammaAndBetaThreadReduce(const int& col, const int& row_dim, const int& col_dim,
+                                                const T& epsilon, const T* dy, const T* x, const T* mean, const T* var,
+                                                T* dg, T* db) {
+  int loop_num = (row_dim + NUM_PER_THREAD_REDUCE - 1) / NUM_PER_THREAD_REDUCE;
+  for (int i = threadIdx.x; i < loop_num; i += blockDim.x) {
+    for (int j = 0; j < NUM_PER_THREAD_REDUCE; j++) {
+      int row = NUM_PER_THREAD_REDUCE * i + j;
+      if (row >= row_dim) {
+        return;
+      }
+
+      int pos = row * col_dim + col;
+      dg[0] += dy[pos] * pow(var[row] + epsilon, -0.5) * (x[pos] - mean[row]);
+      db[0] += dy[pos];
+    }
+  }
+}
+
+template <typename T>
+inline __device__ void GammaAndBetaWarpReduce(T* dg, T* db) {
+  for (int delta = (WARP_SIZE >> 1); delta > 0; delta >>= 1) {
+    dg[0] += __shfl_down_sync(0xffffffff, dg[0], delta);
+    db[0] += __shfl_down_sync(0xffffffff, db[0], delta);
+  }
+}
+
+template <typename T>
+inline __device__ void GammaAndBetaBlockReduce(const int& col, const int& row_dim, T* dg, T* db, T* dg_addr,
+                                               T* db_addr) {
+  if (threadIdx.x >= row_dim) {
+    return;
+  }
+
+  // load data to share memory
+  // thread(0, 32, 64, 96, ...) keep the data
+  extern __shared__ T share_mem[];
+  if (threadIdx.x % WARP_SIZE == 0) {
+    int offset = threadIdx.x / WARP_SIZE * 2;
+    share_mem[offset] = dg[0];
+    share_mem[offset + 1] = db[0];
+  }
+  __syncthreads();
+
+  for (int stride = blockDim.x / WARP_SIZE / 2; stride > 0; stride >>= 1) {
+    if (threadIdx.x < stride) {
+      int offset = (threadIdx.x + stride) * 2;
+      share_mem[threadIdx.x * 2] += share_mem[offset];
+      share_mem[threadIdx.x * 2 + 1] += share_mem[offset + 1];
+    }
+  }
+  __syncthreads();
+
+  if (threadIdx.x == 0) {
+    dg_addr[col] = share_mem[0];
+    db_addr[col] = share_mem[1];
+  }
+}
+
+template <typename T>
+__global__ void GammaAndBetaPropKernel(const int row_dim, const int col_dim, const T epsilon, const T* dy, const T* x,
+                                       const T* mean_addr, const T* var_addr, T* dg_addr, T* db_addr) {
+  // row: [0:param_axis]
+  // col: [param_axis:]
+  // dg[i][j] = dy[i][j] * (var[i] + epsilon, -0.5) * (x[i][j] - mean[i])
+  // dg[j] = \Sigma_{j}dg[i][j]
+  for (int col = blockIdx.x; col < col_dim; col += gridDim.x) {
+    T dg = 0;
+    T db = 0;
+    GammaAndBetaThreadReduce(col, row_dim, col_dim, epsilon, dy, x, mean_addr, var_addr, &dg, &db);
+    GammaAndBetaWarpReduce(&dg, &db);
+    GammaAndBetaBlockReduce(col, row_dim, &dg, &db, dg_addr, db_addr);
+  }
+}
+
+template <typename T>
+inline __device__ void InputThreadReduce(const int& row, const int& col_dim, const int& param_dim, const T& epsilon,
+                                         T* sum1, T* sum2, T* sum3, const T* dy, const T* x, const T* mean,
+                                         const T* var, const T* gamma) {
+  int loop_num = (col_dim + NUM_PER_THREAD_REDUCE - 1) / NUM_PER_THREAD_REDUCE;
+  for (int i = threadIdx.x; i < loop_num; i += blockDim.x) {
+    for (int j = 0; j < NUM_PER_THREAD_REDUCE; j++) {
+      int col = NUM_PER_THREAD_REDUCE * i + j;
+      if (col >= col_dim) {
+        return;
+      }
+
+      int pos = row * col_dim + col;
+      int gamma_offset = pos % param_dim;
+      T v1 = dy[pos] * gamma[gamma_offset];
+      T v2 = x[pos] - mean[row];
+
+      sum1[0] += -0.5 * v1 * v2 * pow(var[row] + epsilon, -1.5);
+      sum2[0] += v1;
+      sum3[0] += -2.0 * v2;
+    }
+  }
+}
+
+template <typename T>
+inline __device__ void InputWarpReduce(T* sum1, T* sum2, T* sum3) {
+  for (int delta = (WARP_SIZE >> 1); delta > 0; delta >>= 1) {
+    sum1[0] += __shfl_down_sync(0xffffffff, sum1[0], delta);
+    sum2[0] += __shfl_down_sync(0xffffffff, sum2[0], delta);
+    sum3[0] += __shfl_down_sync(0xffffffff, sum3[0], delta);
+  }
+}
+
+template <typename T>
+inline __device__ void InputBlockReduce(const int& col_dim, T* sum1, T* sum2, T* sum3, T* share_mem) {
+  if (threadIdx.x >= col_dim) {
+    return;
+  }
+
+  // load data to share memory
+  // thread(0, 32, 64, 96, ...) keep the data
+  if (threadIdx.x % WARP_SIZE == 0) {
+    int offset = threadIdx.x / WARP_SIZE * 3;
+    share_mem[offset] = sum1[0];
+    share_mem[offset + 1] = sum2[0];
+    share_mem[offset + 2] = sum3[0];
+  }
+  __syncthreads();
+
+  for (int stride = blockDim.x / WARP_SIZE / 2; stride > 0; stride >>= 1) {
+    if (threadIdx.x < stride) {
+      int offset = (threadIdx.x + stride) * 3;
+      share_mem[threadIdx.x * 3] += share_mem[offset];
+      share_mem[threadIdx.x * 3 + 1] += share_mem[offset + 1];
+      share_mem[threadIdx.x * 3 + 2] += share_mem[offset + 2];
+    }
+  }
+  __syncthreads();
+}
+
+template <typename T>
+inline __device__ void InputProp(const int& row, const int& col_dim, const int& param_dim, const T& epsilon,
+                                 const T* dy, const T* x, const T* mean, const T* var, const T* gamma, T* dx,
+                                 const T* share_mem) {
+  for (int col = threadIdx.x; col < col_dim; col += blockDim.x) {
+    int pos = (row * col_dim + col);
+    int gamma_offset = pos % param_dim;
+    T v1 = dy[pos] * gamma[gamma_offset];
+    T v2 = x[pos] - mean[row];
+    T v3 = pow(var[row] + epsilon, -0.5);
+    dx[pos] = v1 * v3 + share_mem[0] * (2.0 / col_dim) * v2 +
+              (-1.0 * v3 * share_mem[1] + (1.0 / col_dim) * share_mem[0] * share_mem[2]) * (1.0 / col_dim);
+  }
+}
+
+template <typename T>
+__global__ void InputPropKernel(const int row_dim, const int col_dim, const int param_dim, const T epsilon, const T* dy,
+                                const T* x, const T* mean, const T* var, const T* gamma, T* dx) {
+  for (int row = blockIdx.x; row < row_dim; row += gridDim.x) {
+    T sum1 = 0;
+    T sum2 = 0;
+    T sum3 = 0;
+    extern __shared__ T share_mem[];
+    InputThreadReduce(row, col_dim, param_dim, epsilon, &sum1, &sum2, &sum3, dy, x, mean, var, gamma);
+    InputWarpReduce(&sum1, &sum2, &sum3);
+    InputBlockReduce(col_dim, &sum1, &sum2, &sum3, share_mem);
+    InputProp(row, col_dim, param_dim, epsilon, dy, x, mean, var, gamma, dx, share_mem);
+  }
+}
+
+template <typename T>
+void LayerNormGrad(const int& row_dim, const int& col_dim, const int& param_dim, const T& epsilon, const T* dy,
+                   const T* x, const T* mean, const T* var, const T* gamma, T* dx, T* dg, T* db, cudaStream_t stream) {
+  int share_mem =
+    ((col_dim + NUM_PER_THREAD_REDUCE - 1) / NUM_PER_THREAD_REDUCE + WARP_SIZE - 1) / WARP_SIZE * 3 * sizeof(T);
+  InputPropKernel<<<row_dim, 256, share_mem, stream>>>(row_dim, col_dim, param_dim, epsilon, dy, x, mean, var, gamma,
+                                                       dx);
+
+  share_mem =
+    ((row_dim + NUM_PER_THREAD_REDUCE - 1) / NUM_PER_THREAD_REDUCE + WARP_SIZE - 1) / WARP_SIZE * 2 * sizeof(T);
+  GammaAndBetaPropKernel<<<col_dim, 256, share_mem, stream>>>(row_dim, col_dim, epsilon, dy, x, mean, var, dg, db);
+}
+
+template void LayerNormGrad(const int& row_dim, const int& col_dim, const int& param_dim, const float& epsilon,
+                            const float* dy, const float* x, const float* mean, const float* var, const float* gamma,
+                            float* dx, float* dg, float* db, cudaStream_t stream);

mindspore/ccsrc/kernel/gpu/cuda_impl/layer_norm_grad_impl.cuh

+/**
+ * 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_KERNEL_GPU_CUDA_IMPL_LAYER_NORM_GRAD_H_
+#define MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_LAYER_NORM_GRAD_H_
+
+#include "device/gpu/cuda_common.h"
+
+template <typename T>
+void LayerNormGrad(const int& row_dim, const int& col_dim, const int& param_dim, const T& epsilon, const T* dy,
+                   const T* x, const T* mean, const T* var, const T* gamma, T* dx, T* dg, T* db, cudaStream_t stream);
+
+#endif  // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_LAYER_NORM_GRAD_H_

mindspore/ccsrc/kernel/gpu/cuda_impl/layer_norm_impl.cu

+/**
+ * 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 <stdio.h>
+#include <stdint.h>
+#include <cuda_runtime.h>
+#include "kernel/gpu/cuda_impl/layer_norm_impl.cuh"
+
+constexpr int NUM_PER_THREAD_REDUCE = 4;
+constexpr int WARP_SIZE = 32;
+
+template <typename T>
+inline __device__ void MeanAndVarAccumulation(T* mean, T* var, T* num, const T& val) {
+  // Welford Algorithm:
+  // \mu_k = \mu_{k-1} + (x_k - \mu_{k-1})/k
+  // \sigma_k^2 = \sigma_{k-1}^2 + (x_k - \mu_{k-1}) * (x_k - \mu_k)
+  num[0]++;
+  T mean_new = mean[0] + (val - mean[0]) / num[0];
+  var[0] = var[0] + (val - mean[0]) * (val - mean_new);
+  mean[0] = mean_new;
+}
+
+template <typename T>
+inline __device__ void MeanAndVarMerge(T* m1, T* v1, T* n1, const T& m2, const T& v2, const T& n2) {
+  if (n2 == 0) {
+    return;
+  }
+
+  T count = n1[0] + n2;
+  v1[0] = v1[0] + v2 + (m1[0] - m2) * (m1[0] - m2) * n1[0] * n2 / count;
+  m1[0] = (n1[0] * m1[0] + n2 * m2) / count;
+  n1[0] = count;
+}
+
+template <typename T>
+inline __device__ void ThreadReduce(const int& col_dim, const T* block_addr, T* mean, T* var, T* num) {
+  int loop_num = (col_dim + NUM_PER_THREAD_REDUCE - 1) / NUM_PER_THREAD_REDUCE;
+  for (int i = threadIdx.x; i < loop_num; i += blockDim.x) {
+    for (int j = 0; j < NUM_PER_THREAD_REDUCE; j++) {
+      int pos = NUM_PER_THREAD_REDUCE * i + j;
+      if (pos >= col_dim) {
+        return;
+      }
+      MeanAndVarAccumulation(mean, var, num, block_addr[pos]);
+    }
+  }
+}
+
+template <typename T>
+inline __device__ void WarpReduce(T* mean, T* var, T* num) {
+  for (int delta = (WARP_SIZE >> 1); delta > 0; delta >>= 1) {
+    T mean_other = __shfl_down_sync(0xffffffff, mean[0], delta);
+    T var_other = __shfl_down_sync(0xffffffff, var[0], delta);
+    T num_other = __shfl_down_sync(0xffffffff, num[0], delta);
+    MeanAndVarMerge(mean, var, num, mean_other, var_other, num_other);
+  }
+}
+
+template <typename T>
+inline __device__ void BlockReduce(const int& col_dim, T* mean, T* var, T* num, T* mean_addr, T* var_addr,
+                                   T* share_mem) {
+  if (threadIdx.x >= col_dim) {
+    return;
+  }
+
+  // load data to share memory
+  // thread(0, 32, 64, 96, ...) keep the data
+  if (threadIdx.x % WARP_SIZE == 0) {
+    int offset = threadIdx.x / WARP_SIZE * 3;
+    share_mem[offset] = mean[0];
+    share_mem[offset + 1] = var[0];
+    share_mem[offset + 2] = num[0];
+  }
+  __syncthreads();
+
+  for (int stride = blockDim.x / WARP_SIZE / 2; stride > 0; stride >>= 1) {
+    if (threadIdx.x < stride) {
+      int offset = (threadIdx.x + stride) * 3;
+      MeanAndVarMerge(&share_mem[threadIdx.x * 3], &share_mem[threadIdx.x * 3 + 1], &share_mem[threadIdx.x * 3 + 2],
+                      share_mem[offset], share_mem[offset + 1], share_mem[offset + 2]);
+    }
+  }
+  __syncthreads();
+
+  if (threadIdx.x == 0) {
+    mean_addr[blockIdx.x] = share_mem[0];  // todo: blockDim.x < row
+    share_mem[1] /= col_dim;
+    var_addr[blockIdx.x] = share_mem[1];
+  }
+}
+
+template <typename T>
+inline __device__ void LayerNorm(const int& row, const int& col_dim, const int& param_dim, const T* x,
+                                 const T* share_mem, const T* gamma, const T* beta, const T epsilon, T* y) {
+  for (int col = threadIdx.x; col < col_dim; col += blockDim.x) {
+    int pos = row * col_dim + col;
+    int i = pos % param_dim;
+    y[pos] = (x[pos] - share_mem[0]) / sqrt(share_mem[1] + epsilon) * gamma[i] + beta[i];
+  }
+}
+
+template <typename T>
+__global__ void LayerNormKernel(const int row_dim, const int col_dim, const int param_dim, const T epsilon, const T* x,
+                                const T* gamma, const T* beta, T* y, T* mean_addr, T* var_addr) {
+  for (auto row = blockIdx.x; row < row_dim; row += gridDim.x) {
+    T mean = 0;
+    T var = 0;
+    T num = 0;
+    const T* block_addr = x + row * col_dim;
+    extern __shared__ T share_mem[];
+
+    ThreadReduce(col_dim, block_addr, &mean, &var, &num);
+    WarpReduce(&mean, &var, &num);
+    BlockReduce(col_dim, &mean, &var, &num, mean_addr, var_addr, share_mem);
+
+    __syncthreads();
+    LayerNorm(row, col_dim, param_dim, x, share_mem, gamma, beta, epsilon, y);
+  }
+}
+
+template <typename T>
+void LayerNorm(const int& row_dim, const int& col_dim, const int& param_dim, const T& epsilon, const T* x,
+               const T* gamma, const T* beta, T* y, T* mean, T* var, cudaStream_t stream) {
+  const dim3 block(row_dim);
+  const dim3 thread(256);
+  // keep the mean/var/num after warp reduce
+  int share_mem =
+    ((col_dim + NUM_PER_THREAD_REDUCE - 1) / NUM_PER_THREAD_REDUCE + WARP_SIZE - 1) / WARP_SIZE * 3 * sizeof(T);
+  LayerNormKernel<<<block, thread, share_mem, stream>>>(row_dim, col_dim, param_dim, epsilon, x, gamma, beta, y, mean,
+                                                        var);
+}
+
+template void LayerNorm(const int& row_dim, const int& col_dim, const int& param_dim, const float& epsilon,
+                        const float* x, const float* gamma, const float* beta, float* y, float* mean, float* var,
+                        cudaStream_t stream);

mindspore/ccsrc/kernel/gpu/cuda_impl/layer_norm_impl.cuh

+/**
+ * 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_KERNEL_GPU_CUDA_IMPL_LAYER_NORM_H_
+#define MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_LAYER_NORM_H_
+
+#include "device/gpu/cuda_common.h"
+
+template <typename T>
+void LayerNorm(const int& outer, const int& inner, const int& param_dim, const T& epsilon, const T* x, const T* gamma,
+               const T* beta, T* y, T* mean, T* var, cudaStream_t stream);
+
+#endif  // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_LAYER_NORM_H_

mindspore/ccsrc/kernel/gpu/nn/layer_norm_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 "kernel/gpu/nn/layer_norm_gpu_kernel.h"
+
+namespace mindspore {
+namespace kernel {
+MS_REG_GPU_KERNEL_ONE(LayerNorm,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      LayerNormGpuKernel, float)
+}  // namespace kernel
+}  // namespace mindspore

mindspore/ccsrc/kernel/gpu/nn/layer_norm_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_KERNEL_GPU_NN_LAYER_NORM_GPU_KERNEL_H_
+#define MINDSPORE_CCSRC_KERNEL_GPU_NN_LAYER_NORM_GPU_KERNEL_H_
+
+#include <vector>
+#include "kernel/gpu/gpu_kernel.h"
+#include "kernel/gpu/gpu_kernel_factory.h"
+#include "kernel/gpu/cuda_impl/layer_norm_impl.cuh"
+
+namespace mindspore {
+namespace kernel {
+template <typename T>
+class LayerNormGpuKernel : public GpuKernel {
+ public:
+  LayerNormGpuKernel() : input_row_(1), input_col_(1), param_dim_(1) {}
+  ~LayerNormGpuKernel() override = default;
+
+  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> &,
+              const std::vector<AddressPtr> &outputs, uintptr_t stream_ptr) override {
+    auto x = GetDeviceAddress<T>(inputs, 0);
+    auto gamma = GetDeviceAddress<T>(inputs, 1);
+    auto beta = GetDeviceAddress<T>(inputs, 2);
+    auto y = GetDeviceAddress<T>(outputs, 0);
+    auto mean = GetDeviceAddress<T>(outputs, 1);
+    auto variance = GetDeviceAddress<T>(outputs, 2);
+
+    T epsilon = 10e-12;
+    LayerNorm(input_row_, input_col_, param_dim_, epsilon, x, gamma, beta, y, mean, variance,
+              reinterpret_cast<cudaStream_t>(stream_ptr));
+    return true;
+  }
+  bool Init(const CNodePtr &kernel_node) override {
+    int begin_norm_axis = GetAttr<int>(kernel_node, "begin_norm_axis");
+    int begin_params_axis = GetAttr<int>(kernel_node, "begin_params_axis");
+
+    auto input_shape = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
+    if (begin_norm_axis < 0) {
+      begin_norm_axis += input_shape.size();
+    }
+
+    if (begin_params_axis < 0) {
+      begin_params_axis += input_shape.size();
+    }
+
+    for (size_t i = 0; i < IntToSize(begin_norm_axis); i++) {
+      input_row_ *= input_shape[i];
+    }
+
+    for (size_t i = begin_norm_axis; i < input_shape.size(); i++) {
+      input_col_ *= input_shape[i];
+    }
+
+    for (size_t i = begin_params_axis; i < input_shape.size(); i++) {
+      param_dim_ *= input_shape[i];
+    }
+
+    InitSizeLists();
+    return true;
+  }
+
+ protected:
+  void InitSizeLists() override {
+    input_size_list_.push_back(input_row_ * input_col_ * sizeof(T));
+    input_size_list_.push_back(param_dim_ * sizeof(T));
+    input_size_list_.push_back(param_dim_ * sizeof(T));
+
+    output_size_list_.push_back(input_row_ * input_col_ * sizeof(T));
+    output_size_list_.push_back(input_row_ * sizeof(T));
+    output_size_list_.push_back(input_row_ * sizeof(T));
+    return;
+  }
+
+ private:
+  std::vector<size_t> input_size_list_;
+  std::vector<size_t> output_size_list_;
+  std::vector<size_t> workspace_size_list_;
+
+  int input_row_;
+  int input_col_;
+  int param_dim_;
+};
+}  // namespace kernel
+}  // namespace mindspore
+#endif  // MINDSPORE_CCSRC_KERNEL_GPU_NN_LAYER_NORM_GPU_KERNEL_H_

mindspore/ccsrc/kernel/gpu/nn/layer_norm_grad_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 "kernel/gpu/nn/layer_norm_grad_gpu_kernel.h"
+
+namespace mindspore {
+namespace kernel {
+MS_REG_GPU_KERNEL_ONE(LayerNormGrad,
+                      KernelAttr()
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddInputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32)
+                        .AddOutputAttr(kNumberTypeFloat32),
+                      LayerNormGradGpuKernel, float)
+}  // namespace kernel
+}  // namespace mindspore

mindspore/ccsrc/kernel/gpu/nn/layer_norm_grad_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_KERNEL_GPU_NN_LAYER_NORM_GRAD_GPU_KERNEL_H_
+#define MINDSPORE_CCSRC_KERNEL_GPU_NN_LAYER_NORM_GRAD_GPU_KERNEL_H_
+
+#include <vector>
+#include "kernel/gpu/gpu_kernel.h"
+#include "kernel/gpu/gpu_kernel_factory.h"
+#include "kernel/gpu/cuda_impl/layer_norm_grad_impl.cuh"
+
+namespace mindspore {
+namespace kernel {
+template <typename T>
+class LayerNormGradGpuKernel : public GpuKernel {
+ public:
+  LayerNormGradGpuKernel() : input_row_(1), input_col_(1), param_dim_(1) {}
+  ~LayerNormGradGpuKernel() override = default;
+
+  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> &,
+              const std::vector<AddressPtr> &outputs, uintptr_t stream_ptr) override {
+    auto dy = GetDeviceAddress<T>(inputs, 0);
+    auto x = GetDeviceAddress<T>(inputs, 1);
+    auto var = GetDeviceAddress<T>(inputs, 2);
+    auto mean = GetDeviceAddress<T>(inputs, 3);
+    auto gamma = GetDeviceAddress<T>(inputs, 4);
+    auto dx = GetDeviceAddress<T>(outputs, 0);
+    auto dg = GetDeviceAddress<T>(outputs, 1);
+    auto db = GetDeviceAddress<T>(outputs, 2);
+
+    T epsilon = 10e-12;
+    LayerNormGrad(input_row_, input_col_, param_dim_, epsilon, dy, x, mean, var, gamma, dx, dg, db,
+                  reinterpret_cast<cudaStream_t>(stream_ptr));
+    return true;
+  }
+  bool Init(const CNodePtr &kernel_node) override {
+    int begin_norm_axis = GetAttr<int>(kernel_node, "begin_norm_axis");
+    int begin_params_axis = GetAttr<int>(kernel_node, "begin_params_axis");
+
+    auto input_shape = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
+    if (begin_norm_axis < 0) {
+      begin_norm_axis += input_shape.size();
+    }
+
+    if (begin_params_axis < 0) {
+      begin_params_axis += input_shape.size();
+    }
+
+    for (size_t i = 0; i < IntToSize(begin_norm_axis); i++) {
+      input_row_ *= input_shape[i];
+    }
+
+    for (size_t i = begin_norm_axis; i < input_shape.size(); i++) {
+      input_col_ *= input_shape[i];
+    }
+
+    for (size_t i = begin_params_axis; i < input_shape.size(); i++) {
+      param_dim_ *= input_shape[i];
+    }
+
+    InitSizeLists();
+    return true;
+  }
+
+ protected:
+  void InitSizeLists() override {
+    input_size_list_.push_back(input_row_ * input_col_ * sizeof(T));
+    input_size_list_.push_back(input_row_ * input_col_ * sizeof(T));
+    input_size_list_.push_back(input_row_ * sizeof(T));
+    input_size_list_.push_back(input_row_ * sizeof(T));
+    input_size_list_.push_back(param_dim_ * sizeof(T));
+
+    output_size_list_.push_back(input_row_ * input_col_ * sizeof(T));
+    output_size_list_.push_back(param_dim_ * sizeof(T));
+    output_size_list_.push_back(param_dim_ * sizeof(T));
+    return;
+  }
+
+ private:
+  std::vector<size_t> input_size_list_;
+  std::vector<size_t> output_size_list_;
+  std::vector<size_t> workspace_size_list_;
+
+  int input_row_;
+  int input_col_;
+  int param_dim_;
+};
+}  // namespace kernel
+}  // namespace mindspore
+#endif  // MINDSPORE_CCSRC_KERNEL_GPU_NN_LAYER_NORM_GRAD_GPU_KERNEL_H_

tests/st/ops/gpu/test_layer_norm_grad_op.py

+# 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.
+# ============================================================================
+
+import pytest
+import numpy as np
+from mindspore import Tensor
+from mindspore.ops import operations as P
+from mindspore.ops.operations import _grad_ops as G
+from mindspore.ops import composite as C
+import mindspore.nn as nn
+import mindspore.context as context
+
+
+context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+
+class LayerNormGradNet(nn.Cell):
+    def __init__(self, begin_norm_axis, begin_params_axis):
+        super(LayerNormGradNet, self).__init__()
+        self.norm = G.LayerNormGrad(begin_norm_axis, begin_params_axis)
+
+    def construct(self, dy, x, var, mean, gamma):
+        return self.norm(dy, x, var, mean, gamma)
+
+def LayerNormGradReference(x, dy, gamma, epsilon, begin_norm_axis, begin_params_axis):
+    begin_norm_axis = begin_norm_axis if begin_norm_axis >=0 else begin_norm_axis + len(x.shape)
+    begin_params_axis = begin_params_axis if begin_params_axis >=0 else begin_params_axis + len(x.shape)
+
+    norm_axis = [i for i in range(begin_norm_axis, len(x.shape))]
+    param_axis = [i for i in range(0, begin_params_axis)]
+    num = 1
+    for i in range(begin_norm_axis, len(x.shape)):
+        num *= x.shape[i]
+
+    mean = np.mean(x, axis=tuple(norm_axis), keepdims=True)
+    var = np.var(x, axis=tuple(norm_axis), keepdims=True)
+
+    gamma = gamma.reshape((*((1,)*begin_params_axis), *x.shape[begin_params_axis:]))
+    dg = np.sum(dy * np.power(var + epsilon, -0.5) * (x - mean), axis=tuple(param_axis), keepdims=True)
+    db = np.sum(dy, axis=tuple(param_axis), keepdims=True)
+
+    sum1 = np.sum((-0.5) * dy * gamma * (x - mean) * np.power(var + epsilon, -1.5), axis=tuple(norm_axis), keepdims=True)
+    sum2 = np.sum(dy * gamma, axis=tuple(norm_axis), keepdims=True)
+    sum3 = np.sum(-2.0 * (x - mean), axis=tuple(norm_axis), keepdims=True)
+
+    dx1 = dy * gamma * np.power(var + epsilon, -0.5)
+    dx2 = sum1 * 2.0 / num * (x - mean)
+    dx3 = ((-1.0) * np.power(var + epsilon, -0.5) * sum2 + (1.0 / num) * sum1 * sum3) * (1.0 / num)
+    dx = dx1 + dx2 + dx3
+    return dx, dg, db, mean, var
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernormgrad0():
+    begin_norm_axis = 1
+    begin_params_axis = 1
+    x_np = np.random.randn(4096, 3072).astype(np.float32)
+    dy_np = np.random.randn(4096, 3072).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    epsilon = 10e-12
+    dx_np, dg_np, db_np, mean_np, var_np = LayerNormGradReference(x_np, dy_np, gamma_np, epsilon, begin_norm_axis, begin_params_axis)
+
+    dy_ms = Tensor(dy_np)
+    x_ms = Tensor(x_np)
+    var_ms = Tensor(var_np)
+    mean_ms = Tensor(mean_np)
+    gamma_ms = Tensor(gamma_np)
+
+    net = LayerNormGradNet(begin_norm_axis, begin_params_axis)
+    dx_ms, dg_ms, db_ms = net(dy_ms, x_ms, var_ms, mean_ms, gamma_ms)
+
+    assert np.allclose(dx_ms.asnumpy(), dx_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(dg_ms.asnumpy(), dg_np, rtol=1e-6, atol=1e-3)
+    assert np.allclose(db_ms.asnumpy(), db_np, rtol=1e-6, atol=1e-3)
+
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernormgrad1():
+    begin_norm_axis = 1
+    begin_params_axis = 1
+    x_np = np.random.randn(640, 768).astype(np.float32)
+    dy_np = np.random.randn(640, 768).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    epsilon = 10e-12
+    dx_np, dg_np, db_np, mean_np, var_np = LayerNormGradReference(x_np, dy_np, gamma_np, epsilon, begin_norm_axis, begin_params_axis)
+
+    dy_ms = Tensor(dy_np)
+    x_ms = Tensor(x_np)
+    var_ms = Tensor(var_np)
+    mean_ms = Tensor(mean_np)
+    gamma_ms = Tensor(gamma_np)
+
+    net = LayerNormGradNet(begin_norm_axis, begin_params_axis)
+    dx_ms, dg_ms, db_ms = net(dy_ms, x_ms, var_ms, mean_ms, gamma_ms)
+
+    assert np.allclose(dx_ms.asnumpy(), dx_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(dg_ms.asnumpy(), dg_np, rtol=1e-6, atol=1e-3)
+    assert np.allclose(db_ms.asnumpy(), db_np, rtol=1e-6, atol=1e-3)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernormgrad2():
+    begin_norm_axis = -1
+    begin_params_axis = -1
+    x_np = np.random.randn(32, 128, 768).astype(np.float32)
+    dy_np = np.random.randn(32, 128, 768).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    epsilon = 10e-12
+    dx_np, dg_np, db_np, mean_np, var_np = LayerNormGradReference(x_np, dy_np, gamma_np, epsilon, begin_norm_axis, begin_params_axis)
+
+    dy_ms = Tensor(dy_np)
+    x_ms = Tensor(x_np)
+    var_ms = Tensor(var_np)
+    mean_ms = Tensor(mean_np)
+    gamma_ms = Tensor(gamma_np)
+
+    net = LayerNormGradNet(begin_norm_axis, begin_params_axis)
+    dx_ms, dg_ms, db_ms = net(dy_ms, x_ms, var_ms, mean_ms, gamma_ms)
+
+    assert np.allclose(dx_ms.asnumpy(), dx_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(dg_ms.asnumpy(), dg_np, rtol=1e-6, atol=1e-3)
+    assert np.allclose(db_ms.asnumpy(), db_np, rtol=1e-6, atol=1e-3)
\ No newline at end of file

tests/st/ops/gpu/test_layer_norm_op.py

+# 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.
+# ============================================================================
+
+import pytest
+import numpy as np
+from mindspore import Tensor
+from mindspore.ops import operations as P
+import mindspore.nn as nn
+import mindspore.context as context
+
+context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+
+
+class LayerNormNet(nn.Cell):
+    def __init__(self, begin_norm_axis, begin_params_axis):
+        super(LayerNormNet, self).__init__()
+        self.norm = P.LayerNorm(begin_norm_axis, begin_params_axis)
+
+    def construct(self, x, gamma, beta):
+        return self.norm(x, gamma, beta)
+
+def LayerNormReference(begin_norm_axis, begin_params_axis, x, gamma, beta):
+    begin_norm_axis = begin_norm_axis if begin_norm_axis >=0 else begin_norm_axis + len(x.shape)
+    begin_params_axis = begin_params_axis if begin_params_axis >=0 else begin_params_axis + len(x.shape)
+
+    axis = [i for i in range(begin_norm_axis, len(x.shape))]
+    mean = np.mean(x, axis=tuple(axis), keepdims=True)
+    var  = np.var(x, axis=tuple(axis), keepdims=True)
+
+    gamma = gamma.reshape((*((1,)*begin_params_axis), *x.shape[begin_params_axis:]))
+    beta = beta.reshape((*((1,)*begin_params_axis), *x.shape[begin_params_axis:]))
+    y = np.subtract(x, mean) / np.sqrt(var + 1e-12) * gamma + beta
+    return y, mean, var
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernorm0():
+    begin_norm_axis = 1
+    begin_params_axis = 1
+    x_np = np.random.randn(4096, 3072).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    beta_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    y_np, mean_np, var_np = LayerNormReference(begin_norm_axis, begin_params_axis, x_np, gamma_np, beta_np)
+
+    x_ms = Tensor(x_np)
+    gamma_ms = Tensor(gamma_np)
+    beta_ms = Tensor(beta_np)
+    net = LayerNormNet(begin_norm_axis, begin_params_axis)
+    y_ms, mean_ms, var_ms = net(x_ms, gamma_ms, beta_ms)
+
+    assert np.allclose(y_ms.asnumpy(), y_np, atol=1e-6)
+    assert np.allclose(mean_ms.asnumpy(), mean_np, atol=1e-6)
+    assert np.allclose(var_ms.asnumpy(), var_np, atol=1e-6)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernorm1():
+    begin_norm_axis = 1
+    begin_params_axis = 1
+    x_np = np.random.randn(640, 768).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    beta_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    y_np, mean_np, var_np = LayerNormReference(begin_norm_axis, begin_params_axis, x_np, gamma_np, beta_np)
+
+    x_ms = Tensor(x_np)
+    gamma_ms = Tensor(gamma_np)
+    beta_ms = Tensor(beta_np)
+    net = LayerNormNet(begin_norm_axis, begin_params_axis)
+    y_ms, mean_ms, var_ms = net(x_ms, gamma_ms, beta_ms)
+
+
+    assert np.allclose(y_ms.asnumpy(), y_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(mean_ms.asnumpy(), mean_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(var_ms.asnumpy(), var_np, rtol=1e-6, atol=1e-6)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernorm3d_1():
+    begin_norm_axis = -1
+    begin_params_axis = -1
+    x_np = np.random.randn(32, 128, 768).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    beta_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    y_np, mean_np, var_np = LayerNormReference(begin_norm_axis, begin_params_axis, x_np, gamma_np, beta_np)
+
+    x_ms = Tensor(x_np)
+    gamma_ms = Tensor(gamma_np)
+    beta_ms = Tensor(beta_np)
+    net = LayerNormNet(begin_norm_axis, begin_params_axis)
+    y_ms, mean_ms, var_ms = net(x_ms, gamma_ms, beta_ms)
+
+    assert np.allclose(y_ms.asnumpy(), y_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(mean_ms.asnumpy(), mean_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(var_ms.asnumpy(), var_np, rtol=1e-6, atol=1e-6)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_layernorm3d_2():
+    begin_norm_axis = -1
+    begin_params_axis = 1
+    x_np = np.random.randn(32, 128, 768).astype(np.float32)
+    gamma_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    beta_np = np.random.randn(*x_np.shape[begin_params_axis:]).astype(np.float32)
+    y_np, mean_np, var_np = LayerNormReference(begin_norm_axis, begin_params_axis, x_np, gamma_np, beta_np)
+
+    x_ms = Tensor(x_np)
+    gamma_ms = Tensor(gamma_np)
+    beta_ms = Tensor(beta_np)
+    net = LayerNormNet(begin_norm_axis, begin_params_axis)
+    y_ms, mean_ms, var_ms = net(x_ms, gamma_ms, beta_ms)
+
+    assert np.allclose(y_ms.asnumpy(), y_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(mean_ms.asnumpy(), mean_np, rtol=1e-6, atol=1e-6)
+    assert np.allclose(var_ms.asnumpy(), var_np, rtol=1e-6, atol=1e-6)