Add EluGrad for CPU.

增加测试用例修改测试用例修改测试用例修改算子实现，alpha 写死为 1.0 修改算子实现，alpha 写死为 1 修改算子注册修改算子实现，dy 和 y 搞反了修改测试代码修改算子实现修改算子实现，float16 单独走一个分支修改算子实现，前端的 float 转换为double 修改算子实现，改回来了修改算子实现，float16 转换为 Eigen 库中的 half 修改算子实现，float16 转换为 Eigen 库中的 half 修改算子实现，float16 转换为 Eigen 库中的 half 修改算子实现在单独的文件中实现 EluGrad 修改算子实现在单独的文件中实现 EluGrad 修改算子实现在单独的文件中实现 EluGrad 修改算子实现在单独的文件中实现 EluGrad 格式化代码修改 pylint 错误更新注释信息删除未使用的私有变量删除未使用的私有变量
4 years ago · d86a8ea367
parent fcc4a4eaea
commit d86a8ea367
4 changed files with 217 additions and 1 deletions
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.cc
@ -0,0 +1,85 @@
+/**
+ * Copyright 2021 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 <cmath>
+#include <string>
+#include <thread>
+#include "backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h"
+#include "runtime/device/cpu/cpu_device_address.h"
+
+namespace mindspore {
+namespace kernel {
+template <typename T>
+void EluGradCPUKernel::EluGrad(const T *input0, const T *input1, T *out, size_t start, size_t end) {
+  const T alpha = static_cast<T>(1);
+  for (size_t i = start; i < end; i++) {
+    out[i] = (input1[i] < static_cast<T>(0)) ? input0[i] * (input1[i] + alpha) : input0[i];
+  }
+}
+
+void EluGradCPUKernel::InitKernel(const CNodePtr &kernel_node) {
+  MS_EXCEPTION_IF_NULL(kernel_node);
+  dtype_ = AnfAlgo::GetPrevNodeOutputInferDataType(kernel_node, 0);
+  if (dtype_ != AnfAlgo::GetPrevNodeOutputInferDataType(kernel_node, 1)) {
+    MS_LOG(EXCEPTION) << "Input0 and input1 must has the same data type";
+  }
+}
+
+bool EluGradCPUKernel::Launch(const std::vector<kernel::AddressPtr> &inputs,
+                              const std::vector<kernel::AddressPtr> & /*workspace*/,
+                              const std::vector<kernel::AddressPtr> &outputs) {
+  if (dtype_ == kNumberTypeFloat32 || dtype_ == kNumberTypeFloat) {
+    LaunchKernel<float>(inputs, outputs);
+  } else if (dtype_ == kNumberTypeFloat16) {
+    LaunchKernel<float16>(inputs, outputs);
+  } else {
+    MS_LOG(EXCEPTION) << "Data type is " << TypeIdLabel(dtype_) << "is not support.";
+  }
+  return true;
+}
+
+template <typename T>
+void EluGradCPUKernel::LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs) {
+  T *input0 = reinterpret_cast<T *>(inputs[0]->addr);
+  T *input1 = reinterpret_cast<T *>(inputs[1]->addr);
+  T *output = reinterpret_cast<T *>(outputs[0]->addr);
+
+  size_t lens = outputs[0]->size > 0 ? static_cast<size_t>(outputs[0]->size / sizeof(T)) : 1;
+  auto max_thread_num = std::thread::hardware_concurrency();
+  size_t thread_num = lens < 128 * max_thread_num ? std::ceil(lens / 128.0) : max_thread_num;
+  MS_LOG(INFO) << "Lens=" << lens << "; use thread_num=" << thread_num << "; max_thread_num: " << max_thread_num;
+  std::vector<std::thread> threads;
+  if (thread_num < 1) {
+    MS_LOG(ERROR) << "Invalid value: thread_num " << thread_num;
+    return;
+  }
+  threads.reserve(thread_num);
+  size_t start = 0;
+  size_t once_compute_size = (lens + thread_num - 1) / thread_num;
+  if (once_compute_size < 1) {
+    MS_LOG(ERROR) << "Invalid value: once_compute_size " << once_compute_size;
+    return;
+  }
+  while (start < lens) {
+    size_t end = (start + once_compute_size) > lens ? lens : (start + once_compute_size);
+    threads.emplace_back(std::thread(&EluGradCPUKernel::EluGrad<T>, this, input0, input1, output, start, end));
+    start += once_compute_size;
+  }
+  for (size_t i = 0; i < threads.size(); ++i) {
+    threads[i].join();
+  }
+}
+}  // namespace kernel
+}  // namespace mindspore
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/elu_grad_cpu_kernel.h
@ -0,0 +1,56 @@
+/**
+ * Copyright 2021 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_CPU_ELU_GRAD_CPU_KERNEL_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_ELU_GRAD_CPU_KERNEL_H_
+#include <memory>
+#include <vector>
+#include "backend/kernel_compiler/cpu/cpu_kernel.h"
+#include "backend/kernel_compiler/cpu/cpu_kernel_factory.h"
+
+namespace mindspore {
+namespace kernel {
+class EluGradCPUKernel : public CPUKernel {
+ public:
+  EluGradCPUKernel() = default;
+  ~EluGradCPUKernel() override = default;
+
+  void InitKernel(const CNodePtr &kernel_node) override;
+
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+              const std::vector<AddressPtr> &outputs) override;
+  template <typename T>
+  void LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs);
+
+ private:
+  template <typename T>
+  void EluGrad(const T *input1, const T *input2, T *out, size_t start, size_t end);
+  TypeId dtype_{kTypeUnknown};
+};
+
+MS_REG_CPU_KERNEL(
+  EluGrad,
+  KernelAttr().AddInputAttr(kNumberTypeFloat32).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
+  EluGradCPUKernel);
+MS_REG_CPU_KERNEL(
+  EluGrad,
+  KernelAttr().AddInputAttr(kNumberTypeFloat16).AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
+  EluGradCPUKernel);
+MS_REG_CPU_KERNEL(
+  EluGrad, KernelAttr().AddInputAttr(kNumberTypeFloat).AddInputAttr(kNumberTypeFloat).AddOutputAttr(kNumberTypeFloat),
+  EluGradCPUKernel);
+}  // namespace kernel
+}  // namespace mindspore
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_ELU_GRAD_CPU_KERNEL_H_
--- a/mindspore/ops/operations/nn_ops.py
+++ b/mindspore/ops/operations/nn_ops.py
@ -462,7 +462,7 @@ class Elu(PrimitiveWithInfer):
        Tensor, has the same shape and data type as `input_x`.

    Supported Platforms:
-        ``Ascend`` ``GPU``
+        ``Ascend`` ``GPU`` ``CPU``

    Examples:
        >>> input_x = Tensor(np.array([[-1.0, 4.0, -8.0], [2.0, -5.0, 9.0]]), mindspore.float32)
--- a/tests/st/ops/cpu/test_elu_grad_op.py
+++ b/tests/st/ops/cpu/test_elu_grad_op.py
@ -0,0 +1,75 @@
+# Copyright 2021 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 numpy as np
+import pytest
+
+import mindspore.context as context
+import mindspore.nn as nn
+from mindspore import Tensor
+from mindspore.ops.operations import _grad_ops as G
+
+context.set_context(mode=context.GRAPH_MODE, device_target='CPU')
+
+
+class NetEluGrad(nn.Cell):
+    def __init__(self):
+        super(NetEluGrad, self).__init__()
+        self.elu_grad = G.EluGrad()
+
+    def construct(self, dy, y):
+        return self.elu_grad(dy, y)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_cpu
+@pytest.mark.env_onecard
+def test_elu_grad_fp32():
+    y = Tensor(np.array([[[[-0.3, 1, 2],
+                           [1, -0.6, 1],
+                           [2, 1, -2]]]]).astype(np.float32))
+    dy = Tensor(np.array([[[[-11, 2, 4],
+                            [-1, 1, -1],
+                            [-4, 4, -4]]]]).astype(np.float32))
+
+    expect = np.array([[[[-7.7, 2, 4],
+                         [-1, 0.4, -1],
+                         [-4, 4, 4]]]]).astype(np.float32)
+
+    error = np.ones(shape=[1, 1, 3, 3]) * 1.0e-6
+
+    elu_grad = NetEluGrad()
+    output = elu_grad(dy, y)
+    print(output)
+    diff = np.abs(output.asnumpy() - expect)
+    double_check = diff / expect
+    assert np.all(double_check < error)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_cpu
+@pytest.mark.env_onecard
+def test_elu_grad_fp16():
+    y = Tensor(np.array([[0.5, 2, 5.5], [4.5, -2, 0]]).astype(np.float16))
+    dy = Tensor(np.array([[2, 1, 1.5], [-0.5, -1, -3]]).astype(np.float16))
+    expect = np.array([[2, 1, 1.5], [-0.5, 1, -3]]).astype(np.float16)
+    error = np.ones(shape=[2, 3]) * 1.0e-3
+
+    elu_grad = NetEluGrad()
+    output = elu_grad(dy, y)
+    print(output)
+    diff = np.abs(output.asnumpy() - expect)
+    double_check = diff / expect
+    assert np.all(double_check < error)