!9043 Add support to op L2Loss on gpu

From: @yuan_shen_zhou
Reviewed-by: @liangchenghui,@liangchenghui
Signed-off-by: @liangchenghui,@liangchenghui

mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/l2_loss.cu

@@ -0,0 +1,38 @@
+/**
+ * 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 "l2_loss.cuh"
+#include "runtime/device/gpu/cuda_common.h"
+#include "backend/kernel_compiler/gpu/cuda_impl/util.cuh"
+
+template <typename T>
+__global__ void L2LossKernel(const size_t input_size, const T *input , T *output) {
+  T ret = 0;
+  for (size_t id = blockIdx.x * blockDim.x + threadIdx.x; id < input_size; id += blockDim.x * gridDim.x) {
+      ret = (input[id] * input[id]);
+      ret /= static_cast<T>(2);
+      MsAtomicAdd(output, ret);
+  }
+  return;
+}
+
+template <typename T>
+void L2Loss(const size_t input_size, const T *input , T *output, cudaStream_t stream) {
+  L2LossKernel<<<GET_BLOCKS(input_size), GET_THREADS, 0, stream>>>(input_size, input, output);
+}
+
+template void L2Loss<float>(const size_t input_size, const float *input , float *output, cudaStream_t stream);
+template void L2Loss<half>(const size_t input_size, const half *input , half *output, cudaStream_t stream);

mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/l2_loss.cuh

@@ -0,0 +1,21 @@
+/**
+ * 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_L2_LOSS_H_
+#define MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_L2_LOSS_H_
+template <typename T>
+void L2Loss(const size_t input_size, const T *input , T *output, cudaStream_t stream);
+#endif  // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_L2_LOSS_H_

mindspore/ccsrc/backend/kernel_compiler/gpu/nn/l2_loss_gpu_kernel.cc

@@ -0,0 +1,26 @@
+/**
+ * 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/l2_loss_gpu_kernel.h"
+
+namespace mindspore {
+namespace kernel {
+MS_REG_GPU_KERNEL_ONE(L2Loss, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
+                      L2LossGpuKernel, float)
+MS_REG_GPU_KERNEL_ONE(L2Loss, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
+                      L2LossGpuKernel, half)
+}  // namespace kernel
+}  // namespace mindspore

mindspore/ccsrc/backend/kernel_compiler/gpu/nn/l2_loss_gpu_kernel.h

@@ -0,0 +1,71 @@
+/**
+ * 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_L2_LOSS_GPU_KERNEL_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_NN_L2_LOSS_GPU_KERNEL_H_
+
+#include <vector>
+#include "backend/kernel_compiler/gpu/gpu_kernel.h"
+#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
+#include "backend/kernel_compiler/gpu/cuda_impl/l2_loss.cuh"
+namespace mindspore {
+namespace kernel {
+template <typename T>
+class L2LossGpuKernel : public GpuKernel {
+ public:
+  L2LossGpuKernel() : input_size_(1) {}
+  ~L2LossGpuKernel() 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> &workspaces,
+              const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
+    T *input = GetDeviceAddress<T>(inputs, 0);
+    T *output = GetDeviceAddress<T>(outputs, 0);
+
+    L2Loss(input_size_, input, output, reinterpret_cast<cudaStream_t>(stream_ptr));
+    return true;
+  }
+
+  bool Init(const CNodePtr &kernel_node) override {
+    auto input_shape = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
+    for (size_t i = 0; i < input_shape.size(); i++) {
+      input_size_ *= input_shape[i];
+    }
+
+    InitSizeLists();
+    return true;
+  }
+
+ protected:
+  void InitSizeLists() override {
+    input_size_list_.push_back(input_size_ * sizeof(T));
+    output_size_list_.push_back(sizeof(T));
+  }
+
+ private:
+  size_t input_size_;
+
+  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_L2_LOSS_GPU_KERNEL_H_

mindspore/ops/operations/nn_ops.py

@@ -2157,9 +2157,7 @@ class L2Loss(PrimitiveWithInfer):
     Set `input_x` as x and output as loss.
 
     .. math::
-        loss = sum(x ** 2) / nelement(x)
+        loss = sum(x ** 2) / 2
-
-    :math:`nelement(x)` represents the number of `input_x`.
 
     Inputs:
         - **input_x** (Tensor) - A input Tensor. Data type must be float16 or float32.
@@ -2168,7 +2166,7 @@ class L2Loss(PrimitiveWithInfer):
         Tensor, has the same dtype as `input_x`. The output tensor is the value of loss which is a scalar tensor.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU``
 
     Examples
         >>> input_x = Tensor(np.array([1, 2, 3]), mindspore.float16)

tests/st/ops/gpu/test_l2loss_op.py

@@ -0,0 +1,100 @@
+# 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 numpy as np
+import pytest
+
+import mindspore.context as context
+import mindspore.nn as nn
+import mindspore as ms
+from mindspore import Tensor
+from mindspore.ops import operations as P
+
+
+class L2LossNet(nn.Cell):
+    def __init__(self):
+        super(L2LossNet, self).__init__()
+        self.l2_loss = P.L2Loss()
+
+    def construct(self, x):
+        return self.l2_loss(x)
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_gather_pynative_fp32_22():
+    context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
+    error = 1e-4
+    x = Tensor(np.array([[1., 2.], [3., 4.]]), ms.float32)
+    expect = np.array(15, np.float32)
+    output = P.L2Loss()(x)
+    diff = output.asnumpy() - expect
+    assert np.all(diff < error)
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_gather_pynative_fp16_22():
+    context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
+    error = 1e-4
+    x = Tensor(np.array([[1., 2.], [3., 4.]]), ms.float16)
+    expect = np.array(15, np.float16)
+    output = P.L2Loss()(x)
+    diff = output.asnumpy() - expect
+    assert np.all(diff < error)
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_gather_pynative_fp32_14():
+    context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
+    error = 1e-4
+    x = Tensor(np.array([1., 2., 3., 4.]), ms.float32)
+    expect = np.array(15, np.float32)
+    output = P.L2Loss()(x)
+    diff = output.asnumpy() - expect
+    assert np.all(diff < error)
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_gather_pynative_fp16_14():
+    context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
+    error = 1e-4
+    x = Tensor(np.array([1., 2., 3., 4.]), ms.float16)
+    expect = np.array(15, np.float16)
+    output = P.L2Loss()(x)
+    diff = output.asnumpy() - expect
+    assert np.all(diff < error)
+
+def test_gather_graph_fp32_14():
+    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+    error = 1e-4
+    x = Tensor(np.array([1., 2., 3., 4.]), ms.float32)
+    expect = np.array(15, np.float32)
+    l2_loss = L2LossNet()
+    output = l2_loss(x)
+    diff = output.asnumpy() - expect
+    assert np.all(diff < error)
+
+def test_gather_graph_fp16_14():
+    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+    error = 1e-4
+    x = Tensor(np.array([1., 2., 3., 4.]), ms.float16)
+    expect = np.array(15, np.float16)
+    l2_loss = L2LossNet()
+    output = l2_loss(x)
+    diff = output.asnumpy() - expect
+    assert np.all(diff < error)