!1487 add cpu kernel "AddN"

Merge pull request !1487 from sunsuodong/addn

mindspore/ccsrc/device/cpu/kernel_select_cpu.cc

@@ -85,7 +85,7 @@ bool IsInputFormatDtypeMatched(const KernelAttr &kernel_attr, const std::vector<
                                const std::vector<TypeId> &input_types,
                                const std::vector<size_t> &input_not_cnode_indexes) {
   if (kernel_attr.GetInputSize() != input_types.size()) {
-    MS_LOG(ERROR) << "required input num:" << kernel_attr.GetInputSize() << ", actual input num:" << input_types.size();
+    MS_LOG(DEBUG) << "required input num:" << kernel_attr.GetInputSize() << ", actual input num:" << input_types.size();
     return false;
   }
   auto input_num = input_types.size();

mindspore/ccsrc/kernel/cpu/addn_cpu_kernel.cc

@@ -0,0 +1,66 @@
+/**
+ * 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/cpu/addn_cpu_kernel.h"
+#include "device/cpu/cpu_device_address.h"
+#include "ir/primitive.h"
+
+namespace mindspore {
+namespace kernel {
+void AddNCPUKernel::InitKernel(const CNodePtr &kernel_node) {
+  CheckParam(kernel_node);
+  input_num_ = AnfAlgo::GetInputTensorNum(kernel_node);
+  output_shape_ = AnfAlgo::GetOutputInferShape(kernel_node, 0);
+  CPUKernelUtils::ExpandDimsTo4(&output_shape_);
+}
+
+bool AddNCPUKernel::Launch(const std::vector<kernel::AddressPtr> &inputs,
+                           const std::vector<kernel::AddressPtr> & /*workspace*/,
+                           const std::vector<kernel::AddressPtr> &outputs) {
+  auto output_addr = reinterpret_cast<float *>(outputs[0]->addr);
+
+  for (size_t i = 0; i < output_shape_[0]; ++i) {
+    for (size_t j = 0; j < output_shape_[1]; ++j) {
+      for (size_t k = 0; k < output_shape_[2]; ++k) {
+        for (size_t m = 0; m < output_shape_[3]; ++m) {
+          auto offset = CPUKernelUtils::CalcOffset(output_shape_, i, j, k, m);
+          float sum = 0;
+          for (size_t index = 0; index < input_num_; ++index) {
+            auto input_addr = reinterpret_cast<float *>(inputs[index]->addr);
+            sum += input_addr[offset];
+          }
+          output_addr[offset] = sum;
+        }
+      }
+    }
+  }
+
+  return true;
+}
+
+void AddNCPUKernel::CheckParam(const CNodePtr &kernel_node) {
+  auto input_shape = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
+  if (input_shape.size() > 4) {
+    MS_LOG(EXCEPTION) << "Input dims is " << input_shape.size() << ", but AddNCPUKernel olny support 4d or lower.";
+  }
+
+  size_t output_num = AnfAlgo::GetOutputTensorNum(kernel_node);
+  if (output_num != 1) {
+    MS_LOG(EXCEPTION) << "Output number is " << output_num << ", but AddNCPUKernel needs 1 output.";
+  }
+}
+}  // namespace kernel
+}  // namespace mindspore

mindspore/ccsrc/kernel/cpu/addn_cpu_kernel.h

@@ -0,0 +1,56 @@
+/**
+ * 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_CPU_ADDN_CPU_KERNEL_H_
+#define MINDSPORE_CCSRC_KERNEL_CPU_ADDN_CPU_KERNEL_H_
+#include <vector>
+#include <memory>
+#include "kernel/cpu/cpu_kernel.h"
+#include "kernel/cpu/cpu_kernel_factory.h"
+
+namespace mindspore {
+namespace kernel {
+class AddNCPUKernel : public CPUKernel {
+ public:
+  AddNCPUKernel() : input_num_(0) {}
+  ~AddNCPUKernel() 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;
+
+ private:
+  void CheckParam(const CNodePtr &kernel_node);
+  size_t input_num_;
+  std::vector<size_t> output_shape_;
+};
+
+MS_REG_CPU_KERNEL(
+  AddN,
+  KernelAttr().AddInputAttr(kNumberTypeFloat32).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
+  AddNCPUKernel);
+MS_REG_CPU_KERNEL(AddN,
+                  KernelAttr()
+                    .AddInputAttr(kNumberTypeFloat32)
+                    .AddInputAttr(kNumberTypeFloat32)
+                    .AddInputAttr(kNumberTypeFloat32)
+                    .AddOutputAttr(kNumberTypeFloat32),
+                  AddNCPUKernel);
+}  // namespace kernel
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_KERNEL_CPU_ADDN_CPU_KERNEL_H_

mindspore/ccsrc/kernel/cpu/cpu_kernel_factory.cc

@@ -42,7 +42,7 @@ std::shared_ptr<CPUKernel> CPUKernelFactory::Create(const std::string &kernel_na
   MS_EXCEPTION_IF_NULL(kernel_info);
   const KernelBuildInfo *kernel_build_Info = kernel_info->select_kernel_build_info();
   MS_EXCEPTION_IF_NULL(kernel_build_Info);
-  std::pair<bool, size_t> ret_pair = CPUKernelAttrCheck(kernel_name, kernel_build_Info);
+  std::pair<bool, size_t> ret_pair = CPUKernelAttrCheck(kernel_name, *kernel_build_Info);
   if (ret_pair.first) {
     return (name_to_attr_creator_.find(kernel_name)->second)[ret_pair.second].second();
   }
@@ -50,7 +50,7 @@ std::shared_ptr<CPUKernel> CPUKernelFactory::Create(const std::string &kernel_na
 }
 
 std::pair<bool, size_t> CPUKernelFactory::CPUKernelAttrCheck(const std::string &kernel_name,
-                                                             const KernelBuildInfo *kernel_info) {
+                                                             const KernelBuildInfo &kernel_info) {
   auto iter = name_to_attr_creator_.find(kernel_name);
   if (iter == name_to_attr_creator_.end()) {
     MS_LOG(INFO) << "Not registered CPU kernel: op[" << kernel_name << "]!";
@@ -59,27 +59,34 @@ std::pair<bool, size_t> CPUKernelFactory::CPUKernelAttrCheck(const std::string &
   auto creators = iter->second;
   for (size_t index = 0; index < creators.size(); ++index) {
     auto attr_creator = creators[index];
-    for (size_t i = 0; i < kernel_info->GetInputNum(); ++i) {
-      if (kernel_info->GetInputDeviceType(i) != attr_creator.first.GetInputAttr(i).first) {
-        MS_LOG(WARNING) << "cpu kernel attr check failed. input index: " << i << ".";
-        MS_LOG(WARNING) << "kernel info type:" << kernel_info->GetInputDeviceType(i) << ", "
-                        << "register type:" << attr_creator.first.GetInputAttr(i).first;
-        return std::make_pair(false, 0);
-      }
+    if (CPUKernelSingleAttrCheck(attr_creator, kernel_info)) {
+      return std::make_pair(true, index);
     }
-    for (size_t i = 0; i < kernel_info->GetOutputNum(); ++i) {
-      if (kernel_info->GetOutputDeviceType(i) != attr_creator.first.GetOutputAttr(i).first) {
-        MS_LOG(WARNING) << "cpu kernel attr check failed. output index: " << i << ".";
-        MS_LOG(WARNING) << "kernel info type:" << kernel_info->GetOutputDeviceType(i) << ", "
-                        << "register type:" << attr_creator.first.GetOutputAttr(i).first;
-        return std::make_pair(false, 0);
-      }
-    }
-    return std::make_pair(true, index);
   }
   return std::make_pair(false, 0);
 }
 
+bool CPUKernelFactory::CPUKernelSingleAttrCheck(const std::pair<KernelAttr, CPUKernelCreator> &attr_creator,
+                                                const KernelBuildInfo &kernel_info) {
+  for (size_t i = 0; i < kernel_info.GetInputNum(); ++i) {
+    if (kernel_info.GetInputDeviceType(i) != attr_creator.first.GetInputAttr(i).first) {
+      MS_LOG(DEBUG) << "cpu kernel attr check failed. input index: " << i << ".";
+      MS_LOG(DEBUG) << "kernel info type:" << kernel_info.GetInputDeviceType(i) << ", "
+                    << "register type:" << attr_creator.first.GetInputAttr(i).first;
+      return false;
+    }
+  }
+  for (size_t i = 0; i < kernel_info.GetOutputNum(); ++i) {
+    if (kernel_info.GetOutputDeviceType(i) != attr_creator.first.GetOutputAttr(i).first) {
+      MS_LOG(DEBUG) << "cpu kernel attr check failed. output index: " << i << ".";
+      MS_LOG(DEBUG) << "kernel info type:" << kernel_info.GetOutputDeviceType(i) << ", "
+                    << "register type:" << attr_creator.first.GetOutputAttr(i).first;
+      return false;
+    }
+  }
+  return true;
+}
+
 std::vector<KernelAttr> CPUKernelFactory::GetSupportedKernelAttrList(const std::string &kernel_name) {
   std::vector<KernelAttr> result;
   auto iter = name_to_attr_creator_.find(kernel_name);

mindspore/ccsrc/kernel/cpu/cpu_kernel_factory.h

@@ -43,7 +43,9 @@ class CPUKernelFactory {
   CPUKernelFactory() = default;
   ~CPUKernelFactory() = default;
   DISABLE_COPY_AND_ASSIGN(CPUKernelFactory)
-  std::pair<bool, size_t> CPUKernelAttrCheck(const std::string &kernel_name, const KernelBuildInfo *kernel_info);
+  std::pair<bool, size_t> CPUKernelAttrCheck(const std::string &kernel_name, const KernelBuildInfo &kernel_info);
+  bool CPUKernelSingleAttrCheck(const std::pair<KernelAttr, CPUKernelCreator> &attr_creator,
+                                const KernelBuildInfo &kernel_info);
   std::map<std::string, std::vector<std::pair<KernelAttr, CPUKernelCreator>>> name_to_attr_creator_;
 };
 

tests/st/ops/cpu/test_addn_op.py

@@ -0,0 +1,78 @@
+# 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
+from mindspore import Tensor
+from mindspore.common import dtype as mstype
+from mindspore.ops import operations as P
+
+context.set_context(mode=context.GRAPH_MODE, device_target='CPU')
+
+class Net2I(nn.Cell):
+    def __init__(self):
+        super(Net2I, self).__init__()
+        self.addn = P.AddN()
+
+    def construct(self, x, y):
+        return self.addn((x, y))
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_cpu
+@pytest.mark.env_onecard
+def test_net_2Input():
+    x = np.arange(2 * 3 * 2).reshape(2, 3, 2).astype(np.float32)
+    y = np.arange(2 * 3 * 2).reshape(2, 3, 2).astype(np.float32)
+    addn = Net2I()
+    output = addn(Tensor(x, mstype.float32), Tensor(y, mstype.float32))
+    print("output:\n", output)
+    expect_result = [[[0., 2.],
+                      [4., 6.],
+                      [8., 10.]],
+                     [[12., 14.],
+                      [16., 18.],
+                      [20., 22.]]]
+
+    assert (output.asnumpy() == expect_result).all()
+
+class Net3I(nn.Cell):
+    def __init__(self):
+        super(Net3I, self).__init__()
+        self.addn = P.AddN()
+
+    def construct(self, x, y, z):
+        return self.addn((x, y, z))
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_cpu
+@pytest.mark.env_onecard
+def test_net_3Input():
+    x = np.arange(2 * 3).reshape(2, 3).astype(np.float32)
+    y = np.arange(2 * 3).reshape(2, 3).astype(np.float32)
+    z = np.arange(2 * 3).reshape(2, 3).astype(np.float32)
+    addn = Net3I()
+    output = addn(Tensor(x, mstype.float32), Tensor(y, mstype.float32), Tensor(z, mstype.float32))
+    print("output:\n", output)
+    expect_result = [[0., 3.,  6.],
+                     [9., 12., 15]]
+
+    assert (output.asnumpy() == expect_result).all()
+
+if __name__ == '__main__':
+    test_net_2Input()
+    test_net_3Input()

tests/st/ops/cpu/test_slice_op.py

@@ -1,4 +1,4 @@
-# Copyright 2019 Huawei Technologies Co., Ltd
+# 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.