[NPU] Support op kernel for Fill constant batch size like op (#34721)

* fix npu compile error, test=develop

* add fill constant batch size lilke op npu,test=develop
Co-authored-by: Nqili93 <qili93@qq.com>

paddle/fluid/operators/fill_constant_batch_size_like_op_npu.cc

+/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+
+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 "paddle/fluid/operators/fill_constant_op.h"
+#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/operators/utils.h"
+
+namespace paddle {
+namespace operators {
+
+using Tensor = framework::Tensor;
+
+template <typename DeviceContext, typename T>
+class FillConstantBatchSizeLikeOpNPUKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext &ctx) const override {
+    auto data_type =
+        static_cast<framework::proto::VarType::Type>(ctx.Attr<int>("dtype"));
+    auto float_value = ctx.Attr<float>("value");
+    auto str_value = ctx.Attr<std::string>("str_value");
+    auto force_cpu = ctx.Attr<bool>("force_cpu");
+
+    auto *out = ctx.Output<Tensor>("Out");
+    auto *input = ctx.Input<Tensor>("Input");
+    if (&ctx.Attr<int>("input_dim_idx") == 0) {
+      // set the correct batch size.
+      auto odims = out->dims();
+      int input_dim_idx = ctx.Attr<int>("input_dim_idx");
+      int output_dim_idx = ctx.Attr<int>("output_dim_idx");
+      odims[output_dim_idx] = input->dims()[input_dim_idx];
+      out->mutable_data<T>(odims, ctx.GetPlace());
+    }
+
+    T value;
+    if (str_value.empty()) {
+      value = static_cast<T>(float_value);
+    } else {
+      std::stringstream convert_stream(str_value);
+      if (std::is_same<int64_t, T>::value) {
+        int64_t tmp_value;
+        convert_stream >> tmp_value;
+        value = static_cast<T>(tmp_value);
+      } else {
+        double tmp_value;
+        convert_stream >> tmp_value;
+        value = static_cast<T>(tmp_value);
+      }
+    }
+
+    platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
+    auto &dev_ctx = *pool.Get(ctx.GetPlace());
+    bool cpu_place = force_cpu || ctx.GetPlace() == platform::CPUPlace();
+    if (cpu_place) {
+      math::SetConstant<platform::CPUDeviceContext, T> functor;
+      out->mutable_data(platform::CPUPlace(), data_type);
+      functor(reinterpret_cast<const platform::CPUDeviceContext &>(dev_ctx),
+              out, static_cast<T>(value));
+    } else {
+      out->mutable_data(ctx.GetPlace(), data_type);
+      Tensor tensor_tmp(data_type);
+      tensor_tmp.mutable_data<T>({1}, ctx.GetPlace());
+      FillNpuTensorWithConstant<T>(&tensor_tmp, value);
+
+      auto stream =
+          ctx.template device_context<paddle::platform::NPUDeviceContext>()
+              .stream();
+      const auto &runner =
+          NpuOpRunner("FillD", {tensor_tmp}, {*out},
+                      {{"dims", framework::vectorize(out->dims())}});
+      runner.Run(stream);
+    }
+  }
+};
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+
+REGISTER_OP_NPU_KERNEL(
+    fill_constant_batch_size_like,
+    ops::FillConstantBatchSizeLikeOpNPUKernel<
+        paddle::platform::NPUDeviceContext, float>,
+    ops::FillConstantBatchSizeLikeOpNPUKernel<
+        paddle::platform::NPUDeviceContext, int>,
+    ops::FillConstantBatchSizeLikeOpNPUKernel<
+        paddle::platform::NPUDeviceContext, paddle::platform::float16>);

python/paddle/fluid/tests/unittests/npu/test_fill_constant_batch_size_like_op_npu.py

+#  Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+from __future__ import print_function
+
+import numpy as np
+import unittest
+import sys
+sys.path.append("..")
+from op_test import OpTest
+import paddle
+import paddle.fluid as fluid
+from paddle.fluid import core
+
+paddle.enable_static()
+SEED = 2021
+
+
+class TestFillConstantBatchSizeLike(OpTest):
+    def setUp(self):
+        self.set_npu()
+        self.place = paddle.NPUPlace(0)
+        self.op_type = "fill_constant_batch_size_like"
+        self.init_shape()
+        self.init_value()
+        self.init_dtype()
+        self.init_force_cpu()
+        self.init_dim_idx()
+
+        self.inputs = {
+            'Input': np.random.random(self.input_shape).astype("float32")
+        }
+        self.attrs = {
+            'shape': self.shape,
+            'value': self.value,
+            'str_value': self.str_value,
+            'dtype': self.dtype,
+            'force_cpu': self.force_cpu,
+            'input_dim_idx': self.input_dim_idx,
+            'output_dim_idx': self.output_dim_idx
+        }
+        self.outputs = {
+            'Out': np.full(self.output_shape, self.output_value,
+                           self.output_dtype)
+        }
+
+    def set_npu(self):
+        self.__class__.use_npu = True
+
+    def init_shape(self):
+        self.input_shape = [4, 5]
+        self.shape = [123, 92]
+        self.output_shape = (4, 92)
+
+    def init_value(self):
+        self.value = 3.8
+        self.str_value = ''
+        self.output_value = 3.8
+
+    def init_dtype(self):
+        self.dtype = core.VarDesc.VarType.FP32
+        self.output_dtype = np.float32
+
+    def init_force_cpu(self):
+        self.force_cpu = False
+
+    def init_dim_idx(self):
+        self.input_dim_idx = 0
+        self.output_dim_idx = 0
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place)
+
+
+class TestFillConstantBatchSizeLike2(TestFillConstantBatchSizeLike):
+    def init_shape(self):
+        # test shape
+        self.input_shape = [4, 5, 6, 7]
+        self.shape = [10, 123, 92]
+        self.output_shape = (4, 123, 92)
+
+
+class TestFillConstantBatchSizeLike3(TestFillConstantBatchSizeLike):
+    def init_value(self):
+        # use 'str_value' rather than 'value'
+        self.value = 3.8
+        self.str_value = '4.5'
+        self.output_value = 4.5
+
+
+class TestFillConstantBatchSizeLike6(TestFillConstantBatchSizeLike):
+    def init_dtype(self):
+        self.dtype = core.VarDesc.VarType.FP16
+        self.output_dtype = np.float16
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place, atol=1e-2)
+
+
+class TestFillConstantBatchSizeLike7(TestFillConstantBatchSizeLike):
+    def init_dtype(self):
+        self.dtype = core.VarDesc.VarType.INT32
+        self.output_dtype = np.int32
+
+
+class TestFillConstantBatchSizeLike8(TestFillConstantBatchSizeLike):
+    def init_force_cpu(self):
+        self.force_cpu = True
+
+
+class TestFillConstantBatchSizeLike9(TestFillConstantBatchSizeLike):
+    def init_shape(self):
+        self.input_shape = [4, 5]
+        self.shape = [123, 92]
+        self.output_shape = (123, 4)
+
+    def init_dim_idx(self):
+        self.input_dim_idx = 0
+        self.output_dim_idx = 1
+
+
+if __name__ == '__main__':
+    unittest.main()