[NPU] Support npu kernel for reduceany op (#31422)

* add reduce any npu op

* add gather python unittest

* update c_plus unittest

* update python unittest

* del c++ unittest

* update c++ unittest

* update c++ unittest

paddle/fluid/operators/reduce_ops/CMakeLists.txt

@@ -38,3 +38,7 @@ if(WITH_GPU)
 	nv_test(check_reduce_rank_test SRCS check_reduce_rank_test.cu DEPS tensor)
     endif()
 endif()
+
+if(WITH_ASCEND_CL)
+    cc_test(reduce_any_op_npu_test SRCS reduce_any_op_npu_test.cc DEPS op_registry reduce_any_op scope device_context enforce executor)
+endif()

paddle/fluid/operators/reduce_ops/reduce_any_op_npu.cc

@@ -0,0 +1,55 @@
+/* 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 <memory>
+#include <string>
+
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/framework/tensor_util.h"
+#include "paddle/fluid/operators/npu_op_runner.h"
+
+namespace paddle {
+namespace operators {
+using Tensor = framework::Tensor;
+
+template <typename T>
+class ReduceAnyNPUKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    const Tensor* x = ctx.Input<Tensor>("X");
+    auto* out = ctx.Output<Tensor>("Out");
+
+    bool keep_dim = ctx.Attr<bool>("keep_dim");
+    auto dims = ctx.Attr<std::vector<int>>("dim");
+
+    out->mutable_data<T>(ctx.GetPlace());
+
+    // set attr
+    NPUAttributeMap attr = {{"keep_dims", keep_dim}, {"axes", dims}};
+
+    auto runner = NpuOpRunner("ReduceAnyD", {*x}, {*out}, attr);
+    auto stream =
+        ctx.template device_context<paddle::platform::NPUDeviceContext>()
+            .stream();
+    runner.Run(stream);
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+namespace plat = paddle::platform;
+
+REGISTER_OP_NPU_KERNEL(reduce_any, ops::ReduceAnyNPUKernel<bool>);

paddle/fluid/operators/reduce_ops/reduce_any_op_npu_test.cc

@@ -0,0 +1,83 @@
+/* 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. */
+
+#ifndef _WIN32
+#include <unistd.h>
+#endif
+
+#include <memory>
+#include <string>
+#include <thread>  // NOLINT
+#include <vector>
+
+#include "gtest/gtest.h"
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/framework/operator.h"
+#include "paddle/fluid/framework/program_desc.h"
+#include "paddle/fluid/memory/malloc.h"
+#include "paddle/fluid/memory/memcpy.h"
+#include "paddle/fluid/operators/math/math_function.h"
+#include "paddle/fluid/string/printf.h"
+
+namespace f = paddle::framework;
+namespace p = paddle::platform;
+namespace m = paddle::operators::math;
+
+using Tensor = paddle::framework::Tensor;
+
+USE_OP(reduce_any);
+USE_OP_DEVICE_KERNEL(reduce_any, NPU);
+
+template <typename T>
+void Compare(f::Scope* scope, const p::DeviceContext& ctx) {
+  // init
+  auto x = scope->Var("X");
+  auto tensor_x = x->GetMutable<f::LoDTensor>();
+  std::vector<bool> init_x = {true, false, false, false};
+  f::TensorFromVector<bool>(init_x, ctx, tensor_x);
+  tensor_x->Resize(paddle::framework::make_ddim({2}));
+
+  ctx.Wait();
+
+  auto place = ctx.GetPlace();
+  auto out = scope->Var("Out");
+  auto tensor_out = out->GetMutable<f::LoDTensor>();
+
+  // run
+  std::vector<int> axes;
+  f::AttributeMap attrs = {{"axes", axes}, {"keep_dims", true}};
+  auto op = f::OpRegistry::CreateOp("reduce_any", {{"X", {"X"}}},
+                                    {{"Out", {"Out"}}}, attrs);
+
+  op->Run(*scope, place);
+
+  ctx.Wait();
+
+  std::vector<bool> out_vec;
+  f::TensorToVector<bool>(*tensor_out, ctx, &out_vec);
+
+  ctx.Wait();
+
+  std::vector<bool> expected_vec = {true};
+  EXPECT_EQ(out_vec.size(), expected_vec.size());
+  for (uint32_t i = 0; i < out_vec.size(); i++) {
+    EXPECT_EQ(out_vec[i], expected_vec[i]);
+  }
+}
+
+TEST(reduce_any, NPU) {
+  f::Scope scope;
+  p::NPUDeviceContext ctx(p::NPUPlace(0));
+  Compare<bool>(&scope, ctx);
+}

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

@@ -0,0 +1,133 @@
+#   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 unittest
+import numpy as np
+from op_test import OpTest, skip_check_grad_ci
+import paddle
+import paddle.fluid.core as core
+import paddle.fluid as fluid
+from paddle.fluid import compiler, Program, program_guard
+from paddle.fluid.framework import convert_np_dtype_to_dtype_
+
+paddle.enable_static()
+
+
+@unittest.skipIf(not paddle.is_compiled_with_npu(),
+                 "core is not compiled with NPU")
+class TestAny8DOp(OpTest):
+    def setUp(self):
+        self.set_npu()
+        self.op_type = "reduce_any"
+        self.place = paddle.NPUPlace(0)
+        self.inputs = {
+            'X': np.random.randint(0, 2,
+                                   (2, 5, 3, 2, 2, 3, 4, 2)).astype("bool")
+        }
+        self.attrs = {'dim': (3, 5, 4)}
+        self.outputs = {'Out': self.inputs['X'].any(axis=self.attrs['dim'])}
+
+    def set_npu(self):
+        self.__class__.use_npu = True
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place, check_dygraph=False)
+
+
+@unittest.skipIf(not paddle.is_compiled_with_npu(),
+                 "core is not compiled with NPU")
+class TestAnyOpWithDim(OpTest):
+    def setUp(self):
+        self.set_npu()
+        self.op_type = "reduce_any"
+        self.place = paddle.NPUPlace(0)
+        self.inputs = {'X': np.random.randint(0, 2, (5, 6, 10)).astype("bool")}
+        self.attrs = {'dim': [1]}
+        self.outputs = {'Out': self.inputs['X'].any(axis=1)}
+
+    def set_npu(self):
+        self.__class__.use_npu = True
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place, check_dygraph=False)
+
+
+@unittest.skipIf(not paddle.is_compiled_with_npu(),
+                 "core is not compiled with NPU")
+class TestAny8DOpWithDim(OpTest):
+    def setUp(self):
+        self.set_npu()
+        self.op_type = "reduce_any"
+        self.place = paddle.NPUPlace(0)
+        self.inputs = {
+            'X': np.random.randint(0, 2,
+                                   (2, 5, 3, 2, 2, 3, 4, 2)).astype("bool")
+        }
+        self.attrs = {'dim': (3, 6)}
+        self.outputs = {'Out': self.inputs['X'].any(axis=self.attrs['dim'])}
+
+    def set_npu(self):
+        self.__class__.use_npu = True
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place, check_dygraph=False)
+
+
+@unittest.skipIf(not paddle.is_compiled_with_npu(),
+                 "core is not compiled with NPU")
+class TestAnyOpWithKeepDim(OpTest):
+    def setUp(self):
+        self.set_npu()
+        self.op_type = "reduce_any"
+        self.place = paddle.NPUPlace(0)
+        self.inputs = {'X': np.random.randint(0, 2, (5, 6, 10)).astype("bool")}
+        self.attrs = {'dim': (1, ), 'keep_dim': True}
+        self.outputs = {
+            'Out': np.expand_dims(
+                self.inputs['X'].any(axis=self.attrs['dim']), axis=1)
+        }
+
+    def set_npu(self):
+        self.__class__.use_npu = True
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place, check_dygraph=False)
+
+
+class TestAny8DOpWithKeepDim(OpTest):
+    def setUp(self):
+        self.set_npu()
+        self.op_type = "reduce_any"
+        self.place = paddle.NPUPlace(0)
+        self.inputs = {
+            'X': np.random.randint(0, 2,
+                                   (2, 5, 3, 2, 2, 3, 4, 2)).astype("bool")
+        }
+        self.attrs = {'dim': (1, ), 'keep_dim': True}
+        self.outputs = {
+            'Out': np.expand_dims(
+                self.inputs['X'].any(axis=self.attrs['dim']), axis=1)
+        }
+
+    def set_npu(self):
+        self.__class__.use_npu = True
+
+    def test_check_output(self):
+        self.check_output_with_place(self.place, check_dygraph=False)
+
+
+if __name__ == '__main__':
+    unittest.main()