add fill/fill_any for kunlun (#49645)

paddle/phi/backends/xpu/xpu2_op_list.cc

@@ -214,6 +214,18 @@ XPUOpMap& get_kl2_ops() {
                      phi::DataType::BOOL,
                      phi::DataType::FLOAT16,
                      phi::DataType::FLOAT32})},
+      {"fill",
+       XPUKernelSet({phi::DataType::INT64,
+                     phi::DataType::INT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::FLOAT64,
+                     phi::DataType::FLOAT32})},
+      {"fill_any",
+       XPUKernelSet({phi::DataType::INT64,
+                     phi::DataType::INT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::FLOAT64,
+                     phi::DataType::FLOAT32})},
       {"fill_any_like",
        XPUKernelSet({phi::DataType::INT64,
                      phi::DataType::INT32,

paddle/phi/kernels/xpu/fill_kernel.cc

+// Copyright (c) 2022 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/phi/kernels/impl/fill_kernel_impl.h"
+
+#include "paddle/phi/backends/cpu/cpu_context.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+PD_REGISTER_KERNEL(fill,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::FillKernel,
+                   float,
+                   double,
+                   int64_t,
+                   int,
+                   phi::dtype::float16,
+                   bool) {}

python/paddle/fluid/tests/unittests/xpu/test_fill_any_op_xpu.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.
+
+import sys
+
+sys.path.append("..")
+import unittest
+
+import numpy as np
+from op_test_xpu import XPUOpTest
+from xpu.get_test_cover_info import (
+    XPUOpTestWrapper,
+    create_test_class,
+    get_xpu_op_support_types,
+)
+
+import paddle
+
+paddle.enable_static()
+
+
+class XPUTestFillAnyOp(XPUOpTestWrapper):
+    def __init__(self):
+        self.op_name = 'fill_any'
+        self.use_dynamic_create_class = False
+
+    class TestFillAnyOp(XPUOpTest):
+        def setUp(self):
+            self.op_type = "fill_any"
+            self.dtype = 'float64'
+            self.value = 0.0
+            self.init()
+            self.inputs = {'X': np.random.random((20, 30)).astype(self.dtype)}
+            self.attrs = {
+                'value_float': float(self.value),
+                'value_int': int(self.value),
+            }
+            self.outputs = {
+                'Out': self.value
+                * np.ones_like(self.inputs["X"]).astype(self.dtype)
+            }
+
+        def init(self):
+            pass
+
+        def test_check_output(self):
+            self.check_output_with_place(paddle.XPUPlace(0))
+
+        def test_check_grad(self):
+            self.check_grad_with_place(paddle.XPUPlace(0), ['X'], 'Out')
+
+    class TestFillAnyOpFloat32(TestFillAnyOp):
+        def init(self):
+            self.dtype = np.float32
+            self.value = 0.0
+
+    class TestFillAnyOpFloat16(TestFillAnyOp):
+        def init(self):
+            self.dtype = np.float16
+
+    class TestFillAnyOpvalue1(TestFillAnyOp):
+        def init(self):
+            self.dtype = np.float32
+            self.value = 111111555
+
+    class TestFillAnyOpvalue2(TestFillAnyOp):
+        def init(self):
+            self.dtype = np.float32
+            self.value = 11111.1111
+
+    class TestFillAnyInplace(unittest.TestCase):
+        def test_fill_any_version(self):
+            with paddle.fluid.dygraph.guard():
+                var = paddle.to_tensor(np.ones((4, 2, 3)).astype(np.float32))
+                self.assertEqual(var.inplace_version, 0)
+
+                var.fill_(0)
+                self.assertEqual(var.inplace_version, 1)
+
+                var.fill_(0)
+                self.assertEqual(var.inplace_version, 2)
+
+                var.fill_(0)
+                self.assertEqual(var.inplace_version, 3)
+
+        def test_fill_any_eqaul(self):
+            with paddle.fluid.dygraph.guard():
+                tensor = paddle.to_tensor(
+                    np.random.random((20, 30)).astype(np.float32)
+                )
+                target = tensor.numpy()
+                target[...] = 1
+
+                tensor.fill_(1)
+                self.assertEqual((tensor.numpy() == target).all().item(), True)
+
+        def test_backward(self):
+            with paddle.fluid.dygraph.guard():
+                x = paddle.full([10, 10], -1.0, dtype='float32')
+                x.stop_gradient = False
+                y = 2 * x
+                y.fill_(1)
+                y.backward()
+                np.testing.assert_array_equal(
+                    x.grad.numpy(), np.zeros([10, 10])
+                )
+
+
+support_types = get_xpu_op_support_types('fill_any')
+for stype in support_types:
+    create_test_class(globals(), XPUTestFillAnyOp, stype)
+
+if __name__ == "__main__":
+    unittest.main()

python/paddle/fluid/tests/unittests/xpu/test_fill_op_xpu.py

+#   Copyright (c) 2019 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.
+
+import sys
+
+sys.path.append("..")
+import unittest
+
+import numpy as np
+from op_test_xpu import XPUOpTest
+from xpu.get_test_cover_info import (
+    XPUOpTestWrapper,
+    create_test_class,
+    get_xpu_op_support_types,
+)
+
+import paddle
+import paddle.fluid.core as core
+from paddle.fluid.op import Operator
+
+paddle.enable_static()
+
+
+class XPUTestFillOp(XPUOpTestWrapper):
+    def __init__(self):
+        self.op_name = 'fill'
+        self.use_dynamic_create_class = False
+
+    class TestFillOp1(XPUOpTest):
+        def setUp(self):
+            self.op_type = "fill"
+            val = np.random.random(size=[100, 200])
+            self.inputs = {}
+            self.attrs = {
+                'value': val.flatten().tolist(),
+                'shape': [100, 200],
+                'dtype': int(core.VarDesc.VarType.FP64),
+                'force_cpu': False,
+            }
+            self.outputs = {'Out': val.astype('float64')}
+
+        def test_check_output(self):
+            self.check_output_with_place(paddle.XPUPlace(0))
+
+    class TestFillOp2(XPUOpTest):
+        def setUp(self):
+            self.op_type = "fill"
+            val = np.random.random(size=[100, 200])
+            self.inputs = {}
+            self.attrs = {
+                'value': val.flatten().tolist(),
+                'shape': [100, 200],
+                'dtype': int(core.VarDesc.VarType.FP64),
+                'force_cpu': True,
+            }
+            self.outputs = {'Out': val.astype('float64')}
+
+        def test_check_output(self):
+            self.check_output()
+
+    class TestFillOp3(unittest.TestCase):
+        def check_with_place(self, place, f_cpu):
+            scope = core.Scope()
+            # create Out Variable
+            out = scope.var('Out').get_tensor()
+
+            # create and run fill_op operator
+            val = np.random.random(size=[300, 200])
+            fill_op = Operator(
+                "fill",
+                value=val.flatten(),
+                shape=[300, 200],
+                dtype=int(core.VarDesc.VarType.FP32),
+                force_cpu=f_cpu,
+                Out='Out',
+            )
+            fill_op.run(scope, place)
+
+            # get result from Out
+            result_array = np.array(out)
+            full_array = np.array(val, 'float32')
+
+            np.testing.assert_array_equal(result_array, full_array)
+
+        def test_fill_op(self):
+            places = [core.CPUPlace()]
+            if core.is_compiled_with_xpu():
+                places.append(core.XPUPlace(0))
+
+            for place in places:
+                self.check_with_place(place, True)
+                self.check_with_place(place, False)
+
+
+support_types = get_xpu_op_support_types('fill')
+for stype in support_types:
+    create_test_class(globals(), XPUTestFillOp, stype)
+
+if __name__ == '__main__':
+    unittest.main()