add clip_by_norm on kunlun, *test=kunlun (#30862)

cmake/external/xpu.cmake

@@ -11,7 +11,7 @@ if(NOT XPU_SDK_ROOT)
   elseif(WITH_SUNWAY)
       SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/sunway/xpu_2021_01_13.tar.gz" CACHE STRING "" FORCE)
   else()
-      SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2021_01_13.tar.gz" CACHE STRING "" FORCE)
+      SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2021_02_03.tar.gz" CACHE STRING "" FORCE)
   endif()
 
   SET(XPU_SOURCE_DIR              "${THIRD_PARTY_PATH}/xpu")

paddle/fluid/operators/clip_by_norm_op_xpu.cc

@@ -0,0 +1,74 @@
+/* Copyright (c) 2016 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. */
+
+#ifdef PADDLE_WITH_XPU
+#include "paddle/fluid/operators/clip_by_norm_op.h"
+#include <vector>
+
+namespace paddle {
+namespace operators {
+
+template <typename DeviceContext, typename T>
+class XPUClipByNormKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& context) const override {
+    auto max_norm = context.Attr<T>("max_norm");
+    auto in_var = context.InputVar("X");
+
+    Tensor* output = nullptr;
+    const Tensor* input = nullptr;
+    if (in_var->IsType<framework::LoDTensor>()) {
+      input = context.Input<Tensor>("X");
+
+      output = context.Output<Tensor>("Out");
+      output->mutable_data<T>(context.GetPlace());
+    } else {
+      PADDLE_THROW(platform::errors::InvalidArgument(
+          "Invalid input variable type, only support LodTensor"
+          "type, but got type is %s.",
+          framework::ToTypeName(in_var->Type())));
+    }
+
+    PADDLE_ENFORCE_NOT_NULL(input,
+                            platform::errors::InvalidArgument(
+                                "Input(X) of ClipByNormOp should not be null. "
+                                "Please check if it is created correctly."));
+    auto& dev_ctx = context.template device_context<DeviceContext>();
+    const auto& x_dims = input->dims();
+    std::vector<int> xshape(x_dims.size());
+    std::vector<int> rdims(x_dims.size());
+    for (int i = 0; i < x_dims.size(); i++) {
+      xshape[i] = x_dims[i];
+      rdims[i] = i;
+    }
+    int r = xpu::clip_by_norm<T>(dev_ctx.x_context(), input->data<T>(),
+                                 output->data<T>(), max_norm, xshape, rdims);
+    PADDLE_ENFORCE_EQ(
+        r, XPU_SUCCESS,
+        platform::errors::External("XPU API(clip_by_norm) return "
+                                   "wrong value[%d], please check whether "
+                                   "Baidu Kunlun Card is properly installed.",
+                                   r));
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP_XPU_KERNEL(
+    clip_by_norm,
+    ops::XPUClipByNormKernel<paddle::platform::XPUDeviceContext, float>);
+
+#endif  // PADDLE_WITH_XPU

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

@@ -0,0 +1,77 @@
+#   Copyright (c) 2020 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 sys
+sys.path.append("..")
+import unittest
+import numpy as np
+import paddle.fluid.core as core
+import paddle.fluid as fluid
+from op_test_xpu import OpTest, XPUOpTest
+import paddle
+from paddle.fluid import Program, program_guard
+
+
+class TestXPUClipByNormOp(XPUOpTest):
+    def setUp(self):
+        self.op_type = "clip_by_norm"
+        self.dtype = np.float32
+        self.use_xpu = True
+        self.max_relative_error = 0.006
+        self.initTestCase()
+        input = np.random.random(self.shape).astype("float32")
+        input[np.abs(input) < self.max_relative_error] = 0.5
+        self.inputs = {'X': input, }
+        self.attrs = {}
+        self.attrs['max_norm'] = self.max_norm
+        norm = np.sqrt(np.sum(np.square(input)))
+        if norm > self.max_norm:
+            output = self.max_norm * input / norm
+        else:
+            output = input
+        self.outputs = {'Out': output}
+
+    def test_check_output(self):
+        if paddle.is_compiled_with_xpu():
+            paddle.enable_static()
+            place = paddle.XPUPlace(0)
+            self.check_output_with_place(place)
+
+    def initTestCase(self):
+        self.shape = (100, )
+        self.max_norm = 1.0
+
+
+class TestCase1(TestXPUClipByNormOp):
+    def initTestCase(self):
+        self.shape = (100, )
+        self.max_norm = 1e20
+
+
+class TestCase2(TestXPUClipByNormOp):
+    def initTestCase(self):
+        self.shape = (16, 16)
+        self.max_norm = 0.1
+
+
+class TestCase3(TestXPUClipByNormOp):
+    def initTestCase(self):
+        self.shape = (4, 8, 16)
+        self.max_norm = 1.0
+
+
+if __name__ == "__main__":
+    unittest.main()