support bce_loss and bce_loss_grad in XPU, test=kunlun (#41610)

paddle/fluid/operators/bce_loss_op_xpu.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. */
+
+#ifdef PADDLE_WITH_XPU
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/platform/device/device_wrapper.h"
+
+namespace paddle {
+namespace operators {
+
+using Tensor = framework::Tensor;
+
+template <typename DeviceContext, typename T>
+class XPUBCELossKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& context) const override {
+    auto* x = context.Input<Tensor>("X");
+    auto* labels = context.Input<Tensor>("Label");
+    auto* out = context.Output<Tensor>("Out");
+    out->mutable_data<T>(context.GetPlace());
+
+    auto x_numel = x->numel();
+    auto& dev_ctx = context.template device_context<DeviceContext>();
+    int r = xpu::bce_loss<T>(dev_ctx.x_context(), x->data<T>(),
+                             labels->data<T>(), out->data<T>(), x_numel);
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "bce_loss");
+  }
+};
+
+template <typename DeviceContext, typename T>
+class XPUBCELossGradKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& context) const override {
+    auto* x = context.Input<Tensor>("X");
+    auto* labels = context.Input<Tensor>("Label");
+    auto* dout = context.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = context.Output<Tensor>(framework::GradVarName("X"));
+    dx->mutable_data<T>(context.GetPlace());
+
+    auto x_numel = x->numel();
+    auto& dev_ctx = context.template device_context<DeviceContext>();
+    int r = xpu::bce_loss_grad<T>(dev_ctx.x_context(), x->data<T>(),
+                                  labels->data<T>(), dout->data<T>(),
+                                  dx->data<T>(), x_numel);
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "bce_loss_grad");
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP_XPU_KERNEL(
+    bce_loss, ops::XPUBCELossKernel<paddle::platform::XPUDeviceContext, float>);
+REGISTER_OP_XPU_KERNEL(
+    bce_loss_grad,
+    ops::XPUBCELossGradKernel<paddle::platform::XPUDeviceContext, float>);
+
+#endif  // PADDLE_WITH_XPU

paddle/fluid/platform/device/xpu/xpu2_op_list.h

@@ -43,6 +43,9 @@ XPUOpMap& get_kl2_ops() {
       {"batch_norm_grad",
        XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"batch_norm", XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
+      {"bce_loss_grad",
+       XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
+      {"bce_loss", XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"bilinear_interp_v2",
        XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
       {"bilinear_interp_v2_grad",

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

+# 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.
+
+from __future__ import print_function
+
+import sys
+sys.path.append("..")
+import paddle
+import paddle.fluid as fluid
+import numpy as np
+import unittest
+from op_test_xpu import XPUOpTest
+from xpu.get_test_cover_info import create_test_class, get_xpu_op_support_types, XPUOpTestWrapper
+
+paddle.enable_static()
+
+
+def bce_loss(input, label):
+    return -1 * (label * np.log(input) + (1. - label) * np.log(1. - input))
+
+
+class XPUTestBceLossOp(XPUOpTestWrapper):
+    def __init__(self):
+        self.op_name = 'bce_loss'
+        self.use_dynamic_create_class = False
+
+    class TestBceLossOp(XPUOpTest):
+        def setUp(self):
+            self.op_type = "bce_loss"
+            self.dtype = self.in_type
+            self.place = paddle.XPUPlace(0)
+            self.init_test_case()
+            input_np = np.random.uniform(0.1, 0.8,
+                                         self.shape).astype(self.dtype)
+            label_np = np.random.randint(0, 2, self.shape).astype(self.dtype)
+            output_np = bce_loss(input_np, label_np)
+
+            self.inputs = {'X': input_np, 'Label': label_np}
+            self.outputs = {'Out': output_np}
+
+        def test_check_output(self):
+            self.check_output_with_place(self.place)
+
+        def test_check_grad(self):
+            self.check_grad_with_place(self.place, ['X'], 'Out')
+
+        def init_test_case(self):
+            self.shape = [10, 10]
+
+    class TestBceLossOpCase1(TestBceLossOp):
+        def init_test_cast(self):
+            self.shape = [2, 3, 4, 5]
+
+    class TestBceLossOpCase2(TestBceLossOp):
+        def init_test_cast(self):
+            self.shape = [2, 3, 20]
+
+
+support_types = get_xpu_op_support_types('bce_loss')
+for stype in support_types:
+    create_test_class(globals(), XPUTestBceLossOp, stype)
+
+if __name__ == "__main__":
+    paddle.enable_static()
+    unittest.main()