[PHI] bind nll_loss xpu kernel (#54043)

cmake/external/xpu.cmake

@@ -8,7 +8,7 @@ set(XPU_API_LIB_NAME "libxpuapi.so")
 set(XPU_RT_LIB_NAME "libxpurt.so")
 set(XPU_XFT_LIB_NAME "libxft.so")
 
-set(XPU_BASE_DATE "20230519")
+set(XPU_BASE_DATE "20230523")
 set(XPU_XCCL_BASE_VERSION "1.0.49.2")
 set(XPU_XFT_BASE_VERSION "latest")
 

paddle/phi/backends/xpu/xpu2_op_list.cc

@@ -525,6 +525,8 @@ XPUOpMap& get_kl2_ops() {
                      phi::DataType::FLOAT16,
                      phi::DataType::INT64})},
       {"nearest_interp_v2_grad", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"nll_loss", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"nll_loss_grad", XPUKernelSet({phi::DataType::FLOAT32})},
       {"not_equal",
        XPUKernelSet({phi::DataType::INT64,
                      phi::DataType::INT32,

paddle/phi/kernels/xpu/nll_loss_grad_kernel.cc

+// Copyright (c) 2023 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/nll_loss_grad_kernel.h"
+#include "paddle/phi/backends/xpu/enforce_xpu.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+namespace phi {
+
+template <typename T, typename Context>
+void NllLossGradKernel(const Context& dev_ctx,
+                       const DenseTensor& x,
+                       const DenseTensor& label,
+                       const paddle::optional<DenseTensor>& weight,
+                       const DenseTensor& total_weight,
+                       const DenseTensor& d_out,
+                       int64_t ignore_index,
+                       const std::string& reduction,
+                       DenseTensor* d_x) {
+  using XPUType = typename XPUTypeTrait<T>::Type;
+  const auto& label_type = label.dtype();
+  bool label_type_match =
+      label_type == phi::DataType::INT32 || label_type == phi::DataType::INT64;
+  PADDLE_ENFORCE_EQ(label_type_match,
+                    true,
+                    phi::errors::InvalidArgument(
+                        "Input(Label) holds the wrong type, it holds %s, but "
+                        "desires to be %s or %s",
+                        label_type,
+                        phi::DataType::INT32,
+                        phi::DataType::INT64));
+
+  auto d_out_data = d_out.data<XPUType>();
+  auto d_x_data = dev_ctx.template Alloc<XPUType>(d_x);
+
+  auto d_x_dims = d_x->dims();
+  std::vector<int64_t> d_x_shape = phi::vectorize<int64_t>(d_x_dims);
+
+  auto weight_data =
+      weight.get_ptr() ? weight.get_ptr()->data<float>() : nullptr;
+
+  int64_t reduction_id = 0;
+  if (reduction == "none") {
+    reduction_id = 0;
+  } else if (reduction == "mean") {
+    reduction_id = 1;
+  } else if (reduction == "sum") {
+    reduction_id = 2;
+  }
+
+  auto total_weight_data = total_weight.data<XPUType>();
+
+  int r;
+  if (label_type == phi::DataType::INT32) {
+    const int* label_data = label.data<int>();
+    r = xpu::nll_loss_grad(dev_ctx.x_context(),
+                           d_out_data,
+                           d_x_data,
+                           d_x_shape,
+                           label_data,
+                           weight_data,
+                           reduction_id,
+                           ignore_index,
+                           total_weight_data);
+  } else if (label_type == phi::DataType::INT64) {
+    const int64_t* label_data = label.data<int64_t>();
+    r = xpu::nll_loss_grad(dev_ctx.x_context(),
+                           d_out_data,
+                           d_x_data,
+                           d_x_shape,
+                           label_data,
+                           weight_data,
+                           reduction_id,
+                           ignore_index,
+                           total_weight_data);
+  }
+  PADDLE_ENFORCE_XDNN_SUCCESS(r, "nll_loss_grad");
+}
+
+}  // namespace phi
+
+// TODO(xiongkun): add the non-raw kernel register here.
+PD_REGISTER_KERNEL(
+    nll_loss_grad, XPU, ALL_LAYOUT, phi::NllLossGradKernel, float) {}

paddle/phi/kernels/xpu/nll_loss_kernel.cc

+// Copyright (c) 2023 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/nll_loss_kernel.h"
+#include "paddle/phi/backends/xpu/enforce_xpu.h"
+#include "paddle/phi/core/kernel_registry.h"
+
+namespace phi {
+
+template <typename T, typename Context>
+void NllLossRawKernel(const Context& dev_ctx,
+                      const DenseTensor& x,
+                      const DenseTensor& label,
+                      const paddle::optional<DenseTensor>& weight,
+                      int64_t ignore_index,
+                      const std::string& reduction,
+                      DenseTensor* out,
+                      DenseTensor* total_weight) {
+  using XPUType = typename XPUTypeTrait<T>::Type;
+  const auto& label_type = label.dtype();
+  bool label_type_match =
+      label_type == phi::DataType::INT32 || label_type == phi::DataType::INT64;
+  PADDLE_ENFORCE_EQ(label_type_match,
+                    true,
+                    phi::errors::InvalidArgument(
+                        "Input(Label) holds the wrong type, it holds %s, but "
+                        "desires to be %s or %s",
+                        label_type,
+                        phi::DataType::INT32,
+                        phi::DataType::INT64));
+
+  auto x_data = x.data<XPUType>();
+  auto out_data = dev_ctx.template Alloc<XPUType>(out);
+
+  auto weight_data =
+      weight.get_ptr() ? weight.get_ptr()->data<XPUType>() : nullptr;
+
+  auto total_weight_data = dev_ctx.template Alloc<XPUType>(total_weight);
+
+  auto x_dims = x.dims();
+  std::vector<int64_t> x_shape = phi::vectorize<int64_t>(x_dims);
+
+  int64_t reduction_id = 0;
+  if (reduction == "none") {
+    reduction_id = 0;
+  } else if (reduction == "mean") {
+    reduction_id = 1;
+  } else if (reduction == "sum") {
+    reduction_id = 2;
+  }
+
+  int r;
+  if (label_type == phi::DataType::INT32) {
+    const int* label_data = label.data<int>();
+    r = xpu::nll_loss(dev_ctx.x_context(),
+                      x_data,
+                      out_data,
+                      total_weight_data,
+                      x_shape,
+                      label_data,
+                      weight_data,
+                      reduction_id,
+                      ignore_index);
+  } else if (label_type == phi::DataType::INT64) {
+    const int64_t* label_data = label.data<int64_t>();
+    r = xpu::nll_loss(dev_ctx.x_context(),
+                      x_data,
+                      out_data,
+                      total_weight_data,
+                      x_shape,
+                      label_data,
+                      weight_data,
+                      reduction_id,
+                      ignore_index);
+  }
+  PADDLE_ENFORCE_XDNN_SUCCESS(r, "nll_loss");
+}
+
+}  // namespace phi
+
+// TODO(xiongkun): add the non-raw kernel register here.
+PD_REGISTER_KERNEL(nll_loss, XPU, ALL_LAYOUT, phi::NllLossRawKernel, float) {}

test/xpu/test_nll_loss_op_xpu.py

+#   Copyright (c) 2023 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 unittest
+
+import numpy as np
+from get_test_cover_info import (
+    XPUOpTestWrapper,
+    create_test_class,
+    get_xpu_op_support_types,
+)
+from op_test_xpu import XPUOpTest
+
+import paddle
+
+paddle.enable_static()
+
+
+def nll_loss_1d(
+    logs, dtype, targets, weight=None, reduction='mean', ignore_index=-100
+):
+    input_shape = logs.shape
+    N = input_shape[0]
+    C = input_shape[1]
+    out = np.zeros_like(targets).astype(dtype)
+    total_weight = 0
+    for i in range(N):
+        cur_target = targets[i]
+        if cur_target == ignore_index:
+            out[i] = 0
+            continue
+        cur_weight = weight[cur_target] if weight is not None else 1
+        total_weight += cur_weight
+        out[i] = -logs[i][cur_target] * cur_weight
+    if reduction == 'sum':
+        out = np.sum(out)
+        total_weight = np.array([total_weight]).astype(dtype)
+        return {'Out': out, 'Total_weight': total_weight}
+    elif reduction == 'mean':
+        out = np.sum(out)
+        if total_weight != 0:
+            out /= total_weight
+        total_weight = np.array([total_weight]).astype(dtype)
+        return {'Out': out, 'Total_weight': total_weight}
+    elif reduction == 'none':
+        total_weight = np.array([0]).astype(dtype)
+        return {'Out': out, 'Total_weight': total_weight}
+
+
+def nll_loss_2d(
+    logs, dtype, targets, weight=None, reduction='mean', ignore_index=-100
+):
+    input_shape = logs.shape
+    N = input_shape[0]
+    H = input_shape[2]
+    W = input_shape[3]
+    out = np.zeros_like(targets).astype(dtype)
+    total_weight = 0
+    for i in range(N):
+        for h in range(H):
+            for w in range(W):
+                cur_target = targets[i][h][w]
+                if cur_target == ignore_index:
+                    out[i][h][w] = 0
+                    continue
+                cur_weight = weight[cur_target] if weight is not None else 1
+                total_weight += cur_weight
+                out[i][h][w] = -logs[i][cur_target][h][w] * cur_weight
+    if reduction == 'sum':
+        out = np.sum(out)
+        total_weight = np.array([total_weight]).astype(dtype)
+        return {'Out': out, 'Total_weight': total_weight}
+    elif reduction == 'mean':
+        out = np.sum(out)
+        if total_weight != 0:
+            out /= total_weight
+        total_weight = np.array([total_weight]).astype(dtype)
+        return {'Out': out, 'Total_weight': total_weight}
+    elif reduction == 'none':
+        total_weight = np.array([0]).astype(dtype)
+        return {'Out': out, 'Total_weight': total_weight}
+
+
+class XPUTestNLLLossOP(XPUOpTestWrapper):
+    def __init__(self):
+        self.op_name = 'nll_loss'
+        self.use_dynamic_create_class = False
+
+    class TestNLLLossOpBase1D(XPUOpTest):
+        op_type = 'nll_loss'
+
+        def setUp(self):
+            self.dtype = self.in_type
+            self.place = paddle.XPUPlace(0)
+            self.set_attrs()
+            self.set_inputs()
+            self.inputs = {
+                'X': self.x,
+                'Label': self.label,
+            }
+            if self.weight is not None:
+                self.inputs['Weight'] = self.weight
+            self.outputs = nll_loss_1d(
+                self.x,
+                self.dtype,
+                self.label,
+                self.weight,
+                self.attrs['reduction'],
+            )
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'none'}
+
+        def set_inputs(self):
+            self.class_num = 3
+            x_shape = [5, self.class_num]
+            label_shape = [5]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = np.random.random(self.class_num).astype(self.dtype)
+
+        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')
+
+    class TestNLLLossOpWithWeightMean1D(TestNLLLossOpBase1D):
+        def set_attrs(self):
+            self.attrs = {'reduction': 'mean'}
+
+    class TestNLLLossOpWithWeightSum1D(TestNLLLossOpBase1D):
+        def set_attrs(self):
+            self.attrs = {'reduction': 'sum'}
+
+    class TestNLLLossOpWithoutWeightNone1D(TestNLLLossOpBase1D):
+        def set_inputs(self):
+            self.class_num = 3
+            x_shape = [5, self.class_num]
+            label_shape = [5]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = None
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'none'}
+
+    class TestNLLLossOpWithoutWeightMean1D(TestNLLLossOpBase1D):
+        def set_inputs(self):
+            self.class_num = 3
+            x_shape = [5, self.class_num]
+            label_shape = [5]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = None
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'mean'}
+
+    class TestNLLLossOpWithoutWeightSum1D(TestNLLLossOpBase1D):
+        def set_inputs(self):
+            self.class_num = 3
+            x_shape = [5, self.class_num]
+            label_shape = [5]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = None
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'sum'}
+
+    class TestNLLLossOpBase2D(XPUOpTest):
+        op_type = 'nll_loss'
+
+        def setUp(self):
+            self.dtype = self.in_type
+            self.place = paddle.XPUPlace(0)
+            self.set_attrs()
+            self.set_inputs()
+            self.inputs = {'X': self.x, 'Label': self.label}
+            if self.weight is not None:
+                self.inputs['Weight'] = self.weight
+            self.outputs = nll_loss_2d(
+                self.x,
+                self.dtype,
+                self.label,
+                self.weight,
+                self.attrs['reduction'],
+            )
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'none'}
+
+        def set_inputs(self):
+            self.class_num = 3
+            x_shape = [5, self.class_num, 7, 11]
+            label_shape = [5, 7, 11]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = np.random.random(self.class_num).astype(self.dtype)
+
+        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')
+
+    class TestNLLLossOpWithWeightMean2D(TestNLLLossOpBase2D):
+        def set_attrs(self):
+            self.attrs = {'reduction': 'mean'}
+
+    class TestNLLLossOpWithWeightSum2D(TestNLLLossOpBase2D):
+        def set_attrs(self):
+            self.attrs = {'reduction': 'sum'}
+
+    class TestNLLLossOpWithoutWeightNone2D(TestNLLLossOpBase2D):
+        def set_inputs(self):
+            self.dtype = self.in_type
+            self.class_num = 3
+            x_shape = [5, self.class_num, 7, 11]
+            label_shape = [5, 7, 11]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = None
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'none'}
+
+    class TestNLLLossOpWithoutWeightMean2D(TestNLLLossOpBase2D):
+        def set_inputs(self):
+            self.dtype = self.in_type
+            self.class_num = 3
+            x_shape = [5, self.class_num, 7, 11]
+            label_shape = [5, 7, 11]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = None
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'mean'}
+
+    class TestNLLLossOpWithoutWeightSum2D(TestNLLLossOpBase2D):
+        def set_inputs(self):
+            self.dtype = self.in_type
+            self.class_num = 3
+            x_shape = [5, self.class_num, 7, 11]
+            label_shape = [5, 7, 11]
+            self.x = np.random.random(x_shape).astype(self.dtype)
+            self.label = np.random.randint(
+                low=0, high=self.class_num, size=label_shape
+            ).astype(np.int64)
+            self.weight = None
+
+        def set_attrs(self):
+            self.attrs = {'reduction': 'sum'}
+
+
+support_types = get_xpu_op_support_types('nll_loss')
+for stype in support_types:
+    create_test_class(globals(), XPUTestNLLLossOP, stype)
+
+if __name__ == '__main__':
+    unittest.main()