chore: import upstream snapshot with attribution

2026-07-13 12:40:42 +08:00
commit e25996e7db
15472 changed files with 3536181 additions and 0 deletions
@@ -0,0 +1,102 @@
+/* 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. */
+
+#pragma once
+
+#if defined(__NVCC__) || defined(__HIPCC__)
+#include <thrust/device_ptr.h>
+#include <thrust/functional.h>
+#include <thrust/reduce.h>
+#else
+#include <algorithm>
+#endif
+
+#include "paddle/phi/core/tensor_utils.h"
+#include "paddle/phi/kernels/funcs/sequence_mask.h"
+
+namespace phi {
+
+template <typename T, typename Context>
+void SequenceMaskScalarKernel(const Context& dev_ctx,
+                              const DenseTensor& x,
+                              const Scalar& max_len,
+                              DataType out_dtype,
+                              DenseTensor* y) {
+  int maxlen = max_len.to<int>();
+  auto* x_data = x.data<T>();
+  int64_t x_numel = x.numel();
+
+  if (maxlen < 0) {
+    if (x_numel == 0) {
+      maxlen = 0;
+    } else {
+#if defined(__NVCC__) || defined(__HIPCC__)
+      VLOG(10)
+          << "SequenceMaskOp on GPU may be slow when maxlen is not provided.";
+      maxlen = static_cast<int>(
+          thrust::reduce(thrust::device_pointer_cast(x_data),
+                         thrust::device_pointer_cast(x_data) + x_numel,
+                         static_cast<T>(0),
+                         thrust::maximum<T>()));
+#else
+      maxlen = static_cast<int>(*std::max_element(x_data, x_data + x_numel));
+#endif
+    }
+    auto y_dim = vectorize<int64_t>(x.dims());
+    y_dim.push_back(maxlen);
+    y->Resize(y_dim);
+  }
+  if (x_numel == 0) {
+    dev_ctx.Alloc(y, out_dtype);
+    return;
+  }
+
+  VisitDataType(out_dtype,
+                funcs::SequenceMaskFunctor<Context, T>(
+                    dev_ctx, x_data, y, x_numel * maxlen, maxlen));
+}
+
+template <typename T, typename Context>
+void SequenceMaskKernel(const Context& dev_ctx,
+                        const DenseTensor& x,
+                        const optional<DenseTensor>& max_len_tensor,
+                        int maxlen,
+                        DataType out_dtype,
+                        DenseTensor* y) {
+  if (max_len_tensor) {
+    bool is_gpu_place = dev_ctx.GetPlace().GetType() == AllocationType::GPU;
+    if (is_gpu_place) {
+      DenseTensor temp;
+      Copy(dev_ctx, *max_len_tensor.get_ptr(), CPUPlace(), false, &temp);
+      maxlen = *temp.data<int32_t>();
+    } else {
+      maxlen = *max_len_tensor.get_ptr()->data<int32_t>();
+    }
+
+    auto y_dim = vectorize<int64_t>(x.dims());
+    y_dim.push_back(maxlen);
+    y->Resize(y_dim);
+
+    PADDLE_ENFORCE_GT(
+        maxlen,
+        0,
+        common::errors::InvalidArgument(
+            "Input(MaxLenTensor) value should be greater than 0. But "
+            "received Input(MaxLenTensor) value = %d.",
+            maxlen));
+  }
+  SequenceMaskScalarKernel<T, Context>(
+      dev_ctx, x, Scalar(maxlen), out_dtype, y);
+}
+}  // namespace phi