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 +1,86 @@
+/* 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/softmax_kernel.h"
+#include "paddle/phi/backends/xpu/enforce_xpu.h"
+#include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/kernels/funcs/axis_utils.h"
+#include "paddle/phi/kernels/funcs/math_function.h"
+
+namespace phi {
+
+template <typename T, typename Context>
+void SoftmaxKernel(const Context& dev_ctx,
+                   const DenseTensor& x,
+                   int axis,
+                   DenseTensor* out) {
+  using XPUType = typename XPUTypeTrait<T>::Type;
+  const int rank = x.dims().size();
+  const int calc_axis = funcs::CanonicalAxis(axis, rank);
+
+  // allocate memory on device.
+  dev_ctx.template Alloc<T>(out);
+  // For 0-Sized Tensor
+  if (out->numel() == 0) {
+    return;
+  }
+  // For 0D Tensor
+  if (rank == 0) {
+    funcs::set_constant(dev_ctx, out, static_cast<T>(1.0));
+    return;
+  }
+
+  std::vector<int64_t> x_dims;
+  for (int i = 0; i < rank; i++) {
+    x_dims.push_back(x.dims()[i]);
+  }
+
+  int r = 0;
+  auto version =
+      backends::xpu::get_xpu_version(dev_ctx.GetPlace().GetDeviceId());
+  if (version == backends::xpu::XPUVersion::XPU1) {
+    xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
+    XPUType* clip_x_data_l3 = RAII_GUARD.alloc_l3_or_gm<XPUType>(x.numel());
+    r = xpu::clamp(dev_ctx.x_context(),
+                   reinterpret_cast<const XPUType*>(x.data<T>()),
+                   clip_x_data_l3,
+                   x.numel(),
+                   static_cast<XPUType>(-1e20),
+                   static_cast<XPUType>(1e20));
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "clamp");
+    r = xpu::softmax<XPUType>(dev_ctx.x_context(),
+                              clip_x_data_l3,
+                              reinterpret_cast<XPUType*>(out->data<T>()),
+                              x_dims,
+                              static_cast<int64_t>(calc_axis));
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "softmax");
+  } else {
+    r = xpu::softmax<XPUType>(dev_ctx.x_context(),
+                              reinterpret_cast<const XPUType*>(x.data<T>()),
+                              reinterpret_cast<XPUType*>(out->data<T>()),
+                              x_dims,
+                              static_cast<int64_t>(calc_axis));
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "softmax");
+  }
+}
+
+}  // namespace phi
+
+PD_REGISTER_KERNEL(softmax,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::SoftmaxKernel,
+                   float,
+                   phi::float16,
+                   phi::bfloat16) {}