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 +1,115 @@
+// 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/batch_norm_kernel.h"
+
+#include "paddle/phi/backends/gpu/gpu_dnn.h"
+#include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/kernels/empty_kernel.h"
+
+namespace phi {
+
+template <typename T, typename Context>
+void BatchNormInferKernel(const Context& dev_ctx,
+                          const DenseTensor& x,
+                          const DenseTensor& mean,
+                          const DenseTensor& variance,
+                          const DenseTensor& scale,
+                          const DenseTensor& bias,
+                          float momentum,
+                          float epsilon,
+                          const std::string& data_layout,
+                          DenseTensor* y,
+                          DenseTensor* mean_out,
+                          DenseTensor* variance_out) {
+  // Since saved_mean and saved_variance are used regardless of whether
+  // they are in test mode, temporary variables need to be created here
+  // to be compatible
+  auto saved_mean = EmptyLike<T, Context>(dev_ctx, *mean_out);
+  auto saved_variance = EmptyLike<T, Context>(dev_ctx, *variance_out);
+  BatchNormKernel<T, Context>(dev_ctx,
+                              x,
+                              mean,
+                              variance,
+                              scale,
+                              bias,
+                              /*is_test=*/true,
+                              momentum,
+                              epsilon,
+                              data_layout,
+                              /*use_global_stats=*/false,
+                              /*trainable_statistics=*/false,
+                              y,
+                              mean_out,
+                              variance_out,
+                              &saved_mean,
+                              &saved_variance,
+                              /*reserve_space=*/nullptr);
+}
+
+}  // namespace phi
+
+PD_REGISTER_KERNEL(batch_norm_infer,
+                   CPU,
+                   ALL_LAYOUT,
+                   phi::BatchNormInferKernel,
+                   float,
+                   double) {}
+#ifdef PADDLE_WITH_CUDA
+#if CUDNN_VERSION_MIN(8, 1, 0)
+PD_REGISTER_KERNEL(batch_norm_infer,
+                   GPU,
+                   ALL_LAYOUT,
+                   phi::BatchNormInferKernel,
+                   float,
+                   double,
+                   phi::bfloat16,
+                   phi::float16) {
+  if (kernel_key.dtype() == phi::DataType::FLOAT16 ||
+      kernel_key.dtype() == phi::DataType::BFLOAT16) {
+    kernel->OutputAt(1).SetDataType(phi::DataType::FLOAT32);
+    kernel->OutputAt(2).SetDataType(phi::DataType::FLOAT32);
+  }
+}
+#else
+PD_REGISTER_KERNEL(batch_norm_infer,
+                   GPU,
+                   ALL_LAYOUT,
+                   phi::BatchNormInferKernel,
+                   float,
+                   double,
+                   phi::float16) {
+  if (kernel_key.dtype() == phi::DataType::FLOAT16) {
+    kernel->OutputAt(1).SetDataType(phi::DataType::FLOAT32);
+    kernel->OutputAt(2).SetDataType(phi::DataType::FLOAT32);
+  }
+}
+#endif
+#endif
+#ifdef PADDLE_WITH_HIP
+PD_REGISTER_KERNEL(batch_norm_infer,
+                   GPU,
+                   ALL_LAYOUT,
+                   phi::BatchNormInferKernel,
+                   float,
+                   phi::float16) {}
+#endif
+#ifdef PADDLE_WITH_XPU
+PD_REGISTER_KERNEL(batch_norm_infer,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::BatchNormInferKernel,
+                   float,
+                   phi::float16) {}
+#endif