wehub-resource-sync
127 lines
4.6 KiB
Plaintext
127 lines
4.6 KiB
Plaintext
/* Copyright (c) 2025 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/common/enforce.h"
|
|
#include "paddle/phi/backends/gpu/gpu_launch_config.h"
|
|
#include "paddle/phi/common/memory_utils.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/funcs/index_elementwise.cu.h"
|
|
#include "paddle/phi/kernels/funcs/slice_utils.h"
|
|
#include "paddle/phi/kernels/strided_copy_kernel.h"
|
|
|
|
namespace phi {
|
|
template <typename T, typename Context>
|
|
void StridedElementwiseCopyKernel(const Context& dev_ctx,
|
|
const DenseTensor& input,
|
|
const std::vector<int64_t>& out_dims,
|
|
const std::vector<int64_t>& out_strides,
|
|
int64_t out_offset,
|
|
DenseTensor* out) {
|
|
DenseTensorMeta meta = input.meta();
|
|
meta.strides = make_ddim(out_strides);
|
|
meta.dims = make_ddim(out_dims);
|
|
meta.offset = out_offset;
|
|
out->set_meta(meta);
|
|
auto numel = out->numel();
|
|
T* output_data = out->data<T>();
|
|
PADDLE_ENFORCE_NOT_NULL(
|
|
output_data,
|
|
common::errors::InvalidArgument(
|
|
"StridedElementwiseCopyKernel's out tensor must complete "
|
|
"mutable data before call kernel."));
|
|
|
|
const T* input_data = input.data<T>();
|
|
|
|
if (numel == 1) {
|
|
#ifdef PADDLE_WITH_HIP
|
|
hipMemcpy(output_data,
|
|
input_data,
|
|
SizeOf(input.dtype()),
|
|
hipMemcpyDeviceToDevice);
|
|
#else
|
|
cudaMemcpy(output_data,
|
|
input_data,
|
|
SizeOf(input.dtype()),
|
|
cudaMemcpyDeviceToDevice);
|
|
#endif
|
|
|
|
return;
|
|
}
|
|
|
|
bool can_expand = funcs::CheckIsLastDimsMatch(input.dims(), out->dims());
|
|
PADDLE_ENFORCE_EQ(can_expand || input.numel() == 1,
|
|
true,
|
|
common::errors::InvalidArgument(
|
|
"Input shape(%s) must expand to out shape(%s).",
|
|
input.dims(),
|
|
out->dims()));
|
|
|
|
std::array<int64_t*, 2> strides_array;
|
|
std::vector<int64_t> desired_shape;
|
|
std::array<std::vector<int64_t>, 2> strides_vec;
|
|
|
|
funcs::CopyStride<2>(out_dims,
|
|
out_strides,
|
|
SizeOf(out->dtype()),
|
|
vectorize<int64_t>(input.dims()),
|
|
vectorize<int64_t>(input.strides()),
|
|
SizeOf(input.dtype()),
|
|
&desired_shape,
|
|
&strides_array,
|
|
&numel,
|
|
strides_vec);
|
|
|
|
auto offset_calc =
|
|
funcs::make_offset_calculator_put<2, true>(desired_shape, strides_array);
|
|
|
|
constexpr int block_size = 128;
|
|
constexpr int loop_size = 4;
|
|
const dim3 block(block_size);
|
|
const int64_t grid_x =
|
|
(numel + block.x * loop_size - 1) / (block.x * loop_size);
|
|
PADDLE_ENFORCE_LE_UINT32_MAX(grid_x, "grid.x");
|
|
const dim3 grid(static_cast<uint32_t>(grid_x));
|
|
auto stream = dev_ctx.stream();
|
|
using dtype = funcs::OpaqueType<sizeof(T)>;
|
|
const char* in_ptr = reinterpret_cast<const char*>(input.data<T>());
|
|
char* out_ptr = reinterpret_cast<char*>(out->data<T>());
|
|
funcs::index_elementwise_with_tensor_kernel<block_size, loop_size>
|
|
<<<grid, block, 0, stream>>>(numel, [=] __device__(int64_t idx) {
|
|
const auto offsets = offset_calc.get(idx);
|
|
char* const out_data = out_ptr + offsets[0];
|
|
const char* const in_data = in_ptr + offsets[1];
|
|
|
|
*reinterpret_cast<dtype*>(out_data) =
|
|
*reinterpret_cast<const dtype*>(in_data);
|
|
});
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(strided_elementwise_copy,
|
|
GPU,
|
|
ALL_LAYOUT,
|
|
phi::StridedElementwiseCopyKernel,
|
|
bool,
|
|
uint8_t,
|
|
int8_t,
|
|
int16_t,
|
|
int32_t,
|
|
int64_t,
|
|
float,
|
|
double,
|
|
phi::float16,
|
|
phi::bfloat16,
|
|
phi::complex64,
|
|
phi::complex128,
|
|
phi::float8_e4m3fn,
|
|
phi::float8_e5m2) {}
|