mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
57 lines
1.9 KiB
C++
57 lines
1.9 KiB
C++
/*
|
|
* Copyright (c) 2025, NVIDIA CORPORATION. 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 "userbuffersTensor.h"
|
|
|
|
TRTLLM_NAMESPACE_BEGIN
|
|
|
|
namespace torch_ext
|
|
{
|
|
|
|
std::pair<torch::Tensor, tensorrt_llm::runtime::ub::UBBuffer> create_userbuffers_tensor(
|
|
at::IntArrayRef shape, torch::ScalarType dtype)
|
|
{
|
|
int64_t buffer_size
|
|
= std::accumulate(shape.begin(), shape.end(), 1, std::multiplies<int64_t>()) * torch::elementSize(dtype);
|
|
|
|
std::vector<int64_t> strides_vec(shape.size());
|
|
strides_vec[shape.size() - 1] = 1;
|
|
for (int64_t i = static_cast<int64_t>(shape.size()) - 1; i >= 1; --i)
|
|
{
|
|
strides_vec[i - 1] = strides_vec[i] * shape[i];
|
|
}
|
|
|
|
auto [ptr, ub] = tensorrt_llm::runtime::ub::UserBuffersManager::get_instance().allocate_userbuffers(buffer_size);
|
|
auto& deleter = ptr.get_deleter();
|
|
return std::make_pair(
|
|
torch::from_blob(ptr.release(), shape, strides_vec, deleter, torch::dtype(dtype).device(torch::kCUDA)), ub);
|
|
}
|
|
|
|
// Custom op interface for create_userbuffers_tensor.
|
|
// Python side does not need the UBBuffer object.
|
|
torch::Tensor create_userbuffers_tensor_op(at::IntArrayRef shape, torch::ScalarType dtype)
|
|
{
|
|
return create_userbuffers_tensor(shape, dtype).first;
|
|
}
|
|
|
|
} // namespace torch_ext
|
|
|
|
TRTLLM_NAMESPACE_END
|
|
|
|
TORCH_LIBRARY_FRAGMENT(trtllm, m)
|
|
{
|
|
m.def("create_userbuffers_tensor", &tensorrt_llm::torch_ext::create_userbuffers_tensor_op);
|
|
}
|