mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
deaae40bd7
* Update TensorRT-LLM --------- Co-authored-by: Shixiaowei02 <39303645+Shixiaowei02@users.noreply.github.com>
111 lines
4.3 KiB
C++
111 lines
4.3 KiB
C++
/*
|
|
* Copyright (c) 2022-2024, 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 "tensorrt_llm/runtime/iBuffer.h"
|
|
#include "tensorrt_llm/runtime/iTensor.h"
|
|
|
|
#include "tensorrt_llm/common/assert.h"
|
|
#include "tensorrt_llm/common/cudaUtils.h"
|
|
#include "tensorrt_llm/runtime/bufferView.h"
|
|
|
|
#include <cuda_runtime_api.h>
|
|
|
|
#include <memory>
|
|
|
|
using namespace tensorrt_llm::runtime;
|
|
|
|
MemoryType IBuffer::memoryType(void const* data)
|
|
{
|
|
cudaPointerAttributes attributes{};
|
|
TLLM_CUDA_CHECK(::cudaPointerGetAttributes(&attributes, data));
|
|
switch (attributes.type)
|
|
{
|
|
case cudaMemoryTypeHost: return MemoryType::kPINNED;
|
|
case cudaMemoryTypeDevice:
|
|
case cudaMemoryTypeManaged: return MemoryType::kGPU;
|
|
case cudaMemoryTypeUnregistered: return MemoryType::kCPU;
|
|
default: TLLM_THROW("Unsupported memory type");
|
|
}
|
|
}
|
|
|
|
IBuffer::UniquePtr IBuffer::slice(IBuffer::SharedPtr buffer, std::size_t offset, std::size_t size)
|
|
{
|
|
return std::make_unique<BufferView>(std::move(buffer), offset, size);
|
|
}
|
|
|
|
IBuffer::UniquePtr IBuffer::wrap(void* data, nvinfer1::DataType type, std::size_t size, std::size_t capacity)
|
|
{
|
|
TLLM_CHECK_WITH_INFO(size <= capacity, "Requested size is larger than capacity");
|
|
auto memoryType = IBuffer::memoryType(data);
|
|
|
|
IBuffer::UniquePtr result;
|
|
auto const capacityInBytes = capacity * BufferDataType(type).getSize();
|
|
switch (memoryType)
|
|
{
|
|
case MemoryType::kPINNED:
|
|
result.reset(new GenericBuffer<PinnedBorrowingAllocator>( // NOLINT(modernize-make-unique)
|
|
capacity, type, PinnedBorrowingAllocator(data, capacityInBytes)));
|
|
break;
|
|
case MemoryType::kCPU:
|
|
result.reset( // NOLINT(modernize-make-unique)
|
|
new GenericBuffer<CpuBorrowingAllocator>(capacity, type, CpuBorrowingAllocator(data, capacityInBytes)));
|
|
break;
|
|
case MemoryType::kGPU:
|
|
result.reset( // NOLINT(modernize-make-unique)
|
|
new GenericBuffer<GpuBorrowingAllocator>(capacity, type, GpuBorrowingAllocator(data, capacityInBytes)));
|
|
break;
|
|
default: TLLM_THROW("Unknown memory type");
|
|
}
|
|
result->resize(size);
|
|
return result;
|
|
}
|
|
|
|
std::ostream& tensorrt_llm::runtime::operator<<(std::ostream& output, IBuffer const& buffer)
|
|
{
|
|
auto data = const_cast<IBuffer&>(buffer).data();
|
|
auto tensor = ITensor::wrap(data, buffer.getDataType(),
|
|
ITensor::makeShape({static_cast<SizeType>(buffer.getSize())}), buffer.getCapacity());
|
|
return output << *tensor;
|
|
}
|
|
|
|
char const* IBuffer::getDataTypeName() const
|
|
{
|
|
switch (getDataType())
|
|
{
|
|
case nvinfer1::DataType::kINT64: return DataTypeTraits<nvinfer1::DataType::kINT64>::name;
|
|
case nvinfer1::DataType::kINT32: return DataTypeTraits<nvinfer1::DataType::kINT32>::name;
|
|
case nvinfer1::DataType::kFLOAT: return DataTypeTraits<nvinfer1::DataType::kFLOAT>::name;
|
|
case nvinfer1::DataType::kBF16: return DataTypeTraits<nvinfer1::DataType::kBF16>::name;
|
|
case nvinfer1::DataType::kHALF: return DataTypeTraits<nvinfer1::DataType::kHALF>::name;
|
|
case nvinfer1::DataType::kBOOL: return DataTypeTraits<nvinfer1::DataType::kBOOL>::name;
|
|
case nvinfer1::DataType::kUINT8: return DataTypeTraits<nvinfer1::DataType::kUINT8>::name;
|
|
case nvinfer1::DataType::kINT8: return DataTypeTraits<nvinfer1::DataType::kINT8>::name;
|
|
case nvinfer1::DataType::kFP8: return DataTypeTraits<nvinfer1::DataType::kFP8>::name;
|
|
}
|
|
TLLM_THROW("Unknown data type");
|
|
}
|
|
|
|
char const* IBuffer::getMemoryTypeName() const
|
|
{
|
|
switch (getMemoryType())
|
|
{
|
|
case MemoryType::kPINNED: return MemoryTypeString<MemoryType::kPINNED>::value;
|
|
case MemoryType::kCPU: return MemoryTypeString<MemoryType::kCPU>::value;
|
|
case MemoryType::kGPU: return MemoryTypeString<MemoryType::kGPU>::value;
|
|
}
|
|
TLLM_THROW("Unknown memory type");
|
|
}
|