mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
55 lines
2.7 KiB
C++
55 lines
2.7 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 <gtest/gtest.h>
|
|
#include <memory>
|
|
|
|
#include "tensorrt_llm/common/cudaUtils.h"
|
|
#include "tensorrt_llm/runtime/tllmBuffers.h"
|
|
|
|
namespace tc = tensorrt_llm::common;
|
|
namespace tr = tensorrt_llm::runtime;
|
|
|
|
TEST(TestGetPtrCudaMemoryType, TestMemoryTypesAreAsExpected)
|
|
{
|
|
auto const cpuBuffer = std::make_unique<tr::HostBuffer>(1024, tr::TRTDataType<float>::value);
|
|
|
|
// Note: this I think will change with hardware HMM support. To be confirmed. If it does, check for this
|
|
// support and change expected memory type value accordingly.
|
|
ASSERT_EQ(tc::getPtrCudaMemoryType(cpuBuffer->data()), cudaMemoryType::cudaMemoryTypeUnregistered)
|
|
<< "Host paged memory should appear as unregistered to CUDA.";
|
|
|
|
auto const pinnedCpuBuffer = std::make_unique<tr::PinnedBuffer>(1024, tr::TRTDataType<float>::value);
|
|
ASSERT_EQ(tc::getPtrCudaMemoryType(pinnedCpuBuffer->data()), cudaMemoryType::cudaMemoryTypeHost)
|
|
<< "The memory type of a pinned CPU buffer was not 'host'. Is this system using Confidential Computing?";
|
|
|
|
auto const pinnedPoolCpuBuffer = std::make_unique<tr::PinnedPoolBuffer>(1024, tr::TRTDataType<float>::value);
|
|
ASSERT_EQ(tc::getPtrCudaMemoryType(pinnedPoolCpuBuffer->data()), cudaMemoryType::cudaMemoryTypeHost)
|
|
<< "The memory type of a pinned CPU buffer was not 'host'. Is this system using Confidential Computing?";
|
|
|
|
if (tc::getDeviceCount() <= 0)
|
|
{
|
|
GTEST_SKIP() << "This test cannot run further when no devices are present on the system.";
|
|
}
|
|
auto const stream = std::make_shared<tr::CudaStream>();
|
|
auto const pool = tensorrt_llm::runtime::CudaMemPool::getPrimaryPoolForDevice(stream->getDevice());
|
|
auto const deviceBuffer
|
|
= std::make_unique<tr::DeviceBuffer>(1024, tr::TRTDataType<float>::value, tr::CudaAllocatorAsync{stream, pool});
|
|
ASSERT_EQ(tc::getPtrCudaMemoryType(deviceBuffer->data()), cudaMemoryType::cudaMemoryTypeDevice);
|
|
auto const deviceSyncBuffer = std::make_unique<tr::StaticDeviceBuffer>(1024, tr::TRTDataType<float>::value);
|
|
ASSERT_EQ(tc::getPtrCudaMemoryType(deviceSyncBuffer->data()), cudaMemoryType::cudaMemoryTypeDevice);
|
|
}
|