mirror of
https://github.com/microsoft/mscclpp.git
synced 2026-05-12 17:26:04 +00:00
34945fb107
* Renamed and moved mem alloc functions into the `mscclpp::detail::` namespace (now `mscclpp::detail::gpuCalloc*<T>()`) * Deprecated constructor-calling mem alloc functions (`mscclpp::makeShared*<T>()` and `mscclpp::makeUnique*<T>()`) * Added a new `mscclpp::GpuBuffer<T>()` class that should be used in general for allocating communication buffers * Added a new `mscclpp.utils.GpuBuffer` Python class that inherits `cupy.ndarray` and allocates using `mscclpp::gpuMemAlloc` * Renamed `mscclpp::memcpyCuda*<T>()` functions into `mscclpp::gpuMemcpy*<T>()` for name consistency * A few fixes in NVLS memory allocation * Tackled minor compiler warnings
43 lines
1.3 KiB
C++
43 lines
1.3 KiB
C++
// Copyright (c) Microsoft Corporation.
|
|
// Licensed under the MIT license.
|
|
|
|
#include <gtest/gtest.h>
|
|
|
|
#include <mscclpp/gpu_utils.hpp>
|
|
|
|
TEST(CudaUtilsTest, AllocShared) {
|
|
auto p1 = mscclpp::detail::gpuCallocShared<uint32_t>();
|
|
auto p2 = mscclpp::detail::gpuCallocShared<int64_t>(5);
|
|
}
|
|
|
|
TEST(CudaUtilsTest, AllocUnique) {
|
|
auto p1 = mscclpp::detail::gpuCallocUnique<uint32_t>();
|
|
auto p2 = mscclpp::detail::gpuCallocUnique<int64_t>(5);
|
|
}
|
|
|
|
TEST(CudaUtilsTest, MakeSharedHost) {
|
|
auto p1 = mscclpp::detail::gpuCallocHostShared<uint32_t>();
|
|
auto p2 = mscclpp::detail::gpuCallocHostShared<int64_t>(5);
|
|
}
|
|
|
|
TEST(CudaUtilsTest, MakeUniqueHost) {
|
|
auto p1 = mscclpp::detail::gpuCallocHostUnique<uint32_t>();
|
|
auto p2 = mscclpp::detail::gpuCallocHostUnique<int64_t>(5);
|
|
}
|
|
|
|
TEST(CudaUtilsTest, Memcpy) {
|
|
const int nElem = 1024;
|
|
std::vector<int> hostBuff(nElem);
|
|
for (int i = 0; i < nElem; ++i) {
|
|
hostBuff[i] = i + 1;
|
|
}
|
|
std::vector<int> hostBuffTmp(nElem, 0);
|
|
auto devBuff = mscclpp::detail::gpuCallocShared<int>(nElem);
|
|
mscclpp::gpuMemcpy<int>(devBuff.get(), hostBuff.data(), nElem, cudaMemcpyHostToDevice);
|
|
mscclpp::gpuMemcpy<int>(hostBuffTmp.data(), devBuff.get(), nElem, cudaMemcpyDeviceToHost);
|
|
|
|
for (int i = 0; i < nElem; ++i) {
|
|
EXPECT_EQ(hostBuff[i], hostBuffTmp[i]);
|
|
}
|
|
}
|