microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-03 05:01:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

Pytest (#162)

Browse Source 
Port python tests to mscclpp.
Please run
`mpirun -tag-output -np 8 pytest ./python/test/test_mscclpp.py -x` to start pytest

---------

Co-authored-by: Saeed Maleki <saemal@microsoft.com>
Co-authored-by: Changho Hwang <changhohwang@microsoft.com>
Co-authored-by: Saeed Maleki <30272783+saeedmaleki@users.noreply.github.com>
This commit is contained in:
Binyang2014
2023-09-01 21:22:11 +08:00
committed by GitHub
parent 3df18d20a3
commit 858e381829
39 changed files with 1186 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
python/test/simple_proxy_channel_test.cu Normal file
View File

@@ -0,0 +1,35 @@
// Copyright (c) Microsoft Corporation.
// Licensed under the MIT license.
#include <mscclpp/packet.hpp>
#include <mscclpp/proxy_channel_device.hpp>
// be careful about using channels[my_rank] as it is inavlie and it is there just for simplicity of indexing
extern "C" __global__ void __launch_bounds__(1024, 1)
simple_proxy_channel(mscclpp::SimpleProxyChannelDeviceHandle* channels, int my_rank, int nranks, int* data,
int* scratch, int num_elements, int use_packet) {
int tid = threadIdx.x;
int nthreads = blockDim.x;
uint64_t size_per_rank = (num_elements * sizeof(int)) / nranks;
uint64_t my_offset = size_per_rank * my_rank;
int nthreads_per_rank = nthreads / nranks;
int my_nghr = tid / nthreads_per_rank;
uint64_t my_nghr_offset = size_per_rank * my_nghr;
__syncthreads();
int flag = 123;
if (use_packet) {
mscclpp::putPackets(scratch, 2 * my_offset, data, my_offset, size_per_rank, tid, nthreads, flag);
__syncthreads();
if (tid < nranks && tid != my_rank) {
channels[tid].put(2 * my_offset, 2 * my_offset, 2 * size_per_rank);
}
if (my_nghr != my_rank && my_nghr < nranks)
mscclpp::getPackets(data, my_nghr_offset, scratch, 2 * my_nghr_offset, size_per_rank, tid % nthreads_per_rank,
nthreads_per_rank, flag);
} else {
if (tid < nranks && tid != my_rank) {
channels[tid].putWithSignalAndFlush(my_offset, my_offset, size_per_rank);
channels[tid].wait();
}
}
}