microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-13 01:36:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

bug fix for allgather0

Browse Source
This commit is contained in:
Saeed Maleki
2023-04-03 04:36:20 +00:00
parent bfbdaf6b05
commit 2c6460ce72
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/allgather_test.cu
View File

@@ -53,16 +53,16 @@ __device__ void allgather0(mscclppDevConn_t devConn, int rank, int world_size, i
// this thread's role is a sender role
// put your data asynchronously
if (threadIdx.x % 32 != 0)
if ((threadIdx.x % 32) == 0)
devConn.putWithSignal(rank * nelemsPerGPU * sizeof(int), nelemsPerGPU * sizeof(int));
// make sure everyone is put their data before some thread randomly blocks everyone else in signal
__syncthreads();
// push with flag and sync to make sure the data is received
if (threadIdx.x % 32 != 0)
if ((threadIdx.x % 32) == 0)
devConn.flush();
// this thread's role is a receiver role. wait on the semaphore to make sure the data is ready
if (threadIdx.x % 32 != 0)
if ((threadIdx.x % 32) == 0)
devConn.wait();
}