AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixes to all tests that divide buffers by nranks so that they trim buffer sizes to be multiples of 16 bytes.

Browse Source 
This ensures non-pow2 ranks have buffer addresses aligned suitably for performance.


[ROCm/rccl-tests commit: 29f4114f02]
This commit is contained in:
John Bachan
2024-12-18 11:14:18 -08:00
parent 637acfba05
commit 69b9a05e71
12 changed files with 36 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl-tests/src/alltoall.cu
+5 -5
View File
@@ -7,12 +7,12 @@
#include "cuda_runtime.h"
#include "common.h"
void AlltoAllGetCollByteCount(size_t *sendcount, size_t *recvcount, size_t *paramcount, size_t *sendInplaceOffset, size_t *recvInplaceOffset, size_t count, int nranks) {
*sendcount = (count/nranks)*nranks;
*recvcount = (count/nranks)*nranks;
void AlltoAllGetCollByteCount(size_t *sendcount, size_t *recvcount, size_t *paramcount, size_t *sendInplaceOffset, size_t *recvInplaceOffset, size_t count, size_t eltSize, int nranks) {
*paramcount = (count/nranks) & -(16/eltSize);
*sendcount = nranks*(*paramcount);
*recvcount = *sendcount;
*sendInplaceOffset = 0;
*recvInplaceOffset = 0;
*paramcount = count/nranks;
}
testResult_t AlltoAllInitData(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t op, int root, int rep, int in_place) {
@@ -74,7 +74,7 @@ struct testColl alltoAllTest = {
void AlltoAllGetBuffSize(size_t *sendcount, size_t *recvcount, size_t count, int nranks) {
size_t paramcount, sendInplaceOffset, recvInplaceOffset;
AlltoAllGetCollByteCount(sendcount, recvcount, &paramcount, &sendInplaceOffset, &recvInplaceOffset, count, nranks);
AlltoAllGetCollByteCount(sendcount, recvcount, &paramcount, &sendInplaceOffset, &recvInplaceOffset, count, /*eltSize=*/1, nranks);
}
testResult_t AlltoAllRunTest(struct threadArgs* args, int root, ncclDataType_t type, const char* typeName, ncclRedOp_t op, const char* opName) {