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

Fix message size logging (#115)

Browse Source 
Previously, the logger was logging the number of expected bytes a node was to recieve.
This differs from the stdout logging, where the reported message size is the total size of a message.

Signed-off-by: Grant Pinkert <gpinkert@amd.com>
This commit is contained in:
Grant Pinkert
2025-04-25 06:05:21 -10:00
committed by GitHub
parent 83d38d91b6
commit f611dbd49a
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/common.cu
+4 -4
View File
@@ -760,12 +760,12 @@ testResult_t BenchTime(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t
PRINT(" %7s %6.2f %6.2f %5s", timeStr, algBw, busBw, "N/A");
}
auto largestMessageSize = std::max(args->sendBytes, args->expectedBytes);
if (args->reporter) {
if (args->reportErrors) {
args->reporter->addResult((args->nThreads * args->nGpus), args->nProcs, args->totalProcs, args->expectedBytes, in_place, timeUsec, algBw, busBw, wrongElts);
}
else {
args->reporter->addResult((args->nThreads * args->nGpus), args->nProcs, args->totalProcs, args->expectedBytes, in_place, timeUsec, algBw, busBw);
args->reporter->addResult((args->nThreads * args->nGpus), args->nProcs, args->totalProcs, largestMessageSize, in_place, timeUsec, algBw, busBw, wrongElts);
} else {
args->reporter->addResult((args->nThreads * args->nGpus), args->nProcs, args->totalProcs, largestMessageSize, in_place, timeUsec, algBw, busBw);
}
}