Relaxing default timeout limit, add error log (#1052)

Signed-off-by: Tim Hu <timhu102@amd.com>

[ROCm/rccl commit: 05850e89f2]
This commit is contained in:
Tim
2024-01-18 15:09:08 -05:00
zatwierdzone przez GitHub
rodzic 5f7ef6b671
commit 0343d9ccac
2 zmienionych plików z 7 dodań i 4 usunięć
+1 -1
Wyświetl plik
@@ -63,7 +63,7 @@ namespace RcclUnitTesting
maxRanksPerGpu = GetEnvVar("UT_MAX_RANKS_PER_GPU", 1);
showTiming = GetEnvVar("UT_SHOW_TIMING", 1);
useInteractive = GetEnvVar("UT_INTERACTIVE", 0);
timeoutUs = GetEnvVar("UT_TIMEOUT_US" , 200000);
timeoutUs = GetEnvVar("UT_TIMEOUT_US" , 5000000);
// Limit number of supported reduction operators to just ncclSum if only allReduce is built
#ifdef BUILD_ALLREDUCE_ONLY