Fix crash when setting NCCL_MAX_P2P_NCHANNELS below nchannels.
Fix hang during sendrecv dynamic NVB connection establishment on
cubemesh topologies.
Add environment variable to only use SHARP on communicators beyond
a given number of ranks.
Add debug subsystem to trace memory allocations.
Fix compilation with TRACE=1. (Issue #505)
This commit is contained in:
Sylvain Jeaugey
2021-05-11 18:16:30 -07:00
parent ca8485b0d0
commit 3fec2fa5ee
14 changed files with 142 additions and 52 deletions
+1
View File
@@ -159,6 +159,7 @@ struct ncclComm {
struct ncclQueueInfo* enqueueInfo;
cudaGraphNode_t lastSetupNode;
unsigned long long lastCudaGraphId;
int driverVersion;
};
#endif