NCCL 2.25.1-1

Add Blackwell/SM100 support
 * Add compilation for sm100
 * Add graph search speeds for Blackwell
 * Optimize graph search to converge on large NVLink domains
 * Limit NVLS heads to 32
 * Increase various limits to fit large NVLink domains
 * Add extra checks for IMEX setup, needed for MNNVL
 * Increase MAXCHANNELS to 64

Extend NVTX instrumentation to track NCCL communicators
 * Add communicator ID to NVTX traces to allow for correlation
   between ranks.

RAS fixes


[ROCm/rccl commit: 80f6bda437]
Este commit está contenido en:
Sylvain Jeaugey
2025-01-27 03:30:22 -08:00
padre 7c60e5e3fe
commit 70bd1af305
Se han modificado 30 ficheros con 603 adiciones y 330 borrados
+2
Ver fichero
@@ -363,6 +363,8 @@ ncclResult_t ncclProxyStart(struct ncclComm* comm);
ncclResult_t ncclProxyInit(struct ncclComm* comm, struct ncclSocket* sock, union ncclSocketAddress* peerAddresses, uint64_t *peerAddressesUDS);
ncclResult_t ncclProxyCreate(struct ncclComm* comm);
ncclResult_t ncclProxyConnect(struct ncclComm* comm, int transport, int send, int proxyRank, struct ncclProxyConnector* proxyConn);
// NB: ncclProxyMsgTypeStr[] in proxy.cc needs to match
enum ncclProxyMsgType {
ncclProxyMsgInit = 1,
ncclProxyMsgSharedInit = 2,