2
0

Merge remote-tracking branch 'nccl/master' into develop

Este cometimento está contido em:
BertanDogancay
2025-01-23 11:48:18 -06:00
cometido por Corey Derochie
ascendente b6377e0b8c 178b6b7590
cometimento 36343be84f
122 ficheiros modificados com 9009 adições e 4702 eliminações
+3 -2
Ver ficheiro
@@ -202,7 +202,6 @@ ncclResult_t bootstrapCreateRoot(struct ncclBootstrapHandle* handle, bool idFrom
ncclResult_t bootstrapGetUniqueId(struct ncclBootstrapHandle* handle) {
memset(handle, 0, sizeof(ncclBootstrapHandle));
NCCLCHECK(getRandomData(&handle->magic, sizeof(handle->magic)));
const char* env = ncclGetEnv("NCCL_COMM_ID");
if (env) {
@@ -211,7 +210,9 @@ ncclResult_t bootstrapGetUniqueId(struct ncclBootstrapHandle* handle) {
WARN("Invalid NCCL_COMM_ID, please use format: <ipv4>:<port> or [<ipv6>]:<port> or <hostname>:<port>");
return ncclInvalidArgument;
}
handle->magic = NCCL_MAGIC;
} else {
NCCLCHECK(getRandomData(&handle->magic, sizeof(handle->magic)));
memcpy(&handle->addr, &bootstrapNetIfAddr, sizeof(union ncclSocketAddress));
NCCLCHECK(bootstrapCreateRoot(handle, false));
}
@@ -631,7 +632,7 @@ ncclResult_t bootstrapClose(void* commState) {
struct bootstrapState* state = (struct bootstrapState*)commState;
if (state->unexpectedConnections != NULL) {
unexpectedFree(state);
if (__atomic_load_n(state->abortFlag, __ATOMIC_RELAXED) == 0) {
if (__atomic_load_n(state->abortFlag, __ATOMIC_ACQUIRE) == 0) {
WARN("Unexpected connections are not empty");
return ncclInternalError;
}