Add support for external network.

Dynamically load external network from libnccl-net.so.
Add init function in networks.
Move PCI scoring to net.cu, only ask transport to provide a path.
Simplify CUDA PCI path detection.
Add dummy external network


[ROCm/rccl commit: 0d3a20f96d]
Esse commit está contido em:
Sylvain Jeaugey
2018-11-13 10:37:20 -08:00
commit 378a103248
29 arquivos alterados com 437 adições e 361 exclusões
+3 -3
Ver Arquivo
@@ -271,7 +271,7 @@ struct ncclComm {
while (ret == -1) { \
SYSCHECKVAL(call, name, ret); \
if (ret == -1) { \
INFO(ALL,"Got %s, retrying", strerror(errno)); \
INFO(NCCL_ALL,"Got %s, retrying", strerror(errno)); \
}\
} \
} while (0);
@@ -313,7 +313,7 @@ struct ncclComm {
ncclResult_t res = call; \
if (res != ncclSuccess) { \
/* Print the back trace*/ \
INFO(ALL,"%s:%d -> %d", __FILE__, __LINE__, res); \
INFO(NCCL_ALL,"%s:%d -> %d", __FILE__, __LINE__, res); \
return res; \
} \
} while (0);
@@ -322,7 +322,7 @@ struct ncclComm {
res = call; \
if (res != ncclSuccess) { \
/* Print the back trace*/ \
INFO(ALL,"%s:%d -> %d", __FILE__, __LINE__, res); \
INFO(NCCL_ALL,"%s:%d -> %d", __FILE__, __LINE__, res); \
goto label; \
} \
} while (0);