[gfx950] Make bypassing __threadfence the default for multinode. (#1947)

* Gate based on ROCM version, safe for ROCm 7.0.2 and beyond.
* Updates naming to gfx9CheapFenceOff since we use this for gfx942 and gfx950.  Thanks Nilesh.
* Add info logging statement to NCCL_INIT to print whether enabled when INFO logging is enabled.

[ROCm/rccl commit: c70f5b4621]
Dieser Commit ist enthalten in:
alex-breslow-amd
2025-10-15 09:15:36 -07:00
committet von GitHub
Ursprung 6d151d4e21
Commit 455d516dc4
5 geänderte Dateien mit 12 neuen und 11 gelöschten Zeilen
+1 -1
Datei anzeigen
@@ -875,7 +875,7 @@ public:
patBarrier();
}
if(collWork){
skip_fence = !collWork -> gfx942CheapFenceOff;
skip_fence = !collWork -> gfx9CheapFenceOff;
}
}