Update RCCL/MSCCL work FIFO depth to 256K (#1091)

src/init.cc

@@ -443,7 +443,7 @@ NCCL_PARAM(AggChannelSize, "AGG_CHANNEL_SIZE", -2);
 NCCL_PARAM(DisableGraphHelper, "GRAPH_HELPER_DISABLE", 0);
 // GDRCOPY support: FIFO_ENABLE when enabled locates a workFifo in CUDA memory
 NCCL_PARAM(GdrCopyFifoEnable, "GDRCOPY_FIFO_ENABLE", 1);
-NCCL_PARAM(WorkFifoDepth, "WORK_FIFO_DEPTH", 64<<10);
+NCCL_PARAM(WorkFifoDepth, "WORK_FIFO_DEPTH", 256<<10);
 enum ncclLaunchMode ncclParamLaunchMode;
 
 

src/misc/msccl/msccl_setup.cc

@@ -20,7 +20,7 @@
 RCCL_PARAM(MscclEnableDoneEvent, "MSCCL_ENABLE_DONE_EVENT", 1);
 #endif
 
-RCCL_PARAM(MscclWorkFifoDepth, "MSCCL_WORK_FIFO_DEPTH", 64<<10);
+RCCL_PARAM(MscclWorkFifoDepth, "MSCCL_WORK_FIFO_DEPTH", 256<<10);
 
 static inline size_t computeSizeNeeded(size_t nBytes, int nScratchChunks, int nChunksPerLoop) {
   return (nBytes * (size_t)nScratchChunks) / (size_t)nChunksPerLoop;