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

src/include/collectives.h

@@ -10,7 +10,7 @@
 #define NCCL_COLLECTIVES_H_
 
 #include "nccl.h"
-#include "nccl_common.h"
+#include "nccl_tuner.h"
 #include "device.h"
 
 #define NCCL_MAX_NET_SIZE (1024*1024*1024L) // Rather than send INT_MAX which is 2G-1, send a power of two.
@@ -25,11 +25,17 @@
 #define ALLGATHER_SLICESTEPS (NCCL_STEPS/4)
 #define ALLGATHER_SLICESTEPS_SINGLE_NODE (NCCL_STEPS/2)
 #define ALLGATHER_CHUNKSTEPS (NCCL_STEPS/2)
+#define ALLTOALL_SLICESTEPS 1
+#define ALLTOALL_CHUNKSTEPS 1
 #define REDUCESCATTER_SLICESTEPS (NCCL_STEPS/4)
 #define REDUCESCATTER_SLICESTEPS_SINGLE_NODE (NCCL_STEPS/2)
 #define REDUCESCATTER_CHUNKSTEPS (NCCL_STEPS/2)
 #define BROADCAST_SLICESTEPS 1
 #define BROADCAST_CHUNKSTEPS 1
+#define GATHER_SLICESTEPS 1
+#define GATHER_CHUNKSTEPS 1
+#define SCATTER_SLICESTEPS 1
+#define SCATTER_CHUNKSTEPS 1
 #define REDUCE_SLICESTEPS 1
 #define REDUCE_CHUNKSTEPS 1
 #define NCCL_MAX_SLICE_PER_CHUNK 2  // max value for CHUNKSTEPS/SLICESTEPS, must accord with above