Merge pull request #775 from nusislam/unroll-1

device: use unroll factor based on platforms

[ROCm/rccl commit: fd6553bcf8]

projects/rccl/src/collectives/device/common.h

@@ -11,7 +11,12 @@
 #include "collectives.h"
 #include "devcomm.h"
 
+#if defined(__gfx940__)
 #define COLL_UNROLL 4
+#else
+#define COLL_UNROLL 2
+#endif
+
 #define NCCL_MAX_DEV_ARITY (NCCL_MAX_TREE_ARITY-1)  // Using balanced tree instead of split tree
 
 #define __syncwarp()