diff --git a/ext-tuner/example/rccl_tuner_gfx950.conf b/ext-tuner/example/rccl_tuner_gfx950.conf index 8cb0ac6e2b..9a23eaa163 100644 --- a/ext-tuner/example/rccl_tuner_gfx950.conf +++ b/ext-tuner/example/rccl_tuner_gfx950.conf @@ -2,6 +2,7 @@ # Optimization metric: latency_us # Format: collective_type,min_bytes,max_bytes,algorithm,protocol,channels,nNodes,nRanks,numPipeOps,regBuff # This file contains configurations for allgather, allreduce, and reducescatter on gfx950 with 2,4,8 nodes. +# Also, this file contains configurations for broadcast on gfx950 with 2,4,8,16 nodes. # The config for allgather upto 4MB will be bypassed by rccl as rccl runtime # forced direct implementation for msgs upto 4MB allgather,0,1536,ring,ll128,32,2,16,-1,-1 @@ -130,3 +131,51 @@ reducescatter,201326593,402653184,ring,ll128,56,8,64,-1,-1 reducescatter,402653185,805306368,ring,simple,40,8,64,-1,-1 reducescatter,805306369,1610612736,ring,simple,56,8,64,-1,-1 reducescatter,1610612737,17179869184,ring,simple,64,8,64,-1,-1 +broadcast,0,1536,ring,ll,56,2,16,-1,-1 +broadcast,1537,3072,ring,ll,1,2,16,-1,-1 +broadcast,3073,6144,ring,ll,56,2,16,-1,-1 +broadcast,6145,12288,ring,ll,1,2,16,-1,-1 +broadcast,12289,24576,ring,ll128,16,2,16,-1,-1 +broadcast,24577,393216,ring,ll128,1,2,16,-1,-1 +broadcast,393217,786432,ring,ll128,32,2,16,-1,-1 +broadcast,786433,6291456,ring,ll128,64,2,16,-1,-1 +broadcast,6291457,25165824,ring,ll128,56,2,16,-1,-1 +broadcast,25165825,805306368,ring,ll128,64,2,16,-1,-1 +broadcast,805306369,1610612736,ring,simple,48,2,16,-1,-1 +broadcast,1610612737,17179869184,ring,simple,32,2,16,-1,-1 +broadcast,0,1536,ring,ll,8,4,32,-1,-1 +broadcast,1537,3072,ring,ll,32,4,32,-1,-1 +broadcast,3073,12288,ring,ll,1,4,32,-1,-1 +broadcast,12289,24576,ring,ll128,2,4,32,-1,-1 +broadcast,24577,786432,ring,ll128,1,4,32,-1,-1 +broadcast,786433,6291456,ring,ll128,64,4,32,-1,-1 +broadcast,6291457,25165824,ring,ll128,48,4,32,-1,-1 +broadcast,25165825,50331648,ring,ll128,64,4,32,-1,-1 +broadcast,50331649,100663296,ring,ll128,56,4,32,-1,-1 +broadcast,100663297,3221225472,ring,ll128,64,4,32,-1,-1 +broadcast,3221225473,17179869184,ring,simple,32,4,32,-1,-1 +broadcast,0,3072,ring,ll,16,8,64,-1,-1 +broadcast,3073,6144,ring,ll,32,8,64,-1,-1 +broadcast,6145,12288,ring,ll128,2,8,64,-1,-1 +broadcast,12289,24576,ring,ll128,16,8,64,-1,-1 +broadcast,24577,1572864,ring,ll128,1,8,64,-1,-1 +broadcast,1572865,3145728,ring,simple,1,8,64,-1,-1 +broadcast,3145729,6291456,ring,ll,32,8,64,-1,-1 +broadcast,6291457,12582912,ring,ll128,32,8,64,-1,-1 +broadcast,12582913,25165824,ring,ll128,40,8,64,-1,-1 +broadcast,25165825,50331648,ring,ll128,56,8,64,-1,-1 +broadcast,50331649,6442450944,ring,ll128,64,8,64,-1,-1 +broadcast,6442450945,17179869184,ring,simple,32,8,64,-1,-1 +broadcast,0,1536,ring,ll,24,16,128,-1,-1 +broadcast,1537,3072,ring,ll,64,16,128,-1,-1 +broadcast,3073,6144,ring,ll,24,16,128,-1,-1 +broadcast,6145,12288,ring,ll128,16,16,128,-1,-1 +broadcast,12289,24576,ring,ll128,56,16,128,-1,-1 +broadcast,24577,1572864,ring,ll128,1,16,128,-1,-1 +broadcast,1572865,6291456,ring,simple,1,16,128,-1,-1 +broadcast,6291457,12582912,ring,ll,56,16,128,-1,-1 +broadcast,12582913,25165824,ring,ll128,32,16,128,-1,-1 +broadcast,25165825,50331648,ring,ll128,48,16,128,-1,-1 +broadcast,50331649,100663296,ring,ll128,56,16,128,-1,-1 +broadcast,100663297,12884901888,ring,ll128,64,16,128,-1,-1 +broadcast,12884901889,17179869184,ring,simple,32,16,128,-1,-1