Support JSON output format

[zhiyongww]

Hi NCCL Team,

I'd like to share an enhancement I've made to nccl-tests and would appreciate your review.

Thank you!

sample output:

[root@rtptest1621.ftw6 ~]# NCCL_P2P_DISABLE=1 NCCL_DEBUG=WARN ./nccl_alltoall_perf -b 32M -e 128M -f 2 -n 1000 -g 8 --json
[rtptest1621][[31817,1],0][btl_openib_component.c:1704:init_one_device] error obtaining device attributes for mlx5_0 errno says No space left on device
--------------------------------------------------------------------------
WARNING: There was an error initializing an OpenFabrics device.

  Local host:   rtptest1621
  Local device: mlx5_0
--------------------------------------------------------------------------
NCCL version 2.26.6+cudaCUDA_MAJOR.CUDA_MINOR
{"identifier": "nccl-tests structured output", "version": "2.26.6", "benchmark": [{"33554432": {"count": 1048576, "type": "float", "redop": "none", "root": "-1", "oop_time": 1494.278386, "oop_algbw": 22.455275, "oop_busbw": 19.648366, "oop_error": 0.000000, "ip_time": 1459.561667, "ip_algbw": 22.989390, "ip_busbw": 20.115716, "ip_error": "N/A"}, "67108864": {"count": 2097152, "type": "float", "redop": "none", "root": "-1", "oop_time": 2846.555628, "oop_algbw": 23.575462, "oop_busbw": 20.628529, "oop_error": 0.000000, "ip_time": 2849.237796, "ip_algbw": 23.553269, "ip_busbw": 20.609110, "ip_error": "N/A"}, "134217728": {"count": 4194304, "type": "float", "redop": "none", "root": "-1", "oop_time": 5613.751383, "oop_algbw": 23.908741, "oop_busbw": 20.920148, "oop_error": 0.000000, "ip_time": 5616.690443, "ip_algbw": 23.896230, "ip_busbw": 20.909201, "ip_error": "N/A"}}], "out_of_bounds_value": 0, "out_of_bounds_value_status": "OK", "avg_bus_bw": 20.471845, "avg_bus_bw_status": "OK"}

formatted json:

{
  "identifier": "nccl-tests structured output",
  "version": "2.26.6",
  "benchmark": [
    {
      "33554432": {
        "count": 1048576,
        "type": "float",
        "redop": "none",
        "root": "-1",
        "oop_time": 1494.278386,
        "oop_algbw": 22.455275,
        "oop_busbw": 19.648366,
        "oop_error": 0,
        "ip_time": 1459.561667,
        "ip_algbw": 22.98939,
        "ip_busbw": 20.115716,
        "ip_error": "N/A"
      },
      "67108864": {
        "count": 2097152,
        "type": "float",
        "redop": "none",
        "root": "-1",
        "oop_time": 2846.555628,
        "oop_algbw": 23.575462,
        "oop_busbw": 20.628529,
        "oop_error": 0,
        "ip_time": 2849.237796,
        "ip_algbw": 23.553269,
        "ip_busbw": 20.60911,
        "ip_error": "N/A"
      },
      "134217728": {
        "count": 4194304,
        "type": "float",
        "redop": "none",
        "root": "-1",
        "oop_time": 5613.751383,
        "oop_algbw": 23.908741,
        "oop_busbw": 20.920148,
        "oop_error": 0,
        "ip_time": 5616.690443,
        "ip_algbw": 23.89623,
        "ip_busbw": 20.909201,
        "ip_error": "N/A"
      }
    }
  ],
  "out_of_bounds_value": 0,
  "out_of_bounds_value_status": "OK",
  "avg_bus_bw": 20.471845,
  "avg_bus_bw_status": "OK"
}

with -N 2, after formatting json output

NCCL_P2P_DISABLE=1 NCCL_DEBUG=WARN ./nccl_gather_perf -b 32M -e 128M -f 2 -n 1000 -g 8 -d f8e4m3 -N 2 --json

{
  "identifier": "nccl-tests structured output",
  "version": "2.26.6",
  "benchmark": [
    {
      "33554432": {
        "count": 4194304,
        "type": "f8e4m3",
        "redop": "none",
        "root": "0",
        "oop_time": 1172.362777,
        "oop_algbw": 28.621202,
        "oop_busbw": 25.043552,
        "oop_error": 0,
        "ip_time": 1165.302078,
        "ip_algbw": 28.794621,
        "ip_busbw": 25.195294,
        "ip_error": 0
      },
      "67108864": {
        "count": 8388608,
        "type": "f8e4m3",
        "redop": "none",
        "root": "0",
        "oop_time": 2248.297517,
        "oop_algbw": 29.848747,
        "oop_busbw": 26.117654,
        "oop_error": 0,
        "ip_time": 2252.372999,
        "ip_algbw": 29.794738,
        "ip_busbw": 26.070396,
        "ip_error": 0
      },
      "134217728": {
        "count": 16777216,
        "type": "f8e4m3",
        "redop": "none",
        "root": "0",
        "oop_time": 4423.854938,
        "oop_algbw": 30.339541,
        "oop_busbw": 26.547098,
        "oop_error": 0,
        "ip_time": 4435.595384,
        "ip_algbw": 30.259236,
        "ip_busbw": 26.476832,
        "ip_error": 0
      }
    },
    {
      "33554432": {
        "count": 4194304,
        "type": "f8e4m3",
        "redop": "none",
        "root": "0",
        "oop_time": 1154.07066,
        "oop_algbw": 29.074851,
        "oop_busbw": 25.440494,
        "oop_error": 0,
        "ip_time": 1166.898772,
        "ip_algbw": 28.755221,
        "ip_busbw": 25.160818,
        "ip_error": 0
      },
      "67108864": {
        "count": 8388608,
        "type": "f8e4m3",
        "redop": "none",
        "root": "0",
        "oop_time": 2248.849333,
        "oop_algbw": 29.841423,
        "oop_busbw": 26.111245,
        "oop_error": 0,
        "ip_time": 2250.891268,
        "ip_algbw": 29.814352,
        "ip_busbw": 26.087558,
        "ip_error": 0
      },
      "134217728": {
        "count": 16777216,
        "type": "f8e4m3",
        "redop": "none",
        "root": "0",
        "oop_time": 4422.531622,
        "oop_algbw": 30.348619,
        "oop_busbw": 26.555042,
        "oop_error": 0,
        "ip_time": 4435.042864,
        "ip_algbw": 30.263006,
        "ip_busbw": 26.48013,
        "ip_error": 0
      }
    }
  ],
  "out_of_bounds_value": 0,
  "out_of_bounds_value_status": "OK",
  "avg_bus_bw": 25.940509,
  "avg_bus_bw_status": "OK"
}

[AddyLaddy]

Thanks for this suggestion. We had JSON support in some of our internal code so I ported that to keep the repos consistent.
I hope this features meets your requirements.