[Bug]: total_time more than gemm/comm time added up. #264
Description
For example 20, when I do --gemm_only why is the total_ms greater than gemm_ms? Similar behavior for --comm_only.
Fix the timings in relevant examples so it accurately reflects the correct time or identify where the time is being spent?
{
"world_size": 8,
"m": 3840,
"n": 3840,
"k": 4352,
"m_comm": 3840,
"n_comm": 3840,
"debug": false,
"validate": false,
"trace_tiles": false,
"benchmark": true,
"datatype": "bf16",
"output_file": "log.json",
"BLK_M": 256,
"BLK_N": 64,
"BLK_K": 64,
"gsize_m": 6,
"heap_size": 8589934592,
"gemm_sms": 256,
"comm_sms": 64,
"num_ranks": 8,
"csv": null,
"only_gemm": true,
"only_comm": false,
"tflops": 96.64808160566294,
"total_ms": 1.3279694771766664,
"gemm_ms": 0.48436891938012744,
"gemm_experiments": 126
}