rocm-systems/tests/workloads_old/vcopy/mi200/pmc_perf_1.csv

1	Dispatch_ID	Kernel_Name	GPU_ID	queue-id	queue-index	pid	tid	grd	Workgroup_Size	LDS_Per_Workgroup	scr	arch_vgpr	accum_vgpr	SGPR	wave_size	sig	obj	SQ_INSTS_SALU	SQ_INSTS_VSKIPPED	SQ_INSTS_SMEM	SQ_INSTS_FLAT	SQ_INSTS_LDS	SQ_INSTS_GDS	SQ_INSTS_EXP_GDS	SQ_INSTS_BRANCH	GRBM_SPI_BUSY	TCP_READ_TAGCONFLICT_STALL_CYCLES_sum	TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum	TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum	TCP_TA_TCP_STATE_READ_sum	TA_BUFFER_READ_WAVEFRONTS_sum	TA_BUFFER_WRITE_WAVEFRONTS_sum	TD_COALESCABLE_WAVEFRONT_sum	TD_LOAD_WAVEFRONT_sum	SPI_CSN_NUM_THREADGROUPS	SPI_CSN_WAVE	CPC_CPC_TCIU_BUSY	CPC_CPC_TCIU_IDLE	CPF_CPF_TCIU_BUSY	CPF_CPF_TCIU_STALL	TCC_NC_REQ_sum	TCC_UC_REQ_sum	TCC_CC_REQ_sum	TCC_RW_REQ_sum	DispatchNs	Start_Timestamp	End_Timestamp	CompleteNs
2	0	vecCopy(double*, double*, double*, int, int)	2	0	0	182745	182745	1046016	256	0	0	8	8	16	64	0x0	0x7fc67c952900	49032	0	65372	32684	0	0	0	16344	31572	0.0	0.0	0.0	32684.0	0.0	0.0	0.0	32684.0	4086	16344	512	47418	1751	0	48.0	385.0	0.0	261614.0	170459144419482	170459569892517	170459569917477	170459152290652