rocm-systems/test/metrics.xml

<gfx8>
  <metric name=SQ_CYCLES block=SQ event=2 ></metric>
  <metric name=SQ_WAVES block=SQ event=4 ></metric>
  <metric name=SQ_ITEMS block=SQ event=14 ></metric>
  <metric name=SQ_INSTS_VALU block=SQ event=26 ></metric>
  <metric name=SQ_INSTS_VMEM_WR block=SQ event=27 ></metric>
  <metric name=SQ_INSTS_VMEM_RD block=SQ event=28 ></metric>
  <metric name=SQ_INSTS_SMEM block=SQ event=31 ></metric>
  <metric name=SQ_INSTS_FLAT_LDS_ONLY block=SQ event=33 ></metric>
  <metric name=SQ_INSTS_LDS block=SQ event=34 ></metric>
  <metric name=SQ_WAVE_READY block=SQ event=47 ></metric>

  <metric name=TA_FLAT_READ_WAVEFRONTS block=TA event=101 ></metric>
  <metric name=TA_FLAT_WRITE_WAVEFRONTS block=TA event=102 ></metric>

  <metric name=TCC_CYCLE block=TCC event=1 ></metric>
  <metric name=TCC_REQ block=TCC event=3 ></metric>
  <metric name=TCC_WRITEBACK block=TCC event=22 ></metric>

  <metric name=CPC_ALWAYS_COUNT block=CPC event=0 ></metric>
  <metric name=CPC_ME1_STALL_WAIT_ON_RCIU_READ block=CPC event=8 ></metric>

  <metric
    name=TA_FLAT_WRITE_WAVEFRONTS_sum
    expr=TA_FLAT_WRITE_WAVEFRONTS[0]+TA_FLAT_WRITE_WAVEFRONTS[1]+TA_FLAT_WRITE_WAVEFRONTS[2]+TA_FLAT_WRITE_WAVEFRONTS[3]
  ></metric>
</gfx8>

<global>
  # VWRITE_INSTS
  # The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory.
  <metric name=VWRITE_INSTS expr=1000*(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES ></metric>

  # SFETCH_INSTS
  # The average number of scalar fetch instructions from the video memory executed per work-item (affected by flow control).
  <metric name=SFETCH_INSTS expr=1000*SQ_INSTS_SMEM/SQ_WAVES ></metric>

  # VALU_INSTS
  # The average number of vector ALU instructions executed per work-item (affected by flow control).
  <metric name=VALU_INSTS expr=SQ_INSTS_VALU/SQ_WAVES ></metric>

</global>