What's the diff between X-stage, M-stage and T-stage

Hi, I have a problem about pipeline stage in some metrics.

l1tex__t_bytes_pipe_tex                                                     # of bytes requested for TEX pipe
l1tex__t_output_wavefronts                                                  # of wavefronts sent to Data-Stage from T-Stage
l1tex__f_wavefronts                                                         # of wavefronts processed by L1TEX F-Stage
l1tex__t_requests_pipe_tex                                                  # of requests sent to T-Stage from the L1TEX X-Stage
l1tex__m_xbar2l1tex_read_bytes                                              # of bytes read from L2 into L1TEX M-Stage
  1. Why some interpretation of metrics have T/F/X/M - stage (like last 4 metrics), but others don’t (like first metric). Metrics about wavefronts are related to pipeline stage, but why last metric(byte count) also has stage in its interpretation.
  2. What’s the diff between T-stage, F-stage, X-stage or M-stage?

Thanks in advance!

You can find a lot of info on several of the stages in this GTC video. We will also be extending the information in the kernel profiling guide for a future release.

1 Like

Thanks very much, felix!

I gained a lot of information from this video.
I tested some metrics mentioned in this video. And I have a confusion about naming of metris.

Can you tell me why I test global load, but get output from tex, not l1?