altanh commented on pull request #7500:
URL: https://github.com/apache/tvm/pull/7500#issuecomment-785329002
I updated the printer to additionally show time spent in the pass itself
(excluding sub-passes), along with percentage relative to the total time. I
described the exact formatting in the Python docstring, but open to changing.
Also, long term I'm planning on exposing the profiling data through the Object
FFI so that users can customize output/analysis, but I'll do that in a separate
PR.
Here's new example output:
```
InferType: 242us [242us] (0.03%; 0.03%)
InferType: 278us [278us] (0.04%; 0.04%)
InferType: 2501us [2501us] (0.34%; 0.34%)
sequential: 1us [1us] (0.00%; 0.00%)
sequential: 678773us [90us] (91.82%; 91.82%)
RemoveUnusedFunctions: 92us [92us] (0.01%; 0.01%)
ToBasicBlockNormalForm: 1219us [1219us] (0.16%; 0.18%)
sequential: 11724us [12us] (1.59%; 1.73%)
InferType: 2573us [2573us] (0.35%; 21.95%)
Legalize: 3059us [738us] (0.41%; 26.09%)
InferType: 2322us [2322us] (0.31%; 75.89%)
InferType: 2619us [2619us] (0.35%; 22.34%)
Legalize: 3460us [865us] (0.47%; 29.51%)
InferType: 2595us [2595us] (0.35%; 75.00%)
InferType: 2783us [2783us] (0.38%; 0.41%)
Legalize: 4525us [2064us] (0.61%; 0.67%)
InferType: 2461us [2461us] (0.33%; 54.38%)
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
