Tensor Expression Debug Display (TEDD) #4651

yongfeng-nv · 2020-01-08T05:17:04Z

This is the PR for a Tensor Expression visualization tool (RFC). Any feedback or comment are welcome.

…hedule Tree. 3. Add examples to top level APIs' comments. 4. Top level APIs don't print Dot string by default, unless outputdotstring is True.

yongfeng-nv · 2020-01-08T05:20:06Z

@Hzfengsy, could you help reviewing this PR?

Hzfengsy

Good job! Very impressive work! Can you write a tutorial which may help people to get started with TEDD.

Hzfengsy · 2020-01-08T07:04:24Z

python/tvm/contrib/tedd.py

+
+
+def viz_schedule_tree(sch,
+                      showsvg=False,


Can you please change the argument name to show_svg, dot_file_path and output_dot_string?

I am making these changes.

@Hzfengsy, I have updated the variable names and avoided some lint warnings from the tests. But the test will still fail because of the ci-cpu doesn't have graphviz. I haven't seen response to my question below. Do you have any suggestion? Thanks

I am also working on a tutorial. How about putting it under tutorial/language?

Sounds great! As for the CI test, I would like to skip this test on the CI until the CI updating.

Also, is there a way to allow contributors to modify the docker image on CI? @tqchen
Many new features depend on different environments. It is difficult to change the CI docker every time by only one person.

I have disabled TEDD execution and verification and kept TE construction in the test.

yongfeng-nv · 2020-01-08T16:28:04Z

One question about test failure: https://ci.tvm.ai/blue/organizations/jenkins/tvm/detail/PR-4651/2/pipeline

E ModuleNotFoundError: No module named 'graphviz'

It fails, because tedd depends on graphviz, but tvmai/ci-cpu:v0.54 doesn't have it.
How to fix this issue? Let ci-cpu install graphviz as ci-gpu?

…on lint warnings in the tests.

tqchen · 2020-01-12T03:10:46Z

Thanks @yongfeng-nv . One thing that I think worth considering, as in many viz tools, is the separation of visualization data schema(in this case the perhaps a dom tree or similar kind) from the visualization itself(graphviz).

We can have a tool that extracts the spec into json, then have a tool to take that spec and visualize it

tqchen · 2020-01-12T03:12:06Z

re CI It would be great to add a testing function to see if graphviz exists, if not skip the test

yongfeng-nv · 2020-01-14T23:07:13Z

Thanks @yongfeng-nv . One thing that I think worth considering, as in many viz tools, is the separation of visualization data source specification(in this case the perhaps a dom tree or similar kind) from the visualization(graphviz).

We can have a tool that extracts the spec into json, then have a tool to take that spec and visualize it

@tqchen, I see your point. Let me spec the data source spec.

yongfeng-nv · 2020-01-27T16:58:05Z

Thanks @yongfeng-nv . One thing that I think worth considering, as in many viz tools, is the separation of visualization data source specification(in this case the perhaps a dom tree or similar kind) from the visualization(graphviz).
We can have a tool that extracts the spec into json, then have a tool to take that spec and visualize it

@tqchen, I see your point. Let me spec the data source spec.

@tqchen, @Hzfengsy, I have posted proposal DOM tree in the RFC thread: https://discuss.tvm.ai/t/visualize-tensor-expression/5174/6. Please leave your comments and suggestions.

yongfeng-nv · 2020-02-10T14:59:49Z

@tqchen, @Hzfengsy
I have updated this PR with changes listed in the forum:
https://discuss.tvm.ai/t/visualize-tensor-expression/5174/12?u=maplegu

Two questions about the tutorial:

As graphviz is not always available in CI. Shall I use static images in the tutorial? If so, my current submission uses image locations such as https://github.com/dmlc/web-data/raw/master/tvm/tutorial/tedd_st.png. I haven't committed any image yet. Is it a good location?
I would like to inspect the rendered tutorial. How to view it in my repo?

Hzfengsy · 2020-02-10T15:14:01Z

Thank you for the update.

I think the location on dmlc/web-data is good, but maybe need others' help to commit.

You can render the tutorial at local by

python3 -m pip install sphinx-gallery
cd tvm_path/docs
make html

…(schedule) to dump a json string for the schedule data for visualization. 2. Update tests. 3. Add a tutorial. 4. Add range information to IterVars.

… for DFG. 2. Update tutorial about the InferBound failure.

… tutorial publishing faliure. 2. Fix test about IPython availability check.

yongfeng-nv · 2020-02-11T15:49:39Z

The PR is clear from building/testing failures, after qualifying with @Hzfengsy's help on tutorial.
I am attaching the three static images used by the tutorial. If there is no more comments for me to address, can someone help uploading them as: https://github.com/dmlc/web-data/raw/master/tvm/tutorial/tedd_itervar_rel.png, https://github.com/dmlc/web-data/raw/master/tvm/tutorial/tedd_dfg.png, and https://github.com/dmlc/web-data/raw/master/tvm/tutorial/tedd_st.png? Thanks

yongfeng-nv · 2020-02-18T16:11:02Z

@tqchen, this PR has been clear for a week. It still needs uploading three static images for the tutorial. Can you help moving it forward? Thank you.

tqchen · 2020-02-26T04:21:32Z

Sorry about the delay. Thanks @yongfeng-nv @Hzfengsy this PR is now merged!

* Initial TEDD for publishing. * 1. Fix lint issues. 2. Print intrin.body instead of intrin.name in Schedule Tree. 3. Add examples to top level APIs' comments. 4. Top level APIs don't print Dot string by default, unless outputdotstring is True. * Fix more lint issues. * Update top level API argument names and use raw strings to avoid Python lint warnings in the tests. * Disable TEDD verification, but keep TE construction. * Stop importing tedd to avoid failure. * Separate data extraction and visualization. 1. Add API tedd.dump_json(schedule) to dump a json string for the schedule data for visualization. 2. Update tests. 3. Add a tutorial. 4. Add range information to IterVars. * Update TEDD about InferBound failure. 1. TEDD doesn't call inferbound for DFG. 2. Update tutorial about the InferBound failure. * 1. Import IPython only if SVG is requested. This is required to fix a tutorial publishing faliure. 2. Fix test about IPython availability check.

yongfeng-nv added 2 commits January 7, 2020 23:02

Initial TEDD for publishing.

d24ff82

1. Fix lint issues. 2. Print intrin.body instead of intrin.name in Sc…

bef20dd

…hedule Tree. 3. Add examples to top level APIs' comments. 4. Top level APIs don't print Dot string by default, unless outputdotstring is True.

Fix more lint issues.

5fbd023

Hzfengsy reviewed Jan 8, 2020

View reviewed changes

tqchen added status: help wanted status: need review and removed status: help wanted labels Jan 8, 2020

yongfeng-nv added 3 commits January 9, 2020 18:06

Update top level API argument names and use raw strings to avoid Pyth…

597a0f2

…on lint warnings in the tests.

Disable TEDD verification, but keep TE construction.

58ce33e

Stop importing tedd to avoid failure.

65888da

tqchen self-assigned this Feb 6, 2020

yongfeng-nv force-pushed the tedd branch 2 times, most recently from d964732 to 26b91a1 Compare February 10, 2020 07:18

Separate data extraction and visualization. 1. Add API tedd.dump_json…

0daac78

…(schedule) to dump a json string for the schedule data for visualization. 2. Update tests. 3. Add a tutorial. 4. Add range information to IterVars.

yongfeng-nv force-pushed the tedd branch from 26b91a1 to 0daac78 Compare February 10, 2020 16:32

yongfeng-nv added 2 commits February 10, 2020 13:48

Update TEDD about InferBound failure. 1. TEDD doesn't call inferbound…

bb8752d

… for DFG. 2. Update tutorial about the InferBound failure.

1. Import IPython only if SVG is requested. This is required to fix a…

17d3796

… tutorial publishing faliure. 2. Fix test about IPython availability check.

yongfeng-nv force-pushed the tedd branch from 1c4a6e7 to 17d3796 Compare February 11, 2020 04:01

tqchen approved these changes Feb 26, 2020

View reviewed changes

tqchen merged commit b0b1e7d into apache:master Feb 26, 2020

tqchen added status: accepted and removed status: need review labels Feb 26, 2020

yongfeng-nv deleted the tedd branch February 26, 2020 04:28

yongfeng-nv mentioned this pull request Feb 27, 2020

Images for TEDD tutorial. dmlc/web-data#235

Merged

ZihengJiang mentioned this pull request Sep 17, 2020

TVM v0.7 Release Note Candidate #6486

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Tensor Expression Debug Display (TEDD) #4651

Tensor Expression Debug Display (TEDD) #4651

yongfeng-nv commented Jan 8, 2020

yongfeng-nv commented Jan 8, 2020

Hzfengsy left a comment

Hzfengsy Jan 8, 2020

yongfeng-nv Jan 8, 2020

yongfeng-nv Jan 9, 2020

yongfeng-nv Jan 10, 2020

Hzfengsy Jan 10, 2020

yongfeng-nv Jan 10, 2020

yongfeng-nv commented Jan 8, 2020

tqchen commented Jan 12, 2020 •

edited

Loading

tqchen commented Jan 12, 2020

yongfeng-nv commented Jan 14, 2020

yongfeng-nv commented Jan 27, 2020

yongfeng-nv commented Feb 10, 2020

Hzfengsy commented Feb 10, 2020

yongfeng-nv commented Feb 11, 2020

yongfeng-nv commented Feb 18, 2020

tqchen commented Feb 26, 2020 •

edited

Loading

Tensor Expression Debug Display (TEDD) #4651

Tensor Expression Debug Display (TEDD) #4651

Conversation

yongfeng-nv commented Jan 8, 2020

yongfeng-nv commented Jan 8, 2020

Hzfengsy left a comment

Choose a reason for hiding this comment

Hzfengsy Jan 8, 2020

Choose a reason for hiding this comment

yongfeng-nv Jan 8, 2020

Choose a reason for hiding this comment

yongfeng-nv Jan 9, 2020

Choose a reason for hiding this comment

yongfeng-nv Jan 10, 2020

Choose a reason for hiding this comment

Hzfengsy Jan 10, 2020

Choose a reason for hiding this comment

yongfeng-nv Jan 10, 2020

Choose a reason for hiding this comment

yongfeng-nv commented Jan 8, 2020

tqchen commented Jan 12, 2020 • edited Loading

tqchen commented Jan 12, 2020

yongfeng-nv commented Jan 14, 2020

yongfeng-nv commented Jan 27, 2020

yongfeng-nv commented Feb 10, 2020

Hzfengsy commented Feb 10, 2020

yongfeng-nv commented Feb 11, 2020

yongfeng-nv commented Feb 18, 2020

tqchen commented Feb 26, 2020 • edited Loading

tqchen commented Jan 12, 2020 •

edited

Loading

tqchen commented Feb 26, 2020 •

edited

Loading