[API] HLS Report Enhancement

[yn224]

This PR attempts to improve upon the HLS report interface display, specifically targeting the latency information it returns.

Proposed interface:

...
report = f.report()
report.display( loops, levels, cols )

Here, loops represent an array of loop names, levels represent the maximum loop nest levels the report should print, and cols represent an array of latency category names.

An example output

+-----------+--------------+---------------+---------------+-------------------------+-------------------------+---------------+------------------+
|           |   Trip Count |   Min Latency |   Max Latency |   Min Iteration Latency |   Max Iteration Latency |   Pipeline II |   Pipeline Depth |
|-----------+--------------+---------------+---------------+-------------------------+-------------------------+---------------+------------------|
| A_x       |          410 |     104390920 |     105399520 |                  254612 |                  257072 |           N/A |              N/A |
| + A_y     |          615 |        254610 |        257070 |                     414 |                     418 |           N/A |              N/A |
| ++ A_ra0  |            5 |           410 |           410 |                      82 |                      82 |           N/A |              N/A |
| +++ A_ra1 |            5 |            80 |            80 |                      16 |                      16 |           N/A |              N/A |
| Y_x2      |          410 |      38831920 |      39840520 |                   94712 |                   97172 |           N/A |              N/A |
| + Y_y1    |          615 |         94710 |         97170 |                     154 |                     158 |           N/A |              N/A |
| ++ Y_ra4  |            3 |           150 |           150 |                      50 |                      50 |           N/A |              N/A |
| +++ Y_ra5 |            3 |            48 |            48 |                      16 |                      16 |           N/A |              N/A |
| X_x4      |          410 |      38831920 |      39840520 |                   94712 |                   97172 |           N/A |              N/A |
| + X_y2    |          615 |         94710 |         97170 |                     154 |                     158 |           N/A |              N/A |
| ++ X_ra2  |            3 |           150 |           150 |                      50 |                      50 |           N/A |              N/A |
| +++ X_ra3 |            3 |            48 |            48 |                      16 |                      16 |           N/A |              N/A |
| R_x6      |          410 |       6052420 |       6052420 |                   14762 |                   14762 |           N/A |              N/A |
| + R_y3    |          615 |         14760 |         14760 |                      24 |                      24 |           N/A |              N/A |
+-----------+--------------+---------------+---------------+-------------------------+-------------------------+---------------+------------------+
* Units in clock cycles

The current report interface only includes information about resource usage of the HLS simulation. In order to get more information regarding the HLS report (such as latency values), one has to traverse down the chain of directories created by the simulation, which can be cumbersome. However, by adding this new class rptdisp in this PR, we can also display latency information in addition to the default report that gets generated currently.

Possible limitations The proposed report interface enhancement does not support programs without any loops. Also, querying via regular expressions is unavailable as of now.

[hecmay]

It seems that the CI container is container circleci/python:3.6. And you can change it to 3.7 to fix the error.

[seanlatias]

It seems that the CI container is container circleci/python:3.6. And you can change it to 3.7 to fix the error.

I'm confused. Why do we need to change CI env? It seems the error is because the test is terminated.

[seanlatias]

Or more precisely, the test is canceled.

[seanlatias]

Make sure you merge or rebase the upstream. The CI Python env version is changed here. If the CI env needs to be changed, please file a PR instead.

https://github.com/cornell-zhang/heterocl/pull/323

[hecmay]

Or more precisely, the test is canceled.

The previous CI test failed with an error complaining that python 3.7 is required by not found.

I was trying to rerun the CI to pinpoint why. But I suddenly found the issue. So I cancelled it.

[seanlatias]

@yn224 can you change the comment format according to here?

[seanlatias]

Please update this file to include required packages. Also please do the following two things:

1. Add unit tests.
2. Extend your current PR texts to include: a) The expected outputs for your API and b) The main purposes/changes for the files you changed.

[yn224]

@seanlatias I'm not too sure about these TVM errors that are getting raised

[seanlatias]

We don't have vivado_hls on CI. That's why I suggested you directly test on an existing report file instead of generating one.

[seanlatias]

@yn224 Please let me know if you're ready for review.

[yn224]

@seanlatias I think it can be reviewed now.

[yn224]

This is just the first round. Please fix these first and I'll do a second round.

@seanlatias I believe all of your comments are resolved. Further idea I have about testing includes querying with invalid column names and loop names. But they currently seem to raise IndexError, which I still need to investigate more.

[seanlatias]

Great, I'll take a look.