Chapel: Update text of performance graphs landing page

[bradcray]

The current text of the performance graphs landing page was written at a time that we were just starting to do performance studies and now seems out-of-date. Let's reword it now that it's hosted on chapel-lang.org and we point to it in social media with some regularity.

Somewhat related: Would we want to create a "Performance" sidebar link that contained some of the content of the current PERFORMANCE README (most importantly "use the --fast flag") and also linked to the nightly perf graphs landing page?

TODOs:

• [ ] make PERFORMANCE.md into a webpage (performance.html), linked to from the sidebar
• [ ] have links to the 2 most important configurations from that page
• [ ] make perf-nightly complete in terms of list of machines
• [ ] but boil down text into something short and sweet

[ben-albrecht]

Also related: Seems like PERFORMANCE.md should be migrated to somewhere within https://chapel-lang.org/docs/ , e.g. UsingChapel/Performance

[bradcray]

@ben-albrecht : The reason it's at the top-level today is to maximize the chances that someone will see it and avoid being ignorant about --fast. If we put it onto the web, I'd imagine either retiring it altogether (if it's part of the web hierarchy) or implementing it on the web by putting it under docs/ and using Sphinx to render the webpage.

[ben-albrecht]

If we put it onto the web, I'd imagine either retiring it altogether

Yeah, this what I meant by migrating.

The reason it's at the top-level today is to maximize the chances that someone will see it and avoid being ignorant about --fast.

If we retire PERFORMANCE.md from the top-level, we could still include a pointer to the web page from README.md (among other places) for high visibility

implementing it on the web by putting it under docs/ and using Sphinx to render the webpage.

This is my preference.

[bradcray]

@ronawho: Just to avoid duplicated effort, I've started on an update to the text (not yet PR'd anywhere).

[ronawho]

I was reading through PERFORMANCE.md for an unrelated reason and noticed it's really out of date (it's very shy about single-node perf, says stencil perf is bad, etc.). It also still points to the sourceforge perf landing page

[LouisJenkinsCS]

I would like if it could be a 'performance tip' that on Cray-XC systems like Swan, PrgEnv-gnu is the most optimized, and that programming environments like PrgEnv-cray can prove to be over an order of magnitude slower for certain benchmarks (I've personally experienced this when I forgot to swap the out PrgEnv-cray with PrgEnv-gnu); the only way I would know this would be if either A) I experimented with all types of modules and doing a _lot_ of investigation/digging, or B) am told that PrgEnv-gnu is the fastest.

[LouisJenkinsCS]

Just as a heads up, I created my own one as well: https://gist.github.com/LouisJenkinsCS/c94b0cc998328c1a8a578fd2a805cd67

Of course I don't expect it to be used as the standard/official one, but just in case it may be of any assistance.

[bradcray]

w.r.t. this longstanding issue, I finally got around to putting some performance-focused pages onto the website tonight with lots of help from @ronawho: https://chapel-lang.org/performance.html

I haven't yet removed PERFORMANCE.md from the repository, but I think we could/should now.