[llvm] Cache llvm runtime bitcode to reduce start up time in dev mode

[archibate]

Related issue = close #875

I hate debugging script in dev mode, just because this huge long start up time compiling bitcode everytime I restart..

[Click here for the format server]

[archibate]

Feel free to say no without a reason if you don't like this, just give me a walk around to disable these redundant compiling process..
Btw, guess what, I found that I'm becoming the most frequent commiter this week according to https://github.com/taichi-dev/taichi/pulse...

[yuanming-hu]

Feel free to say no without a reason if you don't like this, just give me a walk around to disable these redundant compiling process..

I think maybe we can add an environment variable TI_CACHE_RUNTIME_BITCODE to control this? By default, it shouldn't be cached. Improving your development experience is important, but we also want to simplify the workflow for people who don't care about start up time.

Btw, guess what, I found that I'm becoming the most frequent commiter this week according to https://github.com/taichi-dev/taichi/pulse...

Thanks for everything you have done for Taichi :-)

[archibate]

TI_CACHE_RUNTIME_BITCODE

I'm fine with this approach. But why we neeed compile rtbc each time restart? Can we just let Makefile care this?

[archibate]

Done, btw, I found no test about @ (matmul) in linalg, is this intensional? Also note that z = z @ w seems to fail due to l-value problems.

[archibate]

May I merge this now? I suck the compile time so much that I have to watch bilibili before a program starts, please approve me asap if you feel everything is ok now.

[yuanming-hu]

Thanks. Please feel free to merge once the wording suggestions are applied.

Done, btw, I found no test about @ (matmul) in linalg, is this intensional? Also note that z = z @ w seems to fail due to l-value problems.

Thanks for reporting this! We should add tests and make z = z @ w work.

[k-ye]

This PR has broken the CI, see:

• https://travis-ci.com/github/taichi-dev/taichi/jobs/343699535
• https://github.com/taichi-dev/taichi/runs/744663826
• https://github.com/taichi-dev/taichi/pull/1155/checks?check_run_id=744629126

Please be mindful of [skip ci], abusing it could hide the presubmit errors

[archibate]

Sorry for breaking CI! Maybe I shouldn't add [skip ci] on the last commit of before merge...
Any idea why <filesystem> is not found on OS X? It should be supported since C++17 according to https://en.cppreference.com/w/cpp/header/filesystem, or OS X doesn't support C++17 yet?

[k-ye]

Hmm, my feeling is that it's a Travis CI problem, maybe their clang is stuck at a super old version? Currently the master branch compiles just fine on my local env.

I found this post (https://stackoverflow.com/a/49600206/12003165): <filesystem> used to be in <experimental/filesystem>, so maybe we need to use that on Mac?

Maybe I shouldn't add [skip ci] on the last commit of before merge...

+1...

[Rullec]

2 weeks ago I met with this problem as well. It seems that <filesystem> this header can only be used directly on macOS 10.15 Catalina, instead of my 10.14 Mojave. It's fully about the version. Ye Kuang <notifications@github.com> 于2020年6月6日周六 下午3:52写道：

…

Hmm, my feeling is that it's a Travis CI problem, maybe their clang is stuck at a super old version? Currently the master branch compiles just fine on my local env. I found this post (https://stackoverflow.com/a/49600206/12003165): <filesystem> used to be in <experimental/filesystem>, so maybe we need to use that on Mac? Maybe I shouldn't add [skip ci] on the last commit of before merge... +1... — You are receiving this because your review was requested. Reply to this email directly, view it on GitHub <#1144 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEAYDGXQXDOA2AYIA3LNJZLRVHYT5ANCNFSM4NSZQVVA> .

[k-ye]

It seems that
this header can only be used directly on macOS 10.15 Catalina, instead of
my 10.14 Mojave. It's fully about the version.

Thanks for the info. Yeah I saw that Travis CI is stilling using XCode 10.3, which only covers up to Mac 10.14.6