[MIR] Should translator use LLVM builder more directly?

[nagisa]

Currently we use various build::* functions in order to build up the function, but many of the checks done by the build::* functions are not necessary for us.

Namely, those functions are concerned with checking for termination and unreachability. Both of these concerns are not relevant to MIR – we have well specified terminator system and unreachable blocks are removed before MIR gets to the trans.

Suggestion is to refactor the MIR translator to use the trans::builder::Builder directly.

[Thiez]

Would this change result in noticeable performance gains when compiling without optimizations?

[dotdash]

I'd not expect that. It would just avoid some unnecessary, but probably
very cheap, checks during trans, and stop "repositioning" the builder at
its current position over and over again. Probably a tiny gain in the trans
phase, but nothing spectacular. I'd support the change nevertheless,
because it feels cleaner.

2016-01-20 19:27 GMT+01:00 Thiez notifications@github.com:

Would this change result in noticeable performance gains when compiling
without optimizations?

—
Reply to this email directly or view it on GitHub
#31003 (comment).

[pczarn]

I'd like to start this tomorrow.