[llvm-dev] [NewPM] Purpose of the 'opt' option '-enable-new-pm'

[Markus Lavin via llvm-dev]

Is the ‘opt’ option ‘-enable-new-pm’ intended to be used mostly as a convenience for migrating lit-tests or is it actually intended to be a built-in translation from generic legacy PM command lines into new PM equivalents?

 

That is will ‘-enable-new-pm’ be supported to the same level as ‘-passes’ as an interface for the new PM?

 

-Markus

[Arthur Eubanks via llvm-dev]

`-enable-new-pm` is purely for the new PM transition, ideally at some point it will disappear when all tests are using `-passes`. We'll have to do a mass update of the lit tests at some point after the new PM transition, probably when we start thinking about deprecating the legacy PM (at least for the optimization pipeline).

(Also unrelated IMO having a `-passes` flag instead of specifying each pass as its own flag is a lot cleaner)

_______________________________________________
LLVM Developers mailing list
llvm...@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

[Markus Lavin via llvm-dev]

> `-enable-new-pm` is purely for the new PM transition, ideally at some point it will disappear when all tests are using `-passes`. We'll have to do a mass update of the lit tests at some point after the new PM transition, probably when we start thinking about deprecating the legacy PM (at least for the optimization pipeline).

Makes sense.

> (Also unrelated IMO having a `-passes` flag instead of specifying each pass as its own flag is a lot cleaner)

FWIW, I agree.