From: Nils Wallménius via rockbox-dev <rockbox-dev_at_cool.haxx.se>
Date: Wed, 8 Apr 2020 14:27:02 +0200

Den sön 5 apr. 2020 15:16Solomon Peachy via rockbox-dev <
rockbox-dev_at_cool.haxx.se> skrev:

> On Sun, Apr 05, 2020 at 09:23:49AM +0200, Marcin Bukat via rockbox-dev
> wrote:
> > sh toolchain used to have serious bug in jumptable generation. I may dig
> up
> > test case where it was easy to verify in disasm. This was showstopper
> long
> > time ago when I considered upgrade sh toolchain.
>
> If you could dig up that testcase, it would be quite useful!
>
> I couldn't find any definitive proof that upstream GCC fixed this bug
> (closest I could find is [1]) at some point over the past decade or so,
> but I figured I'd start with a clean slate and only forward-port the
> workaround patch if it turned out to be necessary. And then pester the
> GCC folks about it. :)
>
> [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=4516
> (Closed in 2007 as FIXED before GCC 4.2.3 was released)
>
> - Solomon
> --
> Solomon Peachy pizza at shaftnet dot org
> High Springs, FL ^^ (email/xmpp) ^^
> Quidquid latine dictum sit, altum videtur.
>

Hi, been away from rb for a couple of years but one thing that might be
worthwhile is to "re tune" the compiler flags for the codecs when changing
toolchain. Basically just build the whole bunch with O[123] and run the
test files through test_codec for each arc. Of course it's not necessary up
front but would make it fairly easy to spot serious perf regressions and to
get any benefits from better optimizers.

Glad to see people still working on it.

BR
Nils

>
Received on 2020-04-08