Re: e500 core lacks lswi, lswx, stswi, and stswx instructions.

On 20 Jan, 2015, at 10:10 , matthew green <[email protected]> wrote:
> i don't agree.
> 
> -Os i asking to make small binaries.  you're now avoiding them
> in all cases when a large portion of users don't need this and
> it bloats the binary and slows the run time.
> 
> why isn't this being part of e500 or whatever default enough?
> why do we have to penalise everyone else?

If I understand the change you can put it back by adding -mstring
to -Os.

This seems like a damned if you do and damned if you don't thing.
Having the compiler default to generating code that, with all
the "normal" compiler options, runs on all machines in the architecture
seems like a pretty desirable thing if the costs aren't too high on any
of them.  I think the i386 port makes the compiler generate code and
build libraries for a 486 even though using a more recent target
would have some utility (e.g. 486 floating point has warts that
processors having SSE instructions can avoid) while the arm port,
which went the other way for what I think are very good reasons
and gave one the option to compile for pretty much exactly the
machine you have, attracted some hate mail from pkgsrc people
for other quite sensible reasons.  For user space lowest common
denominator is good unless it is really excessively low (e.g. arm).

Dennis Ferguson