答复: [COMMIT] romcc 0.63 and some fixes...

[YhLu]

I wonder it get bigger. Just change the print_debug to printo_debug.

Because call printo_debug is bigger than inline print_debug, or the
__attirbute__ (noinline) doesn't take effective.??

YH

-----邮件原件-----
发件人: [e-mail] [mailto:[e-mail]] 
发送时间: 2004年6月8日 13:39
收件人: YhLu
抄送: LinuxBIOS
主题: Re: [COMMIT] romcc 0.63 and some fixes...

YhLu  writes:
> linuxbios image is 71069 bytes; only 65536 allowed
>
> after change some print_debug to printo_debug, it gets bigger. Before that
> it about 68000.

Ouch.  I guess it must be some of the missed opportunities for
optimization.  Or the increase in moving values around in
registers.

So in theory it works, but we are having a hard time applying it :(

Do you have any significant size regression with just switching to the
new romcc.

Eric

YhLu  writes:
> I wonder it get bigger. Just change the print_debug to printo_debug.
>
> Because call printo_debug is bigger than inline print_debug, or the
> __attirbute__ (noinline) doesn't take effective.??

__attribute__ ((noinline)) is mandatory is it must take effect.

The best way to see the code difference is to diff the two
version of auto.inc

So it looks like something/somewhere gets bigger.  Possibly
the non-inline call does something really with the register allocation
and we get a lot of register shuffling going on.

I have not seen this case yet.  Tracking down this issue is a very
delicate operation.  Somewhere a heuristic sucks and is causing this.

Eric