Re: [Sdcc-user] Z80 code size reduction

[Harley L. <los...@gm...>]

On Sun, Apr 10, 2011 at 4:36 AM, Philipp Klaus Krause <pk...@sp...> wrote:
> Am 08.04.2011 16:54, schrieb Harley Laue:
>
>> I think I may have found one. I'll attach the code. It appears it
>> doesn't like one of my macros. It works in sdcc (I don't have it in
>> there, but you can add --opt-code-size and any other options you'd
>> like to the CC line in src/Makefile) Just FYI, it's
>> --optralloc-exact-cost that is the breaking option it looks like.
>
> Thanks. It's fixed in the optralloc branch now.
>
> Philipp

Indeed it seems to be :) Now that it works I can elaborate a bit more
on the branch. My code has about a 1% code reduction. I'm not terribly
surprised since it's very much optimized for speed and all the low
level hardware interaction is via macros that inject the code in quite
a few places making it hard to optimize. Then the program itself isn't
very large yet, and extensively uses the library macros. As the
program progresses though, I suspect more of the code will be able to
be optimized since most of the parts that need the low level macros
are in place. So for as small as the program is, 1% isn't bad.

> --max-allocs-per-node : Higher values result in better code, at the cost
> of longer compile time (and higher memory usage during compilation). The
> default value is 10000.

I was a bit surprised however just how much of a difference this
option made on compile time and memory usage with a much high number
(1000000.) From seconds to minutes.

Overall, I think it looks like some great work and will probably prove
useful when I get to the point where the code + data size is getting
close to 32k as I try to avoid using a mapper. :)