Re: Reload Problem

[Jim Wilson <wilson at specifixinc dot com>]

Konark wrote:
>    I am having a problem in legitimizing a reload
> address.

I assume this means you have defined the LEGITIMIZE_RELOAD_ADDRESS 
macro, and are having problems with it.  This is one of the most 
complicated macros in gcc to define.  Getting it right requires an 
understanding of how reload works.  It may not be worth the trouble.

You haven't provided much useful here info.  You haven't shown us the 
LEIGITMIZE_RELOAD_ADDRESS macro you wrote.  We don't have a copy of your 
port.  We don't have a list of the reloads being generated.  Even if you 
did give us all of this info, we may not be able to help you.  Solving 
reload problems is complicated, and is often more difficult than a 
volunteer can do in their free time.

I can over some suggestions.  Learn a bit about how reload works.  Use 
the -da option to generate dumps, and in particular learn how to read 
the .greg dump file.  This will list instructions that need reloads, and 
the reloads that were generated to solve them.  In your case, the end of 
the file should have a list of reloads that are unsolvable.  You then 
need to figure out how to generate a different reloads that is solvable.

Be sure to look at other ports that define LEGITIMIZE_RELOAD_ADDRESS. 
There probably aren't very many that do, but SH is one, and it handles a 
similar case.

You mentioned that you only have available two registers of the right 
class.  It may be that the instruction needs more than two of these 
registers, in which case this is a lost cause.  You may not be able to 
get this to work in such a limited register class.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com