gnupic: Re: [gnupic] Nice to have: GPASM extensions
Subject:
Re: [gnupic] Nice to have: GPASM extensions
From:
David ####@####.####
Date:
21 Mar 2008 17:38:02 -0000
Message-Id: <20080321133437.03bb0b78@DEEPTHOUGHT.BARNET.net>
On Fri, 21 Mar 2008 12:29:15 +0000
Ralph Corderoy ####@####.#### wrote:
>
> Hi,
>
> Mauricio Giovagnini wrote:
> > I think checking the RP0 and RP1 bits itself to check if the code is
> > ok or not can be tricky, and could cause more harm than what it
> > could help.
That should be at least a little bit obvious, because if the assembler
could always tell when you were in the wrong register bank, it could
generate code to set the bits itself and not make the programmer bother
with it.
> What's needed, and I think I've suggested this before, is a way for
> the author to indicate the bank that's intended to always be accessed
> by a particular instruction...
It would be much, much better (but very, very hard) for the assembler
itself to do some slick analysis and determine the possible program
flow. Then it could make guarantees about whether the bank bits were
set correctly or not, and even take care of bank/page issues for you.
Unfortunately, with jump tables and indirection, I don't think there's
any real general solution to the problem, but for most programs it's
probably pretty easy to solve as special cases.
I've gotten excited about trying to solve these problems in the past,
but I've never gotten very far on it.
David
