This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Retargetting GCC for fictitious processors
- From: Mike Stump <mrs at apple dot com>
- To: Marty Hauff <marty dot hauff at rmit dot edu dot au>
- Cc: gcc at gcc dot gnu dot org
- Date: Thu, 6 Mar 2003 14:23:07 -0800
- Subject: Re: Retargetting GCC for fictitious processors
On Thursday, March 6, 2003, at 01:17 AM, Marty Hauff wrote:
In order to do this I need to be able to tweak the machine
descriptions etc. and reasonably easily (one day automatically)
rebuild the compiler for the new machine. I have considered other
compilers however GCC seems to offer the best mix of maturity vs
accessibility.
As I have quickly found out, however, retargeting GCC is not just a
single afternoon affair - It wants your life!
My questions to the group are:
* Does anyone have any minimalistic machine descriptions that might be
useless in a practical sense but at least make it through a full
compilation.
The way to do this, is to pick a nice, normal processor that is closest
to the target processor you are interested in and just modify it as
appropriate.
For example, you want to test 8 gen regs, or 64 general regs, take
rs6000.{md,c,h} and edit in the changes, about 20 lines if you're
lucky, and in five minutes, you're done.
Compare this with, 2 months to do a port, and then 5 minutes to tweak
it.
If it doesn't build and generate code that is close, your very last
change is wrong, and if you just stare at those 10 lines, maybe it will
hit you, which is better than doing a port, and then wondering why it
doesn't work.