Re: new gcc compiler for pdp10 architecture

[Rask Ingemann Lambertsen <rask at sygehus dot dk>]

On Thu, Aug 09, 2007 at 04:58:18PM -0700, Martin Chaney wrote:
> Hi,
> 
> I've been working on a version of the gcc compiler for the PDP10
> architecture for some while now and since it's approaching the stage of
> being generally usable, my managers and I have agreed that it's time to
> make its existence known.
  ^^^^^^^^^^^^^^^^^^^^^^^^^
   URL? ;-)

> It may be a good while before things are at a
> state where my changes can be folded back into the general development
> tree,

   Please see <URL:http://gcc.gnu.org/develop.html>. GCC is currently in
stage 2. Stage 3 begins in a month and is scheduled to last two months
before it returns to stage 1 for two months. It would therefore be ideal if
you're ready to submit the changes in about 3-4 months time. However, don't
let that stop you from submitting patches for review even during stage 3.

> but I'd like to start laying the groundwork for that eventuality
> and discussing with the relevant folks how best to capture the PDP10
> dependencies so that my customizations can be limited to the expected
> places. 

   The gcc mailing list is the place for that, except that actual patches go
to gcc-patches.

> The PDP10 is an unusual architecture in many respects and to
> get the compiler to function has required customization through much of
> the code that would normally be considered machine independent.

   It's the one with 36-bit words, isn't it? Let the fun begin. :-)

> My port is currently based on the GCC4.1.1 distribution.  There are a
> handful of testsuite tests that fail that I hope to debug before rolling
> forward to the 4.2.1 release.

   Expect broken testsuite tests which can't pass unless "int" is exactly 32
bits. I've fixed several to work on 16-bit targets and I'd be surpriced if
you don't run into "issues" on the PDP10.

> My experience so far has been that
> advancing the base version can cause significant problems and of course
> one goal of folding my changes back into the base is to alleviate this
> problem.

   For this reason, I suggest you sync up to mainline as soon as possible.
You'll have to do so anyway before you submit pathes for the PDP10. A new
dataflow implementation was merged into mainline during stage 1 and you
should have a look at <URL:http://gcc.gnu.org/wiki/DataflowPorting> when you
run into problems.

-- 
Rask Ingemann Lambertsen