Re: Bountysource campaign for gcc-vax

["Maciej W. Rozycki" <macro%orcam.me.uk@localhost>]

On Tue, 27 Apr 2021, John Paul Adrian Glaubitz wrote:

> > Since backporting all the changes would probably a huge effort, you have no
> > other choice but to wait, I'm afraid. Maybe the VAX port can be switched
> > over to gcc-11 the moment it has been uploaded to the NetBSD archives.
> 
> gcc-11 just got released today. So it's just a matter of NetBSD importing the
> new version.

 Thank you for the heads-up!

 With GCC 11.1 out and the trunk relieved from the request not to disrupt 
it too much I have committed outstanding fixes for LRA to work with the 
VAX backend.  Therefore the backend is now ready to switch to LRA anytime 
now, cf. <https://gcc.gnu.org/wiki/LRAIsDefault>.  Although I would not 
recommend doing so, because the quality of code produced is awful, with 
LRA much more RISCy in nature than old reload and not currently prepared 
well to handle the complex address modes the VAX architecture, as well as 
other classic CISC ISAs such as PDP-11 or m68k, have.

 One of the changes improves code generation with old reload as well, so I 
have requested approval to have it backported, targetting GCC 11.2.  If I 
get a go-ahead, then you may consider switching to GCC 11.2 when it is out 
(expected in 2-3 months' timeframe, which you'll probably need to use for 
compiler verification anyway).  Otherwise the change is available to 
cherry-pick from trunk as commit c605a8bf9270 ("VAX: Accept ASHIFT in 
address expressions").

 If old reload lives up to GCC 12, then we may add support for the `-mlra' 
option like with some other backends, to let the user choose between LRA 
and old reload.

  Maciej