Re: [avr]: conversion to define_constraint

[Denis Chertykov <denisc at overta dot ru>]

Zack Weinberg <zackw@panix.com> writes:

#> Denis Chertykov wrote:
> > I have converted avr port to define_constraint ...
> 
> Cool.
> 
> (I remind everyone that I don't have approval authority anymore, but ...)
> This patch also converts some #defines and some magic numbers to
> a (define_constants [...]) form.  It would ease review if you did that in
> a separate patch.  I don't see any intrinsic reason why they have to go
> together.

Because 'converts some #defines and some magic numbers' too simple.
I think that I needn't in rewiev I'm already committed the patch.

> You need to update doc/md.texi::Machine Constraints to point readers at
> config/avr/constraints.md instead of config/avr/avr.h.  (Or if there is
> not already an AVR subsection, add one, using the docstrings you
> wrote.)

md.texi already have an AVR section.
I will change avr.h to constraints.md.

> You put the bulk of the logic for constraint "Q" in an avr.c function, which
> is called only from the constraint definition.  Would you prefer to be able
> to write a C code block in the constraint definition, as can be done for
> predicates?  It would not be hard to add support for this to genpreds.
> [I am willing to code this but not till early April.]

I don't like a big C blocks inside RTL because of difficulties with
debugging.
I will convert an AVR port to define_predicate later.

> Other than that looks good, though I have not close-read the conversion
> for semantic equivalence or anything like that.

I have tested the patch with my internal testsuit based on gcc/testsuite/...

While I have committed the patch I have chenged wrong ChangeLog file.
I have changed svn+ssh://denisc@gcc.gnu.org/svn/gcc/trunk/ChangeLog
instead of svn+ssh://denisc@gcc.gnu.org/svn/gcc/trunk/gcc/ChangeLog

Please suggest me what must I do now ?

Denis.