This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: more AGI stall detection improvements...
>
> In message <19990416005521.43007@atrey.karlin.mff.cuni.cz>you write:
> > Also I've removed the if handling non-conflicting moves in x86_adjust_cost.
> > I believe it is redundant (if the instructions are non-conflicting, const
> > ought to be 0 anyway) and it was skipping my tests.
> I don't believe simply deleting the test is the right thing to do. The
> purpose of that code was to indicate that anti and output dependencies need
> no cost adjustment.
>
> It turns out that in some cases they do need cost adjustments. I think you
> need to:
>
> 1. record the type of dependency (found in REG_NOTE_KIND (link)).
>
> a. true (read-after-write)
> b. anti (write-after-read)
> c. output (write-after-write)
>
> 2. For all the tests in x86_adjust_cost only perform adjustments for the
> appropriate kind of dependency.
>
> 3. Then you can delete the code at the start of x86_adjust_cost.
>
>
> That way you will not perform adjustents for anti & output dependencies
> except when they are actually needed by your new code.
>
> Otherwise you're adjusting things that you do not want to adjust.
OK thanks for explaining it.
I will come with updated patch.
Honza
>
> jeff
--
OK. Lets make a signature file.
+-------------------------------------------------------------------------+
| Jan Hubicka (Jan Hubi\v{c}ka in TeX) hubicka@freesoft.cz |
| Czech free software foundation: http://www.freesoft.cz |
|AA project - the new way for computer graphics - http://www.ta.jcu.cz/aa |
| homepage: http://www.paru.cas.cz/~hubicka/, games koules, Xonix, fast |
| fractal zoomer XaoS, index of Czech GNU/Linux/UN*X documentation etc. |
+-------------------------------------------------------------------------+