[PATCH]: Add CSE pass directly after expand / Add register equal notes for divmod4 for AVR
Björn Haase
bjoern.m.haase@web.de
Thu Jun 16 09:51:00 GMT 2005
Hi Steven,
> So you are proposing a patch that
> 2. introduces regressions, in other words adds bugs, on x86-64.
I am of course working at this issue and as you might have identified by
reading my statement that there presently is still a regression:
I never did expect the patch to be committed *as is*.
The point why I have posted the patch never the less is: I don't have time to
waste and I refuse to be working dozens of hours on an approach requiring an
early CSE pass in case that there never is any hope to get the approach
integrated into gcc because some folks favorite target does not benefit much
from it.
My present impression is that it is impossible to get any change into the gcc
mid-end tree. So please understand when I ask whether in this case there is
at least a tiny chance that justifies to consider to continue working on an
issue that precisely requires such a mid-end change as a prerequisite.
I would *not* have suggested any change in the mid-end if I had seen any other
possibility for addressing the issue I have described in my post simply by
modifications on the back-ends!!!
As stated in my previous mail: The main issue IMO is not this shitty divmod4
missed optimization for AVR. The main issue is, how to restructure back-ends
after the cc0->CCmode conversion so that one could establish condition code
re-use for instruction sequences after subreg lowering at expand. As
explained this could IMO also help to improve DImode operatations on x86. The
presently used splitters after reload are, IMO, certainly far from ideal.
So, if it helps you, consider my patch to be in fact a
"RFC: Is there any hope to get an early CSE pass running before a
remove_unneccessary_notes() eliminates all of the reg_equal notes for speed
reasons?"
IMO, having a patch to base the discussion on, is the better solution in
comparison to posting a RFC message to gcc .at. gcc .dot. gnu .dot. org.
Yours,
Björn
More information about the Gcc-patches
mailing list