https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67732
Oleg Endo changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Ever confirmed|0
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67732
--- Comment #4 from Oleg Endo ---
(In reply to Kazumoto Kojima from comment #2)
>
> void
> foo (float mat[])
> {
> float *p = &mat[16];
>
> *--p = 0.0; *--p = 0.0; *--p = 0.0; *--p = 0.0;
> *--p = 0.0; *--p = 0.0; *--p = 0.0; *--p =
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67732
--- Comment #3 from Oleg Endo ---
Hm, those addressing mode issues are supposed to be improved by the AMS pass.
But it's not ready yet.
Compiling CSiBE with the current AMS branch, without LRA:
sum: 879 -> 3318319-15560 / -0.466724 %
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67732
--- Comment #2 from Kazumoto Kojima ---
(In reply to Oleg Endo from comment #1)
> There are some improvements with LRA, but the regression in teem-1.6.0-src
>
> total: 1105728 -> 1122288+16560 / +1.497656 %
>
> outweighs all of the i
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67732
--- Comment #1 from Oleg Endo ---
Created attachment 36402
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=36402&action=edit
CSiBE comparison
This is the detailed CSiBE comparision of no LRA vs. LRA.
There are some improvements with LRA, b
