https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
Jakub Jelinek changed:
What|Removed |Added
Status|ASSIGNED|RESOLVED
Resolution|---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #14 from GCC Commits ---
The master branch has been updated by Jakub Jelinek :
https://gcc.gnu.org/g:2cbd4409bcfaba2bd4200412090fd06db1948369
commit r15-6725-g2cbd4409bcfaba2bd4200412090fd06db1948369
Author: Jakub Jelinek
Date:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #13 from Jakub Jelinek ---
That patch regressed
+FAIL: gcc.dg/pr81192.c scan-tree-dump-times pre "(?n)find_duplicates:
duplicate of " 1
+FAIL: gcc.dg/tree-ssa/cunroll-13.c scan-tree-dump-times cunroll "Exit
condition of peeled itera
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #12 from Andrew Pinski ---
(In reply to Jakub Jelinek from comment #11)
> Created attachment 60064 [details]
> gcc15-pr117927.patch
>
> Untested patch I'm going to test soon.
The only question is vrp is not run at -O1 nor with -fno
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #11 from Jakub Jelinek ---
Created attachment 60064
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60064&action=edit
gcc15-pr117927.patch
Untested patch I'm going to test soon.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #10 from Jakub Jelinek ---
Any progress on this?
Note, the bit_and in there is invalid if prec is not a power of two.
Given that % is generally quite expensive, I think it should be
(if (single_use (@3) && pow2p_hwi (prec))
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #9 from Andrew Pinski ---
Created attachment 59809
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=59809&action=edit
Patch which I am going to test
Note this does not include the testcase changes.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #8 from Andrew Pinski ---
(In reply to Jakub Jelinek from comment #5)
> So expr_not_equal_to (@2, wi::uhwi (32, TYPE_PRECISION (TREE_TYPE (@2)))
> then.
Oh I didn't know about expr_not_equal_to .
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #5 from Jakub Jelinek ---
So expr_not_equal_to (@2, wi::uhwi (32, TYPE_PRECISION (TREE_TYPE (@2))) then.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #7 from Jakub Jelinek ---
And if you go the masking way, it should be single use on the subtraction (or
addition), otherwise it isn't a win from GIMPLE IL POV (trading 2 statements
for 3).
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #6 from Jakub Jelinek ---
s/32/TYPE_PRECISION (type)/
Anyway, if you want to turn it into masking, then it should not covert just the
subtraction, but also addition and the constant operand doesn't have to be
equal to the precision o
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
--- Comment #4 from Andrew Pinski ---
(In reply to Jakub Jelinek from comment #3)
> So
> --- gcc/match.pd.jj 2024-12-05 18:29:46.246399098 +0100
> +++ gcc/match.pd 2024-12-06 09:38:35.670473040 +0100
> @@ -4981,7 +4981,8 @@ DEFINE_INT_AND
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
Jakub Jelinek changed:
What|Removed |Added
CC||jakub at gcc dot gnu.org
--- Comment #3
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
Richard Biener changed:
What|Removed |Added
Priority|P3 |P1
Keywords|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117927
Andrew Pinski changed:
What|Removed |Added
Summary|Invalid rotate optimization |[15 Regression] Invalid
15 matches
Mail list logo
