https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127309
Jue Wang <juewang at linux dot alibaba.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |juewang at linux dot
alibaba.com
--- Comment #1 from Jue Wang <juewang at linux dot alibaba.com> ---
I tested my commit and its parent on a K3 (Spacemit X100), with:
-O2 -march=rv64gc_zicond -mabi=lp64d -mbranch-cost=4 -static
Test Actual branch outcomes Before After
Original 96.138% / 3.862% 8.199s 9.203s
Modified 49.737% / 50.263% 9.790s 9.200s
The modified test changes only 5003 to 60000. Both use a branch before
the patch and czero afterwards.
I think this points to a cost-model issue exposed by the patch.The conversion
can therefore be accepted through cost <= original_cost.
