blob: 43f42083f919d5022c83bc048f261af699bb6989 (
plain) (
blame)
1
2
3
4
5
6
7
8
|
MOVMSKPS optimization suggestion for RISC-V
Instead of vsrl by 31 + vmsne
You can use vmslt with x0. Less than zero means sign bit is set so it saves an instruction.
I'd fix it myself but I'm not familiar with XTheadVector and if it has the instruction.
Ditto for movmskpd and pmovmskb
|