Optimize NEON 64-bit integral packets

  • Implement p{mul,max,min} for {s,u}64.
  • ARMv7-only:
    • Use bit manipulation for pabs instead of using GPRs.
    • Implement pnegate as 0 - x instead of using GPRs.
    • Polyfill pcmp_{eq,le,lt} for {s,u}64.
    • GCC only:
      • Manually emit vsub.i64 for negation to prevent software fallback.
Edited by onalante-ebay

Merge request reports

Loading
Loading