Closed GunnarVB closed 9 months ago
Hi Bebbo I hope you are doing fine.
Please let me allow to point out some possible improvement: The loop construct defined in m68k.md uses cmp.l while cmpa.w would be optimal in this case.
file : gcc/gcc/config/m68k/m68k.md
(define_insn "*dbge_si" [(set (pc) ... if (GET_CODE (operands[0]) == MEM) return "subq%.l bebbo/amiga-gcc#1,%0\;jcc %l1"; return "subq%.l bebbo/amiga-gcc#1,%0\;**cmp%.l #-1,%0**\;jne %l1"; })
here CMP.l is used. while optimal would be "cmpa.w #-1,%0"
Benefit: Size and performance tuning
Hi Bebbo I hope you are doing fine.
Please let me allow to point out some possible improvement: The loop construct defined in m68k.md uses cmp.l while cmpa.w would be optimal in this case.
file : gcc/gcc/config/m68k/m68k.md
here CMP.l is used. while optimal would be "cmpa.w #-1,%0"
Benefit: Size and performance tuning