Hi.
Does anyone has experience with JIT for ARM CPUs? Maybe I can get some ideas how to improve my JIT implementation for UAE4ALL.
I have now a first special version of UAE4ALL with JIT (very experimental version, nothing to publish). It based on the aranym version of the m68k core.
Only in some programs which are not using blitter or copper so heavy like demos or games, you get a noteable improvement on my 1 GHz device. For example the benchmark tool SYSINFO gives you a 3.5 times faster result than the regular newcpu-implementation for m68k CPU.
In games or demos, the interleaving of the handling of cpu, blitter and copper is too much. JIT is only useful if you can run a large amount of cycles without interrupt by any custom chip.
For now, I stay on FAME core for UAE4ALL because on Pandoras with 600 MHz, the newcpu core is too slow. With or without JIT.
Regards,
TomB
Does anyone has experience with JIT for ARM CPUs? Maybe I can get some ideas how to improve my JIT implementation for UAE4ALL.
I have now a first special version of UAE4ALL with JIT (very experimental version, nothing to publish). It based on the aranym version of the m68k core.
Only in some programs which are not using blitter or copper so heavy like demos or games, you get a noteable improvement on my 1 GHz device. For example the benchmark tool SYSINFO gives you a 3.5 times faster result than the regular newcpu-implementation for m68k CPU.
In games or demos, the interleaving of the handling of cpu, blitter and copper is too much. JIT is only useful if you can run a large amount of cycles without interrupt by any custom chip.
For now, I stay on FAME core for UAE4ALL because on Pandoras with 600 MHz, the newcpu core is too slow. With or without JIT.
Regards,
TomB
Last edited by a moderator: