Re: [hatari-devel] Hatari vs Falcon030 timings

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 26/09/2014 15:32, Douglas Little a écrit :
> Here are some more extensive timing comparisons between Hatari 1.8 /
> WinUAE and a real Falcon, for anyone who is interested in that.
>
> More timings will probably follow as I find the time to add them.
> Currently only word access to/from STRAM is tested, except for the DSP
> case. Not all addressing modes are covered. Only a couple of ALU ops
> covered. No pipeline effects are currently tested.
>
> All cases are permuted against relevant CPU cache flags (which is enough
> to show significant differences).
>
>
> Falcon030:
> https://dl.dropboxusercontent.com/u/12947585/bench/nb.log
>
> Hatari:
> https://dl.dropboxusercontent.com/u/12947585/bench/nb-h180-winuae.log

Hi,

that's very interesting, those kind of repeatable per-instruction low 
level benchmarks will be quite useful to compare Hatari with Falcon and 
see if we get some errors only on some specific opcodes.

At the moment, I'm (slowly) updating Hatari's cpu core with the latest 
WinUAE version, so once it's done it will be helpful to check how 
accurate we are (lot of things changed in WinUAE compared to what is in 
Hatari, so it takes more time than just applying a "diff", but so far 
most compilation errors due to amiga specific code are nearly fixed, so 
I hope to have something to test soon)

Nicolas