[hatari-devel] Re: Hatari emulated performance vs Falcon

[Douglas Little <doug694@xxxxxxxxxxxxxx>]

So essentially what I have found from my test results, is this:

1) When both caches are enabled, Hatari emulated performance is the same as with both caches disabled.

2) When just the data cache is disabled (and instruction cache is enabled on its own), performance is considerably higher than a real Falcon under the same conditions.

It seems that having the d-cache bit enabled actually interferes with i-cache performance on Hatari? i.e. it ignores both caches? This must confuse matters when comparing against a real machine, as this is the typical scenario.

Under Hatari I saw 7.07fps with both caches on, 7.08fps with both caches off, and 8.63fps with just the i-cache enabled. On a real Falcon I saw quite different figures for each case - with a significant difference between both caches on and off (and only a small difference with d-cache on/off).

A zipfile is attached containing the results - the '030' jpg's are photos from a real Falcon. The 'hat' jpg's are screenshots from Hatari. The 'nc' indicates 'no caches' and 'ic'/'dc' indicate instruction/data cache enabled for that test.

D.

Attachment: hatarisnaps.zip
Description: Zip archive