Re: [hatari-devel] Hatari profiler updates and DSP cycle questions

[laurent.sallafranque@xxxxxxx]

Thanks a lot,

I'll give it a look tonight

Laurent


----- Mail original -----
De: "Douglas Little" <doug694@xxxxxxxxxxxxxx>
À: hatari-devel@xxxxxxxxxxxxxxxxxxx
Envoyé: Vendredi 1 Février 2013 12:00:21
Objet: Re: [hatari-devel] Hatari profiler updates and DSP cycle questions

Hi Laurent, 


Here is the TOS executable (and source) for the DSP test which behaves strangely under Hatari. 


Sorry it's a bit messy - it's really just a hacked up version of Nembench so most of the 68k code is irrelevant / disabled (i.e. ignore everything except dsp_tests routine). 


The DSP tests (68k side) perform a single host write, and a single host read to perform each test - the DSP is 100% busy between these two events, so only DSP instruction time is measured. The busy time is measured against a 200Hz TimerC elapsed count at reasonably high resolution (each test runs for several seconds). 


The DSP ASM source shows each test is identical, only the internal/external memory addresses change for each round... 


Note: it should be trivial to change the test for other instructions, to verify Hatari timings on unusual cases if necessary... I planned to do this anyway in the new Nembench but specific instructions may be of interest for Hatari verification... 


D.