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

[ Thread Index | Date Index | More lists.tuxfamily.org/hatari-devel Archives ]


Very interresting, I look at it immediatly.

Could you send me your benchmark program, so I can launch it and have a look at  the instructions in detail ?

Regards

Laurent


Le 01/02/2013 20:47, Douglas Little a écrit :
[minor correction - typo in one of the MIPS entries]


I decided to implement a more complete test for all permutations of X: Y: P: internal/external access, to see if there are problems specific to one of those memory spaces...

Here are the results (I think they are interesting).

Hatari (latest):

X:INT, Y:INT, P:INT = 16.0 MIPS = 2 cyc = 100%
X:EXT, Y:INT, P:INT = 16.0 MIPS = 2 cyc = 100%
X:INT, Y:EXT, P:INT = 16.0 MIPS = 2 cyc = 100%
X:EXT, Y:EXT, P:INT =  5.3 MIPS = 6 cyc = 67%

X:INT, Y:INT, P:EXT = 16.0 MIPS = 2 cyc = 100%
X:EXT, Y:INT, P:EXT =  8.0 MIPS = 4 cyc = 100%
X:INT, Y:EXT, P:EXT =  8.0 MIPS = 4 cyc = 100%
X:EXT, Y:EXT, P:EXT =  4.0 MIPS = 8 cyc = 75%


Falcon030:

X:INT, Y:INT, P:INT = 16.0 MIPS = 2 cyc = 100%
X:EXT, Y:INT, P:INT = 16.0 MIPS = 2 cyc = 100%
X:INT, Y:EXT, P:INT = 16.0 MIPS = 2 cyc = 100%
X:EXT, Y:EXT, P:INT =  8.0 MIPS = 4 cyc = 100%

X:INT, Y:INT, P:EXT = 16.0 MIPS = 2 cyc = 100%
X:EXT, Y:INT, P:EXT =  8.0 MIPS = 4 cyc = 100%
X:INT, Y:EXT, P:EXT =  8.0 MIPS = 4 cyc = 100%
X:EXT, Y:EXT, P:EXT =  5.3 MIPS = 6 cyc = 100%


...so I conclude there is a problem in Hatari calculating penalties when both X: and Y: are accessed together.

however...  P: access penalty for X: or Y: independently seems to be correct. 

This suggests the problem is not simply caused by external access penalty calculation, but how those values are derived in the first place?

D.






Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/