pouët.net

1.21 gijacycles?!!?! (megadrive)

category: code [glöplog]
 
For our sega megadrive/genesis prod our z80 is reading about 8.8kb/s off the 68000 memory. That equates to about 1.6K cycles lost off the 68000's clock per ntsc frame do to access. I think that's a lot. Before I go looking at a 68000 manual to tediously average all the cycles I am going to ask you, pouet, how many instructions do you thing 1.6K cycles is?

also, vgm sucks
added on the 2014-02-06 08:07:59 by sigflup sigflup
thing = think
added on the 2014-02-06 08:12:30 by sigflup sigflup
anything from 20 to 400 instructions, probably around 100-200 instructions
added on the 2014-02-06 08:48:53 by britelite britelite
One more issue I learned the hard way: Z80 accessing the 68k bus steals way more 68k cycles on a real megadrive than in emulators. This explains some minor glitches in Overdrive final that are present in emulators but not on real hardware and vice versa.
added on the 2014-02-06 10:22:08 by Kabuto Kabuto
yeah I'm experience a glitch in my production. Sometimes... not often but sometimes I frame skip. I think this is because my vblank code + my main loop + cycles stolen > frame. My code then waits for the next frame. I'm just trying to determine just how many cycles are stolen so I add everything up and prove that the glitch is exactly from this.

btw, kabuto nice work on overdrive
added on the 2014-02-06 11:08:56 by sigflup sigflup
> For our sega megadrive/genesis prod our z80 is reading about 8.8kb/s off the 68000 memory. That equates to about 1.6K cycles lost off the 68000's clock per ntsc frame do to access.

How do you get your 1.6k cycles lost by 68k from 8.8kb/s reading speed of Z80?
added on the 2014-02-06 15:06:21 by lvd lvd
My math could be wrong. I assumed 11 cycles per z80 access. Mask of Destiny has a logic analyzer and has counted about 8-11.

I solved my problem anyway so I'm not that interested. still kinda interesting though
added on the 2014-02-06 15:55:18 by sigflup sigflup
BB Image
added on the 2014-02-06 18:31:05 by Tigrou Tigrou
So what was the problem?
added on the 2014-02-06 19:05:35 by yzi yzi
I think this was (5th post) the problem. Optimizing fixed it, as it should.
added on the 2014-02-06 19:19:01 by sigflup sigflup
sigflup, ok, i've got it. ~11 68k cycles per single z80 access. 10x!
added on the 2014-02-06 21:29:43 by lvd lvd

login