| Re: [hatari-devel] 2 more falcon demos running with the new core |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/hatari-devel Archives
]
Hi,
It appears that the dsp56000 communicates with the rest of the computer
(host) with handshakes, interrupts and DMA. The dsp clock is connected
to the 56000 and DMA only. Thus the 56000 is independent of the host,
and uses buffered communication (latched, fifo'd, DMA'd) for host
synchronization.
That's true for the DSP, and I think that's what I've coded.
The DSP as 2 entrance (3 with the external port, but it's out of purpose)..
The Host Interface allows 7 registers (vue from the 68030) + 1 empty
register.
The DSP SSI communicates with the crossbar in both handshake and non
handshake modes.
The DSP clock is 32 Mhz, the cpu clock is 16 Mhz.
Both CPU's are disconnected in clocks, but many demos synchronize them
via the Host interface to exchange datas.
The main problem is with the demos that synchronize one time and then,
transfers datas without syncing again (Cycles timings are really
important here).
The crossbar can run with 2 clocks : 25,175 or 32 Mhz
For the Videl, there are 2 clocks too (25,175 and 32 Mhz)
I don't know if these clocks are independant or if they're from the same
source (in this case, we should externalise the clock from the crossbar
code, and create a special clock function that would call both crossbar
and videl).
Regards
Laurent
Le 18/01/2012 21:40, David Savinkoff a écrit :
On Wednesday 18 January 2012 10:54, Laurent Sallafranque wrote:
Hi,
While testing, 2 more demos appear to run better with the latest WINUAE
core and the latest timings.
- dementia by Aura : it works until the end (I've verified 100 000 lines
of traces, DSP and CPU are perfectly in sync in this demo).
Hi,
I looked at the Falcon schematics because of the latest conversations.
I also took at quick look at Hatari's videl, dsp and crossbar code to
get some background on the topic.
Here are a few of my observations:
It appears that the dsp56000 communicates with the rest of the computer
(host) with handshakes, interrupts and DMA. The dsp clock is connected
to the 56000 and DMA only. Thus the 56000 is independent of the host,
and uses buffered communication (latched, fifo'd, DMA'd) for host
synchronization.
dsp clock: 32 MHz (dsp frequency is less than a multiple of the cpu)
videl clock: 25.175 MHz
Choose North American OR European Falcon:
NTSC clock: 32.215905 MHz (ntsc cpu freq = 32.215905/2 MHz)
PAL clock: 32.084988 MHz (pal cpu freq = 32.084988/2 MHz) (peritel SCART)
David
The sound is low (8 bit DMA stereo may be the problem to look at)
- Fried bits #3 Thing demo by Aura and Digital Chaos : this one never
run after the first black and white screen.
Now it runs the 5 first effects.
The nemaining problems :
- Sound is OK, except during the 2nd effect (probably related to big
cycle instructions), because just after this effect, the music becomes
perfect again.
- the first effect (a coloured tunnel) is displayed strangely (Videl issue)
- the demo freezes after the 4th effect (black screen). Probably DSP/CPU
Timings problems.
I'll play with it to try to solve some problems.
Regards
Laurent