Wypowiedź Hemiyody na temat przeróbki i pomiarów na oscyloskopie:
"The DBR* line goes low when the Agnus DMA accesses to the chip ram and up again when it does not. This hack distributes the RAS signaling to the correct membanks depending on the A23 and DBR*.
This hack also needs to consider the RAS refresh that happens on every rasterline, otherwise the memorychips will forget what they have stored. It seems that when it is time for this refresh, both RAS0 and RAS1 goes low. ( RAS is an active low signal btw. ) It it also seems that this is the only time both RAS:es are low, critical for this hack to work.
I hooked up my ancient scope to notice that the DBR* and RAS accesses were not aligned at all. (probably because of long propagation delays inside the agnus..) So basicly there is not a big chance for this hack to work as it is in the finnish magazine imho.
The DBR* needs to be aligned and I did it with 8 OR-gates in series. ( 74LS32, that was the slowest chips I had at hand )
A bit weird that this hack still works at all. I thought Pekka had taken the wrong RAS(0) to the signal he calls RAF0 in the schematic, I think it has to be RAS1 . But there is so much inverting logic going on so I may have missed something. (quite optimized I have to say..)
First screenshot
First screenshot, lower line = original DBR*, upper delayed DBR (closer to 100ns delayed)
Second screenshot
Second screenshot shows delayed DBR* in relation with chipmem RAS (display refresh). (It may have to be delayed a bit further still?) I'll post a timing diagram of a generic dram chip later..
Third screenshot
Third, test with some load on the ram, mainly display updates on first 512kB bank, mod playing on 2nd 512kB bank, Powerpacker doing some crunching in 512kB slowfast mem.
Fourth screenshot
Fourth, "the delay bodge"."
I wypowiedź z tego samego wątku @NMI:
It's just that the propagation delay on the RAS-signals must be kept to an absolute minimum.
That's also why "74F"-logic should be used, or even better, a fast PAL or GAL.
I intend to try and implement this using a CPLD, if I get the time...
Anyway, here is a variation of the hack, with a SR-latch instead of the delay line:
kliknij