Exact emulation of the Snow effect.

[Weiv]

In my yesterday's test border color indicates current upper page number, paper color indicates active screen (0 - cyan, 1 - white), ink color indicates from what page the snow bytes are fetched.

So we see the test results in the table below. In this table rows mean upper page number, columns mean active screen number, table cells mean from what page the snow bytes are fetched.

Code: Select all

Upper Page     Screen 0(page 5)             Screen 1(page 7)
    1                1                           3 
    3                1                           3
    5                5                           7 
    7                5                           7

Thank you again, @TheMartian !

[TheMartian]

I was thinking this:

As VELESOFT said, in 128K machines we've got the HAL which gives us these equations:

ULA15 = A15 & !A14 | A15 & A14 & !B0 -> '1' if 8000-BFFF or C000-FFFF on RAM 0,2,4,6 -> That's UNCONTENDED RAM.
ULA14 = !A15 & A14 | A15 & A14 & B0 -> '1' if 4000-7FFF or C000-FFFF on RAM 1,3,5,7 -> That's CONTENDED RAM.

These two lines are fed to the contention manager in the ULA, so it works when ULA15 = 0, ULA14 = 1. This allows for using the same old 48K contention manager on 128K machines.
ULA15 is also fed to the PCF, as it is the A15 line for uncontended RAM. (It looks like RAM is placed in the 4164 chips in order 0,4,2,6).

UA15 = A15 & !A14 | A15 & A14 & !B0 & B1 -> '1' if 8000-BF00 or C000-FFFF on RAM 2,6
UA14 = A15 & A14 & !B0 & B2 -> '1' if C000-FFFF on RAM 4,6.

These two are only fed to the PCF, so I guess these are only relevant for uncontended RAM.

VA15 = A15 & A14 & B0 & B1 -> '1' if C000-FFFF on RAM 3,7.
VA14 = !A15 & A14 | A15 & A14 & B0 & B2 -> '1' on 4000-7FFF or C000-FFFF on RAM 5,7

VA15 and VA14 are only fed to the IC30, which is a 74LS157 selector. The output goes to the bit A7 in the RAM chips. VA15 is output on RAS = '0', VA14 is output on RAS = '1'. (I'm assuming that the ULA generates all RAS/CAS signals for contended RAM).

So we have that, during RAS low, A7-A0 to the RAM are (VA15,A6-A0); during RAS high (?) A7-A0 are VA14,A13-A7.

ULA pins DMA7-0 and RAS# are tri-state, so when the ULA is not reading video they are disconnected and the signals in the RAM pins come from PCF/HAL. There are also resistors; the level in the RAM pins is the one from the ULA, not the PCF, in case of conflict.

Now there is something I don't understand: If the snow effect is still caused by having a wrong latched row address (in this case, VA15,A6-A0), why in Ninja Hamster do we keep getting snow from page 1 even if page 3 is in?

[Weiv]

Related: WoS forum “The I register pointing to $40–$7f causes ULA snow and a crash. Or does it?”

Mark

@balford confirmed during running my tests on his machines that on some 128 machines the snow leads to crash, but on some of them it does not.

[1024MAK]

@balford confirmed during running my tests on his machines that on some 128 machines the snow leads to crash, but on some of them it does not.

Did he investigate any further, such as what the mechanism was that actually caused it?

Mark

[Weiv]

Did he investigate any further, such as what the mechanism was that actually caused it?

Mark

As far I know he did not, so did I. Apparently the hardware reasons that lead to snow also lead to crash on some machines. The same test was crashed on one 128 machine and went ok on another one.

[Weiv]

Interesting thing that the snow reasons lead to crash on early +2A/+3 (but without the snow itself) as I found out recently.

[1024MAK]

For the 128K toastrack, what would be interesting, would be to swap the PAL 10H8 (IC29) between a machine that crashes and a machine that does not. Then retesting to see if the results were different or if they are the same.

Also, if a PAL 10H8 is found to be the reason, to try it in a +2 (grey) in position IC7.

Mark

[Weiv]

For the 128K toastrack, what would be interesting, would be to swap the PAL 10H8 (IC29) between a machine that crashes and a machine that does not. Then retesting to see if the results were different or if they are the same.

Also, if a PAL 10H8 is found to be the reason, to try it in a +2 (grey) in position IC7.

Mark

I can only share the videos that @balford made for me:

[Pegaz]

That Balford test on 48k model (second video), shows a very different pattern than my test from a couple of years ago, with plain 48k rubber Speccy issue 3.
It was loaded directly through the audio jack, without any peripherals attached (like divide etc).

btw, in the meantime I tried the latest SpecIde windows build from 07/10 last year and it shows a nice snow effect in Ninja Hamster, as well as in this latest test example.
It may not be quite perfect as in the latest development, but it is very close.
Obviously, at the moment, we don't seem to have an emulator running all these tricky tests.
Spin and SpecEmu support ULA artifacts, but the snow effects arent as good, as in SpecIde for example.
Also, theres a game "Chromatrons Attack", I can't remember which emulator supports it, but it doesn't work well on most of them.
https://csscgc2013.blogspot.com/2013/06 ... ttack.html
There are certainly other examples, but this is the most common difference I notice between good and highly accurate emulators...

[Pegaz]

...and yes, this emulator should support Chromatrons Attack:
https://github.com/ec429/spiffy

[Weiv]

That Balford test on 48k model (second video), shows a very different pattern than my test from a couple of years ago, with plain 48k rubber Speccy issue 3.

It's my modifications of the Hikaru's snow test on balford's videos.

Obviously, at the moment, we don't seem to have an emulator running all these tricky tests. Spin and SpecEmu support ULA artifacts, but the snow effects arent as good, as in SpecIde for example.

My emulator ZX Miracle runs all the snow tests ok of course. Also the newest version of SpecEmu.

Also, theres a game "Chromatrons Attack", I can't remember which emulator supports it, but it doesn't work well on most of them.
https://csscgc2013.blogspot.com/2013/06 ... ttack.html

That's the thing I still can't do for now.

[Pegaz]

My emulator ZX Miracle runs all the snow tests ok of course. Also the newest version of SpecEmu.

Strange, I use Specemu version from 25/03, but Ninja Hamster doesnt show the snow effect...
Snow option is enabled, ofc.

[Weiv]

Strange, I use Specemu version from 25/03, but Ninja Hamster doesnt show the snow effect...
Snow option is enabled, ofc.

It's today's version, I saw only video from it)

[Pegaz]

It's today's version, I saw only video from it)

Thanks, apparently Woody hasn't made it public yet.
btw, I wrote before that newer versions of SpecEmu do not support all ULA artifact test programs.
Most of them works, but these two do not work well, on revisions in the last couple of years.
https://easyupload.io/6jex4p

[Weiv]

Thanks, apparently Woody hasn't made it public yet.
btw, I wrote before that newer versions of SpecEmu do not support all ULA artifact test programs.
Most of them works, but these two do not work well, on revisions in the last couple of years.
https://easyupload.io/6jex4p

Well, my emulator ZX Miracle emulates them a bit better than SpecEmu)

[Pegaz]

That's a good thing.
ZX Spin also emulates this, better than SpecEmu.
I hope SpecIde also gets support for this ULA behavior soon.
Here's what your test looks like on last year's build.
If I understood correctly, Marta has a fixed version, but it must be compiled from source.

[Weiv]

That's a good thing.
ZX Spin also emulates this, better than SpecEmu.

ZX Miracle emulates ULA artefacts better than ZX Spin, actually I think there are no emualtors for now that emulate flash stripes right except mine.

[TheMartian]

I think I'm retiring for some time from the road of the 100% exact snow effect in the upper pages.

The reason is, my 128K Toastrack is acting weird after the testing session yesterday (Not complaining about your test, @Weiv, it was a bit unstable before, but keep in mind that memory addresses come both from the ULA and the PCF *at the same time* with only a resistor in between, so let's better not burn Speccies to build emulators... be careful, my friends )

SpecIde already has 100% accuracy in snow effect in 48K mode, and it is quite close to it in 128K mode. From @Weiv's test and from my analysis of the schematics we've got from which page the data is coming, and I'm uploading my changes for that today. @Pegaz, I'll try to provide Windows builds this weekend.

I'll try to focus on other functionalities I intend to add to the emulator. Betadisk, a decent UI, a debugger, better Amstrad CPC emulation...

[TheMartian]

(I think I've got PCF, ULA, and memory replacement chips, so it should be okay...)

[Pegaz]

SpecIde already has 100% accuracy in snow effect in 48K mode, and it is quite close to it in 128K mode. From @Weiv's test and from my analysis of the schematics we've got from which page the data is coming, and I'm uploading my changes for that today. @Pegaz, I'll try to provide Windows builds this weekend.

I'll try to focus on other functionalities I intend to add to the emulator. Betadisk, a decent UI, a debugger, better Amstrad CPC emulation...

Thank you very much, I really look forward to try the latest build of your great emulator.

[Weiv]

I think I'm retiring for some time from the road of the 100% exact snow effect in the upper pages.

The reason is, my 128K Toastrack is acting weird after the testing session yesterday (Not complaining about your test, @Weiv, it was a bit unstable before, but keep in mind that memory addresses come both from the ULA and the PCF *at the same time* with only a resistor in between, so let's better not burn Speccies to build emulators... be careful, my friends )

I'm sorry if my test damaged your Speccy I was not awared about such possibility.

Seems like nobody will run my tests anymore)

[1024MAK]

Bus contention (where more than one chip is driving a bus) should not damage any chips as long as the ULA and hence the Z80 clocks are running. As any contention will only last for a very, very short time.

And the data bus has resistors which limit the current flow. In any case, normal operation of a 16K/48K/+/128K/+2 (grey) includes the Z80 accessing either ROM or non-contended RAM at the same time that the ULA is accessing the screen RAM to draw the screen.

Mark

[Weiv]

Maybe the long snow time was a reason.

[1024MAK]

So, even if you left the test running for five minutes, the Z80 only outputs data on the bus for approximately 3½ clock cycles and external devices (DRAM) typically only output data for approximately 1 clock cycle when the Z80 is reading from them.

At 3.5MHz, that’s 286ns to 1000ns (1µs) at most before the next instruction cycle/part of a cycle.

The data bus uses 470Ω resistors. Part of the address bus uses 330Ω resistors.

Even in a worse case, with one logic output high at 4V and another low at 0V, with a 330Ω resistor, that’s only 12.12mA. So a power of only 48mW per pin (8 pins 388mW). That’s so small that it would not make any significant difference to a chip.

Mark

[TheMartian]

@Weiv, I don't think your test damaged anything.

My Toastrack was a little bit unstable at the moment, the ULA had had a miraculous recovery so maybe it has some intermittent failure that I've got to look into.

(If something, I hope that my faulty-ish Toastrack hasn't invalidated the test results!)

I'll look into it and it'll be fine in no time.