Wanted: an old SILOS primitive library (pri150ww.lib)

[exxos]

Curious what you meant by it being a problem back then though, what did you run into?

[alexh]

I think ijor's talking about how designs like this could be susceptible to things such as process, voltage and temperature variations where one chip would work and another wouldn't?

The tools they had back then for timing digital circuits probably didn't yield meaningful results when you did things like this and rules had to be disabled hiding potential mistakes?

Netlists don't always contain the information about layout. Often timing fixes needed to be applied to digital designs during layout. They would change cell drive strength, modify track lengths, add buffers etc. to fix issues in the original design

[ijor]

Curious what you meant by it being a problem back then though, what did you run into?

I meant it generally, not specifically to Blitter. But it does also affect Blitter and the ST chipset as well. The small differences between Blitteer versions, or at least some of them, is due to to synchronization issues. Even some of the "bad DMA" issues are caused by MCU, or DMA, poor internal synchronization. Shifter bad Spectrum 512 artifacts, etc. And these are all off the top of my head. Some non Atari chips are even worse.

Note that you can't really blame them. At the time the whole issue wasn't well known or understood. And even if you would want to implement rigorous synchronous design practices, it would have been economically prohibitive back then.

[exxos]

I meant it generally, not specifically to Blitter. But it does also affect Blitter and the ST chipset as well. The small differences between Blitteer versions, or at least some of them, is due to to synchronization issues. Even some of the "bad DMA" issues are caused by MCU, or DMA, poor internal synchronization. Shifter bad Spectrum 512 artifacts, etc. And these are all off the top of my head. Some non Atari chips are even worse.

That tracks with what I'm finding going through the gate level.. plenty of scope for it Yeah bad DMA with CPU can types can trigger all sorts of chaos.. oh lets not go down the DMA hole again

It is really amazing, the more I look into this stuff, the more I stand by that its a miracle these machines ever even worked out the factory door in the first place. I'm really starting to hate the Atari chipset right now...

Note that you can't really blame them. At the time the whole issue wasn't well known or understood. And even if you would want to implement rigorous synchronous design practices, it would have been economically prohibitive back then.

Makes sense too, they were clearly saving gates wherever they could (clock gating instead of proper enables, that kind of thing), and synchronous design costs silicon they didn't have to spare back then.

[ijor]

It is really amazing, the more I look into this stuff, the more I stand by that its a miracle these machines ever even worked out the factory door in the first place. I'm really starting to hate the Atari chipset right now...

There were other chips worse than the ST chipset. What is unusual in the ST, is the clock division cascade (32 MHz -> 16 MHz -> 8 MHz ...)
That was a bit extreme. The clock skew is so much between the 32 MHz SHIFTER clock and the main CPU/GLUE 8 MHz clock, that they can't be considered really synchronous. But that was taken into consideration, or so it seems. Most of the SHIFTER control logic complexity, is precisely to be immune to clock skew.