Would someone with a Quadra 700/900/950 be willing to test a program for me?

[dougg3]

Q700 with version 2.

Thanks for testing this again so quickly! This confirms that the Mac OS leaves the Q700’s $F29 register at 0 by default even though it doesn’t on the Q950. I have a feeling I might need to look at the $F09 register as well, at least on revision $B0. Maybe it’s the reason the IRQ is still firing even when I turn it off in $F29.

We are definitely seeing evidence of a repetitive EASC interrupt on the Q700 that doesn’t fire constantly, but does seem to keep going forever. But not on on the 950. Fascinating!

Maybe I need to print out all register values as part of the initial test. The Sound Manager must be setting up the 700’s EASC differently than the 950’s.

 

[cgp]

Thanks! That matches Callan's results, so there's definitely something different about the 700. If you would be willing to re-test with the new version as well, that would be awesome too!

Here's version 2 on my Q700 (with 7.6.1). Note not matching Callan's this time:

 

[dougg3]

Here's version 2 on my Q700 (with 7.6.1). Note not matching Callan's this time:

You're right! The difference is interesting. Yours has register $F29 as $01 when we start the test, which is what I would expect and what I've seen on every other model. Whereas Callan's is $00, which I've never seen before. It might be 7.6.1 versus a different OS version. I should probably print the system version in the test output.

The (3 2) versus (3 1) doesn't concern me a whole bunch. The 2 just means that two half-empty IRQs fired in succession really fast during your test, but not Callan's. The total count was still 3.

The interesting thing about this output from both 700s is that 237 + 3 = 240. The time that I wait during this portion of the test is 4 seconds, which is 240 ticks. It's almost like a sound IRQ is firing once per vertical refresh. I did observe in MAME while emulating a Quadra 700 that if I messed around in the Sound control panel a bit, I got it to spam writing 1 followed by 0 to register $F29 a bunch, which should cause the IRQ handler to re-fire. I should check to see if that code is originating from the VBL interrupt.

I am wondering if Apple has some kind of IRQ workaround in the Sound Manager that's implemented for the 700 but not for the 950. I'll keep investigating and thinking, and will eventually come up with a third version of the test to try.

Thank you both again for performing these tests!

 

[Callan]

I actually ran the test on two different quadra 700's, but i used the same boot image. The results were the same on both. I can try running the app on different os versions if you want me to. The 950 was running os 8 I believe (will check tonight).

 

[cgp]

And I also see slight differences across OS versions.
7.1 gives:

And 7.5.5 yields:

 

[dougg3]

I can try running the app on different os versions if you want me to. The 950 was running os 8 I believe (will check tonight).

Thanks! I think I need to brainstorm and research a bit, and make a next version of the app before I can make much more sense of any additional results. Both you and cgp have demonstrated that we're probably getting an ASC interrupt every VBL tick, which I didn't expect and I think may mean that Apple's Sound Manager code is interfering with what I'm trying to check.

And I also see slight differences across OS versions.

Ahh, that's odd! The VIA2 test should always return $0000 on the 700. My guess is that something happened to change in one or more of the VIA2 registers in the middle of the repetition test, and screwed up the results. I have a feeling if you run it a few more times on 7.1 it will usually be $0000.

I'll probably need to change the test to read that address space multiple times until I don't see a change, in order to ensure consistency in the readback.

Out of curiosity, in System 7.1, if you go into the Sound control panel and spam click the Wild Eep sound 10 times, and then run ASCTester afterward, is it still F29 1 ($01) or does it change to F29 1 ($00) like Callan's?

 

[Andy]

Here's results from a PowerBook 180

 

[cgp]

Ahh, that's odd! The VIA2 test should always return $0000 on the 700. My guess is that something happened to change in one or more of the VIA2 registers in the middle of the repetition test, and screwed up the results. I have a feeling if you run it a few more times on 7.1 it will usually be $0000.

It is random. I do see $0000 most often. And while repeating version 1, I also got $01FE and $01AC each on one occasion.

Out of curiosity, in System 7.1, if you go into the Sound control panel and spam click the Wild Eep sound 10 times, and then run ASCTester afterward, is it still F29 1 ($01) or does it change to F29 1 ($00) like Callan's?

Yep, you're correct!

 

[dougg3]

Here's results from a PowerBook 180

Awesome, thanks! Looks just like the Q950.

It is random. I do see $0000 most often. And while repeating version 1, I also got $01FE and $01AC each on one occasion.

Thanks! That confirms my suspicions. I'll attempt to fix this in the next version.

Yep, you're correct!

By correct, I'm assuming you are confirming it changes to $00 instead of $01. This matches what I see in my patched closer-to-real-EASC-behavior MAME simulating a Quadra 700 and System 7.1. In fact, if I run ASCTester twice, it prints $01 the first time but $00 the second time.

I haven't located where/why/when it gets installed, but System 7.1 seems to have code that runs during every VBL on the Quadra 700 that toggles $F09 and $F29 to 1 and then back to 0. I think the intent is to force the sound IRQ handler to run repeatedly. It doesn't happen when I simulate a 950 or a PowerBook 140; only a 700. So I think it's something they did intentionally for that model. And it's not enabled at startup; only after I play a few sounds in the Sound control panel or run ASCTester.

My money is on Callan's Q950 and Andy's PB180 being real indicators for how the EASC's interrupt works: a single IRQ whenever the FIFO reaches half empty, or whenever you toggle $F29 to 0 while the FIFO is already half empty. Probably $F09 as well for channel A; I'll have to test that in the next version. When I patch MAME to cause the EASC to behave like that, I get a similar result with exactly 237 empty IRQs, although only 2 half-empty IRQs instead of 3.

Is the Quadra 900 super rare? I hear more about the 950 than the 900. MAME shows the Quadra 900 having the same periodic IRQ behavior, so I have a feeling it's a special Sound Manager workaround implemented just for those two machines. I bet we would see the same repeating IRQs on the 900.

 

[cgp]

By correct, I'm assuming you are confirming it changes to $00 instead of $01.

Correct again.