YM2612 Envelope generator question

[vecna]

So, I'm working on a YM2612 core based pretty much exclusively on the information in this forum (I'm trying very hard to avoid referencing MAME). When I was first getting the basic shell functioning, I made an extremely basic test rom that played one note:

* Alg#7, only on operator 4 - should be pure sine
* AR $1F, no attack phase. KS=0.
* DR=4
* SL is 4, so normalized to a 10 bit attenuation value, it should switch to sustain at an attenuation value of 128.
* SR is $10 so once it switches to sustain the note ends very rapidly.
* My TL is $13, though I don't believe this should matter.

The issue I noticed was that the note played noticeably longer in my emulator than in KEGA. I kept looking for an error in my code and never found it - It appears to correct based on the information in page 8 about the envelope generator.

In my emulator it took ~7.4 seconds from key-on to the sustain phase.
In KEGA, Regen, and in MAME (genplus-gx version), it took ~5 seconds (to be exact, it took 87040 EG cycles in genplus-gx).
Interestingly enough, in GENS it took ~4.5 seconds.

So, here's my math.
DR=4,KS=0, means my Rate is 8.
That means my shift value is 9, so it takes 512 EG cycles per EG update tick.
At rate 8, there is a total of 4 attenuation gain per 8-cycle group.
SL=4 means we need to gain 128 attenuation.
So it should take (128/4)*8*512 EG cycles to complete the decay phase. That works to 131072 EG cycles or 7.38 seconds.

I tested on my Everdrive and the result was... it took 5 seconds.

Here is my test ROM. https://dl.dropbox.com/u/2253413/simpletest.bin

I just now tested on some version of Exodus that I found buried in these forums somewhere, and it took ~5 seconds. I'm assuming that Exodus should match the behavior described on page 8, so I assume that the information is correct but I'm just misunderstanding it. But I'm totally at a loss as to what I'm not understanding. The rate calcs and example code seem quite clear.

Any guesses at what I'm missing? What part of my math is wrong?

[vecna]

Bleh! Lying in bed, trying to fall asleep... of course the answer comes to me.

The rate is not 8.

RKS is not an abbreviation for Rate/KeyScale. The rate is 10, and then the math all works out.

I'm dumb.

I can't delete the post so I'll leave this record of my dumbness for all to see!

[Stef]

Hehe sometime writing the problem can help to sort it out
Well interesting enough, as you said, Gens timing is a bit different... I wonder why as i heavily used MAME as reference for almost of the timing used. Anyway hopefully that bug does not affect much general sound emulation.