^ thanks for the clarifications marqs!
So I found the missing two V-V signals, simply by measuring all U36 pins, my previous tracing was mistaken. In the end you just can't see if the lines change angle below a chip without desoldering it.
Correct is:
205 (Vsync) -> U36 pin 17
206 (Hsync) -> U36 pin 18
207 was already correct then I suppose? (carrying both V- and Hblank)
And as we can see, if we look at U36 pin 18, the bad Hsync pulse is there, showing that the signal the GP9001 puts out is already busted:
However, as RGB helpfully remarked earlier, Tatsujin Oh works fine with the OSSC. Tried it, and indeed,
it does.
But why is that - Tatsujin Oh as well uses a single GP9001. Then Hsync must be bad there too, right? Wrong.
Let's look at Jamma Csync output first:
And indeed, accumulated
sync and first backporch line lengths (p1-5) do equal to 5 times normal line length (5*64=320)
![Smile :)](./images/smilies/icon_smile.gif)
Is the error caused by something different, or - it has that characteristic spike - does it miraculously get "corrected" somewhere in the path between the GP9001 and Jamma out?
Here are the relevant adjacent pins for easier measurement:
205 (Vsync) -> U81 (SN74LS08N) pin 1
206 (Hsync) -> U81 pin 2 (or single via directly between GP9001 and U81)
207 (blank) -> couldn't identify it definitely, at least found nothing that looks like V-V's U36 pin 3.
And, here is Hsync:
Tatsujin Oh's GP9001 doesn't have the problem
![Surprised :o](./images/smilies/icon_surprised.gif)
It looks like there are variants of the chip which differ in behavior.
For what it's worth, the imprints on mine are:
Tatsujin Oh:
L7A0498
GP9001
TOA PLAN
9044
V-V:
Only the number in the last line is different: 9150
V-V came after Tatsujin Oh, so one would expect that normally, if anything, bugs like that get fixed. Here on the other hand, it seems like a bug was newly introduced. I guess they simply didn't notice it back then, because they naturally only tested on CRTs.
So I think this shows that the problem lies indeed within the GP9001 (but it appears to be only (a) specific revision(s) of the chip), and that an easy solution is likely out of reach, as was already pointed out.