The program I first noticed the problem with was Hangman , which displayed on EightyOne as follows:
- HangmanRotated.png (8.32 KiB) Viewed 1386 times
When ZXpand+ is enabled in EightyOne (but without WRX enabled since it is not compatible with ARX) then the characters became rotated by 2 lines:
- HangmanRotatedZXpand.png (7.85 KiB) Viewed 1386 times
I was able to verify these results on real hardware (using other RAM in the 8K-16K region since the RAM ZXpand+ provides is WRX enabled). I was able to prove it was not ULA version related nor ROM edition related.
The ARX display driver used in Hangman calls into the ROM to generate the VSync pulse and then returns immediately back into the display driver. The NMI generator is now back on and so when the next HSync time is reached, an NMI will be generated. Upon the return into the display driver, a call is made into the ROM to read the BREAK key. This does not take into account the fact that the routine is at different address in the edition 1 ROM but fortunately it harmlessly returns. However, it does mean that the BREAK key functionality will not work in Hangman when using the edition 1 ROM. However, the ROM edition alone was not the cause of the rotation issue.
After reading the BREAK key, the display driver reads MARGIN and adjusts the value to compensate for the 8 line delay that the display driver performs prior to outputting the main picture area. If ZXpand+ is connected then it takes longer to reach the code that reads MARGIN and the next HSync position is reached. This causes a NMI to occur before MARGIN is read. This extra NMI does not get factored into the number of border lines to output and so an additional one gets generated. This increments the LNCTR one extra time and hence results in the character being displayed rotated.
That explained the impact of connecting ZXpand+, but why did EightyOne without having ZXpand+ selected show the characters of Hangman rotated? It wasn't due to EightyOne since I able to prove the rotation happened on a variety of real ZX81s. This makes me wonder whether Andy had developed the driver using one of the early versions of EightyOne which back then was not clock accurate. Or perhaps he was testing using the replacement ZX81 ULA he was developing back then but it wasn't clock accurate at that stage. Whatever the reason, the issue at hand was how to modify the display driver so that it was not susceptible to these issues. That lead to the development of a revised ARX display driver.
The new display driver forces the LNCTR to reset prior to outputting the main picture area and so the characters can now never be rotated. Forcing LNCTR to 0 also means I could remove the 8 line delay, and that has the benefit that the user program has more border lines in which it can run in parallel with. That means the user program gains a performance boost. These changes also mean the reading of MARGIN after the VSYNC is no longer necessary and that avoids the potential for the number of border lines that are output to change based on what external hardware is connected.
The original display driver generated its VSync pulse non-aligned to the HSync pulses, as Eightyone shows (VSync is red, HSync is blue), and notice how the display driver only outputs 303 scanlines:
- HangmanVSync.png (9.5 KiB) Viewed 1386 times
The new display driver now nicely aligns the VSync to the HSyncs, and this has a benefit since good alignment has been found to increase picture stability when using HDMI converters. It also outputs 310 scanlines, which is the same as the number output by the standard ROM display driver.
The new display driver also has the advantage of allowing the I and IY registers to be used by the user program.
To prove the new ARX display driver works as intended, I've created a small test program. This program simply draws a series of lines and demonstrates how the unusual ARX display file structure can be easily converted to from logical display coordinates. This not only makes manipulation of the screen easier but also means the a program can be more easily ported to/from WRX high resolution. In the following screenshot you can see the VSync pulse is aligned with the backporch period (black) and the HSync pulse (blue):
- NewDriverVSync.png (10.73 KiB) Viewed 1386 times
The new display driver is of comparable size to the original and so I don't believe there are any downsides to using it in favour over the original.
