In order to more accurately emulate the VDP, the main draw loop is
now going through each pixel on the screen and calculates the various
cells that should be displayed, gets the exact pixel data, and then
draws lines them all up in priority-order, and whichever is the first
non-mask pixel colour gets put into the frame buffer. It's rather
verbose and duplicative, but I'll fix it up now that I have something
working
It's a bit weirdly implemented because the Addressable trait doesn't
have access to System, so it has to set a flag on the Bus which is
then checked during the step function in System to activate the
breakpoint if a watched memory location was written to
It turned out to be an issue with the interrupts and when the
vertical interrupt was triggered vs when the vertical blanking
bit was set. The interrupt code in sonic 2 tries to read the
status bit after the vint occurs, and if the vblank bit is set,
it runs the Vint_Level function, but the hacky code I wrote
turned the vblanking bit on after 14ms and off when it triggered
the vint, instead of turning it on at 15ms, and off at 1.2ms, with
the int occuring at the *start* of the time the blanking bit is
set. So the code had been waiting for 14ms after the vint until
it actually started processing thing, which didn't complete before
the next vint, so it only ran the main game loop every 33.2ms which
is why doubling the speed of the simulated execution time made it
seem about the right speed
The hscroll table was multiplying by 2 (because scroll a and b values
are next to each other) but it should have multiplied by 4 because
each value is also 2 bytes and the array is of u8.
I added hscroll by-line support by using a different function for the
line scroll vs the cell or whole screen scrolling. There are still
a bunch of glitches in scroll b's scroll values that I need to fix
Previously it was storing data in the registers, which was an array
of u8, but now it's storing eg. full addresses for the scroll tables
so that they don't need to be fetched from the register values and
converted every rendering. I was thinking this would maybe make
DMA debugging easier, in particular.
The controller th count was starting at 1 instead of 0 which caused
inputs to be read incorrectly, as well as the fact that inputs are
inverted.
The sprite fix now draws them in the right order so that ComradeOj's
sprite test renders correctly, but games still render them broken
because of some other bug yet to be found
For systems that legitimately use 0 for black, this causes a
problem, so I instead use 0xFFFFFFFF for the mask colour which
corresponds to transparent white in ARGB format, which minifb
doesn't support anyways
transistor