BGFX renderer requires some effects also when post-processing is disabled. If they are not available, MAME complains about missing effect files, but then goes on and crashes soon after printing the backtrace. This fix just exits gracefully when effect files are incomplete and falls back to a different renderer if the bgfx folder is missing at all, warning the user in advance.
The keyboard should communicate at 1,200 Baud, but due to bugs in the SCC
it's set to 9,600 Baud. The sun4c machines program the Baud rate
generator dividers for the serial ports with 00:0e and the dividers for
the keyboard/mouse ports with 00:7e. Therefore the ratio of Baud rates
should be (0x7e + 2) / (0x0e + 2) = 128 / 16 = 8. However both the RS232
ports and the keyboard/mouse ports run at 9,600 Baud, when the
keyboard/mouse ports should run at 1,200 Baud (which is 1/8 of 9,600 which
matches the ratio of the divider values).
I've artificially limited the rate the keyboard can transmit at to be no
faster than it could at 1,200 Baud using a timer. I can remove this hack
once we get correct SCC divider behaviour.
I attempted to hook up the SCC interrupts to level 12 as specified in the
SPARCstation-1 Programmer's Model in the table on Page 19. No interrupts
seem to be generated, so either I've screwed this up somehow, the Sun
isn't enabling interrupts, or there's another bug in the SCC emulation.
Sorry if I've screwed it up - I won't be offended if someone replaces it
wholesale.
The keyboard receives the reset commands from the Sun and sends back the
self test pass response, but the SCC seems to get a buffer overrun error.
So it appears that the received data isn't being read out. I haven't been
able to work out why.
* fixed copy of texture data to bitmap in BGFX
* changed AVI dimension to a lowest integral multiple of 4
(2 was still to low for most video players, e.g. VLC)
* added audio to AVI record in HLSL
* HLSL AVI record now uses window dimension instead of snap dimension
Initialize interrupt line register with 0xff (prevents initial bogus
interrupts) and clear active interrupts when an interrupt related
register is written (tests would be good to confirm how this works on
real hardware). [Dirk Best]
