These few cases where it gets corrupt fonts like in some builds of NT 4.0 and mode changes in Win95's driver.
Hopefully this keeps everything else intact.
S3 928:
When the pitch is 1280, make sure the proper horizontal resolution is displayed right (in this case, 1280x1024).
S3 Vision864:
Exclude it from the horizontal blank bits due to more issues that are TBD (like the Trio32 and 64 and up), this works around the mode issues under Windows 9x and more as well as fix 32bpp horizontal rendering in said chip.
1. On both Mach'es: direct linear video memory access is now the default for LFB mapping: fixes Mach64 drivers on NeXTSTEP/OPENSTEP and keeps everything else intact.
2. Mach8/32: minor cleanup to mode change code.
1. Text is no longer glitchy under win95 using XGA.
2. Now there's a better way to handle the cursor/sprite X/Y offsets, should fix remaining Win3.x cursor blackness and other stuff using said cursor.
As well as the cursor/sprite being finally fixed when the xoff goes from 0x20 onwards. This makes Win3.x' XGA cursor look normal and everything else as well. (as in, intact).
And cleanups as well.
Htotal/hblankstart are now applicable to the HT216-32 card only.
Re-apply the Radius ISA SVGA extensions port workaround until a proper ISA bios is found/dumped (warning: the HT209 may not work properly on the PS/1 2121 machine).
ET3000AX: the chip in question should not support 1MB of video memory.
ET4000AX: the early TC6058AF revision can support 1MB of video memory (e.g.: Diamond Speedstar BIOS D3.10, undumped anyway), and actually don't update the rowoffset to 256 when using such chip in 320x200x256 mode. Fixes the copper demo in said chip revision.
Apparently some software reprograms the vertical display wrong sometimes (in this case, vdisp + 2).
This should fix software titles that use such techniques...
Now takes constants with human-readable names instead of
Win32 string table IDs, language.h is no longer needed
ui_msgbox*() no longer accepts string IDs as arguments directly,
plat_get_string() must be explicitly called to retrieve the string
Make sure for the 9001st time to display correctly the horizontal side of the display (looking at you, ATI 1992 set for win3.1x and Mach8/32 drivers for OS/2 1.2x).
1. The SPEA specific cards using the 801/5 chip apparently have an ID that's not 0xa0, rather, they either use 0xa2 or greater for the stepping. Fixes wrong colors in 800x600 modes and greater in 8bpp mode.
2. HWCursor addresses for 8bpp are now properly implemented if the 8bpp mode bit (CRTC3a bit 4) is checked, no longer relaying on GDCREG5 bit 6). Fixes garbage cursor in the SPEA 80x cards drivers (BigWin) using 8bpp mode.
