1. Make the RTG VGA series work on XT's.
2. The RTG3105 is limited to up to 512K of memory, not 1M.
3. Fixed rowoffset in 8bpp mode used by the RTG3105.
If the chips are PCI-based, assume linear addressing is always enabled when the linear base is on valid value (e.g.: not 0). Should fix a regression when linear addressing is used by almost everything in the PCI world (e.g.: Win95's drivers).
1. Cleanups and moving the mach8/32 struct to a dedicated header so that would allow for future 8514/A add-on clones (in paper).
2. Mach8/32's port 0x4ae8/9 and shadow set ports (0x5aee and 0x46ee) now account to the mode changes seriously, should fix most of the horizontal/vertical coordinates while entering GUI modes of various stuff.
3. Horizontal/Vertical window coordinates can only be modified if the display enable bit of port 0x22e8 is set as well as bit 0 of port 0x4aee, fixes most problems noted above.
4. Implemented horizontal blanking stuff a la VGA but actually for 8514/A and clones (like ATI).
5. Added some comments regarding the current situation.
6. The Mach8 was actually a 8514/A clone co-processor, not a single solution card of its own. The ATI Graphics Ultra was a single solution card that is actually a Mach8 + ATI 28800-6 in one, so renaming it accordingly.
7. Fixed garbled/distorted acceleration when device bitmap acceleration is enabled in the ATI Mach8 3.0 Win3.1 drivers.
* Fix duplicated logic in vid_svga.c
* Voodoo 3/Banshee emulation now has blanking extensions applied
* S3 ViRGE and Matrox video cards as well
* Tseng ET4000-series cards as well
* Fix off-by-one error in vid_cl54xx.c
Trident cards are yet to be investigated. XGA and 8514/A are yet to be made blanking extensions compliant. Mach64, Paradise/WDC and OAK OTI cards remain as-is for now.
1. Moved the wraparound fix to the respective recalctimings, makes Commander Keen still playable without flickering and, at the same time, making Solaris work right in 640x480x8bpp+ mode (including 1024x768x8bpp).
2. The multifunction index 0x0e bit that can be toggled in 32bpp mode is now initialized properly, should fix the pinkness on Solaris' 32bpp mode.
1. Properly implemented polygon filling in the BitBLT side of the ATI 8514/A compatibles (Mach8/32), this allows games like Mj8514 and demos like HDIDEMO from IBM to run under ATI's hdiload 1.1 properly.
2. Finally figured out the polygon filling command in the IBM one about read and write masks (Command 5 and Command 2 with polygon filling bits on, currently only for the read mask one), this allows the above samples to render properly with IBM's original hdiload 1.0 from 1987.
1. Made the 924 BIOS use the AT&T 491 ramdac since it supports such (tested).
2. Tweaks to the 928 Brooktree ID detection to make sure the cursor is shown correctly in 16bpp+ modes (it uses the BT485 ID detection so that it knows what BT is using to accommodate the cursor model).
3. Refactored the mode (CRTC50) and pitch timing stuff (moved to recalctimings for example) so that drivers/games/operating systems and what not can be used normally. (Warning, more stuff is to be tested due to a gazillion of combinations used by said stuff).
4. VRAM wraparound is now working as it should, fixes Commander Keen games.
5. Indentation fixes.
6. Attempt to fix 15/16bpp mode acceleration used by the 911/924 chips (not perfect and still has bugs).
7. Added the remaining missing stuff of the Sierra SC1502x RAMDAC including its 8BIT setting.
8. Some drivers use FIFO bits in non-FIFO configurations, should fix hang ups in some instances (namely the 928 S3 2.3 NT 3.1 drivers and possibly more).
9. Separated the 911/924 acceleration from the 80x/928+ one though the use of a function pointer.
10. Fixed the inverted colors in some instances using the S3 Trio64 driver in Win9x (mainly on soft reboots).
11. CX/CY (non-Blits) and DX/DY (Blits) wraparound correctly during their respective operations, fixes OS/2 software cursor once again while keeping existing stuff working.
12. Added some comments to keep track of some anomalies.
13. Fixed some badly formatted if's and switches.
14. Limited the SPEA Mercury Lite VRAM to 1MB per real world configurations.
