XFree86 P9000 Server Release Notes Erik Nygren (nygren@mit.edu) 1998 December 29 1. Change Log 1998.10.13: o Fixed a bug that would cause the server to crash when it tried to enable or disable the screen saver while at a VT (Erik Nygren) 1997.01.30: o Added probing for MemBase and IOBase on Diamond Viper PCI cards (Karl Anders Øygard) o Added support for DPMI screen saving (Karl Anders Øygard) 1996.03.31: o Added support for the XFree86-DGA extension (Erik Nygren) 1995.05.24: o Added p9000frect.c: Accelerated solid rectangle fills at 8/16bpp (Henrik Harmsen) o Added stipple fills to p9000frect.c, and a stub for tile fills when p9000ImageFill is fixed (Chris Mason) o Added p9000pntwin.c: Accelerated paint window at 8/16/32bpp (Henrik Harmsen) o Added p9000gc16.c and p9000gc32.c for the higher bpp drawing functions (Henrik Harmsen) o Additions to p9000im.c: p9000Image[Op]Stipple. And p9000ImageFill. Currently, there are small problems with ImageFill, and it is not being used. (Chris Mason) o Added p9000PixAlu and p9000PixOpAlu. miniterm->alu transla- tion for pixel1 opaque and transparent operations. (Chris Mason) o Added p9000text.c: Non-cached poly text and image text func- tions. Image text functions are not used because they are too slow :( (Chris Mason) 1995.05.21: o Fixed p9000init.c to properly deal with the vram_128 option. This should allow the driver to work properly with all Viper's with 1 MB of memory. (Erik Nygren) 1995.01.29: o Updated P9000.sgml to mention using Robin's scanpci rather than PCIDUMP.EXE. 1995.01.15: o Fixed problem with line capping in accelerated line drawing. (Chris Mason) o Fixed p9000QuadAlu[GXset] to be ~0 rather than 1. (Erik Nygren) 1995.01.14: o Clocks line is no longer used in XF86Config file. Operation should now be consistent with the operation of the other servers which use programmable clocks. (Erik Nygren) o Users with 1MB cards can now explicitly specify videoRam in the XF86Config file when autoprobing fails. The new vram_128 option may also be used to force the detection of 128Kx8 SIMM's. (Erik Nygren) o Added p9000line.c and p9000seg.c for accelerated line draw- ing code using the p9000 quad/clipping engine. Blazingly fast for 1 clipping rectangle, could be made faster for mul- tiple clipping regions by using software clipping. There is still a bug which causes xtest to report Cap style incorrect for thin line and CapNotLast for the XDrawLines tests but not for the XDrawLine or XDrawSegments tests [fixed in 1995.01.15 patch]. (Chris Mason) o Changed p9000blt.c, and p9000win.c to wait for the quad/blit engine to be free. Before a quad/blit, check SR_ISSUE_QBN, then blit, then when all blits are done, do a p9000QBNot- Busy. (Chris Mason) o Changed p9000init.c to clear the screen using the quad meta coord drawing mode. Appears the rect mode does not update the CINDEX register correctly. Changed the color to 1 (black) from 0. (Chris) o Added p9000QuadAlu. When drawing a quad, the p9000 equiva- lent to X's source is the foreground mask. When blit- ing/pixel8ing/pixel1ing, it is the p9000 source mask and the p9000alu lookup table should be used. (Chris Mason) o Added some more registers to p9000reg.h. (Chris Mason) 1994.09.20: o Fixed problem which prevented 16 bpp modes from working (Erik Nygren) 1994.09.16: o Added screen blanking support for 16 bpp and 32 bpp modes. Screen blanking now powers down the RAMDAC rather than just changing the planemask. (Chris Mason, Erik Nygren) o Fixed more problems caused by switch to XF86Config (Erik Nygren) o Possible fix to maxclock for Orchid P9000 (Harry Langen- bacher, Erik Nygren) 1994.09.15: o Now almost always works with XF86Config changes (Erik Nygren) o Cursor code looks at VTSema before writing to RAMDAC. This had been causing the x11perf server crash (Erik Nygren) 1994.09.08: o Fixed problem with xdm and restarting the server (Erik Nygren) o Fixed and enabled ImageRead in CopyArea (Chris Mason) o Made informational comments conform to standard :-) (Erik Nygren) 1994.09.05: o Fixed BIOS probe for Viper PCI (Bob Hollinger) o Fixes to Orchid P9000 support (Harry Langenbacher) o Changing of datatypes in clock code (Harry Langenbacher) o Fixed clock and misc reg restoration so now works fine with svgalib (Chris Mason, Harry, Erik) 1994.08.29: o Increased number of memory regions in xf86_OSlib.h from 2 to 3 as needed by the Viper PCI (Erik Nygren) o Changed method of short pauses in p9000vga.c to outb(0x80,0) (Erik) o Rewrote routines to determine sysconfig from horizontal res- olution. Also added check for valid hres to probe. (Erik Nygren) o Added MoveWindow acceleration for all depths. Opaque move even looks nice at 32bpp now! (Chris Mason) o Minor fixes to acceleration. Acceleration is now enabled by default (Chris Mason) o Added "noaccel" option (Erik Nygren) o Added some fixes for Viper PCI (Matt Thomas) 1994.07.21: o Preliminary Viper PCI support - totally untested so disabled (Erik Nygren) o Preliminary Orchid P9000 support - incomplete and totally untested so disabled (Erik Nygren) o Preliminary accelerated support - incomplete and not fully tested so disabled (Erik Nygren and Chris Mason) 1994.07.08: o 16 and 32 bpp TrueColor support (Erik Nygren) o Color restoration hopefully fixed (Erik Nygren) o Changes to how "Modes" line in Xconfig is processed o Removed banking support :-( 2. Supported Cards 2.1 Diamond Viper VLB All Viper VLB's should work with this server, hopefully... :-) Due to Dia- mond's putting the same BIOS in some Viper VLB's as are used in Viper PCI's, the probe may detect you have a Viper PCI when you really have a Viper VLB. If this happens, put chipset "vipervlb" into your XF86Config file. 2.2 Diamond Viper PCI You may need to specify the chipset "viperpci" in your XF86Config file. Previously you had to find out the values for MemBase and IOBase by yourself. These are now autodetected. 3. Orchid P9000 and random clones The Orchid P9000 and other cards based on the Weitek board design (such as the STAR 2000) should now work. Talk to harry@brain.jpl.nasa.gov if you have problems with this. Specify the chipset "orchid_p9000" in the Device section of XF86Config 4. Viper Pro and other P9100 and P9130 cards (UNSUPPORTED!!!) These are NOT supported yet by this server, but are supported in the p9x00 driver of the SVGA server. 5. Acceleration Some of the acceleration code is working, but there are probably still bugs. Only a very small number of accelerated features have been implemented. Before working on any acceleration, please contact nygren@mit.edu so we don't duplicate efforts. Acceleration may be turned off with the "noaccel" option. The following things are now accelerated: o Hardware cursor (8/16/32bpp) o MoveWindow (8/16/32bpp) o CopyArea (8bpp) 6. XFree86-DGA Extension Support The XFree86-DGA extension is now supported. Note that XF86DGASetViewPort command is not fully implemented due to hardware limitations of the P9000. The SetViewPort and SetVidPage commands have been hacked to allow double buffering under certain conditions. For cards with 1MB or modes where xres*yres*Bpp > 1024K, no double buffering is supported. In this case, the bank size returned is equal to the amount of video memory. Using the XF86DGASetViewPort and XF86DGASetVidPage commands have no results. For cards with 2MB and for modes where virtualX*virtualY*Bpp < 1024K, the behaviors of SetViewPort and SetVidPage are modified to allow double buffer- ing. The bank size returned by XF86DGAGetVideo is equal to xres*yres*Bpp. In this mode, there are two buffers which can be written to, read from, and displayed. The XF86DGASetVidPage command can be used to switch between buffers 0 and 1 for I/O. Whichever buffer is selected will be available through the linear aperture with no offset. If XF86DGASetViewPort is called with ypos < yres, it will cause buffer 0 to be displayed. If ypos >= yres, buffer 1 will be displayed. The result of this behavior is that programs which switch banks as necessary and which use two vertically adjacent banks should work with no P9000-specific changes. 7. High Color and TrueColor Support for 16 and 24 bit truecolor is now supported. Note that 24 bit color is really 32 bits per pixel. Use the -bpp option when starting the server. Examples: startx -- -bpp 32 startx -- -bpp 16 startx -- -bpp 16 -weight 555 startx -- -bpp 16 -weight 565 Note that many programs do not yet work properly with these modes. Don't tell me. Tell the authors unless they've already fixed it. It's their fault... :-) Example problems: xv 3.00 Works fine in 32 bpp and in 16 bpp with 24 bit images. Has prob- lems with colors in 8 bit images in 8 bpp mode. Mosaic 2.1 Has problems with colormap in both 16 bpp and 32 bpp. Newer ver- sions of Mosaic such as 2.4 do work. mpeg_play Doesn't work at all in 16 bpp mode. Works fine 24 bpp mode when compiled with -DRS6000 and when run with ``-dither color'' xpaint 2.1 Works great in both modes but has a bug in the color requester for the selection tool. I think later versions may have fixed this. 8. Random Notes Text restoration should now be fixed. Color restoration should also be fixed. You can now even run the server at the same time as svgalib pro- grams!!! Diamond has actually been fairly open and helpful. No NDA's were signed by anyone who wrote code and Diamond claims that none of the information they provided is proprietary. One unresolved issue is the maximum clock speed. It is currently set to 135 MHz with a warning printed over 110 MHz. Diamond claims that this is the max in their docs, but examination has shown some Viper's to contain 110 MHz bt485's. Without 135 MHz, it is not possible for people to with large moni- tors to run at 1280x1024. Diamond claims that all Vipers have 135MHz bt485's or compatibles. If you have something slower, call their tech support and they will send you a RMA to get the board replaced. 9. Operating System Notes Any operating system that can memory map linear regions in really high memory should work. This should include Linux, FreeBSD, SVR4, and more. 9.1 NetBSD If you have NetBSD, you will need to install the aperture driver. Extract the file apNetBSD.shar (in xc/programs/Xserver/hw/xfree86/etc/apNetBSD.shar) and read the README contained therein. 10. XF86Config The modes line in the XF86Config file is now handled differently. The vir- tual line is now ignored entirely. Each mode on the mode line is looked at and the first usable mode is selected (ie the first one which works with available memory, etc). Any other modes which are valid and have the same dimensions are also used. And other modes are ignored. The current supported keywords in the Device section of the XF86Config file are: VideoRAM 1024 or 2048 (use 2048 for ``3MB'' Orchid P9000's) ChipSet "vipervlb" or "viperpci" or "orchid_p9000" MemBase Viper VLB: 0xA0000000 or 0x20000000 or 0x80000000 (0x80000000 is default if none spec'd) Orchid P9000: 0xC0000000 or 0xD0000000 or 0xE0000000 (this MUST be set to correspond to the jumpers) Viper PCI: any value corresponding to the output of PCIDUMP.EXE IOBase Viper PCI: any value corresponding to the output of PCIDUMP.EXE Others: unused Clocks any values between 25 and 135 corresponding to the clocks for the mode entries being used. This line may now be omitted and clocks will be matched automatically. Option "sw_cursor" use software cursor "vram_128" use if you have 1024K VRAM in 128Kx8 SIMMS "sync_on_green" generate sync pulses on the green signal. Most (all?) P9000 based boards don't support this. "noaccel" do not do hardware acceleration if it's causing prob- lems for you Modes almost any valid mode (there are constraints on the horiz res so not all values are possible) The current supported keywords in the Display section of the XF86Config file are: Depth 8: use 8 bits per pixel for 256 colors (default) 15 or 16: use 16 bits per pixel for up to 65K colors 24 or 32: use 32 bits per pixel (sparse 24 bpp) for up to 16 million colors Weight 555 or 565 if Depth is 15 or 16. Otherwise this is ignored. These are the Red, Green, and Blue bits per pixel (default=565) Here's a portion of a sample XF86Config file for the Viper VLB: Section "Device" Identifier "ViperVLB" VendorName "Diamond" BoardName "Viper VLB" Videoram 2048 # This is mandatory Membase 0x80000000 # This is mandatory on non-ViperVLB's IOBase 0xe000 # Use this ONLY on ViperPCI's EndSection Section "Screen" Driver "accel" Device "ViperVLB" Monitor "NEC4FGe" Subsection "Display" Depth 8 # This line is optional Modes "1024x768" "800x600" EndSubsection EndSection 11. Known Bugs There are currently problems with the server when used in conjunction with xdm, olvwm, and VT switching under Linux. If the cursor changes while you're in a VT, the cursor won't look right when you return from the VT until it is moved between windows (and changes color and shape). Memory probing does not work. You will need to explicitly specify the amount of memory you have. If you have a 1 MB card, try put VideoRAM 1024 into the Device section of your XF86Config file. If this doesn't work, try adding Option "vram_128" to the Device section. 12. Credits Major contributors to P9000 code: o Erik Nygren (nygren@mit.edu) o Harry Langenbacher (harry@brain.jpl.nasa.gov) o Chris Mason (clmtch@osfmail.isc.rit.edu) o Henrik Harmsen (harmsen@eritel.se) Thanks to Matt Thomas (thomas@lkg.dec.com) and Bob Hollinger (bob@interac- cess.com) for helping to get the Viper PCI server working. Special thanks to David Moews (dmoews@xraysgi.ims.uconn.edu) whose banking patch could unfortunately not be included. Thanks to Andy, David, Dave, Jon, Michael, Bob, all the XFree86 core team people, and everyone else! During the course of the next few months, people will be working on accelera- tion, etc. Please send any patches to me (nygren@mit.edu). Generated from XFree86: xc/programs/Xserver/hw/xfree86/doc/sgml/P9000.sgml,v 3.18.2.7 1998/12/29 07:54:30 hohndel Exp $ $XConsortium: P9000.sgml /main/9 1996/05/12 20:58:05 kaleb $