path: root/TODO_NOTES

S3 ViRGE 4.0 devel notes



Status
------
1/26/2003
Ver 1.8.6
Pre-4.3.0 release.  320x240 doublescan support w/mouse adjust,
power management printouts, DAC error printout fix, log XVideo status based
on chipset, preliminary support for disabling XV when a mode doesn't
support it.

5/18/2002
Ver 1.8.5
320x240 mode support (doublescan).

2/2/02
Ver 1.8.4
Make 320x240 mode work in depth 15 & 16.
Testing, need to switch streams on/off based on dblscan_v flag and mode.
Virge MX panel_on test (doesn't work.)

1/10/02
Ver 1.8.3
DGA fix, buffer pointer used wrong value.  Submitted for 4.2.0 (late).

Ver 1.8.2
DPMS testing (display DPMS status, disable MX LCD panel), DGA testing.

11/11/01 KJB
Ver 1.8.1
ViRGE MX (& GX2) fix to CR63 for problems with modes < 1024x768 from 
max <sunmax@libero.it>. Submitted for 4.2.0.

11/5/01 KJB
Ver 1.8.0
ViRGE MX & GX2 XVideo changes.  Fix overlap/color keying on MX, fix
and enable GX2 XVideo.
Submitted for 4.2.0.

11/4/01 KJB
Ver 1.7.0
Bump version for 4.2.0.  Update man and README.

11/3/01 KJB
Ver 1.6.16
Pre-4.2.0 patch.  cr3a fix, virge mx xvideo support, xvideo disable option
(helps with high res modes on dx and screen noise), remove accel solid
fill rect for trio3d, bring over some trio3d and mx fixes from 4.0.3 test
drivers.  Includes VERBLEV bump to 5 to remove register dumps in log file.

Ver 1.6.15
Experimental - Scanline color expand - GX2

9/21/01 KJB
Ver 1.6.14
ModeInit - cr3a for MX/GX2, don't clear reserved bit 0x40 (revert change).
Back out previous cr3a patch for testing.

9/18/01 KJB
Ver 1.6.13
Option "xvideo" added.  Add patch provided by Sven Menke
<Menke@SLAC.Stanford.EDU> for XV support on MX (slight changes to logic).

Ver 1.6.12
Adam J. Richter's cr3a fix (possibly temporary).

5/30/01 KJB
Ver 1.6.10
Disable MaxHValue & MaxVValue setting.

1/03/01 KJB
Ver 1.6.0
Revert ViRGE to pre-Xv changes.  ViRGE DX still supports Xv, but changes
added for Xv and ViRGE caused problems with normal display.

11/27/00 KJB
Ver 1.5.0
Clean up, bump version for 4.0.2 submission.
CR3A, bit 0x40 (reserved) cleared for MX/GX2.  Update man page.

11/24/00 KJB
Disable CPU to screen color expansion on GX2, causes lockups on GX2 with
'locate html' in an xterm.  Add WaitCmd code to prevent accelerator and
reg command path activity at the same time on GX2.
XVideo fixes, left side clipping fixed for video windows extending off left
side of screen.  Add horizontal filtering modes for 1-2x and >3x scaling.

11/7/00 KJB
Ver 1.4.0
Enable MX fixes, testing with accel BLT_BUG set is worse on GX2, so left
it out.  BLT_BUG wasn't being enabled on ViRGE & VX because of case usage,
enabled now.  Re-enable silken mouse for GX2.  Add Render/fbPicture support.
Fix a few compiler warnings.

11/5/00 KJB
Continued work on GX2, much stabler now, but I think there is a lockup case
left if you enable pci_burst and pci_retry with accel.  I still see screen
flashes with vertical bars once in a while, and the log reports a GEReset.
Added fbPicture (render) support, untested.  XV code for GX2 is added,
but not working yet.

10/29/00 KJB
Much work on GX2, now SWCursor locks up but HWCursor is stable, go figure.
Loading Netscape a couple times with the mail window was enough to lock it
up twice in a row.  Server was locked but not a box (PCI) lock.  Also, 
vertical barring still occurred a couple times, so HW cursor wasn't causing
that.  Try blt_bug flag again for lock up case.  And attach to debug server
and bt...  Sleep now...

10/21/00 KJB
Ver 1.3.0
Depth 16 Xv support added.  Testing on ViRGE DX.  1280x1024x24 is noisy, so
needs FIFO tuning.

9/27/00 KJB
Initial Xv support in depth 24.

9/10/00 KJB
Convert to FB.  Add option "UseFB", default true.  Added
VerticalRetraceWait timeout back as default.  #if0 out the line accel
code in s3v_accel.c.  The Subsequent...Bresenham line code causes a lockup
when used with fb.  We didn't have any hardware accel in there anyway...

7/25/00 KJB
Started Xv additions.

6/26/00 KJB
GX2 seems to have an accel bug.  I see the entire screen go solid color or
a wide stripe pattern for about 1 second.  The S3VGEReset gets called twice,
and then everything is okay.  Unless SilkenMouse is enabled...  That seems to
get you a server lock instead.  Short term is to disable SilkenMouse for GX2
only.  SilkenMouse isn't really the problem of course, it just happens that if
we move the mouse during the 1 second engine lockup we appear more likely to
never recover.

6/23/00 KJB
Ver 1.2.0 - Fix console corruption on GX2 caused by reserved bit use in CR3A.
Fix GX2 noise on screen in hi-res depth 24 by increasing FIFO fill threshold. 

6/12/00 KJB
Ver 1.1.0 - Add Init for SilkenMouse, add xf86SetBackingStore call, move
int10Symbols[] to remove warnings.

3/3/00 KJB
Ver 1.0.0 - S3VProbeDDC code for X -configure option added, add timeout to
WAITIDLE macro in s3v_macros.h, change ImageWriteFlags adding NO_GXCOPY.
Start of GX2 fixes, use CR regs for FIFO settings.  Update copyrights.

2/11/00 KJB
Ver 0.11.0 - Add cfb16/24BresS to module symbols to remove warnings.
Fix viewport restore problem after EnterVT in 24 bpp.
Clean up log output, removing register dumps from normal console log.

6/26/99 KJB
Make the memory settings for fifo_conservative the default, 'fifo_conservative'
does nothing additional now.  Patch includes DGA2 additions below, non-working.
Expected to be included in 3.9Pu.

Changes in 3.9Pt by others include additions for newer RAC support.  Some reports
say multi-head works now with ViRGE.

6/17/99 KJB
Ver 0.9.0 - Prelim DGA2 support modeled after MGA.

5/28/99 KJB
Ver 0.8.0 - Changes to 3.9Po - Cleaned up debug register printing function, minor
changes to man page, remove S3V.sgml and add new s3virge.sgml in doc/sgml, also
remove README.S3V from doc directory.


4/5/99 KJB
3.9Ph - Ver 0.7.0 - Virge man page added, HW Cursor fixed, rename chipsets removing
slashes in the names.

03/27/99 KJB
Ver 0.6.0 - hwcursor additions, added s3v_hwcurs.c and Option "swcursor".
Default is hwcursor, Option "swcursor" will disable it.

Ver 0.5.0 - patch against 3.9Pf (seq 2615), fix depth 24 and Accel flags, sync pci_burst option to previous changes, remove s3v_comp.h and s3v_pio.c and merge as needed.

03/21/99 KJB
3.9Pf has Matt Grossman's Alpha changes.
For next patch - remove s3v_pio.c and s3v_comp.h.  Include the EnableMMIO and
DisableMMIO functions from s3v_pio.c in s3v_driver.c.

03/02/99 KJB
3.9Pc - depth 24 doesn't work on my ViRGE DX.  NoAccel doesn't start, accel does
but has blocky noise.

03/01/99 KJB
Macro change done, VGAIN/VGAOUT for register access, INREG/OUTREG for s3v_accel.c.
Added Mark Vojkovich's re-write of the accel code.  It may only be clean for ViRGE DX
at the moment.  x11perf showed a couple artifacts in 'move window via parent'.
In progress, attempt to call cfbScreenInit() functions after MapMem/EnterVT.  Not
working yet.  Version stamped 0.4.0.

02/22/99 KJB
Macro change coming to add Mark's accel update.  VGAOUT for old stuff and MEMOUT
for new stuff?  That way there's no confusion with the old INREG/OUTREG macros.
Or maybe just stick with INREG/OUTREG for new stuff.

01/30/99 KJB
Version stamp 0.3.0.  Changed Chipset flags to use PCI IDs exclusively, also use
common/xf86PciInfo.h for PCI IDs rather than coding them in regs3v.h.

11/28/98 KJB
Bumped version stamp to 0.2, expect code in 3.9No.  Cleaned up s3v.h and
s3v_driver.c by removing unused definition & code sections.  Added
options set_mclk (from 3.3.2) and set_lcdclk (3.3.3 MX).  Code support
from 3.3.3 for ViRGE GX2 and MX+ is included.
Disabled call to 32 bpp AccelInit to get -depth 24 -bpp 32 working again.

11/27/98 KJB
More 3.3.3 import.
New registers saved, CR40,CR45,SR8,(for MX) SR29,SR45,SR55,SR56,SR57.
Reviewed s3vdriver.h, rehs3v.h, newmmio.h, s3v_accel.c, s3v_driver.c.
Added ViRGE MX, MX+ & GX2 support.  Re-synced parts of mode save and init
with 3.3.3 versions.  Added timeout ability for WaitIdle() and friends.
Added chipnames and numbers to Chipsets struct.

11/26/98 KJB
Import additions from 3.3.3, newmmio.h, regs3v.h,

Trap fills disabled because they don't match cfb, pixmap cache & ImageWrite
working, fixed depth 8 color loss on VT switches, INREG & OUTREG modified to
use a single offset value instead of adding the base and offset together.

11/18/98 KJB
3.9Nn
Acceleration working for Bitblt, ScreenToScreenCopy, Color 8x8 Rect fills,
and Rect/Trap fills.
Trap fills do not support transparency, so that needs to be exported to XAA.


10/31/98 KJB
Working depth 8, discolored dep 16 but runs, dep 24 screen goes black, 
C-A-Bkspc restores text console.  At 3.9Nk tree level, module would not
load in Loader server.  why?  Static server tested.

10/29/98 KJB
ModeInit() needs work, options are heavily #if'd to try and get 8bpp
working.

10/16/98 KJB
General 4.0 architecture is setup.  Presently at 3.9Nc level, if moved
to a newer tree you will need to add the resource handling functions from
Egbert (I haven't tackled that yet).

At the moment the ScreenInit() function is coded to return FALSE.  On my
ViRGE DX card this version does not lock up, but it does destroy the
video mode.  Make sure you have an external terminal or network connection
if you run it (or blindly do a restart from your main terminal).  I make
no guarantees that it won't hard lock other versions of ViRGE.

I've left out the Alpha memory mapping, along with all option processing.
Those will need to be done once the driver is minimally working.

Note that everything in s3v_driver.c is MMIO only.  There are a pair of
PIO functions in s3v_pio.c, but that is the only place.  See notes below
about my ViRGE DX BIOS and why I needed to do this on my hardware.

Other stuff...
Some test stuff is assuming 8bpp, so 16 & 24 are broken.
On my hardware, I am presently trying to get the Save/Restore sequence to
recover the video mode.  At the moment, when I run this driver, I get:
ScreenInit() runs to completion.
It returns FALSE, so the Server aborts.
LeaveVT() is called, and runs to completion.
The Server exits gracefully, but my monitor goes powersaver and the video
mode is not recovered.



TODO items
----------
1/30/03 General option "videoram" is ignored by the virge driver.
  (Meelis Roos)
3/24/02 Xv reported to not work as secondary in Xinerama multihead.
  (xav on irc)
3/24/02 DPMS doesn't fully disable the screen.  Blue line across the
  center in all modes. (xav on irc)
3/25/02 Secondary reports primary BIOS during int10 detection. ie:

(II) S3VIRGE(1): VESA BIOS detected
(II) S3VIRGE(1): VESA VBE Version 2.0
(II) S3VIRGE(1): VESA VBE Total Mem: 4194240 kB
(II) S3VIRGE(1): VESA VBE OEM: ATI RAGE128
(II) S3VIRGE(1): VESA VBE OEM Software Rev: 1.0
(II) S3VIRGE(1): VESA VBE OEM Vendor: ATI Technologies Inc.
(II) S3VIRGE(1): VESA VBE OEM Product: R128
(II) S3VIRGE(1): VESA VBE OEM Product Rev: 01.00

(xav on irc)


Check CR65 usage, bit 2 set based on S3_EARLY_SC?  In my manual bit 2 is
enable MMIO to RAMDAC registers.

Notes:
----------

/config/cf/xfree86.cf

have to add s3v to XF86CardDrivers for imake to make the 
drivers/s3v Makefile.

To remake makefiles, after editing Imakefile, go to dir above drivers/s3v
and do a 'make Makefiles'.

For debug, make CDEBUGFLAGS='-g -DDEBUG', adding -DMetroLink enables
timeout for VerticalRetraceWait().



S3 ViRGE DX stuff:

Card seems to power up (or BIOS forces) with MMIO disabled.  All flavors are
disabled, because CR53 comes up as 0.  This may preclude using this card
as the second device in a multi-head situation although David D. says that
the new config. management stuff may help here.



$XFree86: xc/programs/Xserver/hw/xfree86/drivers/s3virge/TODO_NOTES,v 1.22 2003/02/04 02:20:49 dawes Exp $