Age | Commit message (Collapse) | Author |
|
|
|
|
|
on CLE266 because Z values are written even for fragments which
fail the test.
|
|
|
|
|
|
Only upload stencil configuration when a stencil buffer exists.
|
|
1. Unlock hardware before aborting and dumping DMA buffers. Otherwise display system may deadlock.
2. Fix DMA init IOCTL call that got mixed up in one of the previous commits.
3. Fix AGP command alignment
|
|
|
|
|
|
|
|
the xf86drmVIA.[ch] files.
Bring in the IRQ handler and Ring buffer code, but ring buffer is disabled
as it is with the Xserver. It certainly locks up the CLE266.
|
|
|
|
|
|
|
|
including Felix's ptex code.
Re-enable some assembly for performance.
|
|
|
|
|
|
Implement masked clears.
|
|
|
|
|
|
|
|
|
|
remaining.
|
|
Fix various mishandling of cliprects.
Allow multiple primitives to be emitted to a single dma buffer, which
was largely impossible previously.
Re-enable the fast unclipped render stage.
|
|
|
|
Make sure fallbacks are wrapped by SpanRenderStart/SpanRenderFinish
|
|
|
|
|
|
|
|
- Remove via duplicates of shared template files
- Update driver to work with current versions of the above
- Rework dma accounting
- Rework emitting to dma to use a consistent set of macros
The handling of cliprects in the driver is still pretty questionable.
|
|
back to using the shared template files.
|
|
Get the drawXoff adjustment working a bit better.
Seems to pass the glean orthoPos tests.
|
|
|
|
Fix 24/8 depth/stencil visuals.
|
|
|
|
Remove the bogus 'current_vmesa' pointer.
|
|
to run together correctly, though scheduling between them still isn't
great.
|
|
Remove debug printf.
|
|
will have a very low cpu utilization (and also a very low framerate).
Fix up the pageflipping code. This works now but is totally oblivious
to the X server (ie. it works but it's broken). Turned off by a #define.
|
|
code if DEBUG not defined.
|
|
Fix for proplem where polygons partly outside the drawing window would get
one vertex clamped to the opposite window side. This bug could also cause
a hardlock. (Bugzilla #2066, Yann Vernier)
|
|
|
|
Updated the driver to the new VIA security mechanisms in DRM. All command
submissions now passes through DRM ioctls. If the DRM AGP ring-buffer is
not enabled, it will use a DRM mechanism for submitting commands to the
hardware via a PCI bus mechanism.
Removed all direct write accesses to the hardware. Among other things the
VQ was previously turned off for the PCI path, apparently for Tuxracer.
That seemed unneeded and was removed. No visible impact on Tuxracer.
Abstracted all buffer blit operations in via_ioctl.c. The blitter context
is now reprogrammed before each blitting operation.
Updated driver date and drm version requirement.
(Bugzilla Bug #1950, Thomas Hellström)
|
|
Stop the driver from giving empty texture levels with erroneous HW addresses
to the hardware. These will get caught and rejected in the DRM command
verifier.
|
|
Fixed an apparent race condition during locking and drawable info updating,
the result of which was the DRI client leaving traces on the screen rendering
where the drawable previously had been.
If the driver hangs X for a second or so and then restarts, this is probably
the place to look. (Thomas Hellström)
|
|
Check for texture allocation failure (Bugzilla #1633, Bartlomiej Krol)
Reenabled AGP textures (Thomas Hellström)
|
|
[Freedesktop BUG 2080]
|
|
|
|
|
|
Now, the driver's Viewport routine should call _mesa_ResizeBuffersMESA()
if necessary.
Cleaned up code related to GLframebuffer width/height initialization.
Set initial viewport/scissor params in _mesa_make_current2(), instead of
in the drivers' MakeCurrent functions.
|