- Implemented HW-assisted depth span functions. All access to the depth
buffer is done via the engine, so we don't have to worry about the
differences between linear and tiled memory.
- Implemented significantly more 3D-friendly static partitioning of
offscreen memory. 2D still gets enough pixmap cache to be performant.
For example, 800x600@16bpp texture memory goes from ~3mb to 11.5mb on a
16mb card.
- Fix texture uploads. Now request/upload a single buffer at a time,
improves available space sharing with other contexts (rather than
grabbing most of it for a single texture upload). Really need larger
buffers for better efficiency, this is coming...
- Fix glean makeCurrent failure.
Need to track down some strangeness in the texture LRU mechanism seen with
Q3 High Quality, but apart from that we're looking in good shape to
merge with the trunk.
turning CCE off when the X server actually has to do some 2D rendering.
We test if the CCE is running at the start of each accel routine, and
if so we shut it down. If we've shut it down, we start it up again in
LeaveServer(). An extra if statement in the accel routines is worth it,
IMNSHO.
Updates to the elt path code. Allow indirect buffer firing to specify start
and end parameters for buffer reuse. DRM ioctls and wrappers adjusted
accordingly. Still disabled by default.
Try and fix Mobility 128 lockups. Seems to lock when shutting down the X
server from non-standard resolutions, so I've probably messed up the
CCE ioctls somewhat. Default panel resolution seems to be rock solid...
locking fix. Usual caveats apply to using development code.
Includes:
- ctx->Texture.Enabled to ctx->Texture.ReallyEnabled fix
- More useful information in GL_RENDERER string
- More indirect buffer support work
blit buffers can be acquired in batches as needed.
Flush vertex buffers via drmR128FlushVertexBuffer. This ioctl no longer
acquires buffers, it merely flushes and/or discards them.
buffes are handled.
The way buffers are acquired will probably be changed back to what it was
originally, using the DRM_DMA-style ioctl. Texture uploads should grab
as many buffers as needed in a single ioctl.
- New security model for i810
- Enable i810 dri by default
- New indexed vertex path for mga
- Mga kernel driver rework
- Removed dead files in i810 driver
