vl: Don't wrap blocks in pipe_user_buffers.

Mallocing/free eat up a noticeable amount of CPU time for no
practical benefit.

vl: Use pipe_surface instead of pipe_texture in interfaces.

vl: Get rid of pipe_video_surface.

gallium: Add common video format enums.

Merge remote branch 'origin/master' into pipe-video

Conflicts:
configure.ac
src/gallium/auxiliary/vl/Makefile
src/gallium/auxiliary/vl/SConscript
src/gallium/auxiliary/vl/vl_compositor.c
src/gallium/auxiliary/vl/vl_compositor.h
src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
src/gallium/drivers/nouveau/nouveau_winsys.h
src/gallium/drivers/softpipe/sp_video_context.c
src/gallium/include/pipe/p_video_state.h
src/gallium/include/state_tracker/drm_api.h
src/gallium/state_trackers/xorg/xvmc/surface.c
src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.h
src/gallium/winsys/drm/radeon/core/radeon_drm.c
src/gallium/winsys/g3dvl/nouveau/Makefile

vl: Subpicture/compositing fixes.

vl: Add some basic debug output for XvMC.

Set the XVMC_DEBUG env var to:
0 for no extra output
1 for error output
2 for warning output
3 for tracing output

vl: Add switches to autoconf.

Pass 'xorg/xvmc' to --with-state-trackers to get the XvMC state tracker.
Pass --enable-gallium-g3dvl to enable the winsys.

Revert "mesa: Remove pointless comparison of unsigned integer with a negative constant."

This reverts commit a05fdbcb719ac64e6be842372813f0f4ca2f4f93.

Removing the comparison is wrong. The comparison with -1 should be changed
to another value (probably PROGRAM_UNDEFINED) along with another
change in the shader assembler.

Conflicts:

src/mesa/shader/prog_execute.c

intel: Remove non-kernel-exec-fencing support.

Shaves 60k off the driver from removing the broken spans code.  This
means we now require 2.6.29, which seems fair given that it's a year
old and we've removed support for non-KMS already in the last release
of 2D.

intel: Remove the unused s8 spans code.  Not hit during no_rast piglit.

Shaves 5.5k off of the driver.

Fix comparison of unsigned value against < 0.

softpipe: minor tweaks to texture filtering code

gallivm: implement texture border color, plus tweaks to some wrap modes

The progs/test/texwrap demo looks pretty good, but there are still some
tiny differences from softpipe.  There may be a sub-pixel texcoord
interpolation error somewhere.

There's some room for optimization.  Many of the wrap modes compute
intermediate values that are constant for the texture size (see the
min/max values).  These could be computed earlier and stored somewhere
for later use.

gallivm: added texture border_color field

gallivm: added question/comment

llvmpipe: new comment

gallivm: comments

dri/nouveau: Remove some CI remnants.

dri/nouveau: Trivially add GL_ARB_texture_mirrored_repeat.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>

dri/nouveau: Enable GL_EXT_stencil_wrap.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>

dri/nouveau: Add GL_EXT_stencil_wrap for nv04.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>

dri/nouveau: Fix rb->DataType for z24s8 renderbuffers.

dri/nouveau: Try to keep client buffers smaller than the scratch VBO length.

dri/nouveau: Pack client arrays as they're copied to the real BO.

xdemos: add OML sync test

Add program to test glXSwapBuffersMscOML behavior.  Based on glsync.c.

r300g: Fix indexbuf upper limits.

Wine tends to pessimistically use ~0 for its max index, but r300s
only can go up to 2^24-1, causing the kernel checker to freak out.

Civ4 is marginally improved now. Still crashes, but not as bad.

gallivm: rework and implement more texture wrap modes

All the texture wrap modes are now implemented for linear and nearest
sampling.

However, texture border color is not yet supported so some wrap modes
will produce int coords outside the texture bounds.  We'll get garbage
values insted of the texture border color for now.

gallivm: added lp_build_fract()

gallivm: added lp_build_set_sign()

gallivm: added lp_build_negate()

gallivm: clarify unsigned vs. signed integer type construction

The lp_int_type() function was creating an unsigned type.  So rename
that function to lp_uint_type() and create new lp_int_type() that
creates a signed type.

i915: Enable texture tiling by default.

This now passes piglit testcases, and shows performance improvements
on openarena.  Blit-heavy apps may see degradation, but the
expectation is that the common case of 3D rendering will see
significant wins.

This regresses gen-teximage on my 855, and no piglit tests on my 945.

i915: Fix up i830 for tiled drawing offsets.

Corresponds to b87406e55f029d29594ae76a4b39a4fe1007fe4f.

fix getprocaddress test

DRI: Put back missing break-statement

The break in the __DRI_ATTRIB_RENDER_TYPE case was accidentally
removed in commit 5cf2c5851bcd29c2d53bb04ab692b4b156f5a74d.  This puts
it back.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

Revert "mesa: Fix unsigned comparison."

This reverts commit f9504e75f02586a8561733e0e2711c65efa2979d.

This patch is incorrect.

glsl: Remove unsigned greater than or equal zero comparison.

mesa: Fix unsigned comparison.

glsl: Remove unsigned greater than or equal zero comparison.

glsl: Return NULL on _slan_gen_asm error path on non-debug builds.

Exit the function early on the error path, instead of continuing, to
prevent a null pointer dereference later on.

glsl: Add assert to check input to strcmp.

mesa: Add asserts to check inputs to memcpy.

r300: make blit code work on swtcl machines.

Only tested this on my rv530 with R300_NO_TCL=1, but it works there
at least. It would be nice to fix the tex_offset by chaning the vertex
program on tcl to read from offset 6 but this isn't really necessary.

Signed-off-by: Dave Airlie <airlied@redhat.com>

r300_blit: initialise whole fp structure to 0

at least on the r500 we get a lockup unless I do this, the compiler
seems to fail to compile and we just get a null fp.

Signed-off-by: Dave Airlie <airlied@redhat.com>

Remove ffb and gamma from configs.

wgl: Fix mingw32 build.

This was missed in commit fcf438e9e03f6e75bca4a49ad372fe7c4b1abbf8.

Fix unmatched parenthesis introduce by previous commits

I wasn't careful enough when removing support for GCC versions earlier
than 3.3.0.  I could have sworn that I compile tested before pushing,
but apparently not.  FAIL.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

Note removal of pre-GCC 3.3.0 support in release notes.

Remove support for GCC older than 3.3.0 from generated sources

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

Remove support for GCC older than 3.3.0 from generator scripts

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

Remove support for GCC older than 3.3.0

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

Add version check for GCC

Mesa now requires GCC 3.3.0 or later.  See the following thread from
the mesa3d-dev mailing list for more details:

http://marc.info/?t=126748568900005&r=1&w=2

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

radeon/r200/r300/r600: add check_blit vtbl function

Check if the native blit formats are supported, if not,
attempt to use an alternate format.

Skip 3, >4 bpp as per comments from mcencora on irc.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>

Note removal of color-index rendering in release notes.

Remove stray defines of HAVE_RGBA

Now that color-index support is removed from t_dd_tritmp.h and
t_dd_unfilled.h, drivers no longer need define HAVE_RGBA.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/tnl_dd: Remove color-index support from t_dd_unfilled.h

The code removal and the re-indent were done together for this one
because the cause of the affected code blocks is much, much smaller
than in t_dd_tritmp.h.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/tnl_dd: White-space / indentation fixes after previous commit

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/tnl_dd: Remove color-index support from t_dd_tritmp.h

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove unused RasterIndex field

With the preceeding changes, gl_current_attrib::RasterIndex is not
used.  Remove it.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Always return default value for CURRENT_RASTER_INDEX

Since there is no color-index rendering, it is impossible to update
this value.  Just return the initial setting and be happy.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Eliminate index parameter to _mesa_feedback_vertex

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove ClearIndex and IndexMask from device-driver interface

These are used to inform the driver of the clear value for color-index
buffers and to control write-masking of bits in color-index buffers.
No driver use or need (not even Nouveau) these interfaces.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove _swrast_logicop_ci_span

After all the recent color-index rendering removal,
_swrast_logicop_ci_span is no longer used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove _mesa_add_color_index_renderbuffers

After all the recent color-index visual support removal,
_mesa_add_color_index_renderbuffers is no longer used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove checks of Visual.rgbMode in Get paths

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove checks of Visual.rgbMode

This must always be true now, so there is no reason to check it.  Ever.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Remove support for creating color-index visuals

Remove the rgbMode and indexBits parameters from _mesa_create_visual
and _mesa_initialize_visual.  These values are now hardcoded to
GL_TRUE and 0.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

tnl: Remove color-index TNL support

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

DRI: Remove support for color-index visuals

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast_setup: Remove more color-index rendering support

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove color-index rendering support from s_spantemp.h

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/xlib: Remove support for color-index rendering

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

osmesa: Remove support for color-index rendering

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

fbdev: Remove support for color-index rendering

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

DRI/swrast: Remove color-index support from DRI swrast driver

It appears that color-index rendering wasn't actually supported
anyway.  swrastFillInModes did not previously create an color-index
configs, so it doesn't seem like there would be any way to get a
color-index visual.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove SPAN_INDEX

Also adjust the bits that appear after it to fill in the gap.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove _swrast_read_index_span

After all the recent color-index rendering removal,
_swrast_read_index_span is no longer used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove _swrast_mask_ci_span

After all the recent color-index rendering removal,
_swrast_mask_ci_span is no longer used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove _swrast_fog_ci_span and associated code

After all the recent color-index rendering removal,
_swrast_fog_ci_span is no longer used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove _swrast_write_index_span and associated code

After all the recent color-index rendering removal,
_swrast_write_index_span is no longer used anywhere.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove remaining color-index state tracking infrastructure

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for ReadPixels from a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for Clear into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for rendering antialiased lines into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for rendering antialiased triangles into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for rendering lines into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for rendering points into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for rendering triangles into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for Bitmap into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove _swrast_write_zoomed_index_span and associated code

After removing support for CopyPixels and DrawPixels involving
color-index buffers, _swrast_write_zoomed_index_span is no longer
used.  Removed it and all the support for COLOR_INDEX formats in zoom_span.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for CopyPixels to / from a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

swrast: Remove support for DrawPixels into a color-index buffer

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

llvmpipe: Store the original triangle coordinates in the debug build.

For debugging purposes only.

llvmpipe: Ensure context is flushed when destroying shaders.

i915: Use x,y drawing offsets instead of changing buffer offsets.

This should fix rendering into mipmaps of tiled textures.

i915: Tell the kernel when we actually need fence registers on our BOs.

This improves tiled texture performance of OA on my 945 from 25.3fps
to 29.0fps, whereas untiled is 28.2fps, by avoiding stalls for fence
register changes.

intel: Use drm_intel_bo_alloc_tiled for region allocs.

This moves the logic for how to align pitches, heights, and sizes of
objects to one central location.  Fixes rendering with texture tiling
on i915.  Note that current libdrm is required for the change for
I915_TILING_NONE pitch alignment.

i915: Don't do the pitch expansion for tiled buffers.

The weirdness that led to the bumping of pitch for those 512/1024
pixels is that in taking a 2x2 subspan or bilinear filtering we'd end
up hitting the same channel in 2 different pages, leading to lower
performance.  With tiling, that doesn't occur, so we don't need to
waste the memory.

intel: Use GTT mapping in the blit fallback path in case there's tiling.