r600g: take alpha ref update out of line.

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

r600g: split resource emit path from main register emit path

Since resources don't generally vary in size, this splits
the emit path, it also takes into a/c that texture and vertex resources
have different number of relocs, and avoids emitting the extra
reloc for vertex resources.

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

r600g: exit bo per reg scanning loop early.

Exit this loop early to avoid pointless iterations later.

Move the resource bos to the first two regs, it actually
doesn't matter which regs we use for this in resource land.

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

r600g: reorder evergreen draw packets to be smaller.

We were always re-emitting lots of unnecessary changes here,
avoid doing that.

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

r600g: inline r600_bo_reference.

This relies on the reference member being first, so document it.

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

r600g: no need to drop the references here.

We drop them when we reference the new objects in the next line.

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

r600g: use memcmp instead of a loop in state_set_resource

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

mesa: Ignore blits to/from missing buffers

The EXT_framebuffer_object spec (and later specs) say:

     "If a buffer is specified in <mask> and does not exist in both
     the read and draw framebuffers, the corresponding bit is silently
     ignored."

Check for color, depth, and stencil that the source and destination
FBOs have the specified buffers.  If the buffer is missing, remove the
bit from the blit request mask and continue.

Fixes the crash in piglit test 'fbo-missing-attachment-blit from', and
fixes 'fbo-missing-attachment-blit es2 from'.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=37739
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
NOTE: This is a candidate for the stable branches.

mesa: Don't try to clear a NULL renderbuffer

In an ES2 context (or if GL_ARB_ES2_compatibility) is supported, the
framebuffer can be complete with some attachments be missing.  In this
case the _ColorDrawBuffers pointer will be NULL.

Fixes the crash in piglit test fbo-missing-attachment-clear.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=37739
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
NOTE: This is a candidate for the stable branches.

mga: enable GL_ARB_vertex_array_object extension

Tested on a Matrox G550 AGP.

Signed-off-by: Nicolas Kaiser <nikai@nikai.net>
Signed-off-by: Brian Paul <brianp@vmware.com>

egl_dri2: Add missing header

egl_dri2: Add missing license

egl_dri2: Use libudev only if available

Broken since 7f881c43dfb4f1aeeab3a84125b5c106c191a43f.

r600g : fix incorrect size computation in r600_query_result

query->num_results already has the size in dwords of the query
buffer.  There no need to multiply again.  We were reading past
the end of the buffer, resulting in reading garbage.

Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=37028

agd5f: clarify the comment.

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

r600g: remove pre-r6xx asic families and pci ids

Not sure why these were included originally.

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

r600g: always clear query memory

According to the hw documentation, the driver needs to:
- allocate 128 bits for each possible DB
- clear the 128 bits for each possible DB
- write 1 to bits 127 and 63 for upper DBs that don't
exist on a particular asic

Previously we were only doing these steps if the
asic had less than the max possible DBs.

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

r300g: consolidate deducing chipset info

Use the new PCI ID table, make it simpler.

r300_pci_ids: set families to match r300g

r600g: add missing r300 families

Wondering why r600g needs to include r300_pci_ids.h

intel: Update intel-decode.c from intel-gpu-tools.

intel: Implement glFinish() correctly by waiting on all previous rendering.

Before, we were waiting for (most of) the current framebuffer to be
done, which is not quite the same thing.

darwin: Fix VG_LIB_GLOB to also match the unversioned symlink

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

darwin: Don't link against libGL when building libOSMesa

Everything should be resolved through glapi.

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

r300_pci_ids: add missing 4B48

Found in r300_chipset.h.

r300g: Remove is_r3xx

Use r300_pci_ids.h instead.

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

r600g: Use radeon pciid list for the family lookup table

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

radeon: Use pciid list to generate PCI_CHIP_<FAMILY>_<ID> defines

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

targets/egl: Support driver name lookup using pci lists

Make use of this in drm and wayland st/egl backends.

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

egl_dri2: Use external driver pci list

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

Add radeon pci id lists

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

Add intel pci id lists

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

tgsi: s/varient/variant/

draw: rename draw_vs_varient.c to draw_vs_variant.c

draw/llvm: whitespace, formatting fixes

draw: s/1/TRUE/

i965: Fix flipped GT1 vs GT2 URB VS entry count limits.

Merge branch 'master' of ssh://git.freedesktop.org/git/mesa/mesa

i915g: Do generic remapping.

With complex shaders there are often "holes" in the fs inputs, and we only
have 8 tex coorsd to map those to. To fix this, we remap fs inputs to [0..8].
This lets us to run many more GLSL programs.

darwin: Set VG_LIB_{NAME,GLOB} to fix make install

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

r600g: optimise the draw emission packets for r600/eg

This just reduces code size a bit for this chunk.

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

r600g: use an enabled list to track enabled blocks.

At the end of flushing we were scanning over 450 blocks
with generally about 50 enabled. This reduces the scanning
to just the list of enabled blocks.

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

r600g/winsys: overhaul resource range/blocks.

There isn't much point taking the overhead of range/block lookups on resources
we aren't going to be getting resource registers at wierd offsets.

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

r600g: split out block init

This just splits this function up as pre-cursor to reusing
the internals of it.

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

r600g: move resource setting to its own structures.

resource setting could be a fair bit more lightweight,
this patch just separates the resource structs from the standard
reg tracking structs in the driver, later patches will improve
the winsys.

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

r600g: only call bo loop if nbos is positive.

we don't need to loop over all the registers unless we have
some bos in the block, also avoid setting the ctx flags,
and move the optional stuff down below this chunk.

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

r600g: don't need to call the packet dirty function if not dirty.

also fix a unneeded dirty check and add a dirty check speedup.

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

i915g: implement more opcodes.

i915g: implement TGSI_OPCODE_SEQ.

i915g: handle varyings properly.

i915g: add missing and unsupported PIPE_CAPs to avoid debug messages.

dri2: protect dri2FlushFrontBuffer against NULL buffers.

Gallium: fix indentation in u_blitter.c

i915g: don't destroy a texture buffer if it's NULL.

llvmpipe: use $(CXX) instead of g++ for linking.

This allows setting the path to the C++ compiler.

i965: Update SURFACE_STATE dumping for Ivybridge.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Update SAMPLER_STATE dumping for Ivybridge.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Update SF_CLIP_VIEWPORT state dumping for Ivybridge.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>

apple: Package applegl source into MesaLib tarball

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

darwin: Define GALLIUM_DRIVERS_DIRS in darwin config

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Finish build fixes for applegl_glx.c

libGL.dylib now *builds* on darwin.

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: More fixes for compilation failures

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

softpipe: Anisotropic filtering extension.

Reference implementation which produces high quality renderings.
Based on Higher Quality Elliptical Weighted Avarage Filter (EWA).

Signed-off-by: Brian Paul <brianp@vmware.com>

st/mesa: fix renderbuffer internal format for accum buffer

Fixes an accum buffer regression since switching to the table-based
format selection code.

util: add \n to debug_checkpoint_full

Signed-off-by: José Fonseca <jfonseca@vmware.com>

nv50: fix nv50_sampler_state_delete array overflow

num_samplers is array of 3 elements, not 5

apple: Fix multiple build failures in applegl_glx.c

It is still not building, but this gets us many steps closer

See https://bugs.freedesktop.org/show_bug.cgi?id=29162

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Build darwin using applegl rather than indirect

This reverts portions of 6849916170c0275c13510251a7b217c20f2b993e that caused
the darwin config to fail to build due to missing implementations in that
commit.

See https://bugs.freedesktop.org/show_bug.cgi?id=29162

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

glx: Dead code removal

Remove a redundant ifndef GLX_USE_APPLEGL

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: ifdef out come glapi-foo on darwin

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Change from XExtDisplayInfo to struct glx_display

Fixes regression introduced by: ab434f6b7641a64d30725a9ac24929240362d466 and
                                c356f5867f2c1fad7155df538b9affa8dbdcf869

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

dri/nouveau: fix gnome-shell segfault

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

apple: Rename GLXcontext

Fixes regression introduced by: c356f5867f2c1fad7155df538b9affa8dbdcf869

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Rename _gl_context_modes_find_visual to glx_config_find_visual

Fixes regression introduced by: 6ddf66e9230ee862ac341c4767cf6b3b2dd2552b

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Re-add driContext and do_destroy

Fixes regression introduced by: c491e585e43d48a2aeec96ccc4008da6c443fb42

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Rename GLXcontext

Fixes regression introduced by: c356f5867f2c1fad7155df538b9affa8dbdcf869

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Rename __GLcontextModes to struct glx_config

Fixes regression introduced by: 6ddf66e9230ee862ac341c4767cf6b3b2dd2552b

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Rename glcontextmodes.[ch] to glxconfig.[ch]

Fixes regression introduced by: 65d98e25770487456eb3d7eb8ec3ec8272f170b1

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

apple: Update GL specs

Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>

meta: Don't do sRGB encode for framebuffer blits on sRGB-enabled framebuffers.

Fixes fbo-srgb-blit.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=35373
Reviewed-by: Brian Paul <brianp@vmware.com>

meta: Don't do srgb to linear decode when blitting srgb textures.

Fixes the GL_SRGB8_ALPHA8 -> GL_RGBA8 blits in fbo-srgb-blit.c

Reviewed-by: Brian Paul <brianp@vmware.com>

glx: Fix use-before-null-check in dri2InvalidateBuffers().

The compiler used our dereference here to skip the NULL check below.
Fixes window resize in "jconsole -J-Dsun.java2d.opengl=True" under
OpenJDK 6.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=37766
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Drop remaining strict conformance fallback for GL_POINT_SMOOTH.

We actually could do this in hardware in the fragment shader using
gl_PointCoord and the point's size.

i965: Drop strict conformance fallback for GL_LINE_STIPPLE.

We implement line stipples, just not *quite* correctly.  We have a
piglit testcase to use when we want to fix it, if we do.  Until then,
don't lie to our test suites.

i965: Drop strict conformance fallback for GL_LINE_SMOOTH.

We do have hardware antialised lines.  If we care, we should actually
fix them to be conformant (or as close as possible) instead of using
this knob to fool testcases using swrast.

For some interesting reading on the state of GL_*_SMOOTH across
several drivers, see:
http://homepage.mac.com/arekkusu/bugs/invariance/HWAA.html

i965: Drop strict conformance fallback for GL_POLYGON_SMOOTH.

From my reading of the GL 2.1 spec, no antialiasing is strictly
conformant for polygon smoothing.  Yes, it's absurd, but then,
hardware doesn't support this so maybe it's not so absurd.

i965: Drop INTEL_CONFORMANCE=2 fallback code.

This was just a duplicate of no_rast=true driconf option, which is
relatively standard across drivers.

glsl: fixed printing of structure constants.

ir_print_visitor::visit(ir_constant *) was failing to index properly
into ir->type->fields.structure, so the first field name was being
reprinted for every field in the structure.

Signed-off-by: Brian Paul <brianp@vmware.com>

AST dump: fixed printing of conditionals.

ast_expression::print() had an incorrect index into the subexpressions
array, so (a ? b : c) was being incorrectly rendered as (a ? b : b).

Signed-off-by: Brian Paul <brianp@vmware.com>

r600g: move spi update to only when states change.

This updates the spi state after ps/vs binding or rasteriser state
change.

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

cso: move cso hashes to a more table driven scheme

this removes a bad branch pain in the hash table lookup fn.

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

u_prim: convert u_trim_pipe_prim to table driven.

This makes this function not be an always miss for the branch predictor.

Noticed using cachegrind, makes a minor difference to gears numbers on r600g.

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

r600g: make conv pipe prim table driven.

This is a lot more branch predictor friendly, it actually
showed up in cachegrind profiles.

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

mesa: fix typo (s/GLGL/GLSL/)

mesa: 80-column wrapping

mesa: consolidate glGetUniform code

mesa: refactor/consolidate uniform lookup code

mesa: add some minor fixes for geometry shaders

dri: add missing files from 873379a8818eed9ab16c24728b7091a3a3705c5b

mesa: add implementation of glDrawElementsInstancedBaseVertex

Signed-off-by: Brian Paul <brianp@vmware.com>

glapi: regenerated files

Signed-off-by: Brian Paul <brianp@vmware.com>