mesa.git
12 years agoi965: Turn on ARB_depth_buffer_float by default.
Eric Anholt [Thu, 5 Jan 2012 02:02:49 +0000 (18:02 -0800)]
i965: Turn on ARB_depth_buffer_float by default.

Everything about this that we have tests for works except for the
deprecated metaops.  The conclusion we came to on IRC sounded like we
were OK with turning it on as long as core functionality works.  The
remaining failures (copypixels, drawpixels) should just be a matter of
finishing the MapRenderbuffer for them.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoswrast: Convert the glBlitFramebuffer(GL_NEAREST) path to MapRenderbuffer().
Eric Anholt [Thu, 5 Jan 2012 01:54:48 +0000 (17:54 -0800)]
swrast: Convert the glBlitFramebuffer(GL_NEAREST) path to MapRenderbuffer().

Fixes on i965:
ARB_depth_buffer_float/fbo-depthstencil-GL_DEPTH32F_STENCIL8-blit
ARB_depth_buffer_float/fbo-stencil-GL_DEPTH32F_STENCIL8-blit

Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agomesa: Fix packing of stencil bits to MESA_FORMAT_Z32_FLOAT_X24S8.
Eric Anholt [Thu, 5 Jan 2012 01:53:39 +0000 (17:53 -0800)]
mesa: Fix packing of stencil bits to MESA_FORMAT_Z32_FLOAT_X24S8.

We were converting our ubyte stencil value to a float.  Just write it
as a uint, which overwrites the X24 part of X24S8 with 0 but shouldn't
matter.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agoswrast: Fix use of uninitialized value in rbmapping changes.
Eric Anholt [Thu, 5 Jan 2012 01:11:33 +0000 (17:11 -0800)]
swrast: Fix use of uninitialized value in rbmapping changes.

I'm so surprised that gcc didn't catch this that I feel like I must be
misreading.  srcMap is what we initialize (along with dstMap) from
this map value right after this check.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agoswrast: Use the fast copypixels code to implement fast glBlitFramebuffer().
Eric Anholt [Thu, 5 Jan 2012 00:10:14 +0000 (16:10 -0800)]
swrast: Use the fast copypixels code to implement fast glBlitFramebuffer().

They were meaning to do the same thing of memcpying rows, so just
write the code once.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agoswrast: Move the check for fast_copy_pixels() ability to caller.
Eric Anholt [Wed, 4 Jan 2012 23:59:39 +0000 (15:59 -0800)]
swrast: Move the check for fast_copy_pixels() ability to caller.

I'm going to reuse this function from glBlitFramebuffer() handling,
which wants to do the same thing.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agoi965: Add sensible disasm for the JMPI instruction.
Eric Anholt [Mon, 19 Dec 2011 18:52:31 +0000 (10:52 -0800)]
i965: Add sensible disasm for the JMPI instruction.

We care about the jump distance, not that the first src is always the
ip register.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965/gen7: Enable transform feedback as long as kernel support is present.
Eric Anholt [Fri, 30 Dec 2011 07:37:17 +0000 (23:37 -0800)]
i965/gen7: Enable transform feedback as long as kernel support is present.

The last major issue (intervening-read) is fixed, so let's turn this
on for real.  The only other known issue is a hardware limitation for
tesselation with flat shading.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
12 years agoi965/gen7: Fix up the transform feedback buffer pointers on later batches.
Eric Anholt [Fri, 30 Dec 2011 07:26:48 +0000 (23:26 -0800)]
i965/gen7: Fix up the transform feedback buffer pointers on later batches.

Fixes piglit EXT_transform_feedback/intervening-read

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965/gen7: Flush the batch between transform feedbacks.
Eric Anholt [Fri, 30 Dec 2011 02:24:02 +0000 (18:24 -0800)]
i965/gen7: Flush the batch between transform feedbacks.

We need the kernel to reset our pointers to 0 in between.  Note that
the initialization of function pointer had to move to after
InitContext since we didn't have intel->gen set up yet.

Fixes piglit EXT_transform_feedback/immediate-reuse

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965/gen7: Use the updated interface for SO write pointer resetting.
Eric Anholt [Fri, 30 Dec 2011 02:12:48 +0000 (18:12 -0800)]
i965/gen7: Use the updated interface for SO write pointer resetting.

The new kernel patch I submitted makes the interface opt-in, so all
batchbuffers aren't preceded by the 4 MI_LOAD_REGISTER_IMMs.  This
requires the updated i915_drm.h present in libdrm 2.4.30.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agor600g: remove obsolete MULTIWRITE comment
Alex Deucher [Fri, 6 Jan 2012 15:15:37 +0000 (10:15 -0500)]
r600g: remove obsolete MULTIWRITE comment

fs writes all is implemented in the shader on eg+.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
12 years agoglsl_to_tgsi: Remove the obsolete remove_output_reads pass.
Kenneth Graunke [Mon, 2 Jan 2012 10:43:41 +0000 (02:43 -0800)]
glsl_to_tgsi: Remove the obsolete remove_output_reads pass.

This is now handled by the GLSL compiler, so this code is dead.

Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agoglsl_to_tgsi: Use the GLSL compiler's new remove-output-reads pass.
Vincent Lejeune [Mon, 2 Jan 2012 19:17:38 +0000 (20:17 +0100)]
glsl_to_tgsi: Use the GLSL compiler's new remove-output-reads pass.

The existing glsl_to_tgsi::remove_output_read pass did not work properly
when indirect addressing was involved; this commit replaces it with a
lowering pass that occurs before TGSI code generation.

Fixes varying-array related piglit tests.

Signed-off-by: Vincent Lejeune <vljn@ovi.com>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agoglsl: Add a lowering pass to remove reads of shader output variables.
Vincent Lejeune [Tue, 3 Jan 2012 10:08:32 +0000 (02:08 -0800)]
glsl: Add a lowering pass to remove reads of shader output variables.

This is similar to Gallium's existing glsl_to_tgsi::remove_output_read
lowering pass, but done entirely inside the GLSL compiler.

Signed-off-by: Vincent Lejeune <vljn@ovi.com>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodraw: fix piglit base vertex + user vertex array tests
Dave Airlie [Mon, 2 Jan 2012 17:41:34 +0000 (17:41 +0000)]
draw: fix piglit base vertex + user vertex array tests

This fixes
draw-elements-base-vertex user_varrays
draw-elements-instanced-base-vertex user_varrays
for softpipe with no llvm support (DRAW_USE_LLVM=false)

I'm not sure if this is the correct answer, but these tests were showing
a max_index of 7, then trying to fetch up to 43, maybe it should be fixing
max_index earlier somewhere to take care of this.

Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agoStrip LLVM svn rev from `llvm-config --version` output.
ojab [Thu, 5 Jan 2012 10:18:34 +0000 (14:18 +0400)]
Strip LLVM svn rev from `llvm-config --version` output.

Reviewed-by: Vinson Lee <vlee@freedesktop.org>
12 years agomesa: Remove 'texelBytes' declarations that are only used in assertions.
Vinson Lee [Thu, 5 Jan 2012 07:52:19 +0000 (23:52 -0800)]
mesa: Remove 'texelBytes' declarations that are only used in assertions.

This patch silences these GCC warnings.
warning: unused variable 'texelBytes'

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agomesa: Fix transform feedback of unsubscripted arrays.
Paul Berry [Wed, 4 Jan 2012 04:41:34 +0000 (20:41 -0800)]
mesa: Fix transform feedback of unsubscripted arrays.

It is not explicitly stated in the GL 3.0 spec that transform feedback
can be performed on a whole varying array (without supplying a
subscript).  However, it seems clear from context that this was the
intent.  Section 2.15 (TransformFeedback) says this:

    When writing varying variables that are arrays, individual array
    elements are written in order.

And section 2.20.3 (Shader Variables), says this, in the description
of GetTransformFeedbackVarying:

    For the selected varying variable, its type is returned into
    type. The size of the varying is returned into size. The value in
    size is in units of the type returned in type.

If it were not possible to perform transform feedback on an
unsubscripted array, the returned size would always be 1.

This patch fixes the linker so that transform feedback on an
unsubscripted array is supported.

Fixes piglit tests "EXT_transform_feedback/builtin-varyings
gl_ClipDistance[{4,8}]-no-subscript" and
"EXT_transform_feedback/output_type *[2]-no-subscript".

Note: on back-ends that set
gl_shader_compiler_options::LowerClipDistance (for example i965),
tests "EXT_transform_feedback/builtin-varyings
gl_ClipDistance[{1,2,3,5,6,7}]" still fail.  I hope to address this in
a later patch.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
12 years agoAdd .gitignore files to exclude unit test build artifacts from git
Paul Berry [Tue, 3 Jan 2012 05:05:43 +0000 (21:05 -0800)]
Add .gitignore files to exclude unit test build artifacts from git

With the addition of unit tests in commit
3ef3ba4d2eee36f64062a21ce030c3f4d8c4cac4, several additional build
artifacts are created:

  bin/depcomp
  bin/missing
  tests/Makefile
  tests/Makefile.in
  tests/glx/Makefile
  tests/glx/Makefile.in
  tests/glx/.deps/
  tests/glx/.gitignore

This patch adds all of these files to .gitignore.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
12 years agomesa: Avoid segfault when getting an unbound transform feedback buffer name.
Paul Berry [Tue, 3 Jan 2012 21:59:13 +0000 (13:59 -0800)]
mesa: Avoid segfault when getting an unbound transform feedback buffer name.

Previously we were using
gl_transform_feedback_object::Buffers[i]->Name to service an indexed
get request for GL_TRANSFORM_FEEDBACK_BUFFER_BINDING.  However, if no
buffer has been bound, gl_transform_feedback_object::Buffers[i] is
NULL, so this was causing a segfault.

This patch switches to using
gl_transform_feedback_object::BufferNames[i], which is equal to
gl_transform_feedback_object::Buffers[i]->Name if
gl_transform_feedback_object::Buffers[i] is not NULL, and 0 if it is
NULL.

Fixes piglit test "EXT_transform_feedback/get-buffer-state
indexed_binding".

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agomesa: Fix transform feedback of gl_ClipDistance.
Paul Berry [Tue, 27 Dec 2011 03:39:25 +0000 (19:39 -0800)]
mesa: Fix transform feedback of gl_ClipDistance.

On drivers that set gl_shader_compiler_options::LowerClipDistance (for
example i965), references to gl_ClipDistance (a float[8] array) will
be converted to references to gl_ClipDistanceMESA (a vec4[2] array).

This patch modifies the linker so that requests for transform feedback
of gl_ClipDistance are similarly converted.

Fixes Piglit test "EXT_transform_feedback/builtin-varyings
gl_ClipDistance".

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agomesa: Make tfeedback_decl::var_name a const char *.
Paul Berry [Tue, 27 Dec 2011 16:24:57 +0000 (08:24 -0800)]
mesa: Make tfeedback_decl::var_name a const char *.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agogallium: Make use of gl_transform_feedback_info::ComponentOffset.
Paul Berry [Tue, 27 Dec 2011 03:31:44 +0000 (19:31 -0800)]
gallium: Make use of gl_transform_feedback_info::ComponentOffset.

Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965: Make use of gl_transform_feedback_info::ComponentOffset.
Paul Berry [Tue, 27 Dec 2011 03:31:20 +0000 (19:31 -0800)]
i965: Make use of gl_transform_feedback_info::ComponentOffset.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agomesa: Add gl_transform_feedback_info::ComponentOffset.
Paul Berry [Tue, 27 Dec 2011 03:30:10 +0000 (19:30 -0800)]
mesa: Add gl_transform_feedback_info::ComponentOffset.

When using transform feedback, there are three circumstances in which
it is useful for Mesa to instruct a driver to stream out just a
portion of a varying slot (rather than the whole vec4):

(a) When a varying is smaller than a vec4, Mesa needs to instruct the
driver to stream out just the first one, two, or three components of
the varying slot.

(b) In the future, when we implement varying packing, some varyings
will be offset within the vec4, so Mesa will have to instruct the
driver to stream out an arbitrary contiguous subset of the components
of the varying slot (e.g. .yzw or .yz).

(c) On drivers that set gl_shader_compiler_options::LowerClipDistance,
if the client requests that an element of gl_ClipDistance be streamed
out using transform feedback, Mesa will have to instruct the driver to
stream out a single component of one of the gl_ClipDistance varying
slots.

Previous to this patch, only (a) was possible, since
gl_transform_feedback_info specified only the number of components of
the varying slot to stream out.  This patch adds
gl_transform_feedback_info::ComponentOffset, which indicates which
components should be streamed out.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965: Fix transform feedback of gl_ClipVertex.
Paul Berry [Mon, 26 Dec 2011 22:20:47 +0000 (14:20 -0800)]
i965: Fix transform feedback of gl_ClipVertex.

Previously, on i965 Gen6 and above, we weren't allocating space for
gl_ClipVertex in the VUE, since the VS was automatically converting it
to clip distances.  This prevented transform feedback from being able
to capture gl_ClipVertex.

This patch goes aheads and allocates space for gl_ClipVertex in the
VUE on Gen6 and above.  The old behavior is retained on Gen5 and
below, since (a) transform feedback is not yet supported on those
platforms, and (b) those platforms don't currently support
gl_ClipVertex anyhow.

Note: this constitutes a slight waste of VUE space for shaders that
use gl_ClipVertex and don't use transform feedback to capture it.
However, that seems preferable to making the VUE map (and all of the
state that depends on it) dependent on transform feedback settings.

Fixes Piglit test "EXT_transform_feedback/builtin-varyings
gl_ClipVertex".

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965: Fix transform feedback of gl_PointSize.
Paul Berry [Mon, 26 Dec 2011 21:58:46 +0000 (13:58 -0800)]
i965: Fix transform feedback of gl_PointSize.

On i965 Gen6 and above, gl_PointSize is stored in component W of the
first VUE slot (which corresponds to VERT_RESULT_PSIZ in the VUE map).
Normally we store varying floats in component X of a VUE slot, so we
need special case logic for gl_PointSize.

For Gen6, we do this with a ".wwww" swizzle in the GS.  For Gen7, we
shift the component mask by 3 to select the W component.

Fixes Piglit test "EXT_transform_feedback/builtin-varyings
gl_PointSize".

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agomesa: Fix extra memset in store_tfeedback_info()
Paul Berry [Thu, 5 Jan 2012 21:06:36 +0000 (13:06 -0800)]
mesa: Fix extra memset in store_tfeedback_info()

Commit 9d36c96d6ec9f2c05c8e0b9ef18c5462cddee8c1 (mesa: Fix
glGetTransformFeedbackVarying()) accidentally added an extra memset()
call to the store_tfeedback_info() function, causing
prog->LinkedTransformFeedback.NumBuffers to be erased.

This patch removes the extra memset and rearranges the other
operations in store_tfeedback_info() to be in the correct order.

Fixes piglit tests "EXT_transform_feedback/api-errors *unbound*"

Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agoswrast: use memmove() instead of memcpy() in the SHIFT_ARRAY macro
Brian Paul [Thu, 5 Jan 2012 18:59:30 +0000 (11:59 -0700)]
swrast: use memmove() instead of memcpy() in the SHIFT_ARRAY macro

The src/dst arrays would overlap but dst was less than src so a simple
version of memcpy() would do the right thing.  But this isn't guaranteed
when memcpy() is optimized.

Fixes demos/copypix when the dest region was clipped by the left side of
the window.

Reviewed-by: Adam Jackson <ajax@redhat.com>
12 years agost/mesa: add support for gl_ClipDistance
Bryan Cain [Mon, 2 Jan 2012 20:49:46 +0000 (14:49 -0600)]
st/mesa: add support for gl_ClipDistance

12 years agogallium: add support for clip distances
Bryan Cain [Mon, 2 Jan 2012 20:48:47 +0000 (14:48 -0600)]
gallium: add support for clip distances

12 years agoglx/dri2: print FPS when env var LIBGL_SHOW_FPS is 1 (v2)
Marek Olšák [Wed, 4 Jan 2012 01:09:44 +0000 (02:09 +0100)]
glx/dri2: print FPS when env var LIBGL_SHOW_FPS is 1 (v2)

This is useful for apps which don't print FPS.
Only enabled in SwapBuffers.

v2: track state per drawable, use libGL prefix

Reviewed-by: Michel Dänzer <michel@daenzer.net>
12 years agor300/compiler: fix buffer underflow when setting SEM_WAIT on last instruction
Marek Olšák [Wed, 4 Jan 2012 04:18:58 +0000 (05:18 +0100)]
r300/compiler: fix buffer underflow when setting SEM_WAIT on last instruction

Do it after we check whether inst_end != -1.
Also move the code structure at the beginning of r300_fragment_shader_code
to detect underflows easily with valgrind.

12 years agou_vbuf: use cso_cache to cache vertex element states
Marek Olšák [Wed, 4 Jan 2012 01:05:51 +0000 (02:05 +0100)]
u_vbuf: use cso_cache to cache vertex element states

Improves performance to 28 fps in Cogs.

12 years agou_vbuf: implement another upload codepath which unrolls indices
Marek Olšák [Tue, 3 Jan 2012 21:01:03 +0000 (22:01 +0100)]
u_vbuf: implement another upload codepath which unrolls indices

Improves performance from cca 1 fps to 23 fps in Cogs.
This new codepath is not always used, instead, there is a heuristic which
determines whether to use it. Using translate for uploads is generally
slower than what we have had already, it's a win only in a few cases.

12 years agou_vbuf: cleanup variable names to be consistent
Marek Olšák [Tue, 3 Jan 2012 20:33:55 +0000 (21:33 +0100)]
u_vbuf: cleanup variable names to be consistent

12 years agou_vbuf: cleanup the computation of how many vertices to upload/translate
Marek Olšák [Tue, 3 Jan 2012 20:28:41 +0000 (21:28 +0100)]
u_vbuf: cleanup the computation of how many vertices to upload/translate

12 years agou_vbuf: convert min_index,max_index to start,count
Marek Olšák [Tue, 3 Jan 2012 15:37:17 +0000 (16:37 +0100)]
u_vbuf: convert min_index,max_index to start,count

12 years agoutil: add helper function util_dump_draw_info
Marek Olšák [Tue, 3 Jan 2012 18:18:45 +0000 (19:18 +0100)]
util: add helper function util_dump_draw_info

12 years agotrace: dump primitive restart info
Marek Olšák [Tue, 3 Jan 2012 18:18:26 +0000 (19:18 +0100)]
trace: dump primitive restart info

12 years agotranslate: implement translation of 10_10_10_2 types
Marek Olšák [Mon, 2 Jan 2012 02:34:09 +0000 (03:34 +0100)]
translate: implement translation of 10_10_10_2 types

This is for GL_ARB_vertex_type_2_10_10_10_rev.
I just took the code from u_format_table.c. It's based on pack_rgba_float.
I had no other choice. The u_format hooks are not exactly compatible
with translate. The cleanup of it is left for future work.

Reviewed-by: Dave Airlie <airlied@redhat.com>
12 years agotranslate: implement translation of (pure) integer formats
Marek Olšák [Mon, 2 Jan 2012 02:04:00 +0000 (03:04 +0100)]
translate: implement translation of (pure) integer formats

The conversion is limited to only a few cases, because converting to any other
type shouldn't happen in any driver.

Reviewed-by: Dave Airlie <airlied@redhat.com>
12 years agou_format: implement fetch_rgba_uint and fetch_rgba_sint for integer formats
Marek Olšák [Mon, 2 Jan 2012 01:59:51 +0000 (02:59 +0100)]
u_format: implement fetch_rgba_uint and fetch_rgba_sint for integer formats

Fetching int as float and vice versa is not allowed.
Fetching unsigned int as signed int and vice versa is not allowed either.
Doing conversions like that isn't allowed for samplers in OpenGL.

The three hooks could be consolidated into one fetch hook, which would fetch
uint as uint32, sint as sint32, and everything else as float. The receiving
parameter would be void*. This would be useful for implementing vertex fetches
for shader model 4.0, which has untyped registers.

Reviewed-by: Dave Airlie <airlied@redhat.com>
12 years agogallium: add flag PIPE_TRANSFER_MAP_PERMANENTLY
Marek Olšák [Mon, 2 Jan 2012 00:13:15 +0000 (01:13 +0100)]
gallium: add flag PIPE_TRANSFER_MAP_PERMANENTLY

Please see the diff for further info.

This paves the way for moving user buffer uploads out of drivers and should
allow to clean up the mess in u_upload_mgr in the meantime.

For now only allowed for buffers on r300 and r600.

Acked-by: Christian König <deathsimple@vodafone.de>
12 years agogallium: remove deprecated PIPE_TRANSFER_DISCARD
Marek Olšák [Sun, 1 Jan 2012 23:02:34 +0000 (00:02 +0100)]
gallium: remove deprecated PIPE_TRANSFER_DISCARD

PIPE_TRANSFER_DISCARD_RANGE is defined the same.

12 years agou_vbuf: translate per-vertex, per-instance, and constant attribs separately
Marek Olšák [Sun, 1 Jan 2012 19:27:40 +0000 (20:27 +0100)]
u_vbuf: translate per-vertex, per-instance, and constant attribs separately

We don't wanna convert per-instance or constant (zero-stride) attribs into
ordinary vertex attribs.

More importantly, the translation of instance attribs now finally works.

12 years agou_vbuf: take start_instance into account when uploading instanced attribs
Marek Olšák [Sun, 1 Jan 2012 18:11:25 +0000 (19:11 +0100)]
u_vbuf: take start_instance into account when uploading instanced attribs

12 years agou_upload_mgr: remove the 'flushed' parameter
Marek Olšák [Sun, 1 Jan 2012 17:23:07 +0000 (18:23 +0100)]
u_upload_mgr: remove the 'flushed' parameter

Not used by anybody.

Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agou_vbuf: don't map user buffers, just obtain a pointer to them
Marek Olšák [Sun, 1 Jan 2012 16:53:55 +0000 (17:53 +0100)]
u_vbuf: don't map user buffers, just obtain a pointer to them

12 years agou_vbuf: only map a subrange of buffers to translate
Marek Olšák [Sun, 1 Jan 2012 16:38:22 +0000 (17:38 +0100)]
u_vbuf: only map a subrange of buffers to translate

12 years agogallium: fix behavior of pipe_buffer_map_range
Marek Olšák [Sun, 1 Jan 2012 16:12:35 +0000 (17:12 +0100)]
gallium: fix behavior of pipe_buffer_map_range

To match what transfer_map returns. Really, subtracting the offset leads
to bugs if someone expects it to work exactly like transfer_map.

Reviewed-by: Brian Paul <brianp@vmware.com>
12 years agou_vbuf: remove the workaround for half floats and translate
Marek Olšák [Sun, 1 Jan 2012 12:05:16 +0000 (13:05 +0100)]
u_vbuf: remove the workaround for half floats and translate

12 years agotranslate: implement translation of half floats in the generic codepath
Marek Olšák [Sun, 1 Jan 2012 11:36:44 +0000 (12:36 +0100)]
translate: implement translation of half floats in the generic codepath

12 years agomesa: Remove the dead Varyings list in the program.
Eric Anholt [Tue, 3 Jan 2012 21:27:18 +0000 (13:27 -0800)]
mesa: Remove the dead Varyings list in the program.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
12 years agomesa: Fix glGetTransformFeedbackVarying().
Eric Anholt [Tue, 3 Jan 2012 01:08:13 +0000 (17:08 -0800)]
mesa: Fix glGetTransformFeedbackVarying().

The current implementation was totally broken -- it was looking in an
unpopulated structure for varyings, and trying to do so using the
current list of varying names, not the list used at link time.

v2: Fix leaking of memory into the program per re-link.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
12 years agosvga: Trim the dri binary a bit on scons release builds
Jakob Bornecrantz [Thu, 5 Jan 2012 16:14:13 +0000 (17:14 +0100)]
svga: Trim the dri binary a bit on scons release builds

Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
12 years agosvga: Fix texture cube param cap
Jakob Bornecrantz [Thu, 15 Dec 2011 12:04:56 +0000 (13:04 +0100)]
svga: Fix texture cube param cap

Spotted by Thomas Hellstrom.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
12 years agotgsi: consolidate TGSI string arrays in new tgsi_strings.h
Brian Paul [Thu, 5 Jan 2012 15:30:00 +0000 (08:30 -0700)]
tgsi: consolidate TGSI string arrays in new tgsi_strings.h

There was some duplication between the tgsi_dump.c and tgsi_text.c
files.  Also use some static assertions to help catch errors when
adding new TGSI values.

v2: put strings in tgsi_strings.c file instead of the .h file.

Reviewed-by: Dave Airlie <airlied@redhat.com>
12 years agogallium: add STATIC_ASSERT macro
Brian Paul [Wed, 4 Jan 2012 15:40:04 +0000 (08:40 -0700)]
gallium: add STATIC_ASSERT macro

12 years agomesa: only map src/dest regions in _mesa_copy_buffer_subdata()
Brian Paul [Wed, 4 Jan 2012 21:53:55 +0000 (14:53 -0700)]
mesa: only map src/dest regions in _mesa_copy_buffer_subdata()

We were wastefully mapping the whole source/dest buffers before.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
12 years agomesa: print more info in buffer_object_subdata_range_good() error message
Brian Paul [Wed, 4 Jan 2012 21:15:53 +0000 (14:15 -0700)]
mesa: print more info in buffer_object_subdata_range_good() error message

12 years agost/mesa: 80-column wrapping
Brian Paul [Wed, 4 Jan 2012 21:15:19 +0000 (14:15 -0700)]
st/mesa: 80-column wrapping

12 years agoRevert "configure.ac: remove deprecated --with-driver="
Kenneth Graunke [Thu, 5 Jan 2012 07:49:18 +0000 (23:49 -0800)]
Revert "configure.ac: remove deprecated --with-driver="

This reverts commit 5a478976ae0bb12dd712648d5a9b988dd29c97ef.

It broke the build.  DRI drivers were no longer being installed by
`make install` (and probably not being built at all).  It appears to be
due to a few small, subtle mistakes, and the fix isn't clear enough to
simply commit without going through review.  In the meantime, revert it.

12 years agoglsl: rename VERSION to VERSION_TOK for automake
Matt Turner [Sat, 24 Dec 2011 00:21:05 +0000 (19:21 -0500)]
glsl: rename VERSION to VERSION_TOK for automake

Signed-off-by: Matt Turner <mattst88@gmail.com>
12 years agoconfigure.ac: bump AC_PREREQ to 2.60
Matt Turner [Sat, 24 Dec 2011 00:21:04 +0000 (19:21 -0500)]
configure.ac: bump AC_PREREQ to 2.60

All other xorg modules require at least 2.60 (released in 2006), so we
may as well increase it to match.  It's also doubtful anyone tests the
build with 2.59 (from 2003), so it may not even work anyway.

12 years agoconfigure.ac: remove deprecated --with-driver=
Matt Turner [Sat, 24 Dec 2011 00:21:03 +0000 (19:21 -0500)]
configure.ac: remove deprecated --with-driver=

See 9e7a4147.

12 years agoi965: Enable EXT_texture_integer by default
Ian Romanick [Thu, 5 Jan 2012 00:51:22 +0000 (16:51 -0800)]
i965: Enable EXT_texture_integer by default

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-with-reservations-by: Eric Anholt <eric@anholt.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agomesa: Add missing GL_RG_INTEGER cases
Ian Romanick [Thu, 5 Jan 2012 00:08:20 +0000 (16:08 -0800)]
mesa: Add missing GL_RG_INTEGER cases

Adds two missing '|| srcFormat == GL_RG_INTEGER' in assertions and a
bunch of missing pixel converions cases.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoi965: Fix infinite loop regression in intel_miptree_all_slices_resolve.
Kenneth Graunke [Wed, 4 Jan 2012 23:45:49 +0000 (15:45 -0800)]
i965: Fix infinite loop regression in intel_miptree_all_slices_resolve.

Commit 0ed11e333147e280208d9d0b3ff3f39970547643 fixed a "use after free"
bug by getting the next pointer before deleting the current node.

Unfortunately, it also made "next" never get updated if i->need != need.

Fixes infinite loops in piglit tests fbo-depth-array and fbo-depthtex.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
12 years agoi965/vs: Use the proper dimensionality for the sampler result register.
Kenneth Graunke [Mon, 2 Jan 2012 23:51:58 +0000 (15:51 -0800)]
i965/vs: Use the proper dimensionality for the sampler result register.

textureSize() returns an int, ivec2, or ivec3, but never an ivec4.
Creating the destination register as an ivec4 triggered later failures,
even though the register did hold the proper values.

For example, piglit test vs-textureSize-compare calls textureSize on a
2D texture and compares the result to an expected value.  Unfortunately,
our generated code also tried to compare the third and fourth components
which were undefined, and failed.

Fixes piglit test vs-textureSize-compare as well as 19 subcases of
oglconform's glsl-bif-tex-size test.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=44339
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agoi965: Add missing _NEW_TEXTURE dirty bit to brw_vs_prog state atom.
Kenneth Graunke [Wed, 4 Jan 2012 11:14:01 +0000 (03:14 -0800)]
i965: Add missing _NEW_TEXTURE dirty bit to brw_vs_prog state atom.

Commit d45814c925dd6c479cfd383b9b59458fc4359cf7 totally added a data
dependency on _NEW_TEXTURE, even including the comment, but didn't
actually add the dirty bit.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agomesa: Additional error checks for transform feedback.
Paul Berry [Fri, 30 Dec 2011 18:14:35 +0000 (10:14 -0800)]
mesa: Additional error checks for transform feedback.

From the EXT_transform_feedback spec:

    The error INVALID_OPERATION is also generated by BeginTransformFeedbackEXT
    if no binding points would be used, either because no program object is
    active or because the active program object has specified no varying
    variables to record.

    ...

    The error INVALID_VALUE is generated by BindBufferRangeEXT or
    BindBufferOffsetEXT if <offset> is not word-aligned.

Fixes Piglit tests:
- EXT_transform_feedback/api-errors no_prog_active
- EXT_transform_feedback/api-errors interleaved_no_varyings
- EXT_transform_feedback/api-errors separate_no_varyings
- EXT_transform_feedback/api-errors bind_offset_offset_1
- EXT_transform_feedback/api-errors bind_offset_offset_2
- EXT_transform_feedback/api-errors bind_offset_offset_3
- EXT_transform_feedback/api-errors bind_offset_offset_5

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agomesa: Check that all buffers are bound in BeginTransformFeedback.
Paul Berry [Thu, 29 Dec 2011 23:55:01 +0000 (15:55 -0800)]
mesa: Check that all buffers are bound in BeginTransformFeedback.

From the EXT_transform_feedback spec:

    The error INVALID_OPERATION is generated by
    BeginTransformFeedbackEXT if any transform feedback buffer object
    binding point used in transform feedback mode does not have a
    buffer object bound.

This required adding a new NumBuffers field to the
gl_transform_feedback_info struct, to keep track of how many transform
feedback buffers are required by the current program.

Fixes Piglit tests:
- EXT_transform_feedback/api-errors interleaved_unbound
- EXT_transform_feedback/api-errors separate_unbound_0_1
- EXT_transform_feedback/api-errors separate_unbound_0_2
- EXT_transform_feedback/api-errors separate_unbound_1_2

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agomesa: Fix typos in transform feedback error messages.
Paul Berry [Fri, 30 Dec 2011 19:02:31 +0000 (11:02 -0800)]
mesa: Fix typos in transform feedback error messages.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agoglsl: Don't use base type for bit-not when there's an error
Ian Romanick [Sat, 24 Dec 2011 01:16:43 +0000 (17:16 -0800)]
glsl: Don't use base type for bit-not when there's an error

Other parts of the compiler assume that expressions will have
well-formed types or the error type.  Just using the type of the thing
being operated on can cause expressions like ~3.14 or ~false to not
have a well-formed type.  This could then result in an assertion
failure in the context epxression handler.

If there is an error processing the expression, set the type of the IR
expression to error.

Fixes piglit's bit-not-0[789].frag tests.

NOTE: This is a candidate for the 7.11 branch.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42755
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Vinson Lee <vlee@vmware.com>
12 years agoglx: More hacking around versions of XCB that lack GLX_ARB_create_context support
Ian Romanick [Tue, 3 Jan 2012 19:35:59 +0000 (11:35 -0800)]
glx: More hacking around versions of XCB that lack GLX_ARB_create_context support

Detect whether a new enough version of XCB is installed at configure
time.  If it is not, don't enable the extension and don't build the
unit tests.

v2: Move the AM_CONDIATION outside the case-statement so that it is
invoked even for non-GLX builds.  This prevents build failures with
osmesa, for example.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Robert Hooker <robert.hooker@canonical.com>
12 years agogallium/util: fix argument cast in x32_s8_get_tile_rgba() call
Brian Paul [Wed, 4 Jan 2012 20:34:54 +0000 (13:34 -0700)]
gallium/util: fix argument cast in x32_s8_get_tile_rgba() call

12 years agost/mesa: remove st_CompressedTexSubImage1D/2D/3D()
Brian Paul [Wed, 4 Jan 2012 20:30:35 +0000 (13:30 -0700)]
st/mesa: remove st_CompressedTexSubImage1D/2D/3D()

Just use the core Mesa functions instead.

Signed-off-by: Brian Paul <brianp@vmware.com>
12 years agost/mesa: fix indentation in st_copy_texsubimage()
Brian Paul [Wed, 4 Jan 2012 20:30:35 +0000 (13:30 -0700)]
st/mesa: fix indentation in st_copy_texsubimage()

12 years agost/mesa: remove st_TexImage(), use core Mesa code instead
Brian Paul [Wed, 4 Jan 2012 20:30:35 +0000 (13:30 -0700)]
st/mesa: remove st_TexImage(), use core Mesa code instead

The core Mesa code does the equivalent memory allocation, image mapping,
storing and unmapping.  We just need to call prep_teximage() first to
handle the 'surface_based' stuff.

The other change is to always use the level=0 mipmap image when accessing
individual mipmap level images that are stored in resources/buffers.
Apparently, we were always using malloc'd memory for individual mipmap
images, not resource buffers, before.

Signed-off-by: Brian Paul <brianp@vmware.com>
12 years agost/mesa: refactor gl_TexImage() code into prep_teximage()
Brian Paul [Wed, 4 Jan 2012 20:30:35 +0000 (13:30 -0700)]
st/mesa: refactor gl_TexImage() code into prep_teximage()

Preparation for st_TexImage() removal/refactoring.

Signed-off-by: Brian Paul <brianp@vmware.com>
12 years agointel: Re-enable GL_OES_standard_derivatives on GEN4+
Ian Romanick [Mon, 2 Jan 2012 20:48:34 +0000 (12:48 -0800)]
intel: Re-enable GL_OES_standard_derivatives on GEN4+

This extension only needs to be disabled on GEN3.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agointel: Re-allow blitting glCopyBufferSubData() on gen >= 6.
Eric Anholt [Wed, 28 Dec 2011 01:16:10 +0000 (17:16 -0800)]
intel: Re-allow blitting glCopyBufferSubData() on gen >= 6.

This was disabled a year ago due to not having a story for handling
the blitter at the time.  We're fine with using the blitter now.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agointel: Fix pitch handling for linear blits.
Eric Anholt [Wed, 28 Dec 2011 01:22:30 +0000 (17:22 -0800)]
intel: Fix pitch handling for linear blits.

The new assert in intelEmitCopyBlit() gets angry if we don't align to
dwords.  Rather than make the assert have a special case for height ==
1 on the assumption that the hardware doesn't use it in that case,
just supply a correct pitch.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=43214
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agointel: Fix bad read/write flags on self-copies for glCopyBufferSubData().
Eric Anholt [Wed, 28 Dec 2011 00:42:44 +0000 (16:42 -0800)]
intel: Fix bad read/write flags on self-copies for glCopyBufferSubData().

We didn't consume these flags in any way that would produce a
functional difference, but we might have some day.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
12 years agoRemove xmesa_xf86.h
Adam Jackson [Wed, 4 Jan 2012 16:32:38 +0000 (11:32 -0500)]
Remove xmesa_xf86.h

This was a leftover from libGLcore.

Signed-off-by: Adam Jackson <ajax@redhat.com>
12 years agodrisw: Fix drawable creation against non-default screens
Adam Jackson [Tue, 3 Jan 2012 19:39:28 +0000 (14:39 -0500)]
drisw: Fix drawable creation against non-default screens

We don't want to match the visual against the default screen.  If the
drawable is on a non-default screen then the appropriate visual might not
exist on the default screen.  Conversely, if the same visual is
available on multiple screens then simply selecting for the right VID is
sufficient, since the server has promised that the same visual is
compatible with multiple screens.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>
12 years agomesa: remove unused _mesa_init_teximage_fields() target parameter
Brian Paul [Wed, 4 Jan 2012 00:48:12 +0000 (17:48 -0700)]
mesa: remove unused _mesa_init_teximage_fields() target parameter

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
12 years agovl/mpeg2: simple fix to get xine running again
Christian König [Wed, 4 Jan 2012 14:59:29 +0000 (15:59 +0100)]
vl/mpeg2: simple fix to get xine running again

Otherwise xines xxmc plugin will just display green blocks.

Signed-off-by: Christian König <deathsimple@vodafone.de>
12 years agoff_fragment_shader: Don't generate swizzles for scalar combiner inputs
Ian Romanick [Fri, 23 Dec 2011 01:37:26 +0000 (17:37 -0800)]
ff_fragment_shader: Don't generate swizzles for scalar combiner inputs

There are a couple scenarios where the source could be zero and the
operand could be either SRC_ALPHA or ONE_MINUS_SRC_ALPHA.  For
example, if the source was ZERO.  This would result in something like
(0).w, and a later call to ir_validate would get angry.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42517

12 years agoFix read from pointer after free
Anuj Phogat [Wed, 4 Jan 2012 02:12:06 +0000 (18:12 -0800)]
Fix read from pointer after free

Coverity reported a read from pointer after free defect in
src/mesa/drivers/dri/intel/intel_mipmap_tree.c. Bug# 44205
In intel_miptree_all_slices_resolve() function, i = i->next was
executing after freeing i. I have defined a temporary variable
(next) to store the value of i->next before freeing i

Reported-by: Vinson Lee <vlee@vmware.com>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
12 years agoconfigure.ac: fix typo from 3ef3ba4d2
Matt Turner [Wed, 4 Jan 2012 02:58:37 +0000 (21:58 -0500)]
configure.ac: fix typo from 3ef3ba4d2

12 years agost/mesa: use SINT/UINT formats for VertexAttribIPointer
Christoph Bumiller [Tue, 3 Jan 2012 22:00:31 +0000 (23:00 +0100)]
st/mesa: use SINT/UINT formats for VertexAttribIPointer

Reviewed-by: Dave Airlie <airlied@redhat.com>
v2: added assertion that packed formats are not pure integer

12 years agotgsi/softpipe: disable FAST_MATH
Dave Airlie [Tue, 3 Jan 2012 10:38:38 +0000 (10:38 +0000)]
tgsi/softpipe: disable FAST_MATH

In the interest of softpipe preferring correctness over speed and passing more
piglit tests, set this to off by default. For speed you really want llvmpipe.

Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agosoftpipe: remove the 32bits limitation on depth(-stencil) formats
Morgan Armand [Wed, 7 Dec 2011 20:30:48 +0000 (21:30 +0100)]
softpipe: remove the 32bits limitation on depth(-stencil) formats

This patch remove the 32bits limitation. As a side effect, it bring the support for the GL_ARB_depth_buffer_float extension.
No regression have been found on piglit, and all tests for GL_ARB_depth_buffer_float pass successfully.

Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agogallium: use Haiku provided debug_printf in OS.h
Alexander von Gluck [Tue, 3 Jan 2012 15:10:42 +0000 (08:10 -0700)]
gallium: use Haiku provided debug_printf in OS.h

Signed-off-by: Brian Paul <brianp@vmware.com>
12 years agoglsl_to_tgsi: v2 Invalidate and revalidate uniform backing storage
Vadim Girlin [Thu, 22 Dec 2011 14:35:35 +0000 (18:35 +0400)]
glsl_to_tgsi: v2 Invalidate and revalidate uniform backing storage

If glUniform1i and friends are going to dump data directly in
driver-allocated, the pointers have to be updated when the storage
moves.  This should fix the regressions seen with commit 7199096.

I'm not sure if this is the only place that needs this treatment.  I'm
a little uncertain about the various functions in st_glsl_to_tgsi that
modify the TGSI IR and try to propagate changes about that up to the
gl_program.  That seems sketchy to me.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
v2:

Revalidate when shader_program is not NULL.
Update the pointers for all _LinkedShaders.
Init glsl_to_tgsi_visitor::shader_program to NULL in the
get_pixel_transfer_visitor & get_bitmap_visitor.

Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
12 years agoglx: Hack around versions of XCB that lack GLX_ARB_create_context support
Ian Romanick [Tue, 3 Jan 2012 01:44:31 +0000 (17:44 -0800)]
glx: Hack around versions of XCB that lack GLX_ARB_create_context support

A lot of tests in 'make check' will fail under these circumstances,
but at least the build should work.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agomesa: XCB is no longer optional for GLX or DRI
Ian Romanick [Tue, 3 Jan 2012 01:44:30 +0000 (17:44 -0800)]
mesa: XCB is no longer optional for GLX or DRI

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>