mesa.git
9 years agofreedreno: update generated headers
Rob Clark [Sat, 20 Dec 2014 16:49:34 +0000 (11:49 -0500)]
freedreno: update generated headers

9 years agofreedreno/ir3: trans_kill cleanup
Rob Clark [Sat, 20 Dec 2014 16:46:43 +0000 (11:46 -0500)]
freedreno/ir3: trans_kill cleanup

trans_kill() only handles the single opcode.  Drop the remnant of a time
when both KILL and KILL_IF were handled by the same fxn.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: hack for standalone compiler
Rob Clark [Sat, 20 Dec 2014 16:44:28 +0000 (11:44 -0500)]
freedreno/ir3: hack for standalone compiler

Standalone compiler doesn't have screen or context.  We need to come up
with a better way to control the target arch (ie. something that we can
control from cmdline w/ standalone compiler) but for now this hack keeps
it from segfault'ing.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agoi965/fs: Add missing const qualifier.
Matt Turner [Fri, 19 Dec 2014 20:55:13 +0000 (12:55 -0800)]
i965/fs: Add missing const qualifier.

9 years agovc4: Coalesce MOVs into VPM with the instructions generating the values.
Eric Anholt [Thu, 18 Dec 2014 04:35:17 +0000 (20:35 -0800)]
vc4: Coalesce MOVs into VPM with the instructions generating the values.

total instructions in shared programs: 41168 -> 40976 (-0.47%)
instructions in affected programs:     18156 -> 17964 (-1.06%)

9 years agovc4: Redefine VPM writes as a (destination) QIR register file.
Eric Anholt [Thu, 18 Dec 2014 04:23:57 +0000 (20:23 -0800)]
vc4: Redefine VPM writes as a (destination) QIR register file.

This will let me coalesce the VPM writes into the instructions generating
the values.

9 years agodocs: note change in minimum GCC version to 4.2.0
Timothy Arceri [Wed, 17 Dec 2014 20:46:24 +0000 (07:46 +1100)]
docs: note change in minimum GCC version to 4.2.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
9 years agogallium: remove support for GCC older than 4.2.0
Timothy Arceri [Wed, 17 Dec 2014 20:45:04 +0000 (07:45 +1100)]
gallium: remove support for GCC older than 4.2.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agomesa: bump required GCC version to 4.2.0
Timothy Arceri [Wed, 17 Dec 2014 20:29:47 +0000 (07:29 +1100)]
mesa: bump required GCC version to 4.2.0

It turns out Mesa hasn't compiled on less then 4.2 for a while
 so update conf to reflect this.

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agovc4: Add support for turning constant uniforms into small immediates.
Eric Anholt [Wed, 10 Dec 2014 22:56:46 +0000 (14:56 -0800)]
vc4: Add support for turning constant uniforms into small immediates.

Small immediates have the downside of taking over the raddr B field, so
you might have less chance to pack instructions together thanks to raddr B
conflicts.  However, it also reduces some register pressure since it lets
you load 2 "uniform" values in one instruction (avoiding a previous load
of the constant value to a register), and increases some pairing for the
same reason.

total uniforms in shared programs: 16231 -> 13374 (-17.60%)
uniforms in affected programs:     10280 -> 7423 (-27.79%)
total instructions in shared programs: 40795 -> 41168 (0.91%)
instructions in affected programs:     25551 -> 25924 (1.46%)

In a previous version of this patch I had a reduction in instruction count
by forcing the other args alongside a SMALL_IMM to be in the A file or
accumulators, but that increases register pressure and had a bug in
handling FRAG_Z.  In this patch is I just use raddr conflict resolution,
which is more expensive.  I think I'd rather tweak allocation to have some
way to slightly prefer good choices for files in general, rather than risk
failing to register allocate by forcing things into register classes.

9 years agovc4: Move follow_movs() to common QIR code.
Eric Anholt [Wed, 10 Dec 2014 23:37:07 +0000 (15:37 -0800)]
vc4: Move follow_movs() to common QIR code.

I want this from other passes.

9 years agovc4: Fix missing newline for load immediate instruction disasm.
Eric Anholt [Thu, 11 Dec 2014 00:20:36 +0000 (16:20 -0800)]
vc4: Fix missing newline for load immediate instruction disasm.

9 years agomesa: Remove unnecessary -f from $(RM).
Matt Turner [Wed, 17 Dec 2014 21:19:37 +0000 (13:19 -0800)]
mesa: Remove unnecessary -f from $(RM).

$(RM) includes -f.

9 years agomesa: Remove tarballs/checksum rules.
Matt Turner [Mon, 8 Dec 2014 02:09:49 +0000 (18:09 -0800)]
mesa: Remove tarballs/checksum rules.

9 years agogallium: Add egl and gbm to distribution.
Matt Turner [Wed, 17 Dec 2014 21:37:38 +0000 (13:37 -0800)]
gallium: Add egl and gbm to distribution.

9 years agomesa: Set DISTCHECK_CONFIGURE_FLAGS.
Matt Turner [Wed, 17 Dec 2014 20:41:02 +0000 (12:41 -0800)]
mesa: Set DISTCHECK_CONFIGURE_FLAGS.

Enable some non-default options that distros are likely to use.

9 years agotargets/xvmc: Add uninstall hooks to handle megadriver hardlinks.
Matt Turner [Wed, 17 Dec 2014 20:40:43 +0000 (12:40 -0800)]
targets/xvmc: Add uninstall hooks to handle megadriver hardlinks.

9 years agotargets/vdpau: Add uninstall hooks to handle megadriver hardlinks.
Matt Turner [Wed, 17 Dec 2014 20:40:30 +0000 (12:40 -0800)]
targets/vdpau: Add uninstall hooks to handle megadriver hardlinks.

9 years agotargets/vdpau: Add clean-local rule to remove .lib links.
Matt Turner [Wed, 17 Dec 2014 20:39:59 +0000 (12:39 -0800)]
targets/vdpau: Add clean-local rule to remove .lib links.

9 years agovc4: Add a userspace BO cache.
Eric Anholt [Sat, 13 Dec 2014 23:27:39 +0000 (15:27 -0800)]
vc4: Add a userspace BO cache.

Since our kernel BOs require CMA allocation, and the use of them requires
new mmaps, it's pretty expensive and we should avoid it if possible.
Copying my original design for Intel, make a userspace cache that reuses
BOs that haven't been shared to other processes but frees BOs that have
sat in the cache for over a second.

Improves glxgears framerate on RPi by around 30%.

9 years agovc4: Add dmabuf support.
Eric Anholt [Tue, 14 Oct 2014 10:21:04 +0000 (11:21 +0100)]
vc4: Add dmabuf support.

This gets DRI3 working on modesetting with glamor.  It's not enabled under
simulation, because it looks like handing our dumb-allocated buffers off
to the server doesn't actually work for the server's rendering.

9 years agovc4: Drop a weird argument in the BOs-from-handles API.
Eric Anholt [Wed, 17 Dec 2014 00:11:27 +0000 (16:11 -0800)]
vc4: Drop a weird argument in the BOs-from-handles API.

9 years agodraw: revert using correct order for prim decomposition.
Roland Scheidegger [Wed, 17 Dec 2014 19:16:07 +0000 (20:16 +0100)]
draw: revert using correct order for prim decomposition.

This reverts db3dfcfe90a3d27e6020e0d3642f8ab0330e57be.
The commit was correct but we've got some precision problems later in
llvmpipe (or possibly in draw clip) due to the vertices coming in in
different order, causing some internal test failures. So revert for now.
(Will only affect drivers which actually support constant-interpolated
attributes and not just flatshading.)

9 years agoutil: Silence signed-unsigned comparison warnings
Jan Vesely [Thu, 11 Dec 2014 20:05:17 +0000 (15:05 -0500)]
util: Silence signed-unsigned comparison warnings

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoi965: Require pixel alignment for GPU copy blit
Cody Northrop [Mon, 15 Sep 2014 22:14:20 +0000 (16:14 -0600)]
i965: Require pixel alignment for GPU copy blit

The blitter will start at a pixel's natural alignment. For PBOs, if the
provided offset if not aligned, bits will get dropped.

This change adds offset alignment check for src and dst, kicking back if
the requirements are not met.

The change is based on following verbiage from BSPEC:
 Color pixel sizes supported are 8, 16, and 32 bits per pixel (bpp).
 All pixels are naturally aligned.

Found in the following locations:
page 35 of intel-gfx-prm-osrc-hsw-blitter.pdf
page 29 of ivb_ihd_os_vol1_part4.pdf
page 29 of snb_ihd_os_vol1_part5.pdf

This behavior was observed with Steam Big Picture rendering incorrect
icon colors.  The fix has been tested on Ubuntu and SteamOS on Haswell.

Signed-off-by: Cody Northrop <cody@lunarg.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=83908
Reviewed-by: Neil Roberts <neil@linux.intel.com>
9 years agoi965: remove includes of sampler.h from extern "C" blocks
Mark Janes [Tue, 16 Dec 2014 22:29:28 +0000 (14:29 -0800)]
i965: remove includes of sampler.h from extern "C" blocks

C linkage was removed from functions in program/sampler.cpp.  However,
some cpp files include program/sampler.h within extern "C" blocks,
causing link errors for test_vec4_copy_propagation.

Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
9 years agoi965/query: Cache whether the batch references the query BO.
Kenneth Graunke [Thu, 11 Dec 2014 10:26:39 +0000 (02:26 -0800)]
i965/query: Cache whether the batch references the query BO.

Chris Wilson noted that repeated calls to CheckQuery() would call
drm_intel_bo_references(brw->batch.bo, query->bo) on each invocation,
which is expensive.  Once we've flushed, we know that future batches
won't reference query->bo, so there's no point in asking more than once.

This patch adds a brw_query_object::flushed flag, which is a
conservative estimate of whether the batch has been flushed.

On the first call to CheckQuery() or WaitQuery(), we check if the
batch references query->bo.  If not, it must have been flushed for
some reason (such as being full).  We record that it was flushed.
If it does reference query->bo, we explicitly flush, and record that
we did so.

Any subsequent checks will simply see that query->flushed is set,
and skip the drm_intel_bo_references() call.

Inspired by a patch from Chris Wilson.

According to Eero, this does not affect the performance of Witcher 2
on Haswell, but approximately halves the userspace CPU usage.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86969
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
9 years agoi965/query: Use brw_bo_map to handle stall warnings.
Kenneth Graunke [Thu, 11 Dec 2014 09:57:39 +0000 (01:57 -0800)]
i965/query: Use brw_bo_map to handle stall warnings.

This is less code and also measures the duration of the stall for us.

Our old code predates the existance of brw_bo_map().

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
9 years agoi965/query: Remove redundant drm_intel_bo_references call in CheckQuery.
Kenneth Graunke [Thu, 11 Dec 2014 09:43:52 +0000 (01:43 -0800)]
i965/query: Remove redundant drm_intel_bo_references call in CheckQuery.

CheckQuery calls drm_intel_bo_references to see if the batch references
the query BO, and if so, flushes.  It then checks if the query BO is
busy, and if not, calls gen6_queryobj_get_results().

Stupidly, gen6_queryobj_get_results() immediately did a second redundant
drm_intel_bo_references check, even though we know the buffer is not
referenced and in fact idle.

This patch moves the batch-flush check out of gen6_queryobj_get_results
and into WaitQuery() (the other caller).  That way, both callers do a
single batch-flush check.

This should only be a minor improvement, since it would only affect
the first CheckQuery call where the result is actually available.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86969
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
9 years agoi965/query: Add query->bo == NULL early return in CheckQuery hook.
Kenneth Graunke [Thu, 11 Dec 2014 09:40:28 +0000 (01:40 -0800)]
i965/query: Add query->bo == NULL early return in CheckQuery hook.

If query->bo == NULL, this is a redundant CheckQuery call, and we
should simply return.  We didn't do anything anyway - we skipped the
batch flushing block, and although we called get_results(), it has an
early return and does nothing.  Why bother?

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
9 years agoi965/query: Set Ready flag in gen6_queryobj_get_results().
Kenneth Graunke [Thu, 11 Dec 2014 09:29:48 +0000 (01:29 -0800)]
i965/query: Set Ready flag in gen6_queryobj_get_results().

q->Ready means that the results are in, and core Mesa is free to return
them to the application.  gen6_queryobj_get_results() is a natural place
to set that flag; doing so means callers don't have to.

The older non-hardware-context aware code couldn't do this, because we
had to call brw_queryobj_get_results() to gather intermediate results
when we ran out of space for snapshots in the query buffer.  We only
gather complete results in the Gen6+ code, however.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
9 years agovc4: Add support for turning add-based MOVs to muls for pairing.
Eric Anholt [Tue, 16 Dec 2014 19:58:58 +0000 (11:58 -0800)]
vc4: Add support for turning add-based MOVs to muls for pairing.

total instructions in shared programs: 43053 -> 40795 (-5.24%)
instructions in affected programs:     37996 -> 35738 (-5.94%)

9 years agovc4: Add a helper for changing a field in an instruction.
Eric Anholt [Tue, 16 Dec 2014 19:22:53 +0000 (11:22 -0800)]
vc4: Add a helper for changing a field in an instruction.

9 years agovc4: Fix the name of qpu_waddr_ignores_ws().
Eric Anholt [Tue, 16 Dec 2014 19:29:15 +0000 (11:29 -0800)]
vc4: Fix the name of qpu_waddr_ignores_ws().

We're deciding about the WS bit, not PM.

9 years agodocs: note change in minimum GCC version to 4.1.0
Timothy Arceri [Sat, 13 Dec 2014 10:22:39 +0000 (21:22 +1100)]
docs: note change in minimum GCC version to 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoutil: remove support for GCC older than 4.1.0
Timothy Arceri [Fri, 12 Dec 2014 09:47:43 +0000 (20:47 +1100)]
util: remove support for GCC older than 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agomesa: remove support for GCC older than 4.1.0
Timothy Arceri [Fri, 12 Dec 2014 09:44:26 +0000 (20:44 +1100)]
mesa: remove support for GCC older than 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agogbm: remove support for GCC older than 4.1.0
Timothy Arceri [Fri, 12 Dec 2014 09:38:22 +0000 (20:38 +1100)]
gbm: remove support for GCC older than 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agogallium: remove support for GCC older than 4.1.0
Timothy Arceri [Fri, 12 Dec 2014 09:36:25 +0000 (20:36 +1100)]
gallium: remove support for GCC older than 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoegl: remove support for GCC older than 4.1.0
Timothy Arceri [Fri, 12 Dec 2014 09:27:44 +0000 (20:27 +1100)]
egl: remove support for GCC older than 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agomesa: bump required GCC version to 4.1.0
Timothy Arceri [Fri, 12 Dec 2014 08:12:48 +0000 (19:12 +1100)]
mesa: bump required GCC version to 4.1.0

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agomesa: remove support for GCC older than 3.3.0
Timothy Arceri [Fri, 12 Dec 2014 08:05:22 +0000 (19:05 +1100)]
mesa: remove support for GCC older than 3.3.0

GCC >=3.3 has been required since 9aa3aa71386394725ce88df463d6183f62777ee5

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-By: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoi965/fs: Add a comment explaining what saturate propagation does.
Matt Turner [Tue, 16 Dec 2014 19:30:12 +0000 (11:30 -0800)]
i965/fs: Add a comment explaining what saturate propagation does.

9 years agovc4: Add support for enabling early Z discards.
Eric Anholt [Mon, 15 Dec 2014 23:11:07 +0000 (15:11 -0800)]
vc4: Add support for enabling early Z discards.

This is the same basic logic from the original Broadcom driver.

9 years agost/mesa: remove extern "C" around #includes in st_glsl_to_tgsi.cpp
Brian Paul [Tue, 16 Dec 2014 00:05:53 +0000 (17:05 -0700)]
st/mesa: remove extern "C" around #includes in st_glsl_to_tgsi.cpp

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoprogram: remove extern "C" usage in sampler.cpp
Brian Paul [Tue, 16 Dec 2014 00:05:13 +0000 (17:05 -0700)]
program: remove extern "C" usage in sampler.cpp

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoprogram: remove extern "C" around #includes
Brian Paul [Tue, 16 Dec 2014 00:04:38 +0000 (17:04 -0700)]
program: remove extern "C" around #includes

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoglsl: remove extern "C" around #includes
Brian Paul [Mon, 15 Dec 2014 23:41:58 +0000 (16:41 -0700)]
glsl: remove extern "C" around #includes

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agost/mesa: add extern "C" to st_context.h
Brian Paul [Mon, 15 Dec 2014 23:49:54 +0000 (16:49 -0700)]
st/mesa: add extern "C" to st_context.h

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agost/mesa: add extern "C" to st_program.h
Brian Paul [Mon, 15 Dec 2014 23:46:46 +0000 (16:46 -0700)]
st/mesa: add extern "C" to st_program.h

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomain: remove extern C around #includes in ff_fragment_shader.cpp
Brian Paul [Mon, 15 Dec 2014 23:45:57 +0000 (16:45 -0700)]
main: remove extern C around #includes in ff_fragment_shader.cpp

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomesa: move #include of mtypes.h outside __cplusplus check
Brian Paul [Mon, 15 Dec 2014 23:41:29 +0000 (16:41 -0700)]
mesa: move #include of mtypes.h outside __cplusplus check

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoprogram: add #ifndef SAMPLER_H wrapper
Brian Paul [Tue, 16 Dec 2014 00:01:22 +0000 (17:01 -0700)]
program: add #ifndef SAMPLER_H wrapper

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomesa: put extern "C" in src/mesa/program/*h header files
Brian Paul [Mon, 15 Dec 2014 23:37:35 +0000 (16:37 -0700)]
mesa: put extern "C" in src/mesa/program/*h header files

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomesa: put extern "C" in header files
Brian Paul [Mon, 15 Dec 2014 23:36:27 +0000 (16:36 -0700)]
mesa: put extern "C" in header files

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomapi: add glapi-test and shared-glapi-test to .gitignore
Juha-Pekka Heikkila [Tue, 16 Dec 2014 10:28:49 +0000 (12:28 +0200)]
mapi: add glapi-test and shared-glapi-test to .gitignore

On the same go remove src/mapi/shared-glapi/tests/.gitignore
and src/mapi/glapi/tests/.gitignore as useless.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoutil: add u_atomic_test to .gitignore
Juha-Pekka Heikkila [Tue, 16 Dec 2014 10:28:48 +0000 (12:28 +0200)]
util: add u_atomic_test to .gitignore

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoglx: remove __glXstrdup()
Juha-Pekka Heikkila [Tue, 16 Dec 2014 10:28:47 +0000 (12:28 +0200)]
glx: remove __glXstrdup()

I didn't find this being used anywhere

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoi965: add test_vf_float_conversions to .gitignore
Juha-Pekka Heikkila [Tue, 16 Dec 2014 10:28:46 +0000 (12:28 +0200)]
i965: add test_vf_float_conversions to .gitignore

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoi965: Make validate_reg tables constant
Juha-Pekka Heikkila [Tue, 16 Dec 2014 10:28:45 +0000 (12:28 +0200)]
i965: Make validate_reg tables constant

Declare local tables constant.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
9 years agoglsl: remove commented out code
Timothy Arceri [Mon, 15 Dec 2014 07:15:34 +0000 (18:15 +1100)]
glsl: remove commented out code

MaxGeometryOutputComponents is used as the value
for gl_MaxGeometryVaryingComponents

Acked-by: Matt Turner <mattst88@gmail.com>
9 years agoi965: remove commented out code
Timothy Arceri [Mon, 15 Dec 2014 07:10:12 +0000 (18:10 +1100)]
i965: remove commented out code

Acked-by: Matt Turner <mattst88@gmail.com>
9 years agonvc0: add missed PIPE_CAP_VERTEXID_NOBASE
Ilia Mirkin [Tue, 16 Dec 2014 04:17:28 +0000 (23:17 -0500)]
nvc0: add missed PIPE_CAP_VERTEXID_NOBASE

Commit ade8b26bf missed adding this cap to nvc0.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
9 years agost/mesa: use vertex id lowering according to pipe cap bit.
Roland Scheidegger [Fri, 12 Dec 2014 03:14:02 +0000 (04:14 +0100)]
st/mesa: use vertex id lowering according to pipe cap bit.

Tested with llvmpipe by setting the cap bit temporarily, seems to work,
though no driver requests it for now.

9 years agodraw: implement support for the VERTEXID_NOBASE and BASEVERTEX semantics.
Roland Scheidegger [Fri, 12 Dec 2014 03:13:54 +0000 (04:13 +0100)]
draw: implement support for the VERTEXID_NOBASE and BASEVERTEX semantics.

This fixes 4 vertexid related piglit tests with llvmpipe due to switching
behavior of vertexid to the one gl expects.
(Won't fix non-llvm draw path since we don't get the basevertex currently.)

9 years agogallium: add TGSI_SEMANTIC_VERTEXID_NOBASE and TGSI_SEMANTIC_BASEVERTEX
Roland Scheidegger [Fri, 12 Dec 2014 03:13:43 +0000 (04:13 +0100)]
gallium: add TGSI_SEMANTIC_VERTEXID_NOBASE and TGSI_SEMANTIC_BASEVERTEX

Plus a new PIPE_CAP_VERTEXID_NOBASE query. The idea is that drivers not
supporting vertex ids with base vertex offset applied (so, only support
d3d10-style vertex ids) will get such a d3d10-style vertex id instead -
with the caveat they'll also need to handle the basevertex system value
too (this follows what core mesa already does).
Additionally, this is also useful for other state trackers (for instance
llvmpipe / draw right now implement the d3d10 behavior on purpose, but
with different semantics it can just do both).
Doesn't do anything yet.
And fix up the docs wrt similar values.

v2: incorporate feedback from Brian and others, better names, better docs.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
9 years agor600g/sb: implement r600 gpr index workaround. (v3.1)
Dave Airlie [Tue, 9 Dec 2014 06:46:55 +0000 (16:46 +1000)]
r600g/sb: implement r600 gpr index workaround. (v3.1)

r600, rv610 and rv630 all have a bug in their GPR indexing
and how the hw inserts access to PV.

If the base index for the src is the same as the dst gpr
in a previous group, then it will use PV instead of using
the indexed gpr correctly.

The workaround is to insert a NOP when you detect this.

v2: add second part of fix detecting DST rel writes followed
by same src base index reads.

v3: forget adding stuff to structs, just iterate over the
previous node group again, makes it more obvious.
v3.1: drop local_nop.

Fixes ~200 piglit regressions on rv635 since SB was introduced.

Reviewed-By: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agor600g/sb: fix issues with loops created for switch
Vadim Girlin [Wed, 10 Dec 2014 11:41:10 +0000 (14:41 +0300)]
r600g/sb: fix issues with loops created for switch

Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agoRevert "r600g/sb: fix issues cause by GLSL switching to loops for switch"
Dave Airlie [Tue, 9 Dec 2014 00:11:46 +0000 (10:11 +1000)]
Revert "r600g/sb: fix issues cause by GLSL switching to loops for switch"

This reverts commit 7b0067d23a6f64cf83c42e7f11b2cd4100c569fe.

Vadim's patch fixes this a lot better.

9 years agovc4: Add support for 32-bit signed norm/scaled vertex attrs.
Eric Anholt [Mon, 15 Dec 2014 20:37:42 +0000 (12:37 -0800)]
vc4: Add support for 32-bit signed norm/scaled vertex attrs.

32-bit unsigned would require some adjustments to handle values >=
0x80000000.

9 years agovc4: Add support for 16-bit signed/unsigned norm/scaled vertex attrs.
Eric Anholt [Mon, 15 Dec 2014 20:30:26 +0000 (12:30 -0800)]
vc4: Add support for 16-bit signed/unsigned norm/scaled vertex attrs.

9 years agovc4: Rename the 16-bit unpack #define.
Eric Anholt [Mon, 15 Dec 2014 20:26:29 +0000 (12:26 -0800)]
vc4: Rename the 16-bit unpack #define.

It's only an f16 conversion if you're doing a float operation, otherwise
it's 16 bit signed to 32-bit signed.

9 years agovc4: Add support for 8-bit unnormalized vertex attrs.
Eric Anholt [Mon, 15 Dec 2014 19:19:58 +0000 (11:19 -0800)]
vc4: Add support for 8-bit unnormalized vertex attrs.

9 years agovc4: Refactor vertex attribute conversions a bit.
Eric Anholt [Mon, 15 Dec 2014 19:25:23 +0000 (11:25 -0800)]
vc4: Refactor vertex attribute conversions a bit.

There was just way too much indentation.

9 years agovc4: Fix use of r3 as a temp in 8-bit unpacking.
Eric Anholt [Mon, 15 Dec 2014 19:10:14 +0000 (11:10 -0800)]
vc4: Fix use of r3 as a temp in 8-bit unpacking.

We're actually allocating out of r3 now, and I missed it because I'd typed
this one as qpu_rn(3) instead of qpu_r3().

9 years agovc4: Rename UNPACK_8* to UNPACK_8*_F.
Eric Anholt [Mon, 15 Dec 2014 18:45:58 +0000 (10:45 -0800)]
vc4: Rename UNPACK_8* to UNPACK_8*_F.

There is an equivalent unpack function without conversion to float if you
use an integer operation instead.

9 years agovc4: Add support for UMAD.
Eric Anholt [Mon, 15 Dec 2014 17:24:16 +0000 (09:24 -0800)]
vc4: Add support for UMAD.

9 years agovc4: 0-initialize the screen again.
Eric Anholt [Mon, 15 Dec 2014 17:08:09 +0000 (09:08 -0800)]
vc4: 0-initialize the screen again.

I typoed this when rebasing the memory leak fixes.

9 years agoglsl: Add gl_MaxViewports to available builtin constants
Maxence Le Doré [Wed, 10 Dec 2014 07:09:24 +0000 (08:09 +0100)]
glsl: Add gl_MaxViewports to available builtin constants

It seems to have been forgotten during viewports array implementation time.

Cc: "10.4 10.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoi965/brw_reg: struct constructor now needs explicit negate and abs values.
Andres Gomez [Fri, 12 Dec 2014 16:19:07 +0000 (17:19 +0100)]
i965/brw_reg: struct constructor now needs explicit negate and abs values.

We were assuming, when constructing a new brw_reg struct, that the
negate and abs register modifiers would not be present by default in
the new register.

Now, we force explicitly setting these values when constructing a new
register.

This will avoid problems like forgetting to properly set them when we
are using a previous register to generate this new register, as it was
happening in the dFdx and dFdy generation functions.

Fixes piglit test shaders/glsl-deriv-varyings

Cc: "10.4 10.3" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=82991
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agovc4: Fix leaks of the compiled shaders' keys.
Eric Anholt [Mon, 15 Dec 2014 04:50:15 +0000 (20:50 -0800)]
vc4: Fix leaks of the compiled shaders' keys.

9 years agovc4: Fix leaks of the CL contents.
Eric Anholt [Mon, 15 Dec 2014 04:41:25 +0000 (20:41 -0800)]
vc4: Fix leaks of the CL contents.

9 years agovc4: Fix leak of vc4_bos stashed in the context.
Eric Anholt [Mon, 15 Dec 2014 04:39:49 +0000 (20:39 -0800)]
vc4: Fix leak of vc4_bos stashed in the context.

9 years agovc4: Fix leak of the compiled shader programs in the cache.
Eric Anholt [Mon, 15 Dec 2014 04:29:10 +0000 (20:29 -0800)]
vc4: Fix leak of the compiled shader programs in the cache.

9 years agovc4: Fix leak of a copy of the scheduled QPU instructions.
Eric Anholt [Mon, 15 Dec 2014 04:28:13 +0000 (20:28 -0800)]
vc4: Fix leak of a copy of the scheduled QPU instructions.

They're copied into a vc4_bo after compiling is done.

9 years agovc4: Switch to using the util/ hash table.
Eric Anholt [Mon, 15 Dec 2014 03:48:51 +0000 (19:48 -0800)]
vc4: Switch to using the util/ hash table.

No performance difference on a microbenchmark with norast that should hit it
enough to have mattered, n=220.

9 years agovc4: Fix leak of simulator memory on screen cleanup.
Eric Anholt [Sun, 14 Dec 2014 00:40:54 +0000 (16:40 -0800)]
vc4: Fix leak of simulator memory on screen cleanup.

9 years agovc4: Fix a leak of the simulator's exec BO's actual vc4_bo.
Eric Anholt [Sun, 14 Dec 2014 00:37:28 +0000 (16:37 -0800)]
vc4: Fix a leak of the simulator's exec BO's actual vc4_bo.

9 years agohash_table: Fix compiler warnings from the renaming.
Eric Anholt [Mon, 15 Dec 2014 04:21:32 +0000 (20:21 -0800)]
hash_table: Fix compiler warnings from the renaming.

Not sure how we both missed this.  None of the callers were using the
return value, though.

9 years agoutil/hash_table: Rework the API to know about hashing
Jason Ekstrand [Tue, 25 Nov 2014 06:19:50 +0000 (22:19 -0800)]
util/hash_table: Rework the API to know about hashing

Previously, the hash_table API required the user to do all of the hashing
of keys as it passed them in.  Since the hashing function is intrinsically
tied to the comparison function, it makes sense for the hash table to know
about it.  Also, it makes for a somewhat clumsy API as the user is
constantly calling hashing functions many of which have long names.  This
is especially bad when the standard call looks something like

_mesa_hash_table_insert(ht, _mesa_pointer_hash(key), key, data);

In the above case, there is no reason why the hash table shouldn't do the
hashing for you.  We leave the option for you to do your own hashing if
it's more efficient, but it's no longer needed.  Also, if you do do your
own hashing, the hash table will assert that your hash matches what it
expects out of the hashing function.  This should make it harder to mess up
your hashing.

v2: change to call the old entrypoint "pre_hashed" rather than
    "with_hash", like cworth's equivalent change upstream (change by
    anholt, acked-in-general by Jason).

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
9 years agoglx/dri3: Don't fail on glXSwapBuffersMscOML(dpy, window, 0, 0, 0) (v2)
Mario Kleiner [Fri, 5 Dec 2014 07:42:32 +0000 (08:42 +0100)]
glx/dri3: Don't fail on glXSwapBuffersMscOML(dpy, window, 0, 0, 0) (v2)

glXSwapBuffersMscOML() with target_msc=divisor=remainder=0 gets
translated into target_msc=divisor=0 but remainder=1 by the mesa
api. This is done for server DRI2 where there needs to be a way
to tell the server-side DRI2ScheduleSwap implementation if a call
to glXSwapBuffers() or glXSwapBuffersMscOML(dpy,window,0,0,0) was
done. remainder = 1 was (ab)used as a flag to tell the server to
select proper semantic. The DRI3/Present backend ignored this
signalling, treated any target_msc=0 as glXSwapBuffers() request,
and called xcb_present_pixmap with invalid divisor=0, remainder=1
combo. The present extension responded kindly to this with a
BadValue error and dropped the request, but mesa's DRI3/Present
backend doesn't check for error codes. From there on stuff went
downhill quickly for the calling OpenGL client...

This patch fixes the problem.

v2: Change comments to be more clear, with reference to
relevant spec, as suggested by Eric Anholt.

Cc: "10.3 10.4" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Eric Anholt <eric@anholt.net>
9 years agoglx/dri3: Request non-vsynced Present for swapinterval zero. (v3)
Mario Kleiner [Fri, 5 Dec 2014 07:42:31 +0000 (08:42 +0100)]
glx/dri3: Request non-vsynced Present for swapinterval zero. (v3)

Restores proper immediate tearing swap behaviour for
OpenGL bufferswap under DRI3/Present.

Cc: "10.3 10.4" <mesa-stable@lists.freedesktop.org>
v2: Add Frank Binns signed off by for his original earlier
patch from April 2014, which is identical to this one, and
Chris Wilsons reviewed tag from May 2014 for that patch, ergo
also for this one.

v3: Incorporate comment about triple buffering as suggested
by Axel Davy, and reference to relevant spec provided by
Eric Anholt.

Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Eric Anholt <eric@anholt.net>
9 years agoglx/dri3: Track separate (ust, msc) for PresentPixmap vs. PresentNotifyMsc (v2)
Mario Kleiner [Fri, 5 Dec 2014 07:42:30 +0000 (08:42 +0100)]
glx/dri3: Track separate (ust, msc) for PresentPixmap vs. PresentNotifyMsc (v2)

Prevent calls to glXGetSyncValuesOML() and glXWaitForMscOML()
from overwriting the (ust,msc) values of the last successfull
swapbuffers call (PresentPixmapCompleteNotify event), as
glXWaitForSbcOML() relies on those values corresponding to
the most recent completed swap, not to whatever was last
returned from the server.

Problematic call sequence without this patch would have been, e.g.,

glXSwapBuffers()
... wait ...
swap completes -> PresentPixmapComplete event -> (ust,msc)
updated to reflect swap completion time and count.
... wait for at least 1 video refresh cycle/vblank increment.

glXGetSyncValuesOML()
-> PresentNotifyMsc event overwrites (ust,msc) of swap
completion with (ust,msc) of most recent vblank

glXWaitForSbcOML()
-> Returns sbc of last completed swap but (ust,msc) of last
completed vblank, not of last completed swap.
-> Client is confused.

Do this by tracking a separate set of (ust, msc) for the
dri3_wait_for_msc() call than for the dri3_wait_for_sbc()
call.

This makes the glXWaitForSbcOML() call robust again and restores
consistent behaviour with the DRI2 implementation.

Fixes applications originally written and tested against
DRI2 which also rely on this not regressing under DRI3/Present,
e.g., Neuro-Science software like Psychtoolbox-3.

This patch fixes the problem.

v2: Rename vblank_msc/ust to notify_msc/ust as suggested by
Axel Davy for better clarity.

Cc: "10.3 10.4" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
9 years agoglx/dri3: Fix glXWaitForSbcOML() to handle targetSBC==0 correctly. (v2)
Mario Kleiner [Fri, 5 Dec 2014 07:42:29 +0000 (08:42 +0100)]
glx/dri3: Fix glXWaitForSbcOML() to handle targetSBC==0 correctly. (v2)

targetSBC == 0 is a special case, which asks the function
to block until all pending OpenGL bufferswap requests have
completed.

Currently the function just falls through for targetSBC == 0,
returning bogus results.

This breaks applications originally written and tested against
DRI2 which also rely on this not regressing under DRI3/Present,
e.g., Neuro-Science software like Psychtoolbox-3.

This patch fixes the problem.

v2: Simplify as suggested by Axel Davy. Add comments proposed
by Eric Anholt.

Cc: "10.3 10.4" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Eric Anholt <eric@anholt.net>
9 years agodocs: Add 10.4 sha256 sums, news item and link release notes
Emil Velikov [Sun, 14 Dec 2014 13:57:54 +0000 (13:57 +0000)]
docs: Add 10.4 sha256 sums, news item and link release notes

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit af0c82099b955f10471c15cb7a4dc8db29b84963)

Conflicts:
docs/index.html
docs/relnotes.html

9 years agodocs: Update 10.4.0 release notes
Emil Velikov [Sun, 14 Dec 2014 13:45:54 +0000 (13:45 +0000)]
docs: Update 10.4.0 release notes

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit 5fe79b0b1272d530371a5d7248ed378cff4f6d21)

9 years agofreedreno/a4xx: mipmaps
Rob Clark [Fri, 12 Dec 2014 18:38:05 +0000 (13:38 -0500)]
freedreno/a4xx: mipmaps

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno: update generated headers
Rob Clark [Fri, 12 Dec 2014 18:34:57 +0000 (13:34 -0500)]
freedreno: update generated headers

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno: add is_a3xx()/is_a4xx() helpers
Rob Clark [Fri, 12 Dec 2014 23:51:36 +0000 (18:51 -0500)]
freedreno: add is_a3xx()/is_a4xx() helpers

A bunch of open-coded 'gpu_id > 300's seems like it will eventually
cause problems with future generations.  There were already a few minor
problems with caps for features that still need additional work on a4xx.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno: helper to calc layer/level offset
Rob Clark [Fri, 12 Dec 2014 22:20:50 +0000 (17:20 -0500)]
freedreno: helper to calc layer/level offset

Rather than duplicating this everywhere.  Especially as on a4xx the
layout of layers and levels differs based on texture type.

Signed-off-by: Rob Clark <robclark@freedesktop.org>