mesa.git
10 years agoglsl: Use the without_array predicate to simplify some code
Ian Romanick [Sat, 19 Jul 2014 21:41:04 +0000 (14:41 -0700)]
glsl: Use the without_array predicate to simplify some code

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com> [v1]
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
10 years agoglsl: Add without_array type predicate
Ian Romanick [Sat, 19 Jul 2014 21:07:18 +0000 (14:07 -0700)]
glsl: Add without_array type predicate

Returns the type without any arrays.

This will be used in later patches in this series.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Suggested-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoglsl: Use constant_expression_value instead of as_constant
Ian Romanick [Fri, 18 Jul 2014 23:10:16 +0000 (16:10 -0700)]
glsl: Use constant_expression_value instead of as_constant

Just a few lines earlier we may have wrapped the index expression with
ir_unop_i2u expression.  Whenever that happens, as_constant will return
NULL, and that almost always happens.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
10 years agotargets/graw-gdi: link with mesautil, not mesautils
Brian Paul [Mon, 4 Aug 2014 20:15:48 +0000 (14:15 -0600)]
targets/graw-gdi: link with mesautil, not mesautils

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agowmesa: link with mesautil
Brian Paul [Mon, 4 Aug 2014 20:15:05 +0000 (14:15 -0600)]
wmesa: link with mesautil

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agoosmesa: link with mesautil
Brian Paul [Mon, 4 Aug 2014 20:14:27 +0000 (14:14 -0600)]
osmesa: link with mesautil

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agotargets/libgl-gdi: link with mesautil
Brian Paul [Mon, 4 Aug 2014 20:14:11 +0000 (14:14 -0600)]
targets/libgl-gdi: link with mesautil

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agotargets/egl-static: link with libmesautil.la
Brian Paul [Mon, 4 Aug 2014 20:13:07 +0000 (14:13 -0600)]
targets/egl-static: link with libmesautil.la

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agomesa/x86: put code in braces to silence declarations after code warning
Brian Paul [Mon, 4 Aug 2014 20:29:04 +0000 (14:29 -0600)]
mesa/x86: put code in braces to silence declarations after code warning

Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agosrc/Makefile.am: Move gtest before util
Jason Ekstrand [Mon, 4 Aug 2014 19:25:36 +0000 (12:25 -0700)]
src/Makefile.am: Move gtest before util

Since the ralloc test in util/tests needs gtest, we need to make sure that
the gtest subdir is loaded first.  This fixes bug #82148.

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoutil: include c99_compat.h in format_srgb.h to get 'inline' definition
Brian Paul [Mon, 4 Aug 2014 19:38:02 +0000 (13:38 -0600)]
util: include c99_compat.h in format_srgb.h to get 'inline' definition

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agoutil: include c99_compat.h in hash_table.h to get 'inline' definition
Brian Paul [Mon, 4 Aug 2014 19:34:54 +0000 (13:34 -0600)]
util: include c99_compat.h in hash_table.h to get 'inline' definition

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agotargets/vdpau: link with libmesautil.la to fix build breakage
Brian Paul [Mon, 4 Aug 2014 19:09:43 +0000 (13:09 -0600)]
targets/vdpau: link with libmesautil.la to fix build breakage

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agoxlib: fix missing mesautil build breakage
Brian Paul [Mon, 4 Aug 2014 19:08:58 +0000 (13:08 -0600)]
xlib: fix missing mesautil build breakage

Fixes the non-DRI build.

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agosvga: SVGA_3D_CMD_BIND_GB_SHADER needs to reserve two relocations.
Matthew McClure [Fri, 27 Jun 2014 17:24:08 +0000 (10:24 -0700)]
svga: SVGA_3D_CMD_BIND_GB_SHADER needs to reserve two relocations.

With this patch, the SVGA_3D_CMD_BIND_GB_SHADER functionality will reserve
two relocations, one for the shader ID and the second for the MOB ID.

Verified with the WDDM winsys path that the number of relocations and patch
locations required is two.

Fixes Bug 1277406

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
10 years agogallium: Add libmesautil dependency to gdm and xa targets
Jason Ekstrand [Mon, 4 Aug 2014 19:34:04 +0000 (12:34 -0700)]
gallium: Add libmesautil dependency to gdm and xa targets

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agomesa/main: Use the RGB <-> sRGB conversion functions in libmesautil
Jason Ekstrand [Thu, 24 Jul 2014 19:32:49 +0000 (12:32 -0700)]
mesa/main: Use the RGB <-> sRGB conversion functions in libmesautil

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agogallium: Move sRGB <-> RGB handling to libmesautil
Jason Ekstrand [Thu, 24 Jul 2014 18:17:53 +0000 (11:17 -0700)]
gallium: Move sRGB <-> RGB handling to libmesautil

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agoutil: Gather some common macros
Jason Ekstrand [Wed, 23 Jul 2014 21:58:52 +0000 (14:58 -0700)]
util: Gather some common macros

This gathers macros that have been included across components into util so
that the include chain can be more vertical.  In particular, this makes
util stand on its own without any dependence whatsoever on the rest of
mesa.

Signed-off-by: "Jason Ekstrand" <jason.ekstrand@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agoutil: Move the open-addressing linear-probing hash_table to src/util.
Kenneth Graunke [Tue, 25 Feb 2014 09:08:45 +0000 (01:08 -0800)]
util: Move the open-addressing linear-probing hash_table to src/util.

This hash table is used in core Mesa, the GLSL compiler, and the i965
driver, which makes it a good candidate for the new src/util module.

It's much faster than program/hash_table.[ch] (see commit 6991c2922f5
for data), and José's u_hash_table.c has a comment saying Gallium should
probably consider switching to a linear probing hash table at some point.
So this seems like the best candidate for a shared data structure.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
v2 (Jason Ekstrand): Pick up another hash_table use and patch up scons

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agoutil: Move ralloc to a new src/util directory.
Kenneth Graunke [Tue, 25 Feb 2014 07:39:14 +0000 (23:39 -0800)]
util: Move ralloc to a new src/util directory.

For a long time, we've wanted a place to put utility code which isn't
directly tied to Mesa or Gallium internals.  This patch creates a new
src/util directory for exactly that purpose, and builds the contents as
libmesautil.la.

ralloc seemed like a good first candidate.  These days, it's directly
used by mesa/main, i965, i915, and r300g, so keeping it in src/glsl
didn't make much sense.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
v2 (Jason Ekstrand): More realloc uses and some scons fixes

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agomesa/SConscript: Use Makefile.sources instead of duplicating the file lists
Jason Ekstrand [Thu, 24 Jul 2014 00:20:12 +0000 (17:20 -0700)]
mesa/SConscript: Use Makefile.sources instead of duplicating the file lists

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agotargets/dri: resolve the scons build
Emil Velikov [Sat, 2 Aug 2014 03:49:33 +0000 (04:49 +0100)]
targets/dri: resolve the scons build

With earlier commit we've conditionally enabled/added the kms_dri target
for automake builds. Unfortunately the we forgot to add the appropriate
define in the scons build, resulting in a broken library due to the
undefined symbol 'kms_swrast_create_screen'.

Reported-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Tested-by: Roland Scheidegger <sroland@vmware.com>
10 years agomesa/st: Fix compiler warnings
Jan Vesely [Fri, 25 Jul 2014 14:33:00 +0000 (08:33 -0600)]
mesa/st: Fix compiler warnings

both array and index are unsigned types

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agogallium: Fix compiler warning.
Jan Vesely [Fri, 25 Jul 2014 14:33:00 +0000 (08:33 -0600)]
gallium: Fix compiler warning.

warning: type qualifiers ignored on function return type

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoglsl: fix switch statement default case regressions
Tapani Pälli [Fri, 25 Jul 2014 06:40:13 +0000 (09:40 +0300)]
glsl: fix switch statement default case regressions

This patch fixes regressions caused by commit 48deb4d. Regressions
happened because 'run_default' var did not get initialized when default
case was the last one.

Now all the switch tests in es3conform suite are passing.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81857

10 years agost/dri: Fix driver loading if swrast isn't built
Aaron Watry [Sat, 2 Aug 2014 21:10:34 +0000 (16:10 -0500)]
st/dri: Fix driver loading if swrast isn't built

If building hardware drivers only, then kms_swrast_create_screen
won't be defined in inline_drm_helper.h and hardware drivers will
fail to dlopen as a result.

Copy the #if guards from inline_drm_helper.h to dri_kms_init_screen
to make the definition/use of the function match.

Fixes radeonsi_dri.so dlopen with the following configure:

./configure --with-dri-drivers= --with-dri-driverdir=/usr/local/lib/dri/ \
--enable-gbm --enable-gallium-gbm --enable-debug --enable-opencl \
--enable-opencl-icd --with-gallium-drivers=radeonsi \
--with-egl-platforms=drm --enable-glx-tls --enable-texture-float \
--enable-omx

Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agomesa/st: only convert AND(a, NOT(b)) into MAD when not using native integers
Ilia Mirkin [Sun, 3 Aug 2014 05:27:32 +0000 (01:27 -0400)]
mesa/st: only convert AND(a, NOT(b)) into MAD when not using native integers

Native integers imply a somewhat different handling of booleans. Instead
of being 1.0/0.0 floats, they are 0 (true) / -1 (false) integers. As such
the original optimization no longer applies.

Reported-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>
10 years agoRemove XA state tracker support for Radeon
Marek Olšák [Wed, 30 Jul 2014 19:24:59 +0000 (21:24 +0200)]
Remove XA state tracker support for Radeon

We don't support this type of X acceleration and we never did.
Other drivers might want to do the same thing.

Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agodocs: Import 10.2.5 release notes, add news item.
Carl Worth [Sun, 3 Aug 2014 05:54:26 +0000 (22:54 -0700)]
docs: Import 10.2.5 release notes, add news item.

10 years agomesa/st: add support for dynamic ubo selection
Ilia Mirkin [Sun, 6 Jul 2014 05:37:03 +0000 (01:37 -0400)]
mesa/st: add support for dynamic ubo selection

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com> (v1)
v2: fix src register, use index2D for base of 1
Acked-by: Marek Olšák <marek.olsak@amd.com>
10 years agoi965: Delete stale "pre-gen4" comment in texture validation code.
Kenneth Graunke [Sat, 1 Feb 2014 09:10:15 +0000 (01:10 -0800)]
i965: Delete stale "pre-gen4" comment in texture validation code.

In commit 16060c5adcd4d809f97e874fcde763260c17ac18, Eric changed the
code to not relayout just for baselevel changes - only if the range of
miplevels actually increases.  So this comment is now wrong.

Notably, the i915 version of the code actually does what the comment
says.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Delete sampler state structures.
Kenneth Graunke [Sun, 27 Jul 2014 03:30:58 +0000 (20:30 -0700)]
i965: Delete sampler state structures.

We've moved to using bitshifts (like we did for surface state); nothing
uses the structures anymore.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Replace sizeof(struct gen7_sampler_state) with the size itself.
Kenneth Graunke [Sun, 27 Jul 2014 03:36:49 +0000 (20:36 -0700)]
i965: Replace sizeof(struct gen7_sampler_state) with the size itself.

These are the last users of struct gen7_sampler_state.

v2: Use a local sampler_state_size variable, to help distinguish the
    various 16s (suggested by Topi Pohjolainen).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Drop sizeof(struct brw_sampler_state) from estimated prim size.
Kenneth Graunke [Sun, 27 Jul 2014 03:32:41 +0000 (20:32 -0700)]
i965: Drop sizeof(struct brw_sampler_state) from estimated prim size.

This is the last user of the structure.

v2: Use a local variable with a sensible name so people know what 16 is.
    (Suggested by Topi Pohjolainen).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Make BLORP use brw_emit_sampler_state().
Kenneth Graunke [Sun, 27 Jul 2014 03:49:32 +0000 (20:49 -0700)]
i965: Make BLORP use brw_emit_sampler_state().

This simplifies the code, removes use of the old structures, and also
allows us to combine the Gen6 and Gen7+ code.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Delete redundant sampler state dumping code.
Kenneth Graunke [Sun, 27 Jul 2014 03:43:40 +0000 (20:43 -0700)]
i965: Delete redundant sampler state dumping code.

Although the Gen4-6 and Gen7+ variants used different structure types,
they didn't use any of the fields - only the size, which is identical.
So both decoders did exactly the same thing.

Someday we should implement useful decoders for SAMPLER_STATE.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Make some brw_sampler_state.c functions static again.
Kenneth Graunke [Sun, 27 Jul 2014 01:10:53 +0000 (18:10 -0700)]
i965: Make some brw_sampler_state.c functions static again.

Now that gen7_sampler_state.c is gone, everything is once again in a
single file.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Stop using gen7_update_sampler_state; rm gen7_sampler_state.c.
Kenneth Graunke [Sat, 26 Jul 2014 20:15:56 +0000 (13:15 -0700)]
i965: Stop using gen7_update_sampler_state; rm gen7_sampler_state.c.

The code in brw_sampler_state.c now handles all generations; we don't
need the extra Gen7+ only code anymore.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Make brw_update_sampler_state use 8 bits for LOD fields on Gen7+.
Kenneth Graunke [Sat, 26 Jul 2014 20:12:37 +0000 (13:12 -0700)]
i965: Make brw_update_sampler_state use 8 bits for LOD fields on Gen7+.

This was the only actual difference between Gen4-6 and Gen7+ in terms of
the values we program.  The rest was just mechanical structure
rearrangement.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Make brw_update_sampler_state() use brw_emit_sampler_state().
Kenneth Graunke [Sat, 26 Jul 2014 10:04:12 +0000 (03:04 -0700)]
i965: Make brw_update_sampler_state() use brw_emit_sampler_state().

Instead of stuffing bits directly into the brw_sampler_state structure,
we now store them in local variables, then use brw_emit_sampler_state()
to assemble the packet.  This separates the decision about what values
to use from the actual packet emission, which makes the code more
reusable across generations.

v2: Put const on a bunch of local variables and move declarations,
    as suggested by Topi Pohjolainen.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Introduce a function to emit a SAMPLER_STATE structure.
Kenneth Graunke [Sat, 26 Jul 2014 19:58:20 +0000 (12:58 -0700)]
i965: Introduce a function to emit a SAMPLER_STATE structure.

This simply assembles all the SAMPLER_STATE fields into their proper bit
locations.  Making it work on all generations was easy enough; some of
the fields are even in the same place.

Not used by anything yet, but will be soon.  I made it non-static so
BLORP can use it too.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Add const to upload_default_color's sampler parameter.
Kenneth Graunke [Wed, 30 Jul 2014 17:50:12 +0000 (10:50 -0700)]
i965: Add const to upload_default_color's sampler parameter.

It doesn't edit the value, and this lets us use const in more places.

Needed to implement Topi's review comments for the next patch.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Add #defines for SAMPLER_STATE fields.
Kenneth Graunke [Sat, 26 Jul 2014 19:32:31 +0000 (12:32 -0700)]
i965: Add #defines for SAMPLER_STATE fields.

We'll use these to replace the existing structures.

I've adopted the convention that "BRW" applies to all hardware, and
"GENX" applies starting with generation X, but might be replaced by some
later generation.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Convert wrap mode #defines to an enum.
Kenneth Graunke [Sat, 26 Jul 2014 19:06:20 +0000 (12:06 -0700)]
i965: Convert wrap mode #defines to an enum.

This makes it easy to tell that they're grouped together, and also
improves gdb printing.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Delete gen7_upload_sampler_state_table and vtable mechanism.
Kenneth Graunke [Sat, 26 Jul 2014 18:48:50 +0000 (11:48 -0700)]
i965: Delete gen7_upload_sampler_state_table and vtable mechanism.

brw_upload_sampler_state_table now handles all generations, so we don't
need the vtable mechanism either.

There's still a lot of code duplication; the next patches will address
that.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Make brw_upload_sampler_state_table handle Gen7+ as well.
Kenneth Graunke [Sat, 26 Jul 2014 18:36:33 +0000 (11:36 -0700)]
i965: Make brw_upload_sampler_state_table handle Gen7+ as well.

This copies a few changes from gen7_upload_sampler_state_table; the next
patch will delete that function.

Gen7+ has per-stage sampler state pointer update packets, so we emit
them as soon as we emit a new table for a stage.  On Gen6 and earlier,
we have a single packet, so we delay until we've changed everything
that's going to be changed.

v2: Split 3DSTATE_SAMPLER_STATE_POINTERS_XS packet emission into a
    helper function (suggested by Topi Pohjolainen).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Shift brw_upload_sampler_state_table away from structures.
Kenneth Graunke [Sat, 26 Jul 2014 18:27:57 +0000 (11:27 -0700)]
i965: Shift brw_upload_sampler_state_table away from structures.

The Gen4-6 and Gen7+ code is virtually identical, but both use different
structure types.  Switching to use a uint32_t pointer and operate on the
number of DWords will make it possible to share code.

It turns out that SURFACE_STATE is the same number of DWords on every
platform currently; it will be easy to handle a change there, though.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Push computation for sampler state batch offsets up a level.
Kenneth Graunke [Sat, 26 Jul 2014 18:20:02 +0000 (11:20 -0700)]
i965: Push computation for sampler state batch offsets up a level.

Other than this, brw_update_sampler_state only deals with a single
SAMPLER_STATE structure, and doesn't need to know which position it is
in the table.  The caller takes care of dealing with multiple surface
states.

Pushing this up a level allows us to drop the ss_index parameter.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Drop unused 'ss_index' parameter from gen7_update_sampler_state.
Kenneth Graunke [Sat, 26 Jul 2014 18:42:27 +0000 (11:42 -0700)]
i965: Drop unused 'ss_index' parameter from gen7_update_sampler_state.

This was copied from the Gen4-6 code, but is unused.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Stop storing sdc_offset in brw_stage_state.
Kenneth Graunke [Sat, 26 Jul 2014 09:34:02 +0000 (02:34 -0700)]
i965: Stop storing sdc_offset in brw_stage_state.

sdc_offset is produced and consumed in the same function, so there's no
need to store it in the context, nor pass pointers to it through various
call chains.

Saves 128 bytes per brw_stage_state structure, and makes the code
clearer as well.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Drop the degenerate brw_sampler_default_color structure.
Kenneth Graunke [Wed, 7 May 2014 05:35:27 +0000 (22:35 -0700)]
i965: Drop the degenerate brw_sampler_default_color structure.

It's just an array of four floats, and we have an array of four floats,
so this is literally just a memcpy...but with custom structs and strange
macros to give the appearance of doing something more.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Write a better file comment for brw_sampler_state.c.
Kenneth Graunke [Sat, 26 Jul 2014 19:03:21 +0000 (12:03 -0700)]
i965: Write a better file comment for brw_sampler_state.c.

The old one has been inaccurate for years.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965: Rename brw_wm_sampler_state.c to brw_sampler_state.c.
Kenneth Graunke [Sat, 26 Jul 2014 08:26:54 +0000 (01:26 -0700)]
i965: Rename brw_wm_sampler_state.c to brw_sampler_state.c.

When the driver was originally written, it only supported texturing in
the pixel shader backend; vertex and geometry shader texturing came much
later.  Originally, the pixel shader was referred to as "WM" (the
Windowizer/Masker unit).  So, this code happened to only be relevant for
the WM stage, at the time.

However, sampler state really applies to all stages, so putting "wm" in
the filename doesn't make sense.  I dropped it in gen7_sampler_state.c;
at this point the asymmetry just trips people up.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agoi965/blorp: Don't set min_mag_neq bit in Gen6 SAMPLER_STATE.
Kenneth Graunke [Tue, 29 Jul 2014 20:33:43 +0000 (13:33 -0700)]
i965/blorp: Don't set min_mag_neq bit in Gen6 SAMPLER_STATE.

The "Min/Mag State Not Equal" bit is supposed to be set when the min/mag
filters or address rounding modes differ.  BLORP uses identical min/mag
settings, so the bit should be unset.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
10 years agodefine GL_OES_standard_derivatives if extension is supported
Kevin Rogovin [Mon, 19 May 2014 07:52:44 +0000 (10:52 +0300)]
define GL_OES_standard_derivatives if extension is supported

Define the macro GL_OES_standard_derivatives as 1 if the extension
GL_OES_standard_derivatives is supported.

V2 [Chris]: Correct trailing whitespace

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agollvmpipe: don't store number of layers per level
Roland Scheidegger [Fri, 1 Aug 2014 21:22:39 +0000 (23:22 +0200)]
llvmpipe: don't store number of layers per level

This could be recalculated, though it turns out the only use of it after
resource allocation is for calculating whole resource size (for scene size
accounting though that isn't quite ideal neither). Thus, instead just store
the whole resource size and drop it (saving a couple bytes of storage per
resource). It makes things simpler too. Note that for the accounting winsys
resources always come back with size 0 but this is unchanged (we don't actually
know the size in any case).
Also reformat llvmpipe_texture_layout (drop unneded indentation).

v2: adapt to previous changes.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agollvmpipe: integrate memory allocation into llvmpipe_texture_layout
Roland Scheidegger [Fri, 1 Aug 2014 01:08:11 +0000 (03:08 +0200)]
llvmpipe: integrate memory allocation into llvmpipe_texture_layout

Seems pointless to just duplicate some of the calculations (the calculation
of actual memory used compared to what was predicted in llvmpipe_texture_layout
actually could have differed slightly in some cases due to different alignment
rules used though this should have been of no consequence).

v2: keep the previous mip alignment of MAX2(64, cacheline). This was added for
ARB_map_buffer_alignment - I'm not convinced it's needed for textures, but
it was supposed to be cleanup without functional change. Also replace div
with 64bit mul / comparison.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agollvmpipe: get rid of impossible code in alloc_image_data
Roland Scheidegger [Thu, 31 Jul 2014 17:13:57 +0000 (19:13 +0200)]
llvmpipe: get rid of impossible code in alloc_image_data

Only used for non display target resources.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
10 years agoi965/miptree: Layout 1D Array as 2D Array with height of 1
Jordan Justen [Thu, 31 Jul 2014 07:16:19 +0000 (00:16 -0700)]
i965/miptree: Layout 1D Array as 2D Array with height of 1

1D array miptrees were being laid out as a 2D texture with 1 slice.
This happened due to the mesa core storing the 1D array slice count in
the height field. On Intel hardware, we want to create a 2D array with
a height of 1 for the 1D array case.

Fixes assertion failure in piglit (gen6, gen8):
spec/glsl-1.30/execution/tex-miplevel-selection textureOffset 1DArrayShadow

In release builds of Mesa, this test was observed to cause a GPU hang
on gen8.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81450
Tested-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agor600g: Implement gpu_shader5 textureGather
Glenn Kennard [Sun, 20 Jul 2014 13:59:16 +0000 (15:59 +0200)]
r600g: Implement gpu_shader5 textureGather

Adds 0-3 textureGather component selection and non-constant offsets

Caveat: 0 and 1 texture swizzles only work if textureGather component
select is 3 or a component that does not exist in the sampler texture
format. This is a hardware limitation, any other value returns
128/255=0.501961 for both 0 and 1.

Passes all textureGather piglit tests on radeon 6670, except for those
using 0/1 texture swizzles due to aforementioned reason.

Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
10 years agomesa: Add missing atomic buffer bindings and unbindings
Aditya Atluri [Thu, 24 Jul 2014 19:18:49 +0000 (15:18 -0400)]
mesa: Add missing atomic buffer bindings and unbindings

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agor600g/radeonsi: Prefer VRAM for CPU -> GPU streaming buffers
Michel Dänzer [Wed, 25 Jun 2014 09:36:43 +0000 (18:36 +0900)]
r600g/radeonsi: Prefer VRAM for CPU -> GPU streaming buffers

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agor600g/radeonsi: Reduce or even drop special treatment of persistent mappings
Michel Dänzer [Tue, 22 Jul 2014 09:37:59 +0000 (18:37 +0900)]
r600g/radeonsi: Reduce or even drop special treatment of persistent mappings

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agotarget-helpers: Do not build kms_dri on libdrm-less platforms.
Jon TURNEY [Thu, 31 Jul 2014 15:30:37 +0000 (16:30 +0100)]
target-helpers: Do not build kms_dri on libdrm-less platforms.

Fix build since 3b176c441b7ddc5f7d2f891da3f76cf3c1814ce1 for
dri_platform=none hosts.

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agor600g: gpu_shader5 gl_SampleMaskIn support
Glenn Kennard [Wed, 23 Jul 2014 09:57:55 +0000 (11:57 +0200)]
r600g: gpu_shader5 gl_SampleMaskIn support

Map TGSI_SEMANTIC_SAMPLEMASK to register/component.
Enable face register when sample mask is needed by shader.
Requires Evergreen/Cayman

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
10 years agor600g: Implement gpu_shader5 integer ops
Glenn Kennard [Wed, 23 Jul 2014 09:36:36 +0000 (11:36 +0200)]
r600g: Implement gpu_shader5 integer ops

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
10 years agor600g: Add IMUL_HI/UMUL_HI support
Glenn Kennard [Wed, 23 Jul 2014 09:10:06 +0000 (11:10 +0200)]
r600g: Add IMUL_HI/UMUL_HI support

Fixes fs-imulExtended, fs-imulExtended-only-msb, fs-umulExtended,
fs-umulExtended-only-msb piglit tests.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
10 years agor600g: Implement GL_ARB_texture_query_lod
Glenn Kennard [Wed, 23 Jul 2014 08:48:17 +0000 (10:48 +0200)]
r600g: Implement GL_ARB_texture_query_lod

Requires Evergreen or later

v2 (Andreas): Update relnotes/10.3

Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
10 years agogbm: Log at least one dlerror() when we fail to open any drivers.
Eric Anholt [Fri, 20 Jun 2014 13:10:29 +0000 (14:10 +0100)]
gbm: Log at least one dlerror() when we fail to open any drivers.

We don't want to log every single error (such as all the ones where the file
wasn't even present in our list of search paths), but if you didn't find any
driver, then seeing at least one error is useful (since the common case as a
developer is a single DEFAULT_DRIVER_DIR or GBM_DRIVERS_PATH entry).

v2: Rebase on swrast changes.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogbm: Fix a debug log message
Eric Anholt [Fri, 20 Jun 2014 12:54:09 +0000 (13:54 +0100)]
gbm: Fix a debug log message

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium: Add a uif() helper function to complement fui()
Eric Anholt [Fri, 27 Jun 2014 14:09:32 +0000 (15:09 +0100)]
gallium: Add a uif() helper function to complement fui()

I found myself often wanting this when I'm printing out a uint32_t mapping
of some GPU data, and I want to put in an interpretation of that value as
a float.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agoglapi: Do not use backtrace on DragonFly.
Vinson Lee [Thu, 19 Jun 2014 21:31:24 +0000 (14:31 -0700)]
glapi: Do not use backtrace on DragonFly.

execinfo.h is not available on DragonFly.

Fixes this build error.

  CC       glapi_gentable.lo
glapi_gentable.c:44:22: fatal error: execinfo.h: No such file or directory

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agogallivm: fix up out-of-bounds level when using conformant out-of-bound behavior
Roland Scheidegger [Wed, 30 Jul 2014 03:07:20 +0000 (05:07 +0200)]
gallivm: fix up out-of-bounds level when using conformant out-of-bound behavior

When using (d3d10) conformant out-of-bound behavior for texel fetching
(currently always enabled) the level still needs to be set to a safe value
even though the offset in the end won't get used because the level is used
to look up the mip offset itself and the actual strides, which might otherwise
crash.
For simplicity, we'll use level 0 in this case (this ought to be safe, llvmpipe
does not actually fill in level 0 information if first_level is larger, but
some random strides / offsets shouldn't hurt as ultimately we always use
offset 0 in this case).
Fixes a crash in some in-house test where random huge levels appear in
lp_build_fetch_texel() (the test actually uses level 0 always but if the
fetching happens in a block with a execution mask random values may appear).

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
10 years agodri: Add a new capabilities for drivers that can't share buffers
Giovanni Campagna [Wed, 23 Jul 2014 18:28:52 +0000 (19:28 +0100)]
dri: Add a new capabilities for drivers that can't share buffers

The kms-dri swrast driver cannot share buffers using the GEM,
so it must tell the loader to disable extensions relying on
that, without disabling the image DRI extension altogether
(which would prevent the loader from working at all).
This requires a new gallium capability (which is queried on
the pipe_screen and for swrast drivers it's forwarded to the
winsys), and requires a new version of the DRI image extension.

[Emil Velikov]
 - Rebased on top of gallium-dri megadrivers.
 - Drop PIPE_CAP_BUFFER_SHARE and sw_winsys::get_param hook.
The can_share_buffer cap is set at InitScreen. We use a different
InitScreen (and thus value for the cap) function for kms_dri, due to
deeper differences originating from dri megadrivers.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium: Add a dumb drm/kms winsys backed swrast provider
Giovanni Campagna [Wed, 23 Jul 2014 18:37:31 +0000 (19:37 +0100)]
gallium: Add a dumb drm/kms winsys backed swrast provider

Add a new winsys and target that can be used with a dri2 state tracker
and loader instead of drisw. This allows to use gbm as a dri2/image
loader and avoid the extra copy from the backbuffer to the shadow
frontbuffer.

The new driver is called "kms_swrast", and is loaded by gbm as a
fallback, because it is only useful with the gbm platform (as no buffer
sharing is possible)

To force select the driver set the environment variable
GBM_ALWAYS_SOFTWARE

[Emil Velikov]
 - Rebase on top of gallium megadriver.
 - s/text/test/ in configure.ac (Spotted by Andreas Pokorny).
 - Add scons support for winsys/sw/kms-dri and fix the build.
 - Provide separate DriverAPI, due to different InitScreen hook.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agoAdd support for swrast to the DRM EGL platform
Giovanni Campagna [Sun, 15 Jun 2014 11:49:49 +0000 (13:49 +0200)]
Add support for swrast to the DRM EGL platform

Turn GBM into a swrast loader (providing putimage/getimage backed
by a dumb KMS buffer). This allows to run KMS+DRM GL applications
(such as weston or mutter-wayland) unmodified on cards that don't
have any client side HW acceleration component but that can do
modeset (examples include simpledrm and qxl)

[Emil Velikov]
 - Fix make check.
 - Split dri_open_driver() from dri_load_driver().
 - Don't try to bind the swrast extensions when using dri.
 - Handle swrast->CreateNewScreen() failure.
 - strdup the driver_name, as it's free'd at destruction.
 - s/LIBGL_ALWAYS_SOFTWARE/GBM_ALWAYS_SOFTWARE/
 - Move gbm_dri_bo_map/unmap to gbm_driiint.h.
 - Correct swrast fallback logic.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agost/gbm: don't segfault if the fail to create the screen
Emil Velikov [Wed, 23 Jul 2014 17:08:46 +0000 (18:08 +0100)]
st/gbm: don't segfault if the fail to create the screen

Whenever dd_create_screen/pipe_loader_* fails, gdrm->dev may be NULL.
Thus peeking inside the struct will lead to a crash.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agost/gbm: retrieve the driver-name via dd_driver_name()
Emil Velikov [Wed, 23 Jul 2014 17:10:08 +0000 (18:10 +0100)]
st/gbm: retrieve the driver-name via dd_driver_name()

... on static targets. Otherwise we'll crash badly as gdrm->dev is
NULL when we try to copy the string driver_name.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agoglsl/glcpp: rename ERROR to ERROR_TOKEN to fix MSVC build
Brian Paul [Tue, 29 Jul 2014 22:56:06 +0000 (16:56 -0600)]
glsl/glcpp: rename ERROR to ERROR_TOKEN to fix MSVC build

ERROR is a #define in the MSVC WinGDI.h header file.
Add the _TOKEN suffix as we do for a few other lexer tokens.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoconfigure: Don't override user -g or -O options for debug builds
Ian Romanick [Mon, 14 Jul 2014 18:22:20 +0000 (11:22 -0700)]
configure: Don't override user -g or -O options for debug builds

Principle of least surprise: --enable-debug should enable debugging.
Ages ago, Mesa's build system only added -g in dri-debug builds (yay for
the static Makefiles).  If you forgot to change it (or wrap the build
with custom scripts), you would often be disappointed when trying to gdb
Mesa bugs.  New developers, that may not yet have custom scripts, will
have this same issue.

I think we should enable experienced developers to do what they want,
and make things easier for new developers.  I already pass '-ggdb3 -O1'
or '-ggdb3 -Og' for CFLAGS, and I don't want configure to change them
for me.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoglsl: Add flex options to eliminate the default rule
Carl Worth [Tue, 1 Jul 2014 22:10:02 +0000 (15:10 -0700)]
glsl: Add flex options to eliminate the default rule

We've had bugs in the past where we have been inadvertently matching the
default rule.

Just as we did in the pre-processor in the previous commit, we can use:

%option warn nodefault

in the compiler to instruct flex to not generate the default rule, and
further to warn if our set of rules could let any characters go unmatched.

With this warning active, flex actually warns that the catch-all rule we
recently added to the compiler could never be matched. Since that is all
safely determined at compile time now, we can safely drop this run-time
compiler error message, (as we do in this commit).

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agoglsl/glcpp: Add flex options to eliminate the default rule.
Carl Worth [Tue, 1 Jul 2014 22:04:38 +0000 (15:04 -0700)]
glsl/glcpp: Add flex options to eliminate the default rule.

We've had multiple bugs in the past where we have been inadvertently matching
the default rule, (which we never want to do). We recently added a catch-all
rule to avoid this, (and made this rule robust for future start conditions).

Kristian pointed out that flex allows us to go one step better. This syntax:

%option warn nodefault

instructs flex to not generate the default rule at all. Further, flex will
generate a warning at compile time if the set of rules we provide are
inadequate, (such that it would be possible for the default rule to be
matched).

With this warning in place, I found that the catch-all rule was in fact
missing something. The catch-all rule uses a pattern of "." which doesn't
match newlines. So here we extend the newline-matching rule to all start
conditions. That is enough to convince flex that it really doesn't need
any default rule.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agoglsl/glcpp: Combine the two rules matching any character
Carl Worth [Tue, 1 Jul 2014 22:02:14 +0000 (15:02 -0700)]
glsl/glcpp: Combine the two rules matching any character

Using a single rule here means that we can use the <*> syntax to match
all start conditions. This makes the catch-all rule more robust against
the addition of future start conditions, (no need to maintain an ever-
growing list of start conditions for this rul).

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agoglsl/glcpp: Alphabetize lists of start conditions
Carl Worth [Tue, 1 Jul 2014 21:55:56 +0000 (14:55 -0700)]
glsl/glcpp: Alphabetize lists of start conditions

There is no behavioral change here. It's just easier to verify that lists
of start conditions include all expected conditions when they appear in a
consistent order.

The <INITIAL> state is special, so it appears first in all lists. All others
appear in alphabetical order.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agoglsl/glcpp: Add a catch-all rule for unexpected characters.
Carl Worth [Thu, 19 Jun 2014 00:03:19 +0000 (17:03 -0700)]
glsl/glcpp: Add a catch-all rule for unexpected characters.

In some of the recent glcpp bug-fixing, we found that glcpp was emitting
unrecognized characters from the input source file to stdout, and dropping
them from the source passed onto the compiler proper.

This was obviously confusing, and totally undesired.

The bogus behavior comes from an implicit default rule in flex, which is
that any unmatched character is implicitly matched and printed to stdout.

To avoid this implicit matching and printing, here we add an explicit
catch-all rule. If this rule ever matches it prints an internal compiler
error. The correct response for any such error is fixing glcpp to handle
the unexpected character in the correct way.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Treat carriage return as equivalent to line feed.
Carl Worth [Thu, 26 Jun 2014 21:04:01 +0000 (14:04 -0700)]
glsl/glcpp: Treat carriage return as equivalent to line feed.

Previously, the '\r' character was not explicitly matched by any lexer
rule. This means that glcpp would have been using the default flex rule to
match '\r' characters, (where they would have been printed to stdout rather
than actually correctly handled).

With this commit, we treat '\r' as equivalent to '\n'. This is clearly an
improvement the bogus printing to stdout. The resulting behavior is compliant
with the GLSL specification for any source file that uses exclusively '\r' or
'\n' to separate lines.

For shaders that use a multiple-character line separator, (such as "\r\n"),
glcpp won't be precisely compliant with the specification, (treating these as
two newline characters rather than one), but this should not introduce any
semantic changes to the shader programs.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Add test for a multi-line comment within an #if 0 block
Carl Worth [Thu, 26 Jun 2014 17:56:41 +0000 (10:56 -0700)]
glsl/glcpp: Add test for a multi-line comment within an #if 0 block

This test is written to exercise a bug which I recently wrote, (but
fortunately caught and fixed before ever committing it).

For the curious:

  The bug happened when the NEWLINE_CATCHUP code didn't actually return the
  NEWLINE token (due to the skipping). This resulted in the lexer continuing
  on through all the subsequent rules while still in the NEWLINE_CATCHUP start
  condition, (which then triggered the internal-compiler-error catch-all
  rule).

  What is intended is for the return of the NEWLINE token to start a new
  iteration of the lexer loop, at which time the NEWLINE_CATCHUP-handling code
  will reset from the <NEWLINE_CATCHUP> to the <INITIAL> start condition.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Test that macro parameters substitute immediately after periods
Carl Worth [Thu, 26 Jun 2014 17:08:34 +0000 (10:08 -0700)]
glsl/glcpp: Test that macro parameters substitute immediately after periods

At one point while rewriting the lexing rule for pre-processing numbers, I
made it a bit too aggressive and within a replacement list sucked up a
parameter name that appeared immediately after a period. This caused the
parameter name to be unreplaced when the macro was expanded.

It was in some piglit tests that I originally found this issue. Here, I'm
adding a test to "make check" to ensure that this behavior remains correct.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Add (non)-support for ++ and -- operators
Carl Worth [Wed, 25 Jun 2014 21:17:37 +0000 (14:17 -0700)]
glsl/glcpp: Add (non)-support for ++ and -- operators

These operators aren't defined for preprocessor expressions, so we never
implemented them. This led them to be misinterpreted as strings of unary
'+' or '-' operators.

In fact, what is actually desired is to generate an error if these operators
appear in any preprocessor condition.

So this commit looks like it is strictly adding support for these
operators. And it is supporting them as far as passing them through to the
subsequent compiler, (which was already happening anyway).

What's less apparent in the commit is that with these tokens now being lexed,
but with no change to the grammar for preprocessor expressions, these
operators will now trigger errors there.

A new "make check" test is added to verify the desired behavior.

This commit fixes the following Khronos GLES3 CTS test:

invalid_op_1_vertex
invalid_op_1_fragment
invalid_op_2_vertex
invalid_op_2_fragment

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Emit error for duplicate parameter name in function-like macro
Carl Worth [Wed, 25 Jun 2014 20:41:47 +0000 (13:41 -0700)]
glsl/glcpp: Emit error for duplicate parameter name in function-like macro

This will emit an error for something like:

#define FOO(x,x) ...

Obviously, it's not a legal thing to do, and it's easy to check.

Add a "make check" test for this as well.

This fixes the following Khronos GLES3 CTS tests:

invalid_function_definitions.unique_param_name_vertex
invalid_function_definitions.unique_param_name_fragment

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Add an explanatory comment for "loc != NULL" check
Carl Worth [Wed, 25 Jun 2014 20:28:38 +0000 (13:28 -0700)]
glsl/glcpp: Add an explanatory comment for "loc != NULL" check

Just reading the code, it looked like a bug that _define_object_macro had this
check, but _define_function_macro did not. Upon further reading, that's
because the check is to allow for our builtins to be defined, (and there are
no builtin function-like macros).

Add my new understanding as a comment to help the next reader.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Drop the HASH_ prefix from token names like HASH_IF
Carl Worth [Wed, 25 Jun 2014 20:55:18 +0000 (13:55 -0700)]
glsl/glcpp: Drop the HASH_ prefix from token names like HASH_IF

Previously, we had a single token for "#if" but now that we have two separate
tokens, it looks much better to see:

HASH_TOKEN IF

than:

HASH_TOKEN HASH_IF

(Note, that for the same reason we use HASH_TOKEN instead of HASH, we also use
DEFINE_TOKEN instead of DEFINE to avoid a conflict with the <DEFINE> start
condition in the lexer.)

There should be no behavioral change from this commit.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl: Properly lex extra tokens when handling # directives.
Kenneth Graunke [Thu, 12 Jun 2014 05:04:09 +0000 (22:04 -0700)]
glsl: Properly lex extra tokens when handling # directives.

Without this, in the <PP> state, we would hit Flex's default rule, which
prints tokens to stdout, rather than returning them as tokens. (Or, after the
previous commit, we would hit the new catch-all rule and generate an internal
compiler error.)

With this commit in place, we generate the desired syntax error.

This manifested as a weird bug where shaders with semicolons after
extension directives, such as:

   #extension GL_foo_bar : enable;

would print semicolons to the screen, but otherwise compile just fine
(even though this is illegal).

Fixes Piglit's extension-semicolon.frag test.

This also fixes the following Khronos GLES3 conformance tests, (and for real
this time):

invalid_char_in_name_vertex
invalid_char_in_name_fragment

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Carl Worth <cworth@cworth.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl: Add an internal-error catch-all rule
Carl Worth [Wed, 25 Jun 2014 20:07:44 +0000 (13:07 -0700)]
glsl: Add an internal-error catch-all rule

This is to avoid the default, silent flex rule which simply prints the
character to stdout.

For the following Khronos GLES3 conformance tests:

invalid_char_in_name_vertex
invalid_char_in_name_fragment

With this commit, these tests now report Pass where they previously reported
Fail, but Mesa isn't behaving correctly yet. It's now reporting the internal
error where what is really desired is a syntax error.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Correctly parse directives with intervening comments
Carl Worth [Wed, 25 Jun 2014 19:20:22 +0000 (12:20 -0700)]
glsl/glcpp: Correctly parse directives with intervening comments

It's legal (though highly bizarre) for a pre-processor directive to look like
this:

#  /* why? */ define FOO bar

This behavior comes about since the specification defines separate logical
phases in a precise order, and comment-removal occurs in a phase before the
identification of directives.

Our implementation does not use an actual separate phase for comment removal,
so some extra care is necessary to correctly parse this. What we want is for
'#' to introduce a directive iff it is the first token on a line, (ignoring
whitespace and comments). Previously, we had a lexical rule that worked only
for whitespace (not comments) with the following regular expression to find a
directive-introducing '#' at the beginning of a line:

HASH ^{HSPACE}*#{HSPACE}*

In this commit, we switch to instead use a simple literal match of '#' to
return a HASH_TOKEN token and add a new <HASH> start condition for whenever
the HASH_TOKEN is the first non-space token of a line. This requires the
addition of the new bit of state: first_non_space_token_this_line.

This approach has a couple of implications on the glcpp parser:

1. The parser now sees two separate tokens, (such as HASH_TOKEN and
   HASH_DEFINE) where it previously saw one token (HASH_DEFINE) for
   the sequence "#define". This is a straightforward change throughout
   the grammar.

2. The parser may now see a SPACE token before the HASH_TOKEN token of
   a directive. Previously the lexical regular expression for {HASH}
   would eat up the space and there would be no SPACE token.

This second implication is a bit of a nuisance for the parser. It causes a
SPACE token to appear in a production of the grammar with the following two
definitions of a control_line:

control_line
SPACE control_line

This is really ugly, since normally a space would simply be a token
separator, so it wouldn't appear in the tokens of a production. This leads to
a further problem with interleaved spaces and comments:

/* ... */    /* ... */ #define /* ..*/

For this, we must not return several consecutive SPACE tokens, or else we would need an arbitrary number of new productions:

SPACE SPACE control_line
SPACE SPACE SPACE control_line
ad nauseam

To avoid this problem, in this commit we also change the lexer to emit only a
single SPACE token for any series of consecutive spaces, (whether from actual
whitespace or comments). For this compression, we add a new bit of parser
state: last_token_was_space. And we also update the expected results of all
necessary test cases for the new compression of space tokens.

Fortunately, the compression of spaces should not lead to any semantic changes
in terms of what the eventual GLSL compiler sees.

So there's a lot happening in this commit, (particularly for such a tiny
feature). But fortunately, the lexer itself is looking cleaner than ever. The
only ugly bit is all the state updating, but it is at least isolated to a
single shared function.

Of course, a new "make check" test is added for the new feature, (directives
with comments and whitespace interleaved in many combinations).

And this commit fixes the following Khronos GLES3 CTS tests:

function_definition_with_comments_vertex
function_definition_with_comments_fragment

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Rename HASH token to HASH_TOKEN
Carl Worth [Wed, 25 Jun 2014 18:59:55 +0000 (11:59 -0700)]
glsl/glcpp: Rename HASH token to HASH_TOKEN

This is in preparation for the planned addition of a new <HASH> start
condition to the lexer. Both start conditions and token types are, of course,
in the same default C namespace, so a start condition and a token type with
the same name will collide. (And unfortunately, they are both apparently
implemented as equivalent numeric types so the collision is undetected at
compile time and simply leads to unpredictable behavior at run time.)

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Don't use start-condition stack when switching to/from <DEFINE>
Carl Worth [Wed, 25 Jun 2014 18:52:02 +0000 (11:52 -0700)]
glsl/glcpp: Don't use start-condition stack when switching to/from <DEFINE>

This commit does not cause any behavioral change for any valid program. Prior
to entering the <DEFINE> start condition, the only valid start condition is
<INITIAL>, so whether pushing/popping <DEFINE> onto the stack or explicit
returning to <INITIAL> is equivalent.

The reason for this change is that we are planning to soon add a start
condition for <HASH> with the following semantics:

<HASH>: We just saw a directive-introducing '#'

<DEFINE>: We just saw "#define" starting a directive

With these two start conditions in place, the only correct behavior is to
leave <DEFINE> by returning to <INITIAL>. But the old push/pop code would have
returned to the <HASH> start condition which would then cause an error when
the next directive-introducing '#' would be encountered.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Add a -d/--debug option to the standalone glcpp program
Carl Worth [Wed, 25 Jun 2014 18:46:49 +0000 (11:46 -0700)]
glsl/glcpp: Add a -d/--debug option to the standalone glcpp program

The verbose debug output from the parser is quite useful when debugging, and
having this available as a command-line option is much more convenient than
manually forcing this into the code when needed, (which is what I had been
doing for too long previously).

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agoglsl/glcpp: Fix off-by-one error in column in first-line error messages
Carl Worth [Sat, 21 Jun 2014 02:17:28 +0000 (19:17 -0700)]
glsl/glcpp: Fix off-by-one error in column in first-line error messages

For the first line we were initializing the column to 1, but for all
subsequent lines we were initializing the column to 0. The column number is
advanced for each token read before any error message is printed. So the 0
value is the correct initialization, (so that the first column is reported as
column 1).

With this extremely minor change, many of the .expected files are updated such
that error messages for the first line now have the correct column number in
them.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>