mesa.git
7 years agoi965: Drop brw_bo_alloc in ARB_indirect_parameters implementation.
Kenneth Graunke [Tue, 10 Oct 2017 17:19:21 +0000 (10:19 -0700)]
i965: Drop brw_bo_alloc in ARB_indirect_parameters implementation.

The original implementation allocated a new BO here, but we decided to
switch to intel_upload_space, which returns a reference to the current
upload BO.  We accidentally kept the brw_bo_alloc, even though it's no
longer necessary - intel_upload_space will immediately unreference it,
causing us to allocate and immediately free a buffer.

Reviewed-by: Plamena Manolova <plamena.manolova@intel.com>
7 years agoi965: Allow mapped VBOs during drawing in non-debug contexts.
Kenneth Graunke [Sun, 30 Jul 2017 23:15:56 +0000 (16:15 -0700)]
i965: Allow mapped VBOs during drawing in non-debug contexts.

Section 6.3.2 of the GL 4.5 spec says:

   "Any GL command which attempts to read from, write to, or change
    the state of a buffer object may generate an INVALID_OPERATION error
    if all or part of the buffer object is mapped ... However, only
    commands which explicitly describe this error are required to do so.
    If an error is not generated, such commands will have undefined
    results and may result in GL interruption or termination."

Setting this flag allows us to skip walking over the buffer bindings
for every enabled vertex attribute (_mesa_all_buffers_are_unmapped).

Improves performance in GFXBench4's gl_driver2_off microbenchmark by
3.05797% +/- 0.709031% (n=33) on Apollolake.

This breaks KHR-*.draw_elements_base_vertex_tests.invalid_mapped_bos,
but that test is invalid and has been removed from the upstream CTS.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomeson: fix glx test
Dylan Baker [Tue, 10 Oct 2017 21:50:53 +0000 (14:50 -0700)]
meson: fix glx test

That requires a generated header that was rolled into a loop.

fixes: a47c525f3281a27 ("meson: build glx")
Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agonv50,nvc0: fix push hint logic in presence of a start offset
Ilia Mirkin [Wed, 11 Oct 2017 03:50:06 +0000 (23:50 -0400)]
nv50,nvc0: fix push hint logic in presence of a start offset

Previously buffer offsets were passed in explicitly as an offset, which
had to be added to the resource address. Now they are passed in via an
increased 'start' parameter. As a result, we were double-adding the
start offset in this kind of situation.

This condition was triggered by piglit's draw-elements test which has a
requisite glMultiDrawElements in combination with a small enough number
of vertices to go through the immediate push path.

Fixes: 330d0607ed6 ("gallium: remove pipe_index_buffer and set_index_buffer")
Reported-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
7 years agoi965: Make brw_update_texture_surface static.
Kenneth Graunke [Sat, 9 Sep 2017 07:18:39 +0000 (00:18 -0700)]
i965: Make brw_update_texture_surface static.

Trivial.  It's not used in other files.

7 years agoAndroid: fix build break from r600/radeon split
Rob Herring [Tue, 10 Oct 2017 22:01:29 +0000 (17:01 -0500)]
Android: fix build break from r600/radeon split

Commit 06bfb2d28f7a ("r600: fork and import gallium/radeon") broke the
Android build:

external/mesa3d/src/gallium/drivers/radeon/r600_pipe_common.c:43:10: fatal error: 'llvm-c/TargetMachine.h' file not found
         ^~~~~~~~~~~~~~~~~~~~~~~~

Update the Android makefiles so that drivers/radeon is only built when
radeonsi (and therefore LLVM) is enabled.

Fixes: 06bfb2d28f7a (r600: fork and import gallium/radeon)
Acked-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Rob Herring <robh@kernel.org>
7 years agoAndroid: move libraries to /vendor
Rob Herring [Fri, 15 Sep 2017 13:50:09 +0000 (08:50 -0500)]
Android: move libraries to /vendor

As part of Treble project in Android O, all the device specific files have
to be located in a separate vendor partition. This is done by setting
LOCAL_PROPRIETARY_MODULE (the name is misleading). This change will not
break existing platforms without a vendor partition as it will just move
files to /system/vendor.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Rob Herring <robh@kernel.org>
7 years agoi965: Fix output register sizes when multiple variables share a slot.
Kenneth Graunke [Tue, 10 Oct 2017 08:02:44 +0000 (01:02 -0700)]
i965: Fix output register sizes when multiple variables share a slot.

ARB_enhanced_layouts allows multiple output variables to share the same
location - and these variables may not have the same sizes.  For
example, consider these output variables:

   // consume X/Y/Z components of 6 vectors
   layout(location = 0) out vec3 a[6];

   // consumes W component of the first vector
   layout(location = 0, component = 3) out float b;

Looking at the first declaration, we see that VARYING_SLOT_VAR0 needs 24
components worth of space (vec3 padded out to a vec4, 4 * 6 = 24).  But
looking at the second declaration, we would think that VARYING_SLOT_VAR0
needs only 4 components of space (a single float padded out to a vec4).

nir_setup_outputs() only considered the space requirements of the first
declaration it happened to see, so if 'float b' came first, it would
underallocate the output register space, causing brw_fs_validator.cpp
to assert fail about inst->dst.offset exceeding the register size.

Fixes Piglit's tests/spec/arb_enhanced_layouts/execution/component-layout/
vs-to-fs-array-interleave-single-location.shader_test.

Thanks to Tim Arceri for finding this bug and writing a test!

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
7 years agonir: bump loop unroll limit to 96.
Dave Airlie [Tue, 10 Oct 2017 23:48:21 +0000 (09:48 +1000)]
nir: bump loop unroll limit to 96.

With the ssao demo from Vulkan demos:
radv/rx480: 440->440fps
anv/haswell: 24->34 fps

The demo does a 0->32 loop across a ubo with 32 members.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoanv: fix assert in wsi image code.
Dave Airlie [Tue, 10 Oct 2017 23:45:44 +0000 (09:45 +1000)]
anv: fix assert in wsi image code.

This assert was firing just running demos.

Jason said it should be this.

Fixes: 6c7720ed78 (anv/wsi: Allocate enough memory for the entire image)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agomesa/st: fix atomic buffer sizing to align with ssbo.
Dave Airlie [Fri, 15 Sep 2017 03:14:56 +0000 (13:14 +1000)]
mesa/st: fix atomic buffer sizing to align with ssbo.

This respects the size from the range setting like ssbo.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agomesa/bufferobj: consolidate some buffer binding code.
Dave Airlie [Fri, 15 Sep 2017 03:13:20 +0000 (13:13 +1000)]
mesa/bufferobj: consolidate some buffer binding code.

These paths are again 90% the same, consolidate them into
one.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agomesa/bufferobj: consolidate some codepaths between ubo/ssbo/atomics.
Dave Airlie [Fri, 15 Sep 2017 02:55:50 +0000 (12:55 +1000)]
mesa/bufferobj: consolidate some codepaths between ubo/ssbo/atomics.

These are 90% the same code, consolidate them into a couple of
common codepaths.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agomesa: rename various buffer bindings to one struct.
Dave Airlie [Fri, 15 Sep 2017 02:43:55 +0000 (12:43 +1000)]
mesa: rename various buffer bindings to one struct.

One binding to bind them all, these are all the same thing.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agomesa: align atomic buffer handling code with ubo/ssbo (v1.1)
Dave Airlie [Fri, 15 Sep 2017 02:38:18 +0000 (12:38 +1000)]
mesa: align atomic buffer handling code with ubo/ssbo (v1.1)

this adds automatic size support to the atomic buffer code,
but also realigns the code to act like the ubo/ssbo code.

v1.1:
add missing blank lines.
reindent one block properly.
check for NullBufferObj.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoi965: Don't try to decode types for non-existent src1.
Kenneth Graunke [Sat, 7 Oct 2017 07:14:34 +0000 (00:14 -0700)]
i965: Don't try to decode types for non-existent src1.

KHR-GL45.shader_ballot_tests.ShaderBallotBitmasks has a MOV that hits
this validation path.  MOVs don't have a src1 file, but calling
brw_inst_src1_type() was tripping on src1.file being BRW_IMMEDIATE_VALUE
and the hw_type being something invalid for immediates.

To work around this, just pretend src1 is src0 if there isn't a src1.

Fixes: 2572c2771d0cab0b9bc489d354ede44dfc88547b (i965: Validate "Special
       Requirements for Handling Double Precision Data Types")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102680
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
7 years agomain/format: skip format conversion if src and dst format are equal
Karol Herbst [Wed, 16 Aug 2017 18:32:42 +0000 (20:32 +0200)]
main/format: skip format conversion if src and dst format are equal

Fixes 'KHR-GL45.copy_image.functional' on Nouveau and i965.

v2: (by Kenneth Graunke)
    Rewrite patch according to Jason Ekstrand's review feedback.
    This makes it handle differing strides, which i965 needed.

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agomesa: Make _mesa_get_format_bytes handle array formats.
Jason Ekstrand [Sat, 7 Oct 2017 04:11:59 +0000 (21:11 -0700)]
mesa: Make _mesa_get_format_bytes handle array formats.

This is easier than making callers handle a bunch of special cases.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoradv: Only set the MTYPE flags on GFX9+.
Bas Nieuwenhuizen [Sun, 8 Oct 2017 19:58:23 +0000 (21:58 +0200)]
radv: Only set the MTYPE flags on GFX9+.

Older kernels fail the va_op with this flag set. If the kernel
supports GFX9 usefully, it will also support this flag.

Fixes: e8d57802fea "radv/gfx9: allocate events from uncached VA space"
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoi965: Disable auxiliary buffers when there are self-dependencies.
Kenneth Graunke [Fri, 6 Oct 2017 03:31:01 +0000 (20:31 -0700)]
i965: Disable auxiliary buffers when there are self-dependencies.

Jason and I investigated several OpenGL CTS failures where the tests
bind the same texture for rendering and texturing, at the same time.
This has defined results as long as the reads happen before writes,
or the regions are non-overlapping.  Normally, this just works out.

However, CCS can cause problems.  If the shader is reading one set of
pixels, and writing to different pixels that are adjacent, they may end
up being covered by the same CCS block.  So rendering may be writing a
CCS block, while the sampler is trying to read it.  Corruption ensues.

Disabling CCS is unfortunate, but safe.

Fixes several KHR-GL45.texture_barrier.* subtests.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agor600: cleanup llvm ir target selection.
Dave Airlie [Mon, 9 Oct 2017 20:27:40 +0000 (06:27 +1000)]
r600: cleanup llvm ir target selection.

Only r600 target used now for compute IR.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agor600: drop tc_L2_dirty bit, this was SI only.
Dave Airlie [Mon, 9 Oct 2017 20:27:10 +0000 (06:27 +1000)]
r600: drop tc_L2_dirty bit, this was SI only.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradeonsi: lower ffma in nir to mad.
Dave Airlie [Tue, 10 Oct 2017 03:23:40 +0000 (13:23 +1000)]
radeonsi: lower ffma in nir to mad.

This lowers ffma to a * b + c.

This seems like it should keep Marek happiest, so
we'd never get to the fma instruction emission code.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: lower ffma in nir.
Dave Airlie [Tue, 3 Oct 2017 20:33:02 +0000 (06:33 +1000)]
radv: lower ffma in nir.

So it appears the Vulkan SPIR-V fma opcode can be equivalent to a
mad operation, and the fma hw opcode on AMD hw is issued like a double
opcode so is slower. Also the radeonsi stack does this.

This appears to improve performance on a number of games from Feral,
and thanks to Feral for noticing the problem.

I'm reposting this one as Marek indicated he thinks this is what
we should be doing on AMD hw.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Add R16G16B16A16_SNORM fast clear support
Alex Smith [Tue, 10 Oct 2017 16:00:09 +0000 (17:00 +0100)]
radv: Add R16G16B16A16_SNORM fast clear support

Signed-off-by: Alex Smith <asmith@feralinteractive.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agobroadcom/vc5: Fix handling of 5551 textures using the new gallium format.
Eric Anholt [Tue, 10 Oct 2017 18:19:23 +0000 (11:19 -0700)]
broadcom/vc5: Fix handling of 5551 textures using the new gallium format.

Like vc4, we have the alpha in the low bit.  Fixes a bunch of piglit
texwrap failures.

7 years agobroadcom/vc5: Set the RCL's MSAA mode to match the BCL's MSAA state.
Eric Anholt [Fri, 6 Oct 2017 00:18:34 +0000 (17:18 -0700)]
broadcom/vc5: Set the RCL's MSAA mode to match the BCL's MSAA state.

7 years agobraodcom/vc5: Set up clear color for higher-bpp formats.
Eric Anholt [Thu, 5 Oct 2017 22:50:59 +0000 (15:50 -0700)]
braodcom/vc5: Set up clear color for higher-bpp formats.

Fixes arb_color_buffer_float-clear

7 years agobroadcom/vc5: Set up per-MRT clear colors.
Eric Anholt [Thu, 5 Oct 2017 22:40:18 +0000 (15:40 -0700)]
broadcom/vc5: Set up per-MRT clear colors.

Fixes fbo-mrt-alphatest.

7 years agobroadcom/vc5: Fix blendfactor zero handling.
Eric Anholt [Thu, 5 Oct 2017 22:19:49 +0000 (15:19 -0700)]
broadcom/vc5: Fix blendfactor zero handling.

I cut the line out to move it up to the top, when putting "0" in the
switch made the compiler complain that that wasn't a valid enum.

7 years agobroadcom/vc5: Fix Rendering Mode Common Config's color store bitmask.
Eric Anholt [Wed, 4 Oct 2017 23:42:55 +0000 (16:42 -0700)]
broadcom/vc5: Fix Rendering Mode Common Config's color store bitmask.

This controls the RTs that get stored by the default resolved store, the
same way that the extended resolved store packet has a RT bitmask.

7 years agobroadcom/vc5: Add support for f32 render targets.
Eric Anholt [Tue, 3 Oct 2017 00:21:23 +0000 (17:21 -0700)]
broadcom/vc5: Add support for f32 render targets.

The TLB write code is getting ugly and needs a refactoring (that will
hopefully handle TLBU uniform coalescing as well).

7 years agobroadcom/vc5: Fix color masks for non-independent blending.
Eric Anholt [Tue, 3 Oct 2017 00:05:24 +0000 (17:05 -0700)]
broadcom/vc5: Fix color masks for non-independent blending.

This gets fbo-mrt-alphatest working except for the second RT's clear color.

7 years agobroadcom/vc5: Make the BCL's number of render targets setup match the RCL.
Eric Anholt [Mon, 2 Oct 2017 23:55:04 +0000 (16:55 -0700)]
broadcom/vc5: Make the BCL's number of render targets setup match the RCL.

7 years agobraodcom/vc5: Fix tile size setup for MRTs.
Eric Anholt [Mon, 2 Oct 2017 23:54:09 +0000 (16:54 -0700)]
braodcom/vc5: Fix tile size setup for MRTs.

We need to divide the TLB in two for the 2nd color buffer, and again if
the 3rd or 4th are present.

7 years agobroadcom/vc5: Start hooking up multiple render targets support.
Eric Anholt [Mon, 2 Oct 2017 23:43:33 +0000 (16:43 -0700)]
broadcom/vc5: Start hooking up multiple render targets support.

We now emit as many TLB color writes as there are color buffers.

7 years agobroadcom/vc5: Add support for GL_EXT_provoking_vertex.
Eric Anholt [Mon, 2 Oct 2017 21:09:56 +0000 (14:09 -0700)]
broadcom/vc5: Add support for GL_EXT_provoking_vertex.

The bit was missing from the spec, but it's there in the simulator.  Fixes
the piglit clipflat test.

7 years agobraodcom/vc5: Find the actual first TF output for our TF spec.
Eric Anholt [Mon, 2 Oct 2017 19:20:35 +0000 (12:20 -0700)]
braodcom/vc5: Find the actual first TF output for our TF spec.

This doesn't yet support PSIZ, but gets us at least some of TF working.

7 years agobroadcom/vc5: Fix translation of transform feedback's output_register field.
Eric Anholt [Mon, 2 Oct 2017 20:02:32 +0000 (13:02 -0700)]
broadcom/vc5: Fix translation of transform feedback's output_register field.

It's a NIR driver_location, not a slot offset.

7 years agobroadcom/vc5: Mark our primitives as needing TF processing.
Eric Anholt [Mon, 2 Oct 2017 19:17:30 +0000 (12:17 -0700)]
broadcom/vc5: Mark our primitives as needing TF processing.

The TF enable state appears to stick around until the next TF enable
packet is sent, so we only want to request TF when the shader is using it.

7 years agobroadcom/vc5: Fix setup of TF dword output count.
Eric Anholt [Mon, 2 Oct 2017 19:05:30 +0000 (12:05 -0700)]
broadcom/vc5: Fix setup of TF dword output count.

I missed the "- 1" when reading the spec.

7 years agobroadcom/vc5: Fix up a comment from vc4 about the predraw texture setup.
Eric Anholt [Mon, 2 Oct 2017 18:41:57 +0000 (11:41 -0700)]
broadcom/vc5: Fix up a comment from vc4 about the predraw texture setup.

7 years agobroadcom/vc5: Flush the job when mapping a transform feedback buffer.
Eric Anholt [Sat, 30 Sep 2017 23:48:44 +0000 (16:48 -0700)]
broadcom/vc5: Flush the job when mapping a transform feedback buffer.

We will want something fancier for reusing a TF output within the same
frame, but we at least need this in order for piglit tests to work.

7 years agobroadcom/vc5: Fix handling of interp qualifiers on builtin color inputs.
Eric Anholt [Thu, 28 Sep 2017 21:02:05 +0000 (14:02 -0700)]
broadcom/vc5: Fix handling of interp qualifiers on builtin color inputs.

The interpolation qualifier, if specified, is supposed to take precedence
over glShadeModel().

7 years agobroadcom/vc5: Fix CLIF dumping of lists that aren't capped by a HALT.
Eric Anholt [Thu, 28 Sep 2017 20:36:54 +0000 (13:36 -0700)]
broadcom/vc5: Fix CLIF dumping of lists that aren't capped by a HALT.

The HW will halt when you hit a HALT packet, or when you hit the end
address.  Tell CLIF if there's an end address is so that it can stop
correctly.  (There was usually a 0 byte after the CL, so it would stop
anyway).

7 years agobroadcom/vc5: Fix depth and stencil clear values.
Eric Anholt [Thu, 28 Sep 2017 18:41:31 +0000 (11:41 -0700)]
broadcom/vc5: Fix depth and stencil clear values.

I had misread the packet description: We always have a 32f depth, and a
separate u8 stencil.

7 years agobroadcom/vc5: Add missing Z16 format.
Eric Anholt [Thu, 28 Sep 2017 18:47:50 +0000 (11:47 -0700)]
broadcom/vc5: Add missing Z16 format.

We can render to and sample from it just fine.

7 years agobraodcom/vc5: Fix incorrect early Z writes in discard shaders.
Eric Anholt [Thu, 28 Sep 2017 17:39:07 +0000 (10:39 -0700)]
braodcom/vc5: Fix incorrect early Z writes in discard shaders.

Fixes glsl-fs-discard-02.

7 years agobroadcom/compiler: Set up passthrough Z when doing FS discards.
Eric Anholt [Thu, 28 Sep 2017 17:37:02 +0000 (10:37 -0700)]
broadcom/compiler: Set up passthrough Z when doing FS discards.

In order to keep early-Z from writing early in a discard shader, you need
to set the "modifies Z" bit in the shader state (which the new
prog_data.discards will indicate).  Then, in the shader we do a TLB write
to make Z passthrough happen (the QPU result is ignored, so we use a NULL
source).

7 years agobroadcom/compiler: Don't forget the discard state on TLB Z writes.
Eric Anholt [Thu, 28 Sep 2017 18:06:53 +0000 (11:06 -0700)]
broadcom/compiler: Don't forget the discard state on TLB Z writes.

We don't want to write Z for discarded fragments.

7 years agobroadcom/compiler: Use defines instead of magic values in TLB write setup.
Eric Anholt [Thu, 28 Sep 2017 17:56:47 +0000 (10:56 -0700)]
broadcom/compiler: Use defines instead of magic values in TLB write setup.

7 years agobroadcom/vc5: Add proper support for base_vertex and base_instance.
Eric Anholt [Wed, 27 Sep 2017 22:06:09 +0000 (15:06 -0700)]
broadcom/vc5: Add proper support for base_vertex and base_instance.

I had base_vertex hacked into the shader state setup like in vc4, but it's
not correct for big offsets.  Using the proper packet is easier and
hopefully means we can re-emit shader state setup less frequently.

7 years agobroadcom/xml: Add the vc5 Base Vertex/Base Instance packet.
Eric Anholt [Wed, 27 Sep 2017 22:05:14 +0000 (15:05 -0700)]
broadcom/xml: Add the vc5 Base Vertex/Base Instance packet.

This lets us do index_bias and ARB_base_instance.

7 years agobroadcom/vc5: Use supertiles and generic tile lists.
Eric Anholt [Wed, 27 Sep 2017 22:27:31 +0000 (15:27 -0700)]
broadcom/vc5: Use supertiles and generic tile lists.

This massively reduces the size of our RCL setup.  It also gets us closer
to supporting multicore platforms.

7 years agobroadcom/xml: Add a bunch more vc5 tile list management packets.
Eric Anholt [Wed, 27 Sep 2017 21:51:05 +0000 (14:51 -0700)]
broadcom/xml: Add a bunch more vc5 tile list management packets.

We're going to need these for MSAA, and to use the generic per-tile list.

7 years agobroadcom/xml: Remove vc5 base packet for tile bin/render mode config.
Eric Anholt [Wed, 27 Sep 2017 23:38:28 +0000 (16:38 -0700)]
broadcom/xml: Remove vc5 base packet for tile bin/render mode config.

These existed so I could unpack just the sub-id field to switch on in the
old manual CLIF dumper.  The new codegen handles sub-id automatically, but
only if these stub packets aren't there with an implicit sub-id=0.

7 years agobraodcom/xml: Fix a pasteo in vc5 store tile buffer general.
Eric Anholt [Wed, 27 Sep 2017 21:54:05 +0000 (14:54 -0700)]
braodcom/xml: Fix a pasteo in vc5 store tile buffer general.

7 years agobroadcom: Add V3D 3.3 gallium driver called "vc5", for BCM7268.
Eric Anholt [Fri, 3 Feb 2017 00:24:13 +0000 (16:24 -0800)]
broadcom: Add V3D 3.3 gallium driver called "vc5", for BCM7268.

V3D 3.3 is a continuation of the 3D implementation in VC4 (v2.1 and v2.6).
V3D 3.3 introduces an MMU (no more CMA allocations) and support for
GLES3.1.  This driver is not currently conformant, though that will be a
target as soon as possible.

V3D 3.x parts use a new texture tiling layout common across many Broadcom
graphics parts including and the HVS scanout engine.  It also massively
changes the QPU instructions, introducing a common physical register file
(no more A/B split) and half-float instructions, while removing the 4x8
unorm instructions in favor of half-float for talking to fixed function
interfaces.  Because so much has changed, vc5 is implemented in a separate
gallium driver, using only the XML code-generation support from vc4.

v2: Fix tile layout for 64bpp textures.  Fix texture swizzling for 32-bit
    returns.  Fix up a bit of MRT setup.  Sync the simulator to kernel
    behavior a bit more.  Improve uniform debugging code.  Rebase on
    QIR->VIR rename.  Move texture state mostly to the CSOs.  Improve
    cache flushing on the simulator.  Fix program deletion
    use-after-frees.

Acked-by: Dave Airlie <airlied@gmail.com> (uabi plan)
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> (uabi plan)
7 years agobroadcom: Add VC5 NIR compiler.
Eric Anholt [Fri, 3 Feb 2017 18:24:14 +0000 (10:24 -0800)]
broadcom: Add VC5 NIR compiler.

This is a pretty straightforward fork of VC4's NIR compiler to VC5.  The
condition codes, registers, and I/O have all changed, making the backend
hard to share, though their heritage is still recognizable.

v2: Move to src/broadcom/compiler to match intel's layout, rename more
    "vc5" to "v3d", rename QIR to VIR ("V3D IR") to avoid symbol conflicts
    with vc4, use new v3d_debug header, add compiler init/free functions,
    do texture swizzling in NIR to allow optimization.

7 years agobroadcom: Add vc5 CLIF dumping
Eric Anholt [Fri, 3 Feb 2017 00:25:12 +0000 (16:25 -0800)]
broadcom: Add vc5 CLIF dumping

This will be usable with "VC5_DEBUG=cl" on the vc5 driver to stream a CLIF
file (the Broadcom equivalent of i965's AUB) to stderr.  I haven't tested
that this is actually usable with the internal CLIF-consuming tools, but
is close enough as a baseline and is useful for visually inspecting the
command stream.

7 years agobroadcom: Add V3D 3.3 QPU instruction pack, unpack, and disasm.
Eric Anholt [Fri, 3 Feb 2017 00:15:18 +0000 (16:15 -0800)]
broadcom: Add V3D 3.3 QPU instruction pack, unpack, and disasm.

Unlike VC4, I've defined an unpacked instruction format with pack/unpack
functions to convert to 64-bit encoded instructions.  This will let us
incrementally put together our instructions and validate them in a more
natural way than the QPU_GET_FIELD/QPU_SET_FIELD used to.

The pack/unpack unfortuantely are written by hand.  While I could define
genxml for parts of it, there are many special cases (like operand order
of commutative binops choosing which binop is being performed!) and it
probably wouldn't come out much cleaner.

The disasm unit test ensures that we have the same assembly format as
Broadcom's internal tools, other than whitespace changes.

v2: Fix automake variable redefinition complaints, add test to .gitignore

7 years agobroadcom: Introduce a v3d_debug.h header for vc5 and broadcom Vulkan.
Eric Anholt [Fri, 1 Sep 2017 21:38:54 +0000 (14:38 -0700)]
broadcom: Introduce a v3d_debug.h header for vc5 and broadcom Vulkan.

Unlike vc4, where the compiler and gallium driver live together, for vc5
the compiler will live up in the shared broadcom directory, and need
access to the debug flags.  Define a set of debug flags and helpers there,
so it can be shared between compiler, vc5, and vulkan.

7 years agoconfigure: Add the new "vc5" driver to the list, requiring a simulator.
Eric Anholt [Fri, 18 Aug 2017 19:55:43 +0000 (12:55 -0700)]
configure: Add the new "vc5" driver to the list, requiring a simulator.

My intent is to develop the vc5 driver in-tree for some time to build the
CL generation and shader compiler code, and keep out-of-tree patches for
talking to an actual kernel driver until the kernel driver can be
stabilized on the hardware.

v2: Define a HAVE_BROADCOM_DRIVERS, like HAVE_INTEL or HAVE_AMD.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agonir: Move vc4's alpha test lowering to core NIR.
Eric Anholt [Thu, 23 Feb 2017 00:53:18 +0000 (16:53 -0800)]
nir: Move vc4's alpha test lowering to core NIR.

I've been doing this inside of vc4, but vc5 wants it as well and it may be
useful for other drivers (Intel has a related path for pre-gen6 with MRT,
and freedreno had a TGSI path for it at one point).

This required defining a common enum for the standard comparison
functions, but other lowering passes are likely to also want that enum.

v2: Add to meson.build as well.

Acked-by: Rob Clark <robdclark@gmail.com>
7 years agomesa: Alphabetize GL_MESA_tile_raster_order in the extensions list.
Eric Anholt [Tue, 10 Oct 2017 18:41:29 +0000 (11:41 -0700)]
mesa: Alphabetize GL_MESA_tile_raster_order in the extensions list.

trivial, fixes make check.

7 years agomesa: Implement a new GL_MESA_tile_raster_order extension.
Eric Anholt [Tue, 25 Jul 2017 23:13:57 +0000 (16:13 -0700)]
mesa: Implement a new GL_MESA_tile_raster_order extension.

The intent is to use this extension on vc4 to allow X11 to do overlapping
CopyArea() within a pixmap without first blitting the pixmap to a
temporary.  With associated glamor patches, improves x11perf
-copywinwin100 performance on a Raspberry Pi 3 from ~4700/sec to
~5130/sec, and is an even larger boost to uncomposited window movement
performance (most copywinwin100 copies don't overlap).

v2: Fix glIsEnabled() on the new enums.
v3: Drop the local spec since I'm upstreaming the spec.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agobroadcom/vc4: Expose PIPE_CAP_TILE_RASTER_ORDER
Eric Anholt [Thu, 27 Jul 2017 19:05:56 +0000 (12:05 -0700)]
broadcom/vc4: Expose PIPE_CAP_TILE_RASTER_ORDER

Because vc4 can control the order that tiles are rasterized in, we can use
it to implement overlapping blits using normal drawing and
GL_ARB_texture_barrier, as long as we can tell the kernel what order to
render the tiles in.

v2: Fix on the simulator.
v3: Add the cap (disabled) to other drivers, add rst docs for the cap.
v4: Rebase on PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS
v5: Split from the core gallium commit, drop some unnecessary code related
    to glBlitFramebuffer(), fix a crash with clears before state has been
    bound.

7 years agogallium: Create a new PIPE_CAP_TILE_RASTER_ORDER for vc4.
Eric Anholt [Thu, 27 Jul 2017 19:05:56 +0000 (12:05 -0700)]
gallium: Create a new PIPE_CAP_TILE_RASTER_ORDER for vc4.

Because vc4 can control the order that tiles are rasterized in, we can use
it to implement overlapping blits using normal drawing and
GL_ARB_texture_barrier, as long as we can tell the kernel what order to
render the tiles in.

This commit introduces the core gallium support, vc4 changes will follow.

v2: Fix on the simulator.
v3: Add the cap (disabled) to other drivers, add rst docs for the cap.
v4: Rebase on PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS
v5: Drop vc4 changes from this commit, for clarity.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v3)
7 years agobroadcom/vc4: Implement GL_ARB_texture_barrier.
Eric Anholt [Thu, 5 Oct 2017 18:08:23 +0000 (11:08 -0700)]
broadcom/vc4: Implement GL_ARB_texture_barrier.

Improves x11perf -copywinwin100 from ~2000/sec to ~4700/sec.  More
importantly, this is a prerequisite for the new GL_MESA_tile_raster_order
extension.

7 years agodocs: Update the list of used MESA GL enums.
Eric Anholt [Wed, 26 Jul 2017 23:02:07 +0000 (16:02 -0700)]
docs: Update the list of used MESA GL enums.

Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agodocs: Fix a typo in the old MESA_program_debug spec.
Eric Anholt [Wed, 26 Jul 2017 23:00:50 +0000 (16:00 -0700)]
docs: Fix a typo in the old MESA_program_debug spec.

Noticed that we had two 0x8bb4 in the spec while grepping to find an open
slot in the MESA enums set.  gl.xml had the right value.

Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agogit_sha1_gen: accept MESA_GIT_SHA1_OVERRIDE env var
Brian Paul [Mon, 9 Oct 2017 19:47:20 +0000 (13:47 -0600)]
git_sha1_gen: accept MESA_GIT_SHA1_OVERRIDE env var

If one uses a parent build script to download/build Mesa we may not
have a full git repository (maybe a tar archive) so the 'git rev-parse'
command will fail.

This updates the script to look for a MESA_GIT_SHA1_OVERRIDE env var.
If it's set, use that sha1 instead of using git rev-parse.  With this
change we can put a git hash in the GL_VERSION string even when we
don't have a git repo.

v2: incorporate Dylan's suggestions to simplify the code

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
7 years agomesa: move _mesa_half_is_negative() to half_float.h
Brian Paul [Thu, 5 Oct 2017 19:34:52 +0000 (13:34 -0600)]
mesa: move _mesa_half_is_negative() to half_float.h

v2: use !! in the function to be explicit about type conversion.  Though,
gcc generates the same code with or without the logical !!.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agomesa: move _mesa_exec_malloc/free() prototypes to their own header
Brian Paul [Thu, 5 Oct 2017 19:30:32 +0000 (13:30 -0600)]
mesa: move _mesa_exec_malloc/free() prototypes to their own header

Try to start removing things from the cluttered imports.h file.

v2: add new header to Makefile.sources

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
7 years agoi965: minor whitespace fix
Kenneth Graunke [Tue, 10 Oct 2017 17:18:04 +0000 (10:18 -0700)]
i965: minor whitespace fix

7 years agomesa: Set new renderbuffers to RGBA4 on all GLES contexts.
Eric Anholt [Mon, 1 May 2017 17:21:36 +0000 (10:21 -0700)]
mesa: Set new renderbuffers to RGBA4 on all GLES contexts.

Before we were doing RGBA4 on GLES3 only, but as of GLES2 2.0.22 it should
be RGBA4 as well.  Fixes DEQP
functional.state_query.rbo.renderbuffer_internal_format.

Tested-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agomesa: Expose GL_OES_required_internalformat on GLES contexts.
Eric Anholt [Mon, 1 May 2017 16:28:33 +0000 (09:28 -0700)]
mesa: Expose GL_OES_required_internalformat on GLES contexts.

This extension is effectively a backport of GLES3's internalformat
handling to GLES 1/2.  It guarantees that sized internalformats specified
for textures and renderbuffers have at least the specified size stored.
That's a pretty minimal requirement, so I think it can be dummy_true and
exposed as a standard in Mesa.

As a side effect, it also allows GL_RGB565 to be specified as a texture
format, not just as a renderbuffer.  Mesa had previously been allowing 565
textures, which angered DEQP in the absence of this extension being
exposed.

v2: Allow 2101010rev with sized internalformats even on GLES3, citing the
    extension spec.  Extend extension checks for GLES2 contexts exposing
    with texture_float, texture_half_float, and texture_rg.
v3: Fix ALPHA/LUMINANCE/LUMINANCE_ALPHA error checking (GLES3 CTS
    failures)
v4: Mark GL_RGB10 non-color-renderable on ES, fix A/L/LA errors on GLES2
    with float formats.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agomesa: Only expose GLES's EXT_texture_type_2_10_10_10_REV if supported in HW.
Eric Anholt [Mon, 1 May 2017 16:51:01 +0000 (09:51 -0700)]
mesa: Only expose GLES's EXT_texture_type_2_10_10_10_REV if supported in HW.

Previously, we were downconverting to 8888 automatically if the hardware
didn't suport it.  However, with the advent of
GL_OES_required_internalformat, we have to actually store the
internalformats we advertise support for.  And, it seems rather
disingenuous to advertise the extension if we don't actually support it.

v2: Throw an error when using the format on ES2 without the extension present.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agovc4: Add support for 5551 textures.
Eric Anholt [Mon, 1 May 2017 18:25:06 +0000 (11:25 -0700)]
vc4: Add support for 5551 textures.

This keeps us from promoting them up to 8888, at the cost of not being
color-renderable.

7 years agogallium: Add support for 5551 with the 1-bit field in the low bit.
Eric Anholt [Mon, 1 May 2017 18:16:20 +0000 (11:16 -0700)]
gallium: Add support for 5551 with the 1-bit field in the low bit.

This is how VC4 stores 5551 textures, which we need to support for
GL_OES_required_internalformat.

v2: Extend commit message, fix svga driver build, add BE ordering from
    Roland.
v3: Rebase on PIPE_FORMAT_R10G10B10X2_UNORM addition.

Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v2)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v2)
7 years agomesa: Add X1B5G5R5 along with A1B5G5R5.
Eric Anholt [Mon, 1 May 2017 18:21:27 +0000 (11:21 -0700)]
mesa: Add X1B5G5R5 along with A1B5G5R5.

For supporting RGB5 in hardware with A in the low bit (vc4), we need this
format as well.

v2: Add proper _mesa_format_matches_format_and_type() support (from
    Nicolai).

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)
7 years agost_api: remove unused get_resource_for_egl_image
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:48 +0000 (13:58 +0200)]
st_api: remove unused get_resource_for_egl_image

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agost/dri: implement createImageFromRenderbuffer(2)
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:48 +0000 (13:58 +0200)]
st/dri: implement createImageFromRenderbuffer(2)

Tested with dEQP-EGL.functional.image.*renderbuffer* tests.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoegl/dri: remove old left-overs
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:47 +0000 (13:58 +0200)]
egl/dri: remove old left-overs

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agoegl/dri: use createImageFromRenderbuffer2 when available
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:47 +0000 (13:58 +0200)]
egl/dri: use createImageFromRenderbuffer2 when available

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agoegl/dri: factor out egl_error_from_dri_image_error
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:47 +0000 (13:58 +0200)]
egl/dri: factor out egl_error_from_dri_image_error

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agodri_interface: add an error-returning version of createImageFromRenderbuffer
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:46 +0000 (13:58 +0200)]
dri_interface: add an error-returning version of createImageFromRenderbuffer

We ought to be able to distinguish between allocation errors and bad
parameters (non-existent renderbuffer object).

Bumps the version of the DRI Image extension to 17.

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
7 years agost/mesa: don't clobber glGetInternalformat* buffer for GL_NUM_SAMPLE_COUNTS
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:46 +0000 (13:58 +0200)]
st/mesa: don't clobber glGetInternalformat* buffer for GL_NUM_SAMPLE_COUNTS

Applications might pass in a buffer that is sized too large and rely
on the extra space of the buffer not being overwritten.

Fixes dEQP-GLES31.functional.state_query.internal_format.partial_query.num_sample_counts

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agou_threaded_context: fix a memory leak
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:46 +0000 (13:58 +0200)]
u_threaded_context: fix a memory leak

The uploaders can own transfers which need to be unmapped. Destroy them
before the final sync (they're not used from the driver thread anyway)
so that the transfer_unmap call is processed by the driver.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agodisk_cache: remove unnecessary NULL-pointer guards
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:45 +0000 (13:58 +0200)]
disk_cache: remove unnecessary NULL-pointer guards

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agodisk_cache: fix a memory leak
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:45 +0000 (13:58 +0200)]
disk_cache: fix a memory leak

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agost/mesa: whitespace fix
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:44 +0000 (13:58 +0200)]
st/mesa: whitespace fix

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agost/mesa: fix import of EGL images with non-zero level or layer
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:44 +0000 (13:58 +0200)]
st/mesa: fix import of EGL images with non-zero level or layer

In GL state, textures created from EGL images look like plain 2D textures
with a single level, so we use the existing layer_override facility and
add an analogous level_override one.

Fixes dEQP-EGL.functional.image.create.gles2_cubemap_{positive,negative}_{x,y,z}_rgba_texture

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agost/mesa: fix switching from surface-based to non-surface-based textures
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:44 +0000 (13:58 +0200)]
st/mesa: fix switching from surface-based to non-surface-based textures

This can happen with surface-based texture objects derived from EGL
images, since those aren't immutable.

Fixes tests in dEQP-EGL.functional.sharing.gles2.multithread.random.images.teximage2d.* and others

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agoglsl/linker: add check for compute shared memory size
Nicolai Hähnle [Tue, 10 Oct 2017 11:58:43 +0000 (13:58 +0200)]
glsl/linker: add check for compute shared memory size

Unlike uniforms, the limit on shared memory size is not called out
explicitly in the list of things that cause linker errors, but presumably
that's just an oversight in the spec.

Fixes dEQP-GLES31.functional.debug.negative_coverage.{callbacks,get_error,log}.compute.exceed_shared_memory_size_limit

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
7 years agoetnaviv: update HW headers and fix provoking vertex
Lucas Stach [Fri, 29 Sep 2017 12:17:41 +0000 (14:17 +0200)]
etnaviv: update HW headers and fix provoking vertex

Now that the real meaning of the 2 bits in PA_SYSTEM_MODE is known,
we can set them according to the rasterizer state, which fixes uses
that are setting provoking vertex first.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
7 years agoetnaviv: remove flat shading workaround
Lucas Stach [Thu, 8 Jun 2017 15:06:02 +0000 (17:06 +0200)]
etnaviv: remove flat shading workaround

It turned out not to be a hardware bug, but the shader compiler
emitting wrong varying component use information. With that fixed
we can turn flat shading back on.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Wladimir J. van der Laan <laanwj@gmail.com>
7 years agoetnaviv: fix varying interpolation
Lucas Stach [Thu, 8 Jun 2017 15:01:40 +0000 (17:01 +0200)]
etnaviv: fix varying interpolation

It seems that newer cores don't use the PA_ATTRIBUTES to decide if the
varying should bypass the flat shading, but derive this from the component
use. This fixes flat shading on GC880+.

VARYING_COMPONENT_USE_POINTCOORD is a bit of a misnomer now, as it isn't
only used for pointcoords, but missing a better name I left it as-is.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Wladimir J. van der Laan <laanwj@gmail.com>
7 years agoetnaviv: fix bogus flush requests in transfer handling
Lucas Stach [Tue, 26 Sep 2017 16:27:55 +0000 (18:27 +0200)]
etnaviv: fix bogus flush requests in transfer handling

The logic to decide if we need to flush the GPU command stream was broken
and hard to reason about. Fix and clarify this.

Fixes the data sync subtests from piglit arb_vertex_buffer_object.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Wladimir J. van der Laan <laanwj@gmail.com>
7 years agoi965/tes: account for the fact that dvec3/4 inputs take two slots
Iago Toral Quiroga [Mon, 9 Oct 2017 12:17:43 +0000 (14:17 +0200)]
i965/tes: account for the fact that dvec3/4 inputs take two slots

When computing the total size of the URB for tessellation evaluation
inputs we were not accounting for this, and instead we were always
assuming that each input would take a single vec4 slot, which could
lead to computing a smaller read size than required. Specifically, this
is a problem when the last input is a dvec3/4 such that its XY components
are stored in the the second half of a payload register (which can happen
if the offset for the input in the URB is not 64-bit aligned because
there are 32-bit inputs mixed in) and the ZW components in the
first half of the next, as in this case we would fail to account for the
extra slot required for the ZW components.

Fixes (requires another fix in CTS currently in review):
KHR-GL45.enhanced_layouts.varying_locations
KHR-GL45.enhanced_layouts.varying_array_locations

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