mesa.git
7 years agonv50/ir: use sched control codes for gm107 builtins
Samuel Pitoiset [Mon, 19 Dec 2016 23:11:33 +0000 (00:11 +0100)]
nv50/ir: use sched control codes for gm107 builtins

Yes, IMUL/IMAD require dependency barriers and we should
definitely replace these instructions by XMAD but the
different flags need to be figured out. Note that XMAD only
supports 16-bits integers.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
7 years agonv50/ir: improve instruction pipelining on gm107
Samuel Pitoiset [Fri, 16 Dec 2016 13:28:14 +0000 (14:28 +0100)]
nv50/ir: improve instruction pipelining on gm107

This makes use of scheduling control codes which are very useful
for improving the instruction pipelining.

This patch will increase performance on Maxwell GPUs by, at least,
x1.5 up to x3.5 for some benchmarks.

Although this has been fairly well tested, I would not be suprised
if someone hit a corner case somewhere. That way, the scheduler
is enabled by default but it can be deactivated by using
NV50_PROG_SCHED=0.

Thanks to Scott Gray for the reverse engineering work available from
https://github.com/NervanaSystems/maxas/wiki/Control-Codes.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Pierre Moreau <pierre.morrow@free.fr>
Tested-by: Alexandre Courbot <acourbot@nvidia.com>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agonv50/ir: do not insert texture barriers on gm107
Samuel Pitoiset [Fri, 16 Dec 2016 13:22:16 +0000 (14:22 +0100)]
nv50/ir: do not insert texture barriers on gm107

It's actually useless to insert those texture barriers post RA
because the current control code (ie. st 0x0) will wait for all
dependencies before issuing a new instruction.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
7 years agoi965/gen7: expose OpenGL 4.2 on Haswell when supported
Juan A. Suarez Romero [Tue, 11 Oct 2016 15:05:36 +0000 (15:05 +0000)]
i965/gen7: expose OpenGL 4.2 on Haswell when supported

GL_ARB_vertex_attrib_64bit was the last piece missing.

v2: update docs (Jordan)

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965: enable ARB_shader_precision to HSW+
Samuel Iglesias Gonsálvez [Mon, 17 Oct 2016 11:59:18 +0000 (13:59 +0200)]
i965: enable ARB_shader_precision to HSW+

v2: update docs (Jordan)

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965: unify the code to enable of ARB_gpu_shader_fp64 and ARB_vertex_attrib_64bit...
Samuel Iglesias Gonsálvez [Mon, 17 Oct 2016 11:58:27 +0000 (13:58 +0200)]
i965: unify the code to enable of ARB_gpu_shader_fp64 and ARB_vertex_attrib_64bit for HSW+

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965: Enable ARB_vertex_attrib_64bit for Haswell
Alejandro Piñeiro [Thu, 7 Jul 2016 09:49:33 +0000 (11:49 +0200)]
i965: Enable ARB_vertex_attrib_64bit for Haswell

v2: update docs (Jordan)

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965: check for dual slot attributes on any gen
Juan A. Suarez Romero [Wed, 14 Dec 2016 17:12:24 +0000 (17:12 +0000)]
i965: check for dual slot attributes on any gen

Those not supporting 64 bit input vertex attributes will have the
dual_slot value as false.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/vec4: emit correctly load_inputs for 64bit data
Juan A. Suarez Romero [Wed, 6 Jul 2016 10:40:49 +0000 (12:40 +0200)]
i965/vec4: emit correctly load_inputs for 64bit data

For dvec3 and dvec4 types, a single GRF do not have enough space to
allocate two inputs from two different vertices (SIMD4x2).

So the GRF only contains first two components for the two vertices, and
the next GRF has the remaining components.

We want to put all the components for the same vertex in the same
register. Thus, we do a shuffle to reorder the data.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/vec4: take into account doubles when creating attribute mapping
Alejandro Piñeiro [Thu, 21 Apr 2016 12:16:18 +0000 (14:16 +0200)]
i965/vec4: take into account doubles when creating attribute mapping

Doubles needs more that one slot per attribute. So when filling the
attribute_map we check if it is a double in order to allocate one
extra register.

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/vec4/nir: vec4 also needs to remap vs attributes
Alejandro Piñeiro [Fri, 22 Apr 2016 08:49:24 +0000 (10:49 +0200)]
i965/vec4/nir: vec4 also needs to remap vs attributes

Doubles need extra space, so we would need to do a remapping for vec4
too in order to take that into account. We reuse the already
existing remap_vs_attrs, but passing is_scalar, so they could
remap accordingly.

v2: code-format remap_vs_attrs_params initialization (Matt)

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965/vec4: use attribute slots for first non payload GRF
Alejandro Piñeiro [Thu, 21 Apr 2016 10:23:47 +0000 (12:23 +0200)]
i965/vec4: use attribute slots for first non payload GRF

As part of the payload setup, setup_attributes is called with the first
GRF that can be used for the attributes (first ones are used for
uniforms for example) and returns the first GRF that is not part of the
payload. Before this patch, it adds directly the number of attributes.
But as with 64-bit attributes can consume more than one slot, that is
not valid anymore. This patch change the addition to use the number of
slots consumed.

gen >= 8 would not be affected, as they use the scalar mode. For that
case, the vs configuration is done at fs_visitor::assign_vs_urb_setup.

v2: add explanation in commit log (Jordan)

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965: downsize *64*PASSTHRU formats to equivalent *32*FLOAT formats on gen < 8
Alejandro Piñeiro [Mon, 11 Apr 2016 08:05:24 +0000 (10:05 +0200)]
i965: downsize *64*PASSTHRU formats to equivalent *32*FLOAT formats on gen < 8

gen < 8 doesn't support *64*PASSTHRU formats when emitting
vertices. So in order to provide the equivalent functionality, we need
to downsize the format to equivalent *32*FLOAT, and in some cases
(R64G64B64 and R64G64B64A64) submit two 3DSTATE_VERTEX_ELEMENTS for
each vertex element.

Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agoi965: return PASSTHRU surface types also on gen7
Alejandro Piñeiro [Fri, 15 Apr 2016 17:08:04 +0000 (19:08 +0200)]
i965: return PASSTHRU surface types also on gen7

Although gen7 doesn't include surface types as a valid conversion format,
we return it, as it reflects what we want to achieve, even if we need
to workaround it on gen < 8.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agomain/buffers: take into account FRONT_AND_BACK on ReadBuffer
Alejandro Piñeiro [Wed, 11 Jan 2017 20:11:55 +0000 (18:11 -0200)]
main/buffers: take into account FRONT_AND_BACK on ReadBuffer

From OpenGL 3.1 spec, section 4.3.1 "Reading Pixels", page 190 (203 PDF)

  "When READ FRAMEBUFFER BINDING is zero, i.e. the default
   framebuffer, src must be one of the values listed in table 4.4,
   including NONE . FRONT_AND_BACK , FRONT , and LEFT refer to the
   front left buffer."

There is an equivalent text on OpenGL 4.5 spec, section 18.2.1
"Selecting Buffers for Reading", page 502 (524 PDF), so the behaviour
is still the same.

Part of the fix for:
GL45-CTS.direct_state_access.framebuffers_draw_read_buffers_errors

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
7 years agomain/buffers: update error handling on DrawBuffers for 4.5
Alejandro Piñeiro [Tue, 13 Dec 2016 10:58:59 +0000 (08:58 -0200)]
main/buffers: update error handling on DrawBuffers for 4.5

Before 4.5, GL_BACK was not allowed as a value of bufs. Since 4.5 it
is allowed under some circumstances:

From the OpenGL 4.5 specification, Section 17.4.1 "Selecting Buffers
for Writing", page 493 (page 515 of the PDF):
 "An INVALID_ENUM error is generated if any value in bufs is FRONT,
  LEFT, RIGHT, or FRONT_AND_BACK . This restriction applies to both
  the de- fault framebuffer and framebuffer objects, and exists
  because these constants may themselves refer to multiple buffers, as
  shown in table 17.4."

And on page 492 (page 514 of the PDF):
 "If the default framebuffer is affected, then each of the constants
  must be one of the values listed in table 17.6 or the special value
  BACK . When BACK is used, n must be 1 and color values are written
  into the left buffer for single-buffered contexts, or into the back
  left buffer for double-buffered contexts."

This patch keeps the same behaviour if OpenGL version is < 4. We
assume that for 4.x this is the intended behaviour, so a fix, but for
3.x the intended behaviour is the already in place.

Part of the fix for:
GL45-CTS.direct_state_access.framebuffers_draw_read_buffers_errors

v2: remove forgot printf
v3: remove spaces before commas on spec quote, split line too
    long (Anuj)

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
7 years agoradeonsi: num_records is in units of stride for swizzled buffers even on VI
Nicolai Hähnle [Thu, 15 Dec 2016 15:11:54 +0000 (16:11 +0100)]
radeonsi: num_records is in units of stride for swizzled buffers even on VI

The old setting didn't hurt, but this is cleaner.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agodocs: document INTEL_PRECISE_TRIG envvar
Juan A. Suarez Romero [Wed, 11 Jan 2017 17:51:35 +0000 (18:51 +0100)]
docs: document INTEL_PRECISE_TRIG envvar

v2: use more generic description (Jordan)

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agospirv: fix typo in warning message
Iago Toral Quiroga [Mon, 9 Jan 2017 15:09:55 +0000 (16:09 +0100)]
spirv: fix typo in warning message

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
7 years agoi965: Enable predicate support on gen >= 8.
Rafael Antognolli [Thu, 5 Jan 2017 18:33:53 +0000 (10:33 -0800)]
i965: Enable predicate support on gen >= 8.

Predication needs cmd parser only on gen7. For newer platforms, it
should be available without it.

v2 (Ken): rebase on recent changes.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoutil: fix list_is_singular()
Timothy Arceri [Wed, 11 Jan 2017 04:13:35 +0000 (15:13 +1100)]
util: fix list_is_singular()

Currently its dependant on the user calling and checking the result
of list_empty() before using the result of list_is_singular().

Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv/image: Disable HiZ for depth buffer arrays
Nanley Chery [Thu, 5 Jan 2017 14:58:03 +0000 (06:58 -0800)]
anv/image: Disable HiZ for depth buffer arrays

We currently don't perform clears or resolves on multiple array layers
with HiZ.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv/cmd_buffer: Fix programmed HiZ qpitch
Nanley Chery [Thu, 5 Jan 2017 09:23:27 +0000 (01:23 -0800)]
anv/cmd_buffer: Fix programmed HiZ qpitch

Match the comment above the field by using units of pixels and not HiZ
blocks.

Cc: mesa-stable@lists.freedesktop.org
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv/cmd_buffer: Fix arrayed depth/stencil attachments
Nanley Chery [Mon, 19 Dec 2016 17:37:03 +0000 (09:37 -0800)]
anv/cmd_buffer: Fix arrayed depth/stencil attachments

Enable multiple layers of the depth/stencil buffers to be accessible.

Fixes the crucible test, func.depthstencil.arrayed_clear.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoclover: Check for executables before enqueueing a kernel
Pierre Moreau [Thu, 29 Dec 2016 23:29:20 +0000 (00:29 +0100)]
clover: Check for executables before enqueueing a kernel

Without this check, the kernel::bind() method would fail with a
std::out_of_range exception, letting an exception escape from the
library into the client, rather than returning the corresponding error
code CL_INVALID_PROGRAM_EXECUTABLE.

Signed-off-by: Pierre Moreau <pierre.morrow@free.fr>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
7 years agospirv: Shut up unhandled enumeration value warnings.
Kenneth Graunke [Wed, 11 Jan 2017 23:16:11 +0000 (15:16 -0800)]
spirv: Shut up unhandled enumeration value warnings.

We don't want to do anything for the other cases.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agonir: don't turn ieq/ine into inot if used by an if
Timothy Arceri [Sun, 8 Jan 2017 12:52:59 +0000 (23:52 +1100)]
nir: don't turn ieq/ine into inot if used by an if

Otherwise we will end up with an extra instruction to compare the
result of the inot.

On BDW:

total instructions in shared programs: 13060620 -> 13060481 (-0.00%)
instructions in affected programs: 103379 -> 103240 (-0.13%)
helped: 127
HURT: 0

total cycles in shared programs: 256590950 -> 256587408 (-0.00%)
cycles in affected programs: 11324730 -> 11321188 (-0.03%)
helped: 114
HURT: 21

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: add late opt to turn inot/b2f combos back to bcsel
Timothy Arceri [Tue, 3 Jan 2017 11:54:48 +0000 (22:54 +1100)]
nir: add late opt to turn inot/b2f combos back to bcsel

We turn these from bcsel into inot/b2f combos in order for other
optimisation passes to get further. Once we have finished turn
the ones that remain and are used in more than a single expression
back into a bcsel.

On BDW:

total instructions in shared programs: 13060965 -> 13060297 (-0.01%)
instructions in affected programs: 835701 -> 835033 (-0.08%)
helped: 670
HURT: 2

total cycles in shared programs: 256599536 -> 256598006 (-0.00%)
cycles in affected programs: 114655488 -> 114653958 (-0.00%)
helped: 419
HURT: 240

LOST:   0
GAINED: 1

The 2 HURT is because inserting bcsel creates the only use of
const 1.0 in two shaders from tri-of-friendship-and-madness.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: add imprecise flrp optimisation
Timothy Arceri [Tue, 3 Jan 2017 10:10:33 +0000 (21:10 +1100)]
nir: add imprecise flrp optimisation

On BDW:

total instructions in shared programs: 13061890 -> 13061877 (-0.00%)
instructions in affected programs: 2441 -> 2428 (-0.53%)
helped: 13
HURT: 0

total cycles in shared programs: 256612254 -> 256611784 (-0.00%)
cycles in affected programs: 16418 -> 15948 (-2.86%)
helped: 10
HURT: 2

V2: don't use ffma directly

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965: Use the nir_move_comparisons pass.
Kenneth Graunke [Mon, 8 Aug 2016 21:44:06 +0000 (14:44 -0700)]
i965: Use the nir_move_comparisons pass.

While the below stats are encouraging this pass will also become
very usefull for avoiding regression once
brw_do_channel_expressions() and brw_do_vector_splitting() are
disabled.

On Broadwell:

total instructions in shared programs: 13078787 -> 13060898 (-0.14%)
instructions in affected programs: 1809827 -> 1791938 (-0.99%)
helped: 4527
HURT: 157

total cycles in shared programs: 256562762 -> 256590424 (0.01%)
cycles in affected programs: 159749392 -> 159777054 (0.02%)
helped: 5583
HURT: 2289

total spills in shared programs: 14929 -> 14923 (-0.04%)
spills in affected programs: 62 -> 56 (-9.68%)
helped: 1
HURT: 0

total fills in shared programs: 20144 -> 20141 (-0.01%)
fills in affected programs: 253 -> 250 (-1.19%)
helped: 1
HURT: 3

LOST:   0
GAINED: 2

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965: Move nir_lower_locals_to_regs a bit later.
Kenneth Graunke [Wed, 10 Aug 2016 01:02:46 +0000 (18:02 -0700)]
i965: Move nir_lower_locals_to_regs a bit later.

I'm going to add a boolean scheduling pass that I want run late, but
after copy propagation and dead code elimination.  Yet, I don't want
to have to think about registers.  So, move the register conversion
a little later.

No impact on shader-db.  Suggested by Jason Ekstrand.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
7 years agonir: Introduce a nir_opt_move_comparisons() pass.
Kenneth Graunke [Mon, 8 Aug 2016 20:57:29 +0000 (13:57 -0700)]
nir: Introduce a nir_opt_move_comparisons() pass.

This tries to move comparisons (a common source of boolean values)
closer to their first use.  For GPUs which use condition codes,
this can eliminate a lot of temporary booleans and comparisons
which reload the condition code register based on a boolean.

V2: (Timothy Arceri)
 - fix move comparision for phis so we dont end up with:

    vec1 32 ssa_227 = phi block_34: ssa_1, block_38: ssa_240
    vec1 32 ssa_235 = feq ssa_227, ssa_1
    vec1 32 ssa_230 = phi block_34: ssa_221, block_38: ssa_235

 - add nir_op_i2b/nir_op_f2b to the list of comparisons.

V3: (Timothy Arceri)
 - tidy up suggested by Jason.
 - add inot/fnot to move comparison list

V4: (Jason Ekstrand)
 - clean up move_comparison_source
 - get rid of the tuple
 - rework phi handling

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> [v1]
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir/algebraic: add support for conditional helper functions to expressions
Timothy Arceri [Tue, 10 Jan 2017 04:47:31 +0000 (15:47 +1100)]
nir/algebraic: add support for conditional helper functions to expressions

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv/TODO: Check off a bunch of stuff
Jason Ekstrand [Wed, 11 Jan 2017 18:26:23 +0000 (10:26 -0800)]
anv/TODO: Check off a bunch of stuff

7 years agonir/search: Only allow matching SSA values
Jason Ekstrand [Tue, 10 Jan 2017 18:24:55 +0000 (10:24 -0800)]
nir/search: Only allow matching SSA values

This is more correct and should also be a tiny bit faster since we're
just comparing pointers instead of calling nir_src_equal.

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
7 years agoegl/dri2: add image_loader_extension back into loader extensions for wayland
Derek Foreman [Tue, 10 Jan 2017 21:21:47 +0000 (15:21 -0600)]
egl/dri2: add image_loader_extension back into loader extensions for wayland

before commit f871946594129500a67c05a6d9fe99db54b4bb64
image_loader_extension was always present in dri2_dpy->extensions,
after that commit it is only present for render nodes.

Its removal broke partial render based on buffer age on (at least)
raspberry pi.

Fixes: f8719465941 "egl/dri2: rework dri2_egl_display::extensions storage"
Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agogallium/tgsi: fix overflow in parse property
Li Qiang [Tue, 10 Jan 2017 08:56:31 +0000 (03:56 -0500)]
gallium/tgsi: fix overflow in parse property

In parse_identifier, it doesn't stop copying '*pcur'
untill encounter the NULL. As the 'ret' has a
fixed-size buffer, if the '*pcur' has a long string,
there will be a buffer overflow. This patch avoid this.

Signed-off-by: Li Qiang <liq3ea@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
7 years agost/dri: remove trailing whitespace
Mauro Rossi [Tue, 10 Jan 2017 00:52:58 +0000 (01:52 +0100)]
st/dri: remove trailing whitespace

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agoandroid: st/mesa: fix building error in libmesa_st_mesa
Mauro Rossi [Tue, 10 Jan 2017 00:52:57 +0000 (01:52 +0100)]
android: st/mesa: fix building error in libmesa_st_mesa

Fixes building error due to dependency on nir generated headers

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agoradv: fix multi-viewport emission
Dave Airlie [Tue, 10 Jan 2017 04:50:34 +0000 (14:50 +1000)]
radv: fix multi-viewport emission

This set context req seq was in the wrong place.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agonir: change asserts to unreachable in nir_type_conversion_op
Tapani Pälli [Tue, 10 Jan 2017 08:14:03 +0000 (10:14 +0200)]
nir: change asserts to unreachable in nir_type_conversion_op

this is to avoid following compilation error on Android:

   error: control may reach end of non-void function [-Werror,-Wreturn-type]

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
7 years agospirv: gl_PrimitiveID in the fragment shader is handled as an input
Iago Toral Quiroga [Mon, 9 Jan 2017 11:31:53 +0000 (12:31 +0100)]
spirv: gl_PrimitiveID in the fragment shader is handled as an input

Geometry and Tessellation stages do handle this as a system value instead.

Fixes:
dEQP-VK.geometry.basic.primitive_id

Reviewed-by: Dave Airlie <ailried@redhat.com>
7 years agofreedreno: add "nogrow" debug param
Rob Clark [Mon, 2 Jan 2017 22:22:13 +0000 (17:22 -0500)]
freedreno: add "nogrow" debug param

Sometimes it is useful to disable the "growable" cmdstream buffers for
debugging.  (See 419a154d in libdrm)

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: remove hack for glamor
Rob Clark [Fri, 6 Jan 2017 18:50:59 +0000 (13:50 -0500)]
freedreno/a5xx: remove hack for glamor

Now that issues glamor was hitting w/ glsl>=130 (aka missing INSTANCED
bit in vertex attribute state) is fixed, remove hack.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fixed instanced
Rob Clark [Tue, 10 Jan 2017 16:43:12 +0000 (11:43 -0500)]
freedreno/a5xx: fixed instanced

Add missing bit, now that we know where it is.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: use the non-_ZERO_BASE for vertexid
Rob Clark [Mon, 9 Jan 2017 21:12:59 +0000 (16:12 -0500)]
freedreno/a5xx: use the non-_ZERO_BASE for vertexid

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: add texture MIPLVLS
Rob Clark [Mon, 9 Jan 2017 16:21:23 +0000 (11:21 -0500)]
freedreno/a5xx: add texture MIPLVLS

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fix fragcoord related hangs
Rob Clark [Mon, 2 Jan 2017 17:25:08 +0000 (12:25 -0500)]
freedreno/a5xx: fix fragcoord related hangs

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno: update generated headers
Rob Clark [Mon, 2 Jan 2017 17:18:59 +0000 (12:18 -0500)]
freedreno: update generated headers

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agoanv: Enable tessellation shaders.
Kenneth Graunke [Sun, 25 Sep 2016 22:33:03 +0000 (15:33 -0700)]
anv: Enable tessellation shaders.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: Initialize physical device limits for tessellation
Kenneth Graunke [Fri, 30 Sep 2016 01:11:21 +0000 (18:11 -0700)]
anv: Initialize physical device limits for tessellation

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: Clamp depth buffer dimensions to be at least 1.
Kenneth Graunke [Thu, 29 Sep 2016 18:53:17 +0000 (11:53 -0700)]
anv: Clamp depth buffer dimensions to be at least 1.

When there are no framebuffer attachments, fb->width and fb->height will
be 0.  Subtracting 1 results in 4294967295 which is too large for the
field, causing genxml assertions when trying to create the packet.

In this case, we can just program it to 1.

Caught by dEQP-VK.tessellation.tesscoord.triangles_equal_spacing.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: Compile TCS/TES shaders.
Kenneth Graunke [Mon, 26 Sep 2016 00:43:06 +0000 (17:43 -0700)]
anv: Compile TCS/TES shaders.

v2: Merge more TCS/TES info.
v3: Fix caching keys.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: Emit 3DSTATE_HS/TE/DS packets.
Kenneth Graunke [Mon, 21 Nov 2016 08:45:57 +0000 (00:45 -0800)]
anv: Emit 3DSTATE_HS/TE/DS packets.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: Handle patch primitives.
Kenneth Graunke [Sun, 25 Sep 2016 22:29:16 +0000 (15:29 -0700)]
anv: Handle patch primitives.

v2: Use anv_pipeline_has_stage rather than tess_info != NULL.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com> [v1]
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: Add a pass to lower TES patch_vertices intrinsics to a constant.
Kenneth Graunke [Sat, 1 Oct 2016 19:34:36 +0000 (12:34 -0700)]
nir: Add a pass to lower TES patch_vertices intrinsics to a constant.

In Vulkan, we always have both the TCS and TES available in the same
pipeline, so we can simply use the TCS OutputVertices execution mode
value as the TES PatchVertices built-in.

For GLSL, we handle this in the linker.  But we could use this pass
in the case when both TCS and TES are linked together, if we wanted.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: Silence unsupported tessellation capability warnings.
Kenneth Graunke [Sun, 25 Sep 2016 21:23:55 +0000 (14:23 -0700)]
spirv: Silence unsupported tessellation capability warnings.

...when the capability bit is set.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com> [v1]
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: Tidy some repeated if checks by using a switch statement.
Kenneth Graunke [Tue, 10 Jan 2017 05:48:55 +0000 (21:48 -0800)]
spirv: Tidy some repeated if checks by using a switch statement.

Iago suggested tidying this.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: Add tessellation varying and built-in support.
Kenneth Graunke [Sun, 25 Sep 2016 21:23:55 +0000 (14:23 -0700)]
spirv: Add tessellation varying and built-in support.

We need to:
- handle the extra array level for per-vertex varyings
- handle the patch qualifier correctly
- assign varying locations

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: Handle tessellation execution modes.
Kenneth Graunke [Sun, 25 Sep 2016 21:23:55 +0000 (14:23 -0700)]
spirv: Handle tessellation execution modes.

v2: Use info->tess.
v3: Handle more things in either TCS/TES.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com> [v1]
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com> [v1]
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agocompiler: Merge shader_info's tcs and tes structs.
Kenneth Graunke [Mon, 9 Jan 2017 19:37:21 +0000 (11:37 -0800)]
compiler: Merge shader_info's tcs and tes structs.

Annoyingly, SPIR-V lets you specify all of these fields in either the
TCS or TES, which means that we need to be able to store all of them
for either shader stage.  Putting them in a union won't work.

Combining both is an easy solution, and given that the TCS struct only
had a single field, it's pretty inexpensive.

This patch renames the combined struct to "tess" to indicate that it's
for tessellation in general, not one of the two stages.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agogenxml: Rename 3DSTATE_HS::Enable to "Function Enable".
Kenneth Graunke [Tue, 10 Jan 2017 00:36:12 +0000 (16:36 -0800)]
genxml: Rename 3DSTATE_HS::Enable to "Function Enable".

"Function Enable" is what the other stages use.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: set input_slots_valid on brw_wm_prog_key
Lionel Landwerlin [Tue, 10 Jan 2017 12:46:25 +0000 (12:46 +0000)]
anv: set input_slots_valid on brw_wm_prog_key

With shaders using a lot of inputs/outputs, like this (from Gtk+) :

layout(location = 0) in vec2 inPos;
layout(location = 1) in float inGradientPos;
layout(location = 2) in flat int inRepeating;
layout(location = 3) in flat int inStopCount;
layout(location = 4) in flat vec4 inClipBounds;
layout(location = 5) in flat vec4 inClipWidths;
layout(location = 6) in flat ColorStop inStops[8];

layout(location = 0) out vec4 outColor;

we're missing the programming of the input_slots_valid field leading
to an assert further down the backend code.

v2: Use valid slots of the geometry or vertex stage (Jason)

v3: Use helper to find correct vue map (Jason)

v4: Set the valid slots off the previous stages (Jason)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: add helper to get vue map for fragment shader
Lionel Landwerlin [Tue, 10 Jan 2017 12:46:25 +0000 (12:46 +0000)]
anv: add helper to get vue map for fragment shader

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: add get_.*_prog_data for tesselation stages
Lionel Landwerlin [Tue, 10 Jan 2017 17:21:40 +0000 (17:21 +0000)]
anv: add get_.*_prog_data for tesselation stages

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: make get_.*_prog_data take a const pipeline
Lionel Landwerlin [Tue, 10 Jan 2017 17:21:18 +0000 (17:21 +0000)]
anv: make get_.*_prog_data take a const pipeline

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: Fix anonymous union initialization with older GCC.
Vinson Lee [Sun, 8 Jan 2017 17:02:38 +0000 (09:02 -0800)]
nir: Fix anonymous union initialization with older GCC.

Fix this build error with GCC 4.4.7.

  CC     nir/nir_opt_copy_prop_vars.lo
nir/nir_opt_copy_prop_vars.c: In function ‘copy_prop_vars_block’:
nir/nir_opt_copy_prop_vars.c:765: error: unknown field ‘deref’ specified in initializer
nir/nir_opt_copy_prop_vars.c:765: warning: missing braces around initializer
nir/nir_opt_copy_prop_vars.c:765: warning: (near initialization for ‘(anonymous).<anonymous>’)
nir/nir_opt_copy_prop_vars.c:765: warning: initialization from incompatible pointer type

Fixes: 62332d139c8f ("nir: Add a local variable-based copy propagation pass")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agodocs: add Vulkan Float64 capability support for anv driver
Samuel Iglesias Gonsálvez [Mon, 9 Jan 2017 10:35:31 +0000 (11:35 +0100)]
docs: add Vulkan Float64 capability support for anv driver

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoradv/ac: add support for multi sample image coords
Dave Airlie [Fri, 30 Dec 2016 04:24:10 +0000 (14:24 +1000)]
radv/ac: add support for multi sample image coords

This just adds the nir->llvm support, enabling
the extension causes some failures on llvm 3.9 at least,
but this code seems fine.

NIR passes the sampler in src[1].x, and we LLVM/SI requires
it as the last parameters in the coords (coord[2] for 2D,
coord[3] for 2DArray).

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoglsl: Do not allow scalar types in vector relational functions
Boyan Ding [Mon, 31 Oct 2016 06:45:48 +0000 (14:45 +0800)]
glsl: Do not allow scalar types in vector relational functions

According to OpenGL Shading Language 4.50 spec, Section 8.7 "Vector
Relational Functions", functions of this type do not operate on scalar
types, so remove scalar types from signature definitions to make the
behavior consistent with glslangValidator and other drivers.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
7 years agonir: remove duplicated foreach loop
Thomas Hindoe Paaboel Andersen [Mon, 9 Jan 2017 21:45:57 +0000 (22:45 +0100)]
nir: remove duplicated foreach loop

The foreach loop was called both in the else case and right after. The
indentation seems to indicate that the extra call was from a previous
version with an else section with out curly brackets.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965: Fix number of slots in SSO mode when there are no user varyings.
Kenneth Graunke [Tue, 29 Nov 2016 00:59:45 +0000 (16:59 -0800)]
i965: Fix number of slots in SSO mode when there are no user varyings.

We want vue_map->num_slots to be one more than the final slot.

When assigning fixed slots, built-in slots, and non-SSO user varyings,
we do slot++.  This leaves "slot" as one past the most recently assigned
slot.  But for SSO user varyings, we computed slot based on the varying
location value...and left it at that slot value.

To work around this inconsistency, I made num_slots be "slot + 1" if
separate and "slot" otherwise.  The problem is...if there are no user
varyings in SSO mode...then we would have done slot++ when assigning
built-ins, so it would be off by one.  This resulted in loops from 0
to vue_map->num_slots hitting a bonus BRW_VARYING_SLOT_PAD at the end.

This used to break the SIMD8 VS/TES backends, but I fixed that in
commit 480d6c1653713dcae617ac523b2ca5deee01c845.  It's probably safe
at this point, but we should fix it anyway.

To fix this, do slot++ in all cases.  For SSO mode, we overwrite slot
for every varying, so this increment only matters on the last varying.
Because we process varyings in order, this will set slot to 1 more
than the highest assigned slot.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agospirv: Move cursor before calling vtn_ssa_value() in phi 2nd pass.
Kenneth Graunke [Mon, 9 Jan 2017 07:03:25 +0000 (23:03 -0800)]
spirv: Move cursor before calling vtn_ssa_value() in phi 2nd pass.

vtn_ssa_value() can produce variable loads, and the cursor might
be after a return statement, causing nir_builder assert failures
about not inserting instructions after a jump.

This fixes:
dEQP-VK.spirv_assembly.instruction.graphics.barrier.in_if
dEQP-VK.spirv_assembly.instruction.graphics.barrier.in_switch

Cc: "13.0 12.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agomesa: set GLSL 1.20 for the fixed-function fragment shader
Marek Olšák [Sat, 7 Jan 2017 14:36:45 +0000 (15:36 +0100)]
mesa: set GLSL 1.20 for the fixed-function fragment shader

This fixes broken depth texturing after:

commit 22639a6e19f95902aef23474ad672bf489231ea7
Author: Timothy Arceri <timothy.arceri@collabora.com>
Date:   Mon Nov 21 00:29:29 2016 +1100

    st/mesa: get Version from gl_program rather than gl_shader_program

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
7 years agoradv: Create single RADV_DEBUG env var.
Bas Nieuwenhuizen [Mon, 2 Jan 2017 17:57:02 +0000 (18:57 +0100)]
radv: Create single RADV_DEBUG env var.

Also changed RADV_SHOW_QUEUES to a no compute queue option. That
would make more sense later when the compute queue is established,
but the transfer queue still experimental.

v2: Don't include the trace flag.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoac/debug: Dump indirect buffers.
Bas Nieuwenhuizen [Sun, 1 Jan 2017 15:47:12 +0000 (16:47 +0100)]
ac/debug: Dump indirect buffers.

This is for handling chained command buffers and secondary command
buffers. It doesn't handle the trace id for secondary command buffers
yet, but I don't think that is possible in general with just writes,
as we could call a secondary command buffer multiple times.

I think this is good enough for now, as the most useful case is the
chaining when we grow an IB.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Dump command buffer on hang.
Bas Nieuwenhuizen [Fri, 23 Dec 2016 22:51:18 +0000 (23:51 +0100)]
radv: Dump command buffer on hang.

v2:
  - Now use the filename specified by RADV_TRACE_FILE env var.
  - Use the same var to enable tracing.

I thought we could as well always set the filename explicitly
instead of having some arbitrary defaults, and at that point
we don't need a separate feature enable.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoac/debug: Move IB decode to common code.
Bas Nieuwenhuizen [Sat, 24 Dec 2016 12:08:00 +0000 (13:08 +0100)]
ac/debug: Move IB decode to common code.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoac/debug: Move sid_tables.h generation to common code.
Bas Nieuwenhuizen [Sat, 24 Dec 2016 10:53:41 +0000 (11:53 +0100)]
ac/debug: Move sid_tables.h generation to common code.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agorelnotes: Claim OpenGL 4.5 rather than 4.4
Jason Ekstrand [Sun, 8 Jan 2017 04:29:49 +0000 (20:29 -0800)]
relnotes: Claim OpenGL 4.5 rather than 4.4

Acked-by: Matt Turner <mattst88@gmail.com>
7 years agomesa: Bump the version to 17.0
Jason Ekstrand [Sun, 8 Jan 2017 04:26:12 +0000 (20:26 -0800)]
mesa: Bump the version to 17.0

Acked-by: Matt Turner <mattst88@gmail.com>
7 years agoradeonsi: fix the Witcher 2 black transitions
Marek Olšák [Tue, 3 Jan 2017 19:12:28 +0000 (20:12 +0100)]
radeonsi: fix the Witcher 2 black transitions

v2: do it properly

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98238

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: set si_shader_context::input_decls for ranged decls correctly
Marek Olšák [Tue, 3 Jan 2017 19:03:37 +0000 (20:03 +0100)]
radeonsi: set si_shader_context::input_decls for ranged decls correctly

This has no effect because no code uses those members with ranged decls.

Tested-by: Edmondo Tommasina <edmondo.tommasina@gmail.com>
Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: cleanly communicate whether si_shader_dump should check R600_DEBUG
Marek Olšák [Tue, 3 Jan 2017 13:19:37 +0000 (14:19 +0100)]
radeonsi: cleanly communicate whether si_shader_dump should check R600_DEBUG

Tested-by: Edmondo Tommasina <edmondo.tommasina@gmail.com>
Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoisl: render target cube maps should be handled as 2D images, not cubes
Iago Toral Quiroga [Fri, 6 Jan 2017 11:29:45 +0000 (12:29 +0100)]
isl: render target cube maps should be handled as 2D images, not cubes

This fixes layered rendering Vulkan CTS tests with cube (arrays). We
also do this in the GL driver, see this code from gen8_depth_state.c
for example:

case GL_TEXTURE_CUBE_MAP_ARRAY:
case GL_TEXTURE_CUBE_MAP:
   /* The PRM claims that we should use BRW_SURFACE_CUBE for this
    * situation, but experiments show that gl_Layer doesn't work when we do
    * this.  So we use BRW_SURFACE_2D, since for rendering purposes this is
    * equivalent.
    */
   surftype = BRW_SURFACE_2D;
   depth *= 6;
   break;

So I guess we simply forgot to port this workaround to Vulkan.

v2: tweak the conditions so the special case is cube texture sampling
    rather than anything else (Jason)

Fixes:
dEQP-VK.geometry.layered.cube*

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: don't skip the VUE header if we are reading gl_Layer in a fragment shader
Iago Toral Quiroga [Thu, 5 Jan 2017 12:17:53 +0000 (13:17 +0100)]
anv: don't skip the VUE header if we are reading gl_Layer in a fragment shader

This is the same we do in the GL driver: the hardware provides gl_Layer
in the VUE header, so when the fragment shader reads it we can't skip it.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: enable shaderFloat64 feature
Samuel Iglesias Gonsálvez [Thu, 10 Nov 2016 09:06:48 +0000 (10:06 +0100)]
anv: enable shaderFloat64 feature

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv: enable float64 feature on supported platforms
Samuel Iglesias Gonsálvez [Wed, 4 Jan 2017 12:11:35 +0000 (13:11 +0100)]
anv: enable float64 feature on supported platforms

v2:
- Remove image_ms_array initialization (Jason)

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: enable SpvCapabilityFloat64 only to supported platforms
Samuel Iglesias Gonsálvez [Wed, 4 Jan 2017 08:35:36 +0000 (09:35 +0100)]
spirv: enable SpvCapabilityFloat64 only to supported platforms

v2 (Jason):
- Use nir_spirv_supported_extensions to check if the feature is enabled.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir/i965: use two slots from inputs_read for dvec3/dvec4 vertex input attributes
Juan A. Suarez Romero [Fri, 16 Dec 2016 09:24:43 +0000 (10:24 +0100)]
nir/i965: use two slots from inputs_read for dvec3/dvec4 vertex input attributes

So far, input_reads was a bitmap tracking which vertex input locations
were being used.

In OpenGL, an attribute bigger than a vec4 (like a dvec3 or dvec4)
consumes just one location, any other small attribute. So we mark the
proper bit in inputs_read, and also the same bit in double_inputs_read
if the attribute is a dvec3/dvec4.

But in Vulkan, this is slightly different: a dvec3/dvec4 attribute
consumes two locations, not just one. And hence two bits would be marked
in inputs_read for the same vertex input attribute.

To avoid handling two different situations in NIR, we just choose the
latest one: in OpenGL, when creating NIR from GLSL/IR, any dvec3/dvec4
vertex input attribute is marked with two bits in the inputs_read bitmap
(and also in the double_inputs_read), and following attributes are
adjusted accordingly.

As example, if in our GLSL/IR shader we have three attributes:

layout(location = 0) vec3  attr0;
layout(location = 1) dvec4 attr1;
layout(location = 2) dvec3 attr2;

then in our NIR shader we put attr0 in location 0, attr1 in locations 1
and 2, and attr2 in location 3 and 4.

Checking carefully, basically we are using slots rather than locations
in NIR.

When emitting the vertices, we do a inverse map to know the
corresponding location for each slot.

v2 (Jason):
- use two slots from inputs_read for dvec3/dvec4 NIR from GLSL/IR.

v3 (Jason):
- Fix commit log error.
- Use ladder ifs and fix braces.
- elements_double is divisible by 2, don't need DIV_ROUND_UP().
- Use if ladder instead of a switch.
- Add comment about hardware restriction in 64bit vertex attributes.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoisl: fix VA64 support for double and dvecN vertex attributes
Samuel Iglesias Gonsálvez [Tue, 15 Nov 2016 11:49:38 +0000 (12:49 +0100)]
isl: fix VA64 support for double and dvecN vertex attributes

We use *64*_PASSTHRU formats to upload vertex attributes of 64 bits
to avoid conversions. From the BDW PRM, Volume 2d, page 586
(VERTEX_ELEMENT_STATE):

     "When SourceElementFormat is set to one of the *64*_PASSTHRU
     formats, 64-bit components are stored in the URB without any
     conversion. In this case, vertex elements must be written as 128
     or 256 bits, with VFCOMP_STORE_0 being used to pad the output
     as required. E.g., if R64_PASSTHRU is used to copy a 64-bit Red
     component into the URB, Component 1 must be specified as
     VFCOMP_STORE_0 (with Components 2,3 set to VFCOMP_NOSTORE)
     in order to output a 128-bit vertex element, or Components 1-3 must
     be specified as VFCOMP_STORE_0 in order to output a 256-bit vertex
     element. Likewise, use of R64G64B64_PASSTHRU requires Component 3
     to be specified as VFCOMP_STORE_0 in order to output a 256-bit vertex
     element."

v2,v3 (Jason):
- Don't delete unused formats.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoanv/pipeline: get map for double input attributes
Juan A. Suarez Romero [Tue, 15 Nov 2016 16:57:30 +0000 (16:57 +0000)]
anv/pipeline: get map for double input attributes

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: add support for doubles to OpSpecConstant
Samuel Iglesias Gonsálvez [Mon, 14 Nov 2016 11:08:32 +0000 (12:08 +0100)]
spirv: add support for doubles to OpSpecConstant

v2 (Jason):
- Fix indent in radv change
- Add vtn_u64_literal() helper to take 64 bits (Jason)

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv/nir: add (un)packDouble2x32() translation
Samuel Iglesias Gonsálvez [Thu, 10 Nov 2016 09:37:40 +0000 (10:37 +0100)]
spirv/nir: add (un)packDouble2x32() translation

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv/nir: implement DF conversions
Samuel Iglesias Gonsálvez [Wed, 7 Dec 2016 07:10:16 +0000 (08:10 +0100)]
spirv/nir: implement DF conversions

SPIR-V does not have special opcodes for DF conversions. We need to identify
them by checking the bit size of the operand and the result.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: add nir_type_conversion_op()
Samuel Iglesias Gonsálvez [Wed, 7 Dec 2016 07:34:42 +0000 (08:34 +0100)]
nir: add nir_type_conversion_op()

This function returns the nir_op corresponding to the conversion between
the given nir_alu_type arguments.

This function lacks support for integer-based types with bit_size != 32
and for float16 conversion ops.

v2:
- Improve readiness of the code and delete cases that don't happen now (Jason)

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: add nir_get_nir_type_for_glsl_type()
Samuel Iglesias Gonsálvez [Wed, 7 Dec 2016 07:34:02 +0000 (08:34 +0100)]
nir: add nir_get_nir_type_for_glsl_type()

v2 (Jason):
- Refactor nir_get_nir_type_for_glsl_type() to avoid using unneeded helpers (Jason)

v3:
- Use return directly (Jason)

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: add support for doubles on OpComposite{Insert,Extract}
Samuel Iglesias Gonsálvez [Thu, 10 Nov 2016 11:11:03 +0000 (12:11 +0100)]
spirv: add support for doubles on OpComposite{Insert,Extract}

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: Enable double floating points when copying variables in _vtn_variable_copy()
Samuel Iglesias Gonsálvez [Thu, 10 Nov 2016 11:07:53 +0000 (12:07 +0100)]
spirv: Enable double floating points when copying variables in _vtn_variable_copy()

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agospirv: add double support to _vtn_block_load_store()
Samuel Iglesias Gonsálvez [Thu, 10 Nov 2016 09:07:09 +0000 (10:07 +0100)]
spirv: add double support to _vtn_block_load_store()

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>