mesa.git
8 years agoanv: Don't assert-fail if someone asks for a non-existent entrypoint
Jason Ekstrand [Tue, 22 Mar 2016 23:11:53 +0000 (16:11 -0700)]
anv: Don't assert-fail if someone asks for a non-existent entrypoint

8 years agoUpdate to the latest Vulkan header from Khronos
Jason Ekstrand [Tue, 22 Mar 2016 23:06:53 +0000 (16:06 -0700)]
Update to the latest Vulkan header from Khronos

8 years agoanv/batch_chain: Fall back to growing batches when chaining isn't available
Jason Ekstrand [Fri, 18 Mar 2016 23:32:46 +0000 (16:32 -0700)]
anv/batch_chain: Fall back to growing batches when chaining isn't available

8 years agoanv/allocator: Make the bo_pool dynamically sized
Jason Ekstrand [Fri, 18 Mar 2016 20:06:08 +0000 (13:06 -0700)]
anv/allocator: Make the bo_pool dynamically sized

8 years agoanv/allocator: Add a size field to bo_pool_alloc
Jason Ekstrand [Fri, 18 Mar 2016 18:50:53 +0000 (11:50 -0700)]
anv/allocator: Add a size field to bo_pool_alloc

8 years agoMerge remote-tracking branch 'origin/master' into vulkan
Jordan Justen [Thu, 17 Mar 2016 08:38:05 +0000 (01:38 -0700)]
Merge remote-tracking branch 'origin/master' into vulkan

8 years agoi965/nir: Lower nir compute shader shared variables
Jordan Justen [Mon, 18 Jan 2016 17:45:46 +0000 (09:45 -0800)]
i965/nir: Lower nir compute shader shared variables

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agonir: Lower shared var atomics during nir_lower_io
Jordan Justen [Mon, 18 Jan 2016 17:59:19 +0000 (09:59 -0800)]
nir: Lower shared var atomics during nir_lower_io

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agonir: Add support for lowering load/stores of shared variables
Jordan Justen [Mon, 18 Jan 2016 17:44:31 +0000 (09:44 -0800)]
nir: Add support for lowering load/stores of shared variables

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agonir: Add atomic operations on variables
Jordan Justen [Mon, 18 Jan 2016 17:53:44 +0000 (09:53 -0800)]
nir: Add atomic operations on variables

This allows us to first generate atomic operations for shared
variables using these opcodes, and then later we can lower those to
the shared atomics intrinsics with nir_lower_io.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agonir: Add compute shader shared variable storage class
Jordan Justen [Sat, 9 Jan 2016 01:16:29 +0000 (17:16 -0800)]
nir: Add compute shader shared variable storage class

Previously we were receiving shared variable accesses via a lowered
intrinsic function from glsl. This change allows us to send in
variables instead. For example, when converting from SPIR-V.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agonir/print: Add space after shader_storage var mode
Jordan Justen [Sun, 17 Jan 2016 07:11:16 +0000 (23:11 -0800)]
nir/print: Add space after shader_storage var mode

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agoi965: Skip execution size adjustment for instructions of width 4
Iago Toral Quiroga [Thu, 3 Dec 2015 08:50:49 +0000 (09:50 +0100)]
i965: Skip execution size adjustment for instructions of width 4

This code in brw_set_dest adjusts the execution size of any instruction
with a dst.width < 8. However, we don't want to do this with instructions
operating on doubles, since these will have a width of 4, but still
need an execution size of 8 (for SIMD8). Unfortunately, we can't just check
the size of the operands involved to detect if we are doing an operation on
doubles, because we can have instructions that do operations on double
operands interpreted as UD, operating on any of its 2 32-bit components.

Previous commits have made it so we never emit instructions with a horizontal
width of 4 that don't have the correct execution size set for gen6+, so
we can skip it in this case, avoiding the conflicts with fp64 requirements.

Expanding the same fix to other hardware generations requires many more
changes but since we are not targetting fp64 support on them
wer don't really care for now.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/vec4/gen6: fix exec_size for MOV with a width of 4 in generate_gs_ff_sync()
Samuel Iglesias Gonsalvez [Tue, 15 Dec 2015 08:34:38 +0000 (09:34 +0100)]
i965/vec4/gen6: fix exec_size for MOV with a width of 4 in generate_gs_ff_sync()

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/vec4/gen6: fix exec_size for instructions with destination width of 4
Samuel Iglesias Gonsalvez [Fri, 4 Dec 2015 09:23:15 +0000 (10:23 +0100)]
i965/vec4/gen6: fix exec_size for instructions with destination width of 4

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/vec4/gen6: fix exec_size for instructions with width of 4 in generate_gs_svb_write()
Samuel Iglesias Gonsalvez [Thu, 3 Dec 2015 17:27:39 +0000 (18:27 +0100)]
i965/vec4/gen6: fix exec_size for instructions with width of 4 in generate_gs_svb_write()

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/gs/gen6: fix execsize for instructions with width of 4 in gen6_sol_program()
Samuel Iglesias Gonsalvez [Thu, 3 Dec 2015 17:05:39 +0000 (18:05 +0100)]
i965/gs/gen6: fix execsize for instructions with width of 4 in gen6_sol_program()

v2:
- Add assert (Topi).

Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965: set correct execsize for MOVS with a width of 4 in brw_find_live_channel
Iago Toral Quiroga [Thu, 3 Dec 2015 10:11:14 +0000 (11:11 +0100)]
i965: set correct execsize for MOVS with a width of 4 in brw_find_live_channel

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/eu: set execution size for SEND message in brw_send_indirect_message
Iago Toral Quiroga [Thu, 3 Dec 2015 10:10:12 +0000 (11:10 +0100)]
i965/eu: set execution size for SEND message in brw_send_indirect_message

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/fs: Set exec size for gen7 pull const loads
Iago Toral Quiroga [Thu, 3 Dec 2015 09:59:23 +0000 (10:59 +0100)]
i965/fs: Set exec size for gen7 pull const loads

v2 (Topi):
  - No need to set the execsize for the indirect send message,
    the next patch will handle that.
  - Set the execution size explicitly instead of taking it from
    the width of the dst that we set before.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agoi965/eu: set correct execution size in brw_NOP
Iago Toral Quiroga [Thu, 3 Dec 2015 07:49:13 +0000 (08:49 +0100)]
i965/eu: set correct execution size in brw_NOP

v2: NOP should have an execsize of 1 (Matt)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agometa: Don't use integer handles for shaders or programs.
Kenneth Graunke [Tue, 15 Mar 2016 17:51:55 +0000 (10:51 -0700)]
meta: Don't use integer handles for shaders or programs.

Previously, we gave our internal clear/blit shaders actual GL handles
and stored them in the shader/program hash table.  We used ordinary
GL API entrypoints to work with them.

We thought this shouldn't be a problem because GL doesn't allow
applications to invent their own names for shaders or programs.
GL allocates all names via glCreateShader and glCreateProgram.

However, having them in the hash table is a bit risky: if a broken
application guesses the name of our shaders or programs, it could
alter them, potentially screwing up future meta operations.

Also, test cases can observe the programs in the hash table.  Running
a single dEQP process that executes the following test list:

dEQP-GLES3.functional.negative_api.buffer.clear
dEQP-GLES3.functional.negative_api.shader.compile_shader
dEQP-GLES3.functional.negative_api.shader.delete_shader

would result in the last two tests breaking.  The compile_shader test
calls glCompileShader(9) straight away, and since it hasn't even created
any shaders or programs, it expects to get a GL_INVALID_VALUE error
because there's no such name.  However, because the clear test ran
first, it created Meta programs, so an object named "9" did exist.

This patch reworks Meta to work with gl_shader and gl_shader_program
pointers directly.  These internal programs have bogus names, and are
never stored in the hash tables, so they're invisible to applications.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94485
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agomesa: Expose compile_shader() and link_program() beyond the file.
Kenneth Graunke [Tue, 15 Mar 2016 17:51:33 +0000 (10:51 -0700)]
mesa: Expose compile_shader() and link_program() beyond the file.

This will allow me to use them directly from Meta, bypassing the
versions that work with GL integer handles.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
8 years agomesa: Make link_program() take a gl_shader_program, not a GLuint.
Kenneth Graunke [Wed, 16 Mar 2016 00:08:17 +0000 (17:08 -0700)]
mesa: Make link_program() take a gl_shader_program, not a GLuint.

In half the callers, we already have a pointer, and don't need
to look it up again.  This will also help with upcoming meta work.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
8 years agomesa: Make compile_shader() take a gl_shader, not a GLuint.
Kenneth Graunke [Tue, 15 Mar 2016 17:24:09 +0000 (10:24 -0700)]
mesa: Make compile_shader() take a gl_shader, not a GLuint.

In half the callers, we already have a pointer, and don't need
to look it up again.  This will also help with upcoming meta work.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
8 years agometa: Use the _mesa_meta_compile_and_link_program helper more places.
Kenneth Graunke [Tue, 15 Mar 2016 23:21:36 +0000 (16:21 -0700)]
meta: Use the _mesa_meta_compile_and_link_program helper more places.

Less boilerplate.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
8 years agovc4: Move discard handling to the condition flag.
Eric Anholt [Wed, 16 Mar 2016 01:00:22 +0000 (18:00 -0700)]
vc4: Move discard handling to the condition flag.

Now that the field exists in the instruction, we can make discards less
special.  As a bonus, that means that we should be able to merge some more
.sf instructions together when we get around to that.

This causes some scheduling changes, as it allows tlb_color_reads to be
delayed past the discard condition setup.  Since the tlb_color_read ends
up later, this may mean performance improvements, but I haven't tested.

total instructions in shared programs: 78114 -> 78035 (-0.10%)
instructions in affected programs:     1922 -> 1843 (-4.11%)
total estimated cycles in shared programs: 234318 -> 234329 (0.00%)
estimated cycles in affected programs:     8200 -> 8211 (0.13%)

8 years agovc4: Don't make a temporary for setting flags.
Eric Anholt [Wed, 16 Mar 2016 01:58:43 +0000 (18:58 -0700)]
vc4: Don't make a temporary for setting flags.

The register allocator doesn't really do anything about the temp, so it
doesn't seem like it should matter.  However, the scheduler would think
that a new def is being created.

This doesn't change anything yet, but it avoids a bunch of regressions in
the next commit.

8 years agovc4: Add a safety check for setting flags.
Eric Anholt [Wed, 16 Mar 2016 01:57:20 +0000 (18:57 -0700)]
vc4: Add a safety check for setting flags.

If a pack was on the src reg, should it be a float, int, or mul unpack?
Just complain, instead.

8 years agovc4: Reuse list_for_each_entry_safe_rev().
Eric Anholt [Wed, 16 Mar 2016 01:50:32 +0000 (18:50 -0700)]
vc4: Reuse list_for_each_entry_safe_rev().

This didn't exist when I wrote the code.

8 years agoanv/blit: Reduce number of VUE headers being read
Nanley Chery [Thu, 10 Mar 2016 19:06:25 +0000 (11:06 -0800)]
anv/blit: Reduce number of VUE headers being read

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoanv/blit: Remove completed finishme for VkFilter
Nanley Chery [Mon, 14 Mar 2016 18:11:50 +0000 (11:11 -0700)]
anv/blit: Remove completed finishme for VkFilter

This task was finished as of:
d9079648d0f1c380929dea0f3a447ddfdf5dcd27.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoanv/blit2d: Only use one extent in meta_emit_blit2d
Nanley Chery [Mon, 14 Mar 2016 15:15:16 +0000 (08:15 -0700)]
anv/blit2d: Only use one extent in meta_emit_blit2d

Since scaling isn't involved, we don't need multiple extents.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoanv/blit2d: Remove sampler from pipeline
Nanley Chery [Fri, 11 Mar 2016 02:25:10 +0000 (18:25 -0800)]
anv/blit2d: Remove sampler from pipeline

Since we're using texelFetch with a sampled image, a sampler is no
longer needed. This agrees with the Vulkan Spec section 13.2.4
Descriptor Set Updates:

sampler is a sampler handle, and is used in descriptor updates for types
VK_DESCRIPTOR_TYPE_SAMPLER and VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER
if the binding being updated does not use immutable samplers.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoanv/blit2d: Use texel fetch in frag shader
Nanley Chery [Fri, 11 Mar 2016 00:06:14 +0000 (16:06 -0800)]
anv/blit2d: Use texel fetch in frag shader

The texelFetch operation requires that the sampled texture coordinates
be unnormalized integers. This will simplify the copy shader for
w-tiled images (stencil buffers).

v2 (Jason):
   Use f2i for texel coords
   Fix num_components indirectly
   Use float inputs for interpolation
   Nest tex_pos functions

Suggested-by: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoRevert "anv/meta: Make meta_emit_blit() public"
Nanley Chery [Sat, 12 Mar 2016 00:26:28 +0000 (16:26 -0800)]
Revert "anv/meta: Make meta_emit_blit() public"

This reverts commit f39168392243d6dacefbc8708b764c5978ff24df.

Some conflicts had to be resolved in order for this revert to be
successful.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoRevert "anv/meta: Prefix anv_ to meta_emit_blit()"
Nanley Chery [Sat, 12 Mar 2016 00:25:02 +0000 (16:25 -0800)]
Revert "anv/meta: Prefix anv_ to meta_emit_blit()"

This reverts commit 514c0557178b0325c59a28d68b0f250f0eeaddf5.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoanv/blit2d: Customize meta blit structs and functions for blit2d API
Nanley Chery [Wed, 9 Mar 2016 19:31:49 +0000 (11:31 -0800)]
anv/blit2d: Customize meta blit structs and functions for blit2d API

* Add fields in meta struct
* Add support in meta init/teardown
* Switch to custom meta_emit_blit2d()

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agoanv/blit2d: Copy anv_meta_blit.c functions
Nanley Chery [Tue, 8 Mar 2016 22:12:55 +0000 (14:12 -0800)]
anv/blit2d: Copy anv_meta_blit.c functions

These will be customized for blit2d operations.

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
8 years agometa: Use ARB_explicit_attrib_location in the rest of the meta shaders.
Kenneth Graunke [Tue, 15 Mar 2016 18:13:25 +0000 (11:13 -0700)]
meta: Use ARB_explicit_attrib_location in the rest of the meta shaders.

This is cleaner than using glBindAttribLocation().

Not all drivers support the extension, but I don't think those drivers
use GLSL in the first place.  Apparently some Meta shaders already use
GL_ARB_explicit_attrib_location, so I think it should be okay.

Honestly, I'm not sure how the old code worked anyway - we bound the
attribute location for "texcoords", while all the shaders capitalized
or spelled it differently.

v2: Convert another instance in brw_meta_fast_clear.c.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
8 years agomesa: Ignore glPointSize when GL_POINT_SIZE_ARRAY_OES is enabled
Plamena Manolova [Tue, 15 Mar 2016 18:39:49 +0000 (20:39 +0200)]
mesa: Ignore glPointSize when GL_POINT_SIZE_ARRAY_OES is enabled

When a user defines a point size array and enables it, the point
size value set via glPointSize should be ignored. To achieve this,
we can simply toggle ctx->VertexProgram.PointSizeEnabled.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42187
Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agoanv/device: Flush the fence batch rather than the start of the BO
Jason Ekstrand [Tue, 15 Mar 2016 22:24:24 +0000 (15:24 -0700)]
anv/device: Flush the fence batch rather than the start of the BO

8 years agoMerge remote-tracking branch 'public/master' into vulkan
Jason Ekstrand [Tue, 15 Mar 2016 21:09:50 +0000 (14:09 -0700)]
Merge remote-tracking branch 'public/master' into vulkan

8 years agovc4: Coalesce instructions using VPM reads into the VPM read.
Varad Gautam [Mon, 7 Mar 2016 19:31:59 +0000 (01:01 +0530)]
vc4: Coalesce instructions using VPM reads into the VPM read.

This is done instead of copy propagating the VPM reads into the
instructions using them, because VPM reads have to stay in order.

shader-db results:
total instructions in shared programs: 78509 -> 78114 (-0.50%)
instructions in affected programs:     5203 -> 4808 (-7.59%)
total estimated cycles in shared programs: 234670 -> 234318 (-0.15%)
estimated cycles in affected programs:     5345 -> 4993 (-6.59%)

Signed-off-by: Varad Gautam <varadgautam@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Rhys Kidd <rhyskidd@gmail.com>
8 years agovc4: rename file to group vpm optimizations together
Varad Gautam [Mon, 7 Mar 2016 19:31:58 +0000 (01:01 +0530)]
vc4: rename file to group vpm optimizations together

This file will contain optimization passes for both vpm reads
and writes.

Signed-off-by: Varad Gautam <varadgautam@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
8 years agovc4: Fix failures with nir_extract_* since the addition of the opcodes.
Eric Anholt [Tue, 15 Mar 2016 19:48:55 +0000 (12:48 -0700)]
vc4: Fix failures with nir_extract_* since the addition of the opcodes.

8 years agollvmpipe: fix lp_rast_plane alignment on 32bit
Roland Scheidegger [Tue, 15 Mar 2016 15:39:55 +0000 (16:39 +0100)]
llvmpipe: fix lp_rast_plane alignment on 32bit

Some rasterization code relies (for sse) on the first and third planes
(but not the second for now) being 128bit aligned, and we didn't get that
on 32bit - I mistakenly thought the 64bit number in the struct would get
the thing aligned to 64bit even on 32bit archs.
Stephane Marchesin really figured this out.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
CC: <mesa-stable@lists.freedesktop.org>
8 years agodraw: fix line stippling
Roland Scheidegger [Tue, 15 Mar 2016 18:40:44 +0000 (19:40 +0100)]
draw: fix line stippling

The logic was comparing actual ints, not true/false values.
This meant that it was emitting always multiple line segments instead of just
one even if the stipple test had the same result, which looks inefficient, and
the segments also overlapped thus breaking line aa as well.
(In practice, with the no-op default line stipple pattern, for a 10-pixel
long line from 0-9 it was emitting 10 segments, with the individual segments
ranging from 0-1, 0-2, 0-3 and so on.)

This fixes https://bugs.freedesktop.org/show_bug.cgi?id=94193

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
CC: <mesa-stable@lists.freedesktop.org>
8 years agosoftpipe: fix misleading TGSI_QUAD_SIZE usage
Roland Scheidegger [Sun, 13 Mar 2016 18:38:23 +0000 (19:38 +0100)]
softpipe: fix misleading TGSI_QUAD_SIZE usage

All these img filter loops iterate through NUM_CHANNELS, not QUAD_SIZE.
In practice both are of course the same unchangeable value (4), but it
makes the code look a bit confusing. Moreover, some of the functions were
actually given an array of 4 values according to the declaration, yet the
code was addressing values 0/4/8/12 out of it, so fix this by just saying
it's a pointer to floats like the other functions.

While here, also add comment about not quite correct filtering.

There's no actual code difference.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
8 years agosoftpipe: fix anisotropic filtering crash
Roland Scheidegger [Sun, 13 Mar 2016 18:13:09 +0000 (19:13 +0100)]
softpipe: fix anisotropic filtering crash

The filt_args->offset wasn't assigned but was always used later leading
to a crash (as far as I can tell, texel offsets don't actually make much
sense with anisotropic filtering, but because there's no explicit setting
if offsets are enabled there the array is always accessed).

This fixes https://bugs.freedesktop.org/show_bug.cgi?id=94481

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
CC: <mesa-stable@lists.freedesktop.org>
8 years agoradeonsi: set DEPTH_BEFORE_SHADER based on FS_EARLY_DEPTH_STENCIL
Nicolai Hähnle [Fri, 11 Mar 2016 23:20:00 +0000 (18:20 -0500)]
radeonsi: set DEPTH_BEFORE_SHADER based on FS_EARLY_DEPTH_STENCIL

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agotgsi: add tgsi_full_src_register_from_dst helper function
Nicolai Hähnle [Tue, 9 Feb 2016 17:54:10 +0000 (12:54 -0500)]
tgsi: add tgsi_full_src_register_from_dst helper function

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agogallium/u_inlines: add util_copy_image_view
Nicolai Hähnle [Sat, 6 Feb 2016 21:49:17 +0000 (16:49 -0500)]
gallium/u_inlines: add util_copy_image_view

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agost/mesa: set image access flags in st_bind_images
Nicolai Hähnle [Sun, 13 Mar 2016 14:10:53 +0000 (09:10 -0500)]
st/mesa: set image access flags in st_bind_images

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agogallium: add access field to pipe_image_view
Nicolai Hähnle [Sat, 12 Mar 2016 01:04:19 +0000 (20:04 -0500)]
gallium: add access field to pipe_image_view

This allows drivers to make smarter decisions e.g. about whether the image
has to be decompressed.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agost/glsl_to_tgsi: set FS_EARLY_DEPTH_STENCIL when required
Nicolai Hähnle [Fri, 11 Mar 2016 23:11:35 +0000 (18:11 -0500)]
st/glsl_to_tgsi: set FS_EARLY_DEPTH_STENCIL when required

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agotgsi: add TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL
Nicolai Hähnle [Fri, 11 Mar 2016 23:11:22 +0000 (18:11 -0500)]
tgsi: add TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agost/glsl_to_tgsi: set memory access type on image intrinsics
Nicolai Hähnle [Sun, 13 Mar 2016 02:47:35 +0000 (21:47 -0500)]
st/glsl_to_tgsi: set memory access type on image intrinsics

This is required to preserve the image variable's coherent/restrict/volatile
qualifiers in TGSI.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agost/glsl_to_tgsi: provide Texture and Format information for image ops
Nicolai Hähnle [Sun, 7 Feb 2016 18:28:01 +0000 (13:28 -0500)]
st/glsl_to_tgsi: provide Texture and Format information for image ops

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agotgsi: add Texture and Format to tgsi_instruction_memory
Nicolai Hähnle [Thu, 10 Mar 2016 21:30:07 +0000 (16:30 -0500)]
tgsi: add Texture and Format to tgsi_instruction_memory

Frontends should have this information readily available, and it simplifies
image LOAD/STORE/ATOM* handling especially with indirect image access.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agoget: reconcile aliasing enums for MaxCombinedShaderOutputResources
Nicolai Hähnle [Thu, 10 Mar 2016 22:41:29 +0000 (17:41 -0500)]
get: reconcile aliasing enums for MaxCombinedShaderOutputResources

The enums MAX_COMBINED_IMAGE_UNITS_AND_FRAGMENT_OUTPUTS and
MAX_COMBINED_SHADER_OUTPUT_RESOURCES are equal and should therefore only
appear once.

Noticed while implementing ARB_shader_image_load_store without previously
implementing SSBO.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agoi965/fs: Restrict inequality that can only hold equal in saturate propagation.
Francisco Jerez [Mon, 7 Mar 2016 03:03:56 +0000 (19:03 -0800)]
i965/fs: Restrict inequality that can only hold equal in saturate propagation.

Should have no functional change.  The IP value of an instruction that
reads src_var cannot possibly be after the end of the live interval of
the variable it's reading from, by the definition of live interval.
Might save future readers a momentary WTF while trying to understand
this code.

Reviewed-by: Matt Turner <mattst88@gmail.com>
8 years agoi965/vec4: Consider removal of no-op MOVs as progress during register coalesce.
Francisco Jerez [Mon, 14 Mar 2016 02:15:45 +0000 (19:15 -0700)]
i965/vec4: Consider removal of no-op MOVs as progress during register coalesce.

Bug found by the liveness analysis validation pass that will be
introduced in a later commit.  The no-op MOV check in
opt_register_coalesce() was removing instructions which makes the
cached liveness analysis calculation inconsistent with the shader IR.
We were failing to set progress to true in that case though, which
means that invalidate_live_intervals() wouldn't necessarily be called
at the end of the function.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
8 years agoi965/fs: Add missing analysis invalidation in fixup_3src_null_dest().
Francisco Jerez [Fri, 11 Mar 2016 23:27:22 +0000 (15:27 -0800)]
i965/fs: Add missing analysis invalidation in fixup_3src_null_dest().

Bug found by the liveness analysis validation pass that will be
introduced in a later commit.  fixup_3src_null_dest() was allocating
registers which makes the cached liveness analysis calculation
incomplete, so it must be invalidated.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
8 years agoi965/fs: Add missing analysis invalidation in opt_sampler_eot().
Francisco Jerez [Fri, 11 Mar 2016 23:22:56 +0000 (15:22 -0800)]
i965/fs: Add missing analysis invalidation in opt_sampler_eot().

Bug found by the liveness analysis validation pass that will be
introduced in a later commit.  opt_sampler_eot() was allocating
registers and inserting and removing instructions, which makes the
cached liveness analysis calculation inconsistent with the shader IR,
so it must be invalidated.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
8 years agoclover: Fix pipe_grid_info.indirect not being initialized.
Hans de Goede [Mon, 14 Mar 2016 14:01:05 +0000 (15:01 +0100)]
clover: Fix pipe_grid_info.indirect not being initialized.

After pipe_grid_info.indirect was introduced, clover was not modified
to set it causing it to pass uninitialized memory for it to launch_grid.

This commit fixes this by zero-ing the entire pipe_grid_info struct when
declaring it, to avoid similar problems popping-up in the future.

Cc: "11.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
[ Francisco Jerez: Trivial codestyle fix. ]
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
8 years agomesa: docs: Intel i965 hardware limits.
Sarah Sharp [Thu, 29 Oct 2015 23:11:54 +0000 (16:11 -0700)]
mesa: docs: Intel i965 hardware limits.

This should help the next person working on hardware enabling figure out
where in the Intel PRMs to find the magic platform hardware values.

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
8 years agomesa: docs: i965: Use correct doxygen groupings syntax
Sarah Sharp [Thu, 29 Oct 2015 19:12:40 +0000 (12:12 -0700)]
mesa: docs: i965: Use correct doxygen groupings syntax

When reading the source code, it's useful to indicate that a group of
fields in a struct are related in someway. There were several places
where people tried to group related structure members with the {@
syntax, without realizing they also needed to add the \name syntax in
order to generate correct doxygen html.

There are several files with groupings that look like this:

struct foo {
    /**
     * Related fields description
     * @{
     */
    int bar;
    char baz;
    /** @} */
    long qux;
}

However, the doxygen syntax for grouping is:

struct foo {
    /**
     * \name Related fields description
     * @{
     */
    int bar;
    char baz;
    /** @} */
    long qux;
}

https://www.stack.nl/~dimitri/doxygen/manual/grouping.html

Without the group name definition, the fields don't get properly
grouped. Instead, the group description is applied to the first field.

Fix the Intel hardware information structure, brw_device_info to
properly group the GPU hardware limitations and hardware quirks fields.

Once you've run `cd doxygen; make clean; make all`,
updated documentation can be found at

mesa/doxygen/i965/structbrw__device__info.html

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
8 years agogallium/swr: Resource management
Bruce Cherniak [Thu, 10 Mar 2016 01:30:00 +0000 (19:30 -0600)]
gallium/swr: Resource management

Better tracking of resource state and synchronization.
A follow on commit will clean up resource functions into a new
swr_resource.cpp file.

Reviewed-By: George Kyriazis <george.kyriazis@intel.com>
8 years agoconfigure.ac: require libdrm 2.4.66 for drmGetDevice
Marek Olšák [Mon, 14 Mar 2016 15:41:54 +0000 (16:41 +0100)]
configure.ac: require libdrm 2.4.66 for drmGetDevice

since 737b6ed13e8f813987b5566004f0f45e9c55f1e8
src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c no longer compiles:
error: unknown type name ‘drmDevicePtr’

8 years agoi965: Remove useless IR self-destruct backend_shader method.
Francisco Jerez [Wed, 9 Mar 2016 01:23:37 +0000 (17:23 -0800)]
i965: Remove useless IR self-destruct backend_shader method.

From the point it's constructed the CFG contains the only existing
copy of the program IR, and it never becomes invalid.  Calling
backend_shader::invalidate_cfg would have destroyed the program
structure irrecoverably -- We weren't calling it at all for a good
reason.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
8 years agonv50,nvc0: Set only NEW_CP_GLOBALS upon binding
Pierre Moreau [Sun, 13 Mar 2016 21:11:30 +0000 (22:11 +0100)]
nv50,nvc0: Set only NEW_CP_GLOBALS upon binding

Signed-off-by: Pierre Moreau <pierre.morrow@free.fr>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
8 years agofreedreno/ir3: lower extract_byte/word
Rob Clark [Sun, 13 Mar 2016 18:03:04 +0000 (14:03 -0400)]
freedreno/ir3: lower extract_byte/word

The following commits broke things by starting to feed us unhandled
extract_u16/extract_u8 opcodes:

commit 905ff861982450831a56d112036f68a751337441
Author:     Matt Turner <mattst88@gmail.com>
AuthorDate: Wed Feb 3 14:28:31 2016 -0800
Commit:     Matt Turner <mattst88@gmail.com>
CommitDate: Fri Mar 4 11:52:34 2016 -0800

    nir: Recognize open-coded extract_u16.

commit 76289fbfa84a06ef4db8ad44ea0eb88ad0be8d5c
Author:     Matt Turner <mattst88@gmail.com>
AuthorDate: Thu Jan 21 09:09:48 2016 -0800
Commit:     Matt Turner <mattst88@gmail.com>
CommitDate: Fri Mar 4 11:52:34 2016 -0800

    nir: Recognize open-coded extract_u8.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agonv50,nvc0: handle SQRT lowering inside the driver
Ilia Mirkin [Sun, 13 Mar 2016 02:26:21 +0000 (21:26 -0500)]
nv50,nvc0: handle SQRT lowering inside the driver

First off, st/mesa lowers DSQRT incorrectly (it uses CMP to attempt to
find out whether the input is less than 0). Secondly the current
approach (x * rsq(x)) behaves poorly for x = inf - a NaN is produced
instead of inf.

Instead we switch to the less accurate rcp(rsq(x)) method - this behaves
nicely for all valid inputs. We still don't do this for DSQRT since the
RSQ/RCP ops are *really* inaccurate, and don't even have Newton-Raphson
steps right now. Eventually we should have a separate library function
for DSQRT that does it more precisely (and perhaps move this lowering to
the post-opt phase).

This fixes a number of dEQP precision tests that were expecting better
behavior for infinite inputs.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
8 years agonv50/ir: avoid folding mul + add if the mul has a dnz
Ilia Mirkin [Sun, 13 Mar 2016 01:53:34 +0000 (20:53 -0500)]
nv50/ir: avoid folding mul + add if the mul has a dnz

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
8 years agonvc0: fix blit triangle size to fully cover FB's > 8192x8192
Ilia Mirkin [Sat, 12 Mar 2016 02:26:31 +0000 (21:26 -0500)]
nvc0: fix blit triangle size to fully cover FB's > 8192x8192

The idea is that a single triangle will cover the whole area being
drawn, allowing the blit shader to do its work. However the max fb size
is 16384x16384, which means that the triangle we draw needs to be twice
that in order to cover the whole area fully. Increase the size of the
triangle to 32768x32768.

This fixes a number of dEQP tests that were failing because a blit was
involved which would miss some of the resulting texture.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>
8 years agofreedreno: OUT_RELOC vs OUT_RELOCW fixes
Rob Clark [Wed, 9 Mar 2016 09:07:51 +0000 (04:07 -0500)]
freedreno: OUT_RELOC vs OUT_RELOCW fixes

Make sure we use OUT_RELOCW() in cases where the buffer is written to.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a4xx: hw binning
Rob Clark [Sat, 5 Mar 2016 21:53:11 +0000 (16:53 -0500)]
freedreno/a4xx: hw binning

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a4xx: use generated headers for draw initiator
Rob Clark [Sat, 5 Mar 2016 21:50:09 +0000 (16:50 -0500)]
freedreno/a4xx: use generated headers for draw initiator

No need to open-code this.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a4xx: remove RB_RENDER_CONTROL patching
Rob Clark [Sat, 5 Mar 2016 21:47:26 +0000 (16:47 -0500)]
freedreno/a4xx: remove RB_RENDER_CONTROL patching

Bitfields where shuffled around for the better on a4xx, so we don't need
any patching on this one.  It appears to be something we set entirely in
the gmem code so no conflict between tiling and render state like we had
in a3xx.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno: update generated headers
Rob Clark [Sat, 5 Mar 2016 21:41:21 +0000 (16:41 -0500)]
freedreno: update generated headers

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a3xx: move where we deal w/ binning FS
Rob Clark [Sun, 13 Mar 2016 16:15:28 +0000 (12:15 -0400)]
freedreno/a3xx: move where we deal w/ binning FS

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a4xx: move where we deal w/ binning FS
Rob Clark [Mon, 7 Mar 2016 05:52:03 +0000 (00:52 -0500)]
freedreno/a4xx: move where we deal w/ binning FS

Move where we pick dummy FS for binning pass, so the whole driver sees
the same dummy/no-op FS stage.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a3xx: constify the shader variants
Rob Clark [Sun, 13 Mar 2016 16:09:51 +0000 (12:09 -0400)]
freedreno/a3xx: constify the shader variants

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a4xx: constify the shader variants
Rob Clark [Mon, 7 Mar 2016 05:50:21 +0000 (00:50 -0500)]
freedreno/a4xx: constify the shader variants

Most of the driver just needs read-only access, so constify..

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agofreedreno/a3xx: remove duplicate mark of end of binning cmds
Rob Clark [Sun, 6 Mar 2016 03:11:26 +0000 (22:11 -0500)]
freedreno/a3xx: remove duplicate mark of end of binning cmds

Signed-off-by: Rob Clark <robclark@freedesktop.org>
8 years agoradeonsi: avoid crash when a sampler state is bound for a buffer texture
Nicolai Hähnle [Fri, 11 Mar 2016 16:07:38 +0000 (11:07 -0500)]
radeonsi: avoid crash when a sampler state is bound for a buffer texture

Sampler states don't really make sense with buffer textures, but they
can be set anyway, so we need to be defensive here. This bug was lurking
for a while and was finally noticed due to PBO uploads setting sampler
states.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94284
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Laurent Carlier <lordheavym@gmail.com>
Tested-by: Shawn Starr <shawn.starr@rogers.com>
8 years agoi965: Use foreach_in_list_reverse_safe() macro.
Matt Turner [Mon, 15 Feb 2016 06:28:51 +0000 (22:28 -0800)]
i965: Use foreach_in_list_reverse_safe() macro.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agonir/clone: Add support for cloning a single function_impl
Jason Ekstrand [Sat, 13 Feb 2016 04:51:50 +0000 (20:51 -0800)]
nir/clone: Add support for cloning a single function_impl

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir/validate: Better function validation
Jason Ekstrand [Sun, 14 Feb 2016 02:08:30 +0000 (18:08 -0800)]
nir/validate: Better function validation

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir/print: Better function argument printing
Jason Ekstrand [Sat, 13 Feb 2016 04:41:19 +0000 (20:41 -0800)]
nir/print: Better function argument printing

Since we aren't going to put the function parameters or the return variable
in the list of locals, it won't get a proper declaration.  This changes
nir_print to print the type along with each parameter or return variable.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir/print: Factor variable name lookup into a helper
Jason Ekstrand [Sat, 26 Dec 2015 18:26:29 +0000 (10:26 -0800)]
nir/print: Factor variable name lookup into a helper

Otherwise, we have a problem when we go to print functions with arguments
because their names get added to the hash table during declaration which
happens after we print the prototype.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir: Create function parameters in function_impl_create
Jason Ekstrand [Mon, 15 Feb 2016 05:42:34 +0000 (21:42 -0800)]
nir: Create function parameters in function_impl_create

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir: Add a helper for creating a "bare" nir_function_impl
Jason Ekstrand [Wed, 28 Oct 2015 04:34:56 +0000 (21:34 -0700)]
nir: Add a helper for creating a "bare" nir_function_impl

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir: Add a new "param" variable mode for parameters and return variables
Jason Ekstrand [Fri, 12 Feb 2016 19:58:06 +0000 (11:58 -0800)]
nir: Add a new "param" variable mode for parameters and return variables

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agonir/glsl: Remove dead function parameter handling code
Jason Ekstrand [Fri, 12 Feb 2016 18:50:56 +0000 (10:50 -0800)]
nir/glsl: Remove dead function parameter handling code

NIR has never been used on IR where we haven't already done function
inlining so this code has been dead from the beginning.  Let's just get rid
of it for now.  We can always put it back in if we decide to use NIR for
function inlining at some point in the future.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agoanv/gen7: Add stall and flushes before switching pipelines
Jordan Justen [Fri, 11 Mar 2016 01:19:13 +0000 (17:19 -0800)]
anv/gen7: Add stall and flushes before switching pipelines

This is a port of 18c76551ee425b981efefc61f663a7781df17882 from OpenGL
to Vulkan.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agoanv: Add flush_pipeline_before_pipeline_select
Jordan Justen [Fri, 11 Mar 2016 01:25:45 +0000 (17:25 -0800)]
anv: Add flush_pipeline_before_pipeline_select

flush_pipeline_before_pipeline_select adds workarounds required before
switching the pipeline.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agoanv/genX: Add flush_pipeline_select_gpgpu
Jordan Justen [Fri, 11 Mar 2016 01:16:58 +0000 (17:16 -0800)]
anv/genX: Add flush_pipeline_select_gpgpu

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
8 years agoHACK: Don't re-configure L3$ in render stages pre-BDW
Jason Ekstrand [Sat, 12 Mar 2016 16:54:41 +0000 (08:54 -0800)]
HACK: Don't re-configure L3$ in render stages pre-BDW

This fixes a "regression" on Haswell and prior caused by merging the gen7
and gen8 flush_state functions.  Haswell should still work just fine if
you're on a 4.4 kernel, but we really should make it detect the command
parser version and do something intelligent.