mesa.git
9 years agonir/spirv: Add support for various barrier type instructions
Jason Ekstrand [Thu, 22 Oct 2015 01:17:11 +0000 (18:17 -0700)]
nir/spirv: Add support for various barrier type instructions

9 years agoFix a couple of dereferences
Jason Ekstrand [Thu, 22 Oct 2015 01:16:50 +0000 (18:16 -0700)]
Fix a couple of dereferences

9 years agospirv/nir: Add more stage asserts
Jason Ekstrand [Thu, 22 Oct 2015 01:00:05 +0000 (18:00 -0700)]
spirv/nir: Add more stage asserts

9 years agonir/spirv: Add support for GS metadata
Jason Ekstrand [Thu, 22 Oct 2015 00:58:34 +0000 (17:58 -0700)]
nir/spirv: Add support for GS metadata

9 years agonir/gather_info: Add more info for geometry shaders
Jason Ekstrand [Thu, 22 Oct 2015 00:42:47 +0000 (17:42 -0700)]
nir/gather_info: Add more info for geometry shaders

9 years agoMerge remote-tracking branch 'mesa-public/master' into vulkan
Jason Ekstrand [Thu, 22 Oct 2015 00:40:13 +0000 (17:40 -0700)]
Merge remote-tracking branch 'mesa-public/master' into vulkan

9 years agoi965: Implement gl_InvocationID.
Kenneth Graunke [Fri, 10 Jul 2015 07:16:19 +0000 (00:16 -0700)]
i965: Implement gl_InvocationID.

It's stored in bits 31:27 of g1 (along with the URB handles).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Implement nir_intrinsic_load_primitive.
Kenneth Graunke [Thu, 24 Sep 2015 01:59:57 +0000 (18:59 -0700)]
i965: Implement nir_intrinsic_load_primitive.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Add a fs_visitor constructor that takes a brw_gs_compile.
Kenneth Graunke [Tue, 30 Jun 2015 05:50:28 +0000 (22:50 -0700)]
i965: Add a fs_visitor constructor that takes a brw_gs_compile.

Unlike the vs/wm structs, brw_gs_compile is actually useful: it contains
the input VUE map and information about the control data headers.
Passing this in allows us to share that code in brw_gs.c, and calculate
them before deciding on vec4 vs. scalar mode, as it's independent of
that choice.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Add a brw->scalar_gs flag controlled by INTEL_SCALAR_GS=1.
Kenneth Graunke [Fri, 26 Jun 2015 23:20:21 +0000 (16:20 -0700)]
i965: Add a brw->scalar_gs flag controlled by INTEL_SCALAR_GS=1.

This patch introduces a brw->scalar_gs flag, similar to brw->scalar_vs,
which controls whether or not to use SIMD8 geometry shaders.

For now, we control it via a new environment variable, INTEL_SCALAR_GS.
This provides a convenient way to try it out.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Make emit_urb_writes() reserve space for GS header information.
Kenneth Graunke [Wed, 6 May 2015 03:19:04 +0000 (20:19 -0700)]
i965: Make emit_urb_writes() reserve space for GS header information.

Geometry shaders have additional header data at the beginning of their
output URB entries.  Shaders that use EndPrimitive() or multiple streams
have a control data header; shaders with a dynamic vertex count have an
additional vec4 slot to hold the 32-bit vertex count (and 96 bits of
padding).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Make emit_urb_writes() only set EOT for the VS.
Kenneth Graunke [Thu, 12 Mar 2015 08:55:44 +0000 (01:55 -0700)]
i965: Make emit_urb_writes() only set EOT for the VS.

The GS will emit a bunch of vertices, and we don't want to do an EOT
prematurely.  We'll emit GS_OPCODE_THREAD_END when we want to terminate
the thread.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Make fs_visitor::emit_urb_writes reusable for scalar GS.
Kenneth Graunke [Thu, 12 Mar 2015 12:52:13 +0000 (05:52 -0700)]
i965: Make fs_visitor::emit_urb_writes reusable for scalar GS.

GS doesn't have ClampVertexColor, and we don't want to go through VS
structures.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Introduce a brw_vue_prog_data::include_vue_handles flag.
Kenneth Graunke [Tue, 29 Sep 2015 21:43:29 +0000 (14:43 -0700)]
i965: Introduce a brw_vue_prog_data::include_vue_handles flag.

Tessellation shaders and SIMD8 geometry shaders may need to resort to
the pull model for inputs at times.  When set, the state upload code
will tell the hardware to provide URB handles for input data.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Introduce a new SHADER_OPCODE_URB_READ_SIMD8 opcode.
Kenneth Graunke [Tue, 29 Sep 2015 21:32:02 +0000 (14:32 -0700)]
i965: Introduce a new SHADER_OPCODE_URB_READ_SIMD8 opcode.

In scalar mode, geometry shader inputs can easily take up hundreds of
registers.  This makes pushing VUE entries impractical; we'll need to
resort to the pull model in some cases.

To support this, we introduce a new opcode corresponding to the "URB
Read SIMD8" message.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965: Introduce new SHADER_OPCODE_URB_WRITE_SIMD8_MASKED/PER_SLOT opcodes.
Kenneth Graunke [Wed, 6 May 2015 07:04:10 +0000 (00:04 -0700)]
i965: Introduce new SHADER_OPCODE_URB_WRITE_SIMD8_MASKED/PER_SLOT opcodes.

In the vec4 backend, we have a vec4_instruction::urb_write_flags field.
There are many kinds of flags for SIMD4x2 messages.

However, there are really only two (per-slot offset, use channel masks)
for SIMD8 messages.  Rather than adding a boolean flag for per-slot
offsets (polluting all instructions), I decided to just make three new
opcodes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
9 years agoi965/gs: Do prog_data setup and other calculations in brw_compile_gs
Jason Ekstrand [Wed, 21 Oct 2015 19:03:21 +0000 (12:03 -0700)]
i965/gs: Do prog_data setup and other calculations in brw_compile_gs

This commit moves the large pile of setup calculations we have to do for
geometry shaders out of brw_gs_emit and into brw_compile_gs.  This has a
couple of nice implications.  First, it's less work that the caller of
brw_compile_gs has to do.  Second, it's consistent with the vertex and
fragment stages.  Finally, it allows us to put brw_gs_compile back behind
the API boundary where it belongs.

v2 (Jason Ekstrand):
 - Pull the changes to use nir info into a separate patch
 - Put brw_gs_compile into brw_shader.h rather than brw_vec4_gs_visitor.h
   so that we can use it for scalar GS.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoi965/gs: Use NIR info for setting up prog_data
Jason Ekstrand [Wed, 21 Oct 2015 19:02:08 +0000 (12:02 -0700)]
i965/gs: Use NIR info for setting up prog_data

Previously, we were pulling bits from GL data structures in order to set up
the prog_data.  However, in this brave new world of NIR, we want to be
pulling it out of the NIR shader whenever possible.  This way, we can move
all this setup code into brw_compile_gs without depending on the old GL
stuff.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoi965/gs: Pull prog_data out of brw_gs_compile
Jason Ekstrand [Wed, 21 Oct 2015 00:12:03 +0000 (17:12 -0700)]
i965/gs: Pull prog_data out of brw_gs_compile

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoi965/gs: Use NIR instead of the brw_geometry_program for GS metadata
Jason Ekstrand [Tue, 20 Oct 2015 23:40:30 +0000 (16:40 -0700)]
i965/gs: Use NIR instead of the brw_geometry_program for GS metadata

With this, we can remove the geometry program from brw_gs_compile.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoi965/gs: Move the mem_ctx argument to brw_compile_gs
Jason Ekstrand [Tue, 20 Oct 2015 23:21:09 +0000 (16:21 -0700)]
i965/gs: Move the mem_ctx argument to brw_compile_gs

This makes it better match the other brw_compile_* functions.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoi965/gs: Set static_vertex_count unconditionally on GEN8+
Jason Ekstrand [Tue, 20 Oct 2015 23:46:50 +0000 (16:46 -0700)]
i965/gs: Set static_vertex_count unconditionally on GEN8+

We always have NIR, so there's no reason for the check.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agonir: Constify nir_gs_count_vertices
Jason Ekstrand [Wed, 21 Oct 2015 00:40:19 +0000 (17:40 -0700)]
nir: Constify nir_gs_count_vertices

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agonir/info: Add more information about geometry shaders
Jason Ekstrand [Tue, 20 Oct 2015 23:35:44 +0000 (16:35 -0700)]
nir/info: Add more information about geometry shaders

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoanv: Don't teardown uninitialized anv_physical_device
Chad Versace [Wed, 21 Oct 2015 18:36:39 +0000 (11:36 -0700)]
anv: Don't teardown uninitialized anv_physical_device

If the user called vkDestroyDevice but never called
vkEnumeratePhysicalDevices, then the driver tried to ralloc_free() an
unitialized anv_physical_device.

Fixes test 'dEQP-VK.api.device_init.create_instance_name_version'.

9 years agoi965: (trivial) rename computes stencil to gen9
Ben Widawsky [Tue, 20 Oct 2015 21:29:38 +0000 (14:29 -0700)]
i965: (trivial) rename computes stencil to gen9

All the documentation I can find says that this bit (and functionality) only
exists on SKL+. Since the bit isn't yet used, there is no real impact here.

The original code was added by Ken here (a surprisingly long time ago):
commit f3c6d6f1e151f6a44a76038dccebe4434038dcb1
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Thu Nov 29 21:00:27 2012 -0800

    i965: Update 3DSTATE_PS, 3DSTATE_WM, and add 3DSTATE_PS_EXTRA.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoi965: Correct the comment about fb write payload
Ben Widawsky [Tue, 20 Oct 2015 21:29:36 +0000 (14:29 -0700)]
i965: Correct the comment about fb write payload

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
9 years agomesa/glformats: Undo code changes from _mesa_base_tex_format() move
Nanley Chery [Thu, 8 Oct 2015 23:44:30 +0000 (16:44 -0700)]
mesa/glformats: Undo code changes from _mesa_base_tex_format() move

The refactoring commit, c6bf1cd, accidentally reverted cd49b97
and 99b1f47. These changes caused more code to be added to the
function and removed the existing support for ASTC. This patch
reverts those modifications.

v2. Actually include ASTC support again.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92221
Cc: "11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
9 years agoi965: Mark compacted 3-src instructions as Gen8+.
Matt Turner [Tue, 20 Oct 2015 02:02:16 +0000 (19:02 -0700)]
i965: Mark compacted 3-src instructions as Gen8+.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965: Add const to brw_compact_inst_bits.
Matt Turner [Tue, 20 Oct 2015 01:59:53 +0000 (18:59 -0700)]
i965: Add const to brw_compact_inst_bits.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965: Add mask_control_ex field and handle it in compaction.
Matt Turner [Mon, 19 Oct 2015 22:08:28 +0000 (15:08 -0700)]
i965: Add mask_control_ex field and handle it in compaction.

Documentation is sparse, but it appears to have existed on G45 and ILK
as a second bit extension of the mask_control field. Setting the pair of
bits to 0b11 enables "NoCMask".

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965: Add devinfo->gen assertions for acc_wr_control.
Matt Turner [Mon, 19 Oct 2015 21:47:17 +0000 (14:47 -0700)]
i965: Add devinfo->gen assertions for acc_wr_control.

... and for flag_subreg_nr since it's right near by.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965: Prepare for next commit by adding more whitespace.
Matt Turner [Mon, 19 Oct 2015 21:46:18 +0000 (14:46 -0700)]
i965: Prepare for next commit by adding more whitespace.

We're going to add a field with a longer name that wouldn't align with
the rest.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965: Compact acc_wr_control only on Gen6+.
Matt Turner [Mon, 19 Oct 2015 22:01:20 +0000 (15:01 -0700)]
i965: Compact acc_wr_control only on Gen6+.

It only exists on Gen6+, and the next patches will add compaction
support for the (unused) field in the same location on earlier
platforms.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965: Add devinfo parameter to brw_compact_inst_* funcs.
Matt Turner [Mon, 19 Oct 2015 21:41:36 +0000 (14:41 -0700)]
i965: Add devinfo parameter to brw_compact_inst_* funcs.

The next commit will add assertions dependent on devinfo->gen.

Use compact()/uncompact() macros where possible, like the 3-src code
does.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoi965/vec4: Don't emit MOVs for unused URB slots.
Matt Turner [Thu, 15 Oct 2015 22:13:12 +0000 (15:13 -0700)]
i965/vec4: Don't emit MOVs for unused URB slots.

Otherwise we'd emit a MOV from the null register (which isn't allowed).

Helps 24 programs in shader-db (the geometry shaders in GSCloth):

instructions in affected programs:     302 -> 262 (-13.25%)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
9 years agoosmesa: Expose GL entry points for Windows build via DEF file.
Nigel Stewart [Mon, 12 Oct 2015 11:26:37 +0000 (21:26 +1000)]
osmesa: Expose GL entry points for Windows build via DEF file.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92437
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoconfigure.ac: ensure RM is set
Jonathan Gray [Sat, 10 Oct 2015 06:42:40 +0000 (17:42 +1100)]
configure.ac: ensure RM is set

GNU make predefines RM to rm -f but this is not required by POSIX
so ensure that RM is set.  This fixes "make clean" on OpenBSD.

v2: use AC_CHECK_PROG

Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
CC: "10.6 11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
9 years agoi965/fs: Disable opt_sampler_eot for more message types
Neil Roberts [Tue, 20 Oct 2015 09:56:15 +0000 (11:56 +0200)]
i965/fs: Disable opt_sampler_eot for more message types

In bfdae9149e0 I disabled the opt_sampler_eot optimisation for TG4
message types because I found by experimentation that it doesn't work.
I wrote in the comment that I couldn't find any documentation for this
problem. However I've now found the documentation and it has
additional restrictions on further message types so this patch updates
the comment and adds the others.

Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoi965: Remove block arg from foreach_inst_in_block_*_starting_from
Neil Roberts [Tue, 20 Oct 2015 09:16:00 +0000 (11:16 +0200)]
i965: Remove block arg from foreach_inst_in_block_*_starting_from

Since 49374fab5d793 these macros no longer actually use the block
argument. I think this is worth doing to make the macros easier to use
because they already have really long names and a confusing set of
arguments.

Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agoglsl: check for arrays of arrays when assigning explicit locations
Timothy Arceri [Wed, 21 Oct 2015 02:44:10 +0000 (13:44 +1100)]
glsl: check for arrays of arrays when assigning explicit locations

This fixes assigning explicit locations in the CTS test:

ES31-CTS.explicit_uniform_location.uniform-loc-arrays-of-arrays

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
9 years agoglsl: add is_array_of_arrays() helper
Timothy Arceri [Wed, 21 Oct 2015 02:37:11 +0000 (13:37 +1100)]
glsl: add is_array_of_arrays() helper

As suggested by Ian Romanick

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
9 years agoglsl: Fix bad indentation in bit_logic_result_type().
Kenneth Graunke [Wed, 21 Oct 2015 02:51:56 +0000 (19:51 -0700)]
glsl: Fix bad indentation in bit_logic_result_type().

The first level of indentation was using 4 spaces.  Mesa uses 3.

Trivial.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoglsl: add AoA support to subroutines
Timothy Arceri [Wed, 5 Aug 2015 05:49:22 +0000 (15:49 +1000)]
glsl: add AoA support to subroutines

process_parameters() will now be called earlier because we need
actual_parameters processed earlier so we can use it with
match_subroutine_by_name() to get the subroutine variable, we need
to do this inside the recursive function generate_array_index() because
we can't create the ir_dereference_array() until we have gotten to the
outermost array.

For the remainder of the array dimensions the type doesn't matter so we
can just use the existing _mesa_ast_array_index_to_hir() function to
process the ast.

Reviewed-by: Dave Airlie <airlied@redhat.com>
9 years agoglsl: fix record type detection in explicit location assign
Tapani Pälli [Tue, 20 Oct 2015 09:18:51 +0000 (12:18 +0300)]
glsl: fix record type detection in explicit location assign

Check current_var directly instead of using the passed in record_type.

This fixes following failing CTS test:
ES31-CTS.explicit_uniform_location.uniform-loc-types-structs

No Piglit regressions.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
9 years agoglsl: do not try to reserve explicit locations for buffer variables
Tapani Pälli [Tue, 20 Oct 2015 07:24:50 +0000 (10:24 +0300)]
glsl: do not try to reserve explicit locations for buffer variables

Explicit locations are only used with uniform variables.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
9 years agoglsl: skip buffer variables when filling UniformRemapTable
Tapani Pälli [Tue, 20 Oct 2015 05:32:15 +0000 (08:32 +0300)]
glsl: skip buffer variables when filling UniformRemapTable

UniformRemapTable is used only for remapping user specified uniform
locations to driver internally used ones, shader storage buffer
variables should not utilize uniform locations.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
9 years agosvga: add switch case for PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
Brian Paul [Wed, 21 Oct 2015 00:22:43 +0000 (18:22 -0600)]
svga: add switch case for PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT

A third instance of this was needed but missed in the previous commit.
Return 32 as for the two other cases.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
9 years agodraw: fix splitting of line loops (v2)
Brian Paul [Sat, 17 Oct 2015 18:07:32 +0000 (12:07 -0600)]
draw: fix splitting of line loops (v2)

When the draw module splits long line loops, the sections are emitted
as line strips.  But the primitive type wasn't set correctly so each
section was being drawn as a loop, introducing extra line segments.

To fix this, we pass a new DRAW_LINE_LOOP_AS_STRIP flag to the run()
function.  The linear/elt_run() functions have to check for this flag
and set their primitive type accordingly.

No piglit regressions.  Fixes piglit's lineloop with -count 4097 or
higher.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81174

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
9 years agoanv/pipeline: Remove a redundant line
Jason Ekstrand [Tue, 20 Oct 2015 23:02:00 +0000 (16:02 -0700)]
anv/pipeline: Remove a redundant line

We set compute_sample_id based on multisample state two lines below.

9 years agoanv/pipeline: Update a comment
Jason Ekstrand [Tue, 20 Oct 2015 23:00:55 +0000 (16:00 -0700)]
anv/pipeline: Update a comment

9 years agoanv/pipeline: Set key->render_to_fbo to false for fragment shaaders
Jason Ekstrand [Tue, 20 Oct 2015 22:37:14 +0000 (15:37 -0700)]
anv/pipeline: Set key->render_to_fbo to false for fragment shaaders

Vulkan uses the upper-left convention.  This is the same as DX one and what
our hardware does.  We had it flipped around.

9 years agonir/spirv: Fix a typo
Jason Ekstrand [Tue, 20 Oct 2015 22:35:13 +0000 (15:35 -0700)]
nir/spirv: Fix a typo

9 years agonir/spirv: Handle SpvExecutionMode
Jason Ekstrand [Tue, 20 Oct 2015 22:23:56 +0000 (15:23 -0700)]
nir/spirv: Handle SpvExecutionMode

9 years agoi965/gen9: Remove temporary variable 'bpp' in tr_mode_..._texture_alignment()
Anuj Phogat [Thu, 13 Aug 2015 18:19:47 +0000 (11:19 -0700)]
i965/gen9: Remove temporary variable 'bpp' in tr_mode_..._texture_alignment()

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
9 years agoi965/gen9: Remove temporary variable 'align_yf' in tr_mode_..._texture_alignment()
Anuj Phogat [Wed, 12 Aug 2015 23:39:05 +0000 (16:39 -0700)]
i965/gen9: Remove temporary variable 'align_yf' in tr_mode_..._texture_alignment()

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
9 years agoi965/gen9: Remove parameter 'brw' from tr_mode_..._texture_alignment()
Anuj Phogat [Wed, 23 Sep 2015 23:13:00 +0000 (16:13 -0700)]
i965/gen9: Remove parameter 'brw' from tr_mode_..._texture_alignment()

V2: Rebased on master.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
9 years agoi965/gen9: Reuse YF alignment tables in tr_mode_..._texture_alignment()
Anuj Phogat [Wed, 23 Sep 2015 23:12:26 +0000 (16:12 -0700)]
i965/gen9: Reuse YF alignment tables in tr_mode_..._texture_alignment()

Patch just does some refactoring to make the code look better. No
functional changes in here.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
9 years agoanv: Completely rework shader compilation
Jason Ekstrand [Tue, 20 Oct 2015 05:06:59 +0000 (22:06 -0700)]
anv: Completely rework shader compilation

Now that we have a decent interface in upstream mesa, we can get rid of all
our hacks.  As of this commit, we no longer use any fake GL state objects
and all of shader compilation is moved into anv_pipeline.c.  This should
make way for actually implementing a shader cache one of these days.

As a nice side-benifit, this commit also gains us an extra 300 passing CTS
tests because we're actually filling out the texture swizzle information
for vertex shaders.

9 years agonir: Add a pass to gather info from the shader
Jason Ekstrand [Tue, 20 Oct 2015 05:05:36 +0000 (22:05 -0700)]
nir: Add a pass to gather info from the shader

This pass fills out a bunch of the fields in nir_shader_info by inspecting
the shader.

9 years agoanv: Move the brw_compiler from anv_compiler to physical_device
Jason Ekstrand [Tue, 20 Oct 2015 03:21:45 +0000 (20:21 -0700)]
anv: Move the brw_compiler from anv_compiler to physical_device

9 years agoi965: Move brw_compiler_create to brw_compiler.h
Jason Ekstrand [Tue, 20 Oct 2015 03:19:09 +0000 (20:19 -0700)]
i965: Move brw_compiler_create to brw_compiler.h

9 years agoi965: Split process_nir into two haves; pre- and post-
Jason Ekstrand [Tue, 20 Oct 2015 01:46:29 +0000 (18:46 -0700)]
i965: Split process_nir into two haves; pre- and post-

9 years agoanv/compiler: Remove more pre-SNB shader key setup
Jason Ekstrand [Tue, 20 Oct 2015 01:15:45 +0000 (18:15 -0700)]
anv/compiler: Remove more pre-SNB shader key setup

9 years agoanv/compiler: Get rid of GS support.
Jason Ekstrand [Tue, 20 Oct 2015 00:05:40 +0000 (17:05 -0700)]
anv/compiler: Get rid of GS support.

The geometry shader support is currently completely untested.  As I go
through and re-factor the compiler, I'd rather not refactor dead code that
I don't have a way to know if I broke.  Let's just remove it for now.  We
can put it back in easily enough later and then we'll do it properly.

9 years agoanv/meta: Use the actual render pass for creating blit pipelines
Jason Ekstrand [Tue, 20 Oct 2015 20:01:15 +0000 (13:01 -0700)]
anv/meta: Use the actual render pass for creating blit pipelines

9 years agovbo: convert display list GL_LINE_LOOP prims to GL_LINE_STRIP
Brian Paul [Fri, 16 Oct 2015 04:57:08 +0000 (22:57 -0600)]
vbo: convert display list GL_LINE_LOOP prims to GL_LINE_STRIP

When a long GL_LINE_LOOP prim was split across primitives we drew
stray lines.  See previous commit for details.

This patch converts GL_LINE_LOOP prims into GL_LINE_STRIP prims so
that drivers don't have to worry about the _mesa_prim::begin/end flags.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81174

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Acked-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: fix GL_LINE_LOOP stray line bug
Brian Paul [Fri, 16 Oct 2015 04:31:50 +0000 (22:31 -0600)]
vbo: fix GL_LINE_LOOP stray line bug

When long GL_LINE_LOOP primitives don't fit in one vertex buffer they
have to be split across buffers.  The code to do this was basically correct
but drivers had to pay special attention to the _mesa_prim::begin,end flags
in order to draw the sections of the line loop properly.  Apparently, the
only drivers to do this were those using the old 'tnl' module for software
vertex processing.

Now we convert the split pieces of GL_LINE_LOOP prims into GL_LINE_STRIP
primitives so that drivers don't have to worry about the special begin/end
flags.  The only time a driver will get a GL_LINE_LOOP prim is when the
whole thing fits in one vertex buffer.

Mostly fixes bug 81174, but not completely.  There's another bug somewhere
in the src/gallium/auxiliary/draw/ code.  If the piglit lineloop test is
run with -count 4096, rendering is correct, but with -count 4097 there are
stray lines.  4096 is a magic number in the draw code (search for "4096").

Also note that this does not fix long line loops in display lists.  The
next patch fixes that.

v2: fix incorrect -1 in vbo_compute_max_verts(), per Charmaine.  Remove
incorrect assertion which was added in vbo_copy_vertices().

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81174
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=49779
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=28130

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: add new vbo_compute_max_verts() helper function
Brian Paul [Fri, 16 Oct 2015 17:19:40 +0000 (11:19 -0600)]
vbo: add new vbo_compute_max_verts() helper function

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: simplify some code in vbo_exec_End()
Brian Paul [Fri, 16 Oct 2015 02:33:52 +0000 (20:33 -0600)]
vbo: simplify some code in vbo_exec_End()

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: simplify some code in vbo_copy_vertices()
Brian Paul [Fri, 16 Oct 2015 03:25:18 +0000 (21:25 -0600)]
vbo: simplify some code in vbo_copy_vertices()

As before, use a new 'last_prim' pointer to simplify things.  Plus, add
some const qualifiers.

v2: use 'sz' in another place, per Sinclair.  And update subject line.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: simplify some code in vbo_exec_wrap_buffers()
Brian Paul [Fri, 16 Oct 2015 02:18:32 +0000 (20:18 -0600)]
vbo: simplify some code in vbo_exec_wrap_buffers()

Use a new 'last_prim' pointer to simplify things.

v2: remove unneeded assert(exec->vtx.prim_count > 0)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: replace the comment on vbo_copy_vertices()
Brian Paul [Fri, 16 Oct 2015 03:21:56 +0000 (21:21 -0600)]
vbo: replace the comment on vbo_copy_vertices()

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: make vbo_exec_vtx_wrap() static
Brian Paul [Fri, 16 Oct 2015 02:22:25 +0000 (20:22 -0600)]
vbo: make vbo_exec_vtx_wrap() static

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agovbo: remove unneeded ctx parameter for merge_prims()
Brian Paul [Thu, 15 Oct 2015 20:11:54 +0000 (14:11 -0600)]
vbo: remove unneeded ctx parameter for merge_prims()

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agotnl: add some comments in render_line_loop code
Brian Paul [Fri, 16 Oct 2015 04:30:22 +0000 (22:30 -0600)]
tnl: add some comments in render_line_loop code

And remove '(void) flags' line which is not needed.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agomesa: simple whitespace fix in texstore.c
Brian Paul [Thu, 15 Oct 2015 15:52:09 +0000 (09:52 -0600)]
mesa: simple whitespace fix in texstore.c

9 years agovbo: reduce number of vertex buffer mappings for vertex attributes
Brian Paul [Thu, 15 Oct 2015 18:33:00 +0000 (12:33 -0600)]
vbo: reduce number of vertex buffer mappings for vertex attributes

Whenever we got a glColor, glNormal, glTexCoord, etc. call outside a
glBegin/End pair, we'd immediately map a vertex buffer to begin
accumulating vertex data.  In some cases, such as with display lists,
this led to excessive vertex buffer mapping.  For example, if we have
a display list such as:

glNewList(42, GL_COMPILE);
glBegin(prim);
glVertex2f();
...
glVertex2f();
glEnd();
glEndList();

Then did:

glColor3f();
glCallList(42);

We'd map a vertex buffer as soon as we saw glColor3f but we'd never
actually write anything to it.  Note that the vertex position data
was put into a vertex buffer during display list compilation.

With this change, we delay mapping the vertex buffer until we actually
have a vertex to write to it (triggered by a glVertex() call).  In the
above case, we no longer map a vertex buffer when setting the color and
calling the list.

For drivers such as VMware's, reducing buffer mappings gives improved
performance.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
9 years agost/mesa: optimize 4-component ubyte glDrawPixels
Brian Paul [Thu, 15 Oct 2015 17:54:06 +0000 (11:54 -0600)]
st/mesa: optimize 4-component ubyte glDrawPixels

If we didn't find a gallium surface format that exactly matched the
glDrawPixels format/type combination, we used some other 32-bit packed
RGBA format and swizzled the whole image in the mesa texstore/format code.

That slow path can be avoided in some common cases by using the
pipe_samper_view's swizzle terms to do the swizzling at texture sampling
time instead.

For now, only GL_RGBA/ubyte and GL_BGRA/ubyte combinations are supported.
In the future other formats and types like GL_UNSIGNED_INT_8_8_8_8 could
be added.

v2: fix incorrect swizzle setup (need to invert the tex format's swizzle)

Reviewed by: Jose Fonseca <jfonseca@vmware.com>

9 years agomesa: make memcpy_texture() non-static
Brian Paul [Thu, 15 Oct 2015 15:54:29 +0000 (09:54 -0600)]
mesa: make memcpy_texture() non-static

So that we can use it directly from the mesa/gallium state tracker.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agost/mesa: check for out-of-memory in st_DrawPixels()
Brian Paul [Thu, 15 Oct 2015 15:38:00 +0000 (09:38 -0600)]
st/mesa: check for out-of-memory in st_DrawPixels()

Before, if make_texture() or st_create_texture_sampler_view() failed
we silently no-op'd the glDrawPixels.  Now, set GL_OUT_OF_MEMORY.
This also allows us to un-nest a bunch of code.

v2: also check if allocation of sv[1] fails, per Jose.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
9 years agost/mesa: use MAX3() instead of MAX2(MAX2) in draw_textured_quad()
Brian Paul [Thu, 15 Oct 2015 15:32:50 +0000 (09:32 -0600)]
st/mesa: use MAX3() instead of MAX2(MAX2) in draw_textured_quad()

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
9 years agomesa: fix incorrect opcode in save_BlendFunci()
Brian Paul [Thu, 15 Oct 2015 14:43:02 +0000 (08:43 -0600)]
mesa: fix incorrect opcode in save_BlendFunci()

Fixes assertion failure with new piglit
arb_draw_buffers_blend-state_set_get test.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomesa: add more cases to print_list() in dlist.c
Brian Paul [Thu, 15 Oct 2015 14:40:54 +0000 (08:40 -0600)]
mesa: add more cases to print_list() in dlist.c

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agovk: Use consistent names for anv_cmd_state dirty bits
Chad Versace [Sat, 17 Oct 2015 03:03:46 +0000 (20:03 -0700)]
vk: Use consistent names for anv_cmd_state dirty bits

Prefix all anv_cmd_state dirty bit tokens with ANV_CMD_DIRTY. For
example:

    old                           -> new
    ANV_DYNAMIC_VIEWPORT_DIRTY    -> ANV_CMD_DIRTY_DYNAMIC_VIEWPORT
    ANV_CMD_BUFFER_PIPELINE_DIRTY -> ANV_CMD_DIRTY_PIPELINE

Change type of anv_cmd_state::dirty and ::compute_dirty from uint32_t to
the self-documenting type anv_cmd_dirty_mask_t.

9 years agoanv/pipeline: Fix requirement for depthstencil state
Chad Versace [Mon, 19 Oct 2015 18:39:30 +0000 (11:39 -0700)]
anv/pipeline: Fix requirement for depthstencil state

The Vulkan spec allows VkGraphicsPipelineCreateInfo::pDepthStencilState
to be NULL when the pipeline's subpass contains no depthstencil
attachment (see spec quote below). anv_pipeline_init_dynamic_state()
required it unconditionally.

This path fixes anv_pipeline_init_dynamic_state() to access
pDepthStencilState only when there is a depthstencil attachment.

From the Vulkan spec (20 Oct 2015, git-aa308cb)

   pDepthStencilState [...] may only be NULL if renderPass and subpass
   specify a subpass that has no depth/stencil attachment.

9 years agoanv/pipeline: Validate VkGraphicsPipelineCreateInfo
Chad Versace [Sat, 17 Oct 2015 03:31:39 +0000 (20:31 -0700)]
anv/pipeline: Validate VkGraphicsPipelineCreateInfo

The Vulkan spec (20 Oct 2015, git-aa308cb) states that some fields of
VkGraphicsPipelineCreateInfo are required under certain conditions.
Add a new function, anv_pipeline_validate_create_info() that asserts the
requirements hold.

The assertions helped me discover bugs in Crucible and anv_meta.c.

9 years agoanv: Define anv_validate macro
Chad Versace [Tue, 20 Oct 2015 17:52:57 +0000 (10:52 -0700)]
anv: Define anv_validate macro

If a block of code is annotated with anv_validate, then the block runs
only in debug builds.

9 years agovk/meta: Add required renderpass to pipeline
Chad Versace [Mon, 19 Oct 2015 15:44:38 +0000 (08:44 -0700)]
vk/meta: Add required renderpass to pipeline

The Vulkan spec (20 Oct 2015, git-aa308cb) requires that
VkGraphicsPipelineCreateInfo::renderPass be a valid handle. To satisfy
that, define a static dummy render pass used for all meta operations.

9 years agovk/meta: Add required multisample state to pipeline
Chad Versace [Mon, 19 Oct 2015 15:37:03 +0000 (08:37 -0700)]
vk/meta: Add required multisample state to pipeline

The Vulkan spec (20 Oct 2015, git-aa308cb) requires that
VkGraphicsPipelineCreateInfo::pMultisampleState not be NULL.

9 years agoi965: silence incompatible pointer type warning
Emil Velikov [Sat, 17 Oct 2015 22:42:13 +0000 (23:42 +0100)]
i965: silence incompatible pointer type warning

src/mesa/drivers/dri/i965/brw_program.c:94:39:
warning: passing argument 1 of ‘_mesa_init_gl_program’ from incompatible
pointer type [-Wincompatible-pointer-types]
          return _mesa_init_gl_program(&prog->program, target, id);

                                       ^

Runtime was unaffected as brw_geometry_program is subclassed from
gl_geometry_program, thus the address passed was the same.

Fixes: bcb56c2c69d (program: convert _mesa_init_gl_program() to take
struct gl_program *)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agogallium: add PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
Marek Olšák [Tue, 20 Oct 2015 16:26:02 +0000 (18:26 +0200)]
gallium: add PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT

This avoids a serious r600g bug leading to a GPU hang.
The chances this bug will get fixed are pretty low now.

I deeply regret listening to others and not pushing this patch, leaving
other users with a GPU-crashing driver. Yes, it should be fixed
in the compiler and it's ugly, but users couldn't care less about that.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86720

Cc: 11.0 10.6 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
9 years agovc4: Switch our vertex attr lowering to being NIR-based.
Eric Anholt [Sat, 1 Aug 2015 03:58:57 +0000 (20:58 -0700)]
vc4: Switch our vertex attr lowering to being NIR-based.

This exposes more information to NIR's optimization, and should be
particularly useful when we do range-based optimization.

total uniforms in shared programs: 32066 -> 32065 (-0.00%)
uniforms in affected programs:     21 -> 20 (-4.76%)
total instructions in shared programs: 93104 -> 92630 (-0.51%)
instructions in affected programs:     31901 -> 31427 (-1.49%)

9 years agovc4: Add limited support for ibfe/ubfe.
Eric Anholt [Tue, 20 Oct 2015 09:49:10 +0000 (10:49 +0100)]
vc4: Add limited support for ibfe/ubfe.

This is just enough to cover our unpack modes, which will be used by some
new NIR-based lowering in the next commit.

9 years agotgsi/scan: use properties for clip/cull distance writemasks
Marek Olšák [Sat, 17 Oct 2015 21:17:52 +0000 (23:17 +0200)]
tgsi/scan: use properties for clip/cull distance writemasks

No changes needed for drivers already relying on tgsi_shader_info.

Reviewed-by: Brian Paul <brianp@vmware.com>
9 years agost/mesa: pass the clip distance array size to drivers
Marek Olšák [Sat, 17 Oct 2015 21:17:52 +0000 (23:17 +0200)]
st/mesa: pass the clip distance array size to drivers

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
9 years agogallium: add new properties for clip and cull distance usage
Marek Olšák [Sat, 17 Oct 2015 21:15:28 +0000 (23:15 +0200)]
gallium: add new properties for clip and cull distance usage

The TGSI usage mask can't be used, because these are declared as an output
array of 2 elements.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
9 years agomesa: replace UsesClipDistance with ClipDistanceArraySize
Marek Olšák [Sat, 17 Oct 2015 20:50:11 +0000 (22:50 +0200)]
mesa: replace UsesClipDistance with ClipDistanceArraySize

This is more practical and needed by gallium.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
9 years agoradeonsi: enable BC_OPTIMIZE if centroid isn't used
Marek Olšák [Sun, 18 Oct 2015 13:09:24 +0000 (15:09 +0200)]
radeonsi: enable BC_OPTIMIZE if centroid isn't used

This solution was recommended by a Catalyst developer.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
9 years agoradeonsi: fix the export_prim_id field size in the shader key
Marek Olšák [Sat, 17 Oct 2015 21:59:52 +0000 (23:59 +0200)]
radeonsi: fix the export_prim_id field size in the shader key

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>