mesa.git
7 years agost/nine: Avoid crash on empty Draw*Up
Axel Davy [Sun, 23 Oct 2016 20:27:30 +0000 (22:27 +0200)]
st/nine: Avoid crash on empty Draw*Up

Ignore empty draw calls.
Avoid assertion fault when such draw calls happen
in u_upload_mgr.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Capture texturestage states in pixel stateblocks
Axel Davy [Sun, 23 Oct 2016 20:26:23 +0000 (22:26 +0200)]
st/nine: Capture texturestage states in pixel stateblocks

pixels stateblocks need to capture these.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add missing changed states to pixel stateblocks
Axel Davy [Sun, 23 Oct 2016 10:16:38 +0000 (12:16 +0200)]
st/nine: Add missing changed states to pixel stateblocks

Some states were not properly recorded in pixel stateblocks.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add some debug info in stateblocks
Axel Davy [Sun, 23 Oct 2016 10:16:06 +0000 (12:16 +0200)]
st/nine: Add some debug info in stateblocks

This is useful to check what is exactly recorded.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Remove useless check in surface9 ctor
Axel Davy [Fri, 21 Oct 2016 15:14:14 +0000 (17:14 +0200)]
st/nine: Remove useless check in surface9 ctor

Textures already have the check in BaseTexture9.
Non-Textures cannot be in the MANAGED Pool.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix bad light initialization in stateblocks
Axel Davy [Wed, 19 Oct 2016 22:00:57 +0000 (00:00 +0200)]
st/nine: Fix bad light initialization in stateblocks

src was initialized instead of dst.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Remove unused ff.changed.group
Axel Davy [Wed, 19 Oct 2016 20:18:56 +0000 (22:18 +0200)]
st/nine: Remove unused ff.changed.group

It was unused.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix ps multisample check
Axel Davy [Wed, 19 Oct 2016 19:08:57 +0000 (21:08 +0200)]
st/nine: Fix ps multisample check

We want to use centroid for nonmaskable
multisampling as well.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix useless swapchain init checks
Axel Davy [Wed, 19 Oct 2016 18:57:48 +0000 (20:57 +0200)]
st/nine: Fix useless swapchain init checks

In NineDevice9_SetDefaultState we can assume the
implicit swapchain is properly initialized.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Don't update stream_usage_mask in sw path
Axel Davy [Sun, 16 Oct 2016 15:47:56 +0000 (17:47 +0200)]
st/nine: Don't update stream_usage_mask in sw path

The variable is used only in the hw path.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Remove useless call to nine_update_state
Axel Davy [Sun, 16 Oct 2016 08:54:41 +0000 (10:54 +0200)]
st/nine: Remove useless call to nine_update_state

The call was not needed.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add validation to SetSamplerState
Axel Davy [Sat, 16 May 2015 14:46:51 +0000 (16:46 +0200)]
st/nine: Add validation to SetSamplerState

Check value validity and mimick Win behaviour.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Improve doc of D3DPMISCCAPS_POSTBLENDSRGBCONVERT
Axel Davy [Sat, 9 Jan 2016 12:07:27 +0000 (13:07 +0100)]
st/nine: Improve doc of D3DPMISCCAPS_POSTBLENDSRGBCONVERT

The cap should be advertised for d3d10 able cards,
but only for Ex contexts.

Unfortunately at this point Mesa has no way to know if
Ex is used or not (the info is got later).

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agogallium-docs: Add documentation for when using several contexts
Axel Davy [Mon, 19 Dec 2016 19:06:51 +0000 (20:06 +0100)]
gallium-docs: Add documentation for when using several contexts

Add documentation to explicit what can be expected and what is allowed
when using several contexts.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agogallium-docs: Add documentation for threading requirements
Axel Davy [Mon, 19 Dec 2016 18:51:13 +0000 (19:51 +0100)]
gallium-docs: Add documentation for threading requirements

Add documentation for the requirements related to threading
for screens and contexts.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agoegl: Check config's surface types in eglCreate*Surface()
Chad Versace [Fri, 16 Dec 2016 19:00:13 +0000 (11:00 -0800)]
egl: Check config's surface types in eglCreate*Surface()

If the provided EGLConfig does not support the requested surface type,
then emit EGL_BAD_MATCH.

Fixes dEQP-EGL.functional.negative_api.create_pbuffer_surface
on GBM.

Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoglsl: Use ir_var_temporary when generating inline functions.
Kenneth Graunke [Mon, 19 Dec 2016 22:32:57 +0000 (14:32 -0800)]
glsl: Use ir_var_temporary when generating inline functions.

We were using ir_var_auto for the inlined function parameter variables,
which is wrong, as it suggests that those are real variables declared
by the program.

Normally this doesn't matter.  However, if you called built-ins at
global scope, it would pollute the global variable namespace with
these new parameter temporaries.  If the shader already had variables
with those names, the linker might see contradictory global variable
declarations and raise an error.

Making them temporaries indicates that these are just things generated
by the compiler internally.  This avoids confusing the linker.

Fixes a new Piglit test: glsl-fs-multiple-builtins.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99154
Reported-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
7 years agoi965: Don't bail on vertex element processing if we need draw params.
Kenneth Graunke [Mon, 19 Dec 2016 19:21:27 +0000 (11:21 -0800)]
i965: Don't bail on vertex element processing if we need draw params.

BaseVertex, BaseInstance, DrawID, and some edge flag conditions need
vertex buffer and elements structs.  We can't bail early in this case.

Gen4-7 already do this properly.  Gen8+ did not.

Thanks to Ilia Mirkin for helping track this down.

Cc: mesa-stable@lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99144
Reported-by: Pierre-Eric Pelloux-Prayer <pelloux@gmail.com>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agomesa: don't attempt to unlock an unlocked debug state mutex
Jonathan Gray [Sun, 11 Dec 2016 15:42:16 +0000 (02:42 +1100)]
mesa: don't attempt to unlock an unlocked debug state mutex

Commit 929fcee47e46781c57f2a354ce0a013915c033d1 introduced code that
attempts to unlock an unlocked mutex which is undefined behaviour.

On OpenBSD this leads to an abort:

0  0x0000124dadfa96ba in thrkill () at <stdin>:2
1  0x0000124dadf3da39 in *_libc_abort () at /usr/src/lib/libc/stdlib/abort.c:52
2  0x0000124d2c1165b5 in *_libpthread_pthread_mutex_unlock (mutexp=<optimized out>)
    at /usr/src/lib/librthread/rthread_sync.c:221
3  0x0000124d279c02e4 in init_attrib_groups (ctx=0x124df0fda000) at main/context.c:825
4  _mesa_initialize_context (ctx=ctx@entry=0x124df0fda000, api=api@entry=API_OPENGL_CORE,
    visual=visual@entry=0x7f7ffffbdfd0, share_list=share_list@entry=0x0,
    driverFunctions=driverFunctions@entry=0x7f7ffffbda60) at main/context.c:1204
5  0x0000124d27b507ec in st_create_context (api=api@entry=API_OPENGL_CORE,
    pipe=pipe@entry=0x124dc4910000, visual=visual@entry=0x7f7ffffbdfd0,
    share=share@entry=0x0, options=options@entry=0x7f7ffffbe128)
    at state_tracker/st_context.c:545
6  0x0000124d27b8639f in st_api_create_context (stapi=<optimized out>,
    smapi=0x124d1b608800, attribs=0x7f7ffffbe100, error=0x7f7ffffbe0fc, shared_stctxi=0x0)
    at state_tracker/st_manager.c:669
7  0x0000124d27cc5b9c in dri_create_context (api=<optimized out>, visual=0x124d8a0f8a00,
    cPriv=0x124de473f240, major_version=<optimized out>, minor_version=<optimized out>,
    flags=<optimized out>, notify_reset=false, error=0x7f7ffffbe2b4,
    sharedContextPrivate=0x0) at dri_context.c:123
8  0x0000124d27cc5029 in driCreateContextAttribs (screen=0x124d8a0f8400,
    api=<optimized out>, config=0x124d8a0f8a00, shared=<optimized out>,
    num_attribs=<optimized out>, attribs=<optimized out>, error=0x7f7ffffbe2b4,
    data=0x124d77814a00) at dri_util.c:448
9  0x0000124d8e109b00 in drisw_create_context_attribs (base=0x124df3e08700,
    config_base=0x124d7a0e7300, shareList=<optimized out>, num_attribs=<optimized out>,
    attribs=<optimized out>, error=0x7f7ffffbe2b4) at drisw_glx.c:476
10 0x0000124d8e104b4a in glXCreateContextAttribsARB (dpy=0x124d533f0000,
    config=0x124d7a0e7300, share_context=0x0, direct=1, attrib_list=0x7f7ffffbe300)
    at create_context.c:78

Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
7 years agoglsl: allow invariant on fragment shader outputs.
Dave Airlie [Mon, 23 May 2016 04:18:03 +0000 (14:18 +1000)]
glsl: allow invariant on fragment shader outputs.

From page 27 (page 33 of the PDF) of the GLSL 1.20 spec:

    " Only variables output from a vertex shader can be candidates for
      invariance."

But this later changes to:

From page 37 (page 43 of the PDF) of the GLSL 1.30 spec:

    " Only variables output from a shader can be candidates for
      invariance."

We can also find:

From page 37 (page 43 of the PDF) of the GLSL 1.30 spec:

    " Initially, by default, all output variables are allowed to be
      variant. To force all output variables to be invariant, use the
      pragma

        #pragma STDGL invariant(all)

      before all declarations in a shader. If this pragma is used after
      the declaration of any variables or functions, then the set of
      outputs that behave as invariant is undefined. It is an error to
      use this pragma in a fragment shader."

But this needs to be corrected and it is being addressed at:
https://cvs.khronos.org/bugzilla/show_bug.cgi?id=16140

Fixes GL45-CTS.shading_language_420pack.qualifier_order.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Andres Gomez <agomez@igalia.com>
7 years agoi965: keep gl_program shader info in sync after gather info
Timothy Arceri [Thu, 1 Dec 2016 02:37:38 +0000 (13:37 +1100)]
i965: keep gl_program shader info in sync after gather info

It's possible that nir_shader was cloned and it no longer contains
a pointer to the shader_info in gl_program. So we need to copy
shader_info back to gl_program if that is the case.

Fixes a regression with NIR_TEST_CLONE=true

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98840

7 years agonir: Trivial clean ups in the generated nir_constant_expressions.c
Ian Romanick [Tue, 25 Aug 2015 17:11:17 +0000 (18:11 +0100)]
nir: Trivial clean ups in the generated nir_constant_expressions.c

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agonir: Silence unused parameter warnings in nir_constant_expression.c
Ian Romanick [Tue, 25 Aug 2015 16:44:00 +0000 (17:44 +0100)]
nir: Silence unused parameter warnings in nir_constant_expression.c

nir/nir_constant_expressions.c:290:25: warning: unused parameter 'num_components' [-Wunused-parameter]
 evaluate_ball3(unsigned num_components, nir_const_value *_src)
                         ^
nir/nir_constant_expressions.c: In function 'evaluate_fddx':
nir/nir_constant_expressions.c:1282:57: warning: unused parameter '_src' [-Wunused-parameter]
 evaluate_fddx(unsigned num_components, nir_const_value *_src)
                                                         ^

v2: Unconditionally mark the parameters as MAYBE_UNUSED instead of
conditionally adding (void) casts to keep the generator simple.
Suggested by Jason.

Number of total warnings in my build reduced from 1575 to 1485
(reduction of 89).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agonir: Silence missing field initializer warnings for vectors in nir_constant_expressions
Ian Romanick [Tue, 25 Aug 2015 16:21:47 +0000 (17:21 +0100)]
nir: Silence missing field initializer warnings for vectors in nir_constant_expressions

nir/nir_constant_expressions.c: In function 'evaluate_ball2':
nir/nir_constant_expressions.c:279:7: warning: missing initializer for field 'z' of 'struct bool_vec' [-Wmissing-field-initializers]
       };
       ^
nir/nir_constant_expressions.c:234:10: note: 'z' declared here
    bool z;
          ^

Number of total warnings in my build reduced from 2532 to 2304
(reduction of 228).

v2: Initialize bool vectors with 0 instead of false to keep the
generator simpler.  Suggested by Ken.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Silence "unused parameter" warnings in ast_type.cpp
Ian Romanick [Mon, 12 Dec 2016 21:35:37 +0000 (13:35 -0800)]
glsl: Silence "unused parameter" warnings in ast_type.cpp

glsl/ast_type.cpp: In function ‘bool validate_point_mode(YYLTYPE*, _mesa_glsl_parse_state*, const ast_type_qualifier&, const ast_type_qualifier&)’:
glsl/ast_type.cpp:173:30: warning: unused parameter ‘loc’ [-Wunused-parameter]
 validate_point_mode(YYLTYPE *loc,
                              ^~~
glsl/ast_type.cpp:174:45: warning: unused parameter ‘state’ [-Wunused-parameter]
                     _mesa_glsl_parse_state *state,
                                             ^~~~~

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Trivial whitespace fixes in link_uniforms.cpp
Ian Romanick [Mon, 12 Dec 2016 20:13:13 +0000 (12:13 -0800)]
glsl: Trivial whitespace fixes in link_uniforms.cpp

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Silence unused parameter warning in propagate_invariance.cpp
Ian Romanick [Mon, 12 Dec 2016 19:19:20 +0000 (11:19 -0800)]
glsl: Silence unused parameter warning in propagate_invariance.cpp

glsl/propagate_invariance.cpp: In member function ‘virtual ir_visitor_status {anonymous}::ir_invariance_propagation_visitor::visit_leave(ir_assignment*)’:
glsl/propagate_invariance.cpp:86:63: warning: unused parameter ‘ir’ [-Wunused-parameter]
 ir_invariance_propagation_visitor::visit_leave(ir_assignment *ir)
                                                               ^~

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Minor formatting fixes in link_uniform_blocks.cpp
Ian Romanick [Mon, 12 Dec 2016 18:56:05 +0000 (10:56 -0800)]
glsl: Minor formatting fixes in link_uniform_blocks.cpp

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Fix all the whitespace errors in link_uniform_block_active_visitor.cpp
Ian Romanick [Fri, 9 Dec 2016 23:38:24 +0000 (15:38 -0800)]
glsl: Fix all the whitespace errors in link_uniform_block_active_visitor.cpp

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agomesa: Silence numerous "unused parameter" warnings in dlist.c
Ian Romanick [Fri, 9 Dec 2016 21:44:09 +0000 (13:44 -0800)]
mesa: Silence numerous "unused parameter" warnings in dlist.c

main/dlist.c: In function ‘save_DrawArraysInstancedARB’:
main/dlist.c:1748:36: warning: unused parameter ‘mode’ [-Wunused-parameter]
 save_DrawArraysInstancedARB(GLenum mode,
                                    ^~~~
main/dlist.c:1749:35: warning: unused parameter ‘first’ [-Wunused-parameter]
                             GLint first,
                                   ^~~~~
main/dlist.c:1750:37: warning: unused parameter ‘count’ [-Wunused-parameter]
                             GLsizei count,
                                     ^~~~~
main/dlist.c:1751:37: warning: unused parameter ‘primcount’ [-Wunused-parameter]
                             GLsizei primcount)
                                     ^~~~~~~~~
main/dlist.c: In function ‘save_DrawElementsInstancedARB’:
main/dlist.c:1759:38: warning: unused parameter ‘mode’ [-Wunused-parameter]
 save_DrawElementsInstancedARB(GLenum mode,
                                      ^~~~
main/dlist.c:1760:39: warning: unused parameter ‘count’ [-Wunused-parameter]
                               GLsizei count,
                                       ^~~~~
main/dlist.c:1761:38: warning: unused parameter ‘type’ [-Wunused-parameter]
                               GLenum type,
                                      ^~~~
main/dlist.c:1762:45: warning: unused parameter ‘indices’ [-Wunused-parameter]
                               const GLvoid *indices,
                                             ^~~~~~~
main/dlist.c:1763:39: warning: unused parameter ‘primcount’ [-Wunused-parameter]
                               GLsizei primcount)
                                       ^~~~~~~~~
main/dlist.c: In function ‘save_DrawElementsInstancedBaseVertexARB’:
main/dlist.c:1771:48: warning: unused parameter ‘mode’ [-Wunused-parameter]
 save_DrawElementsInstancedBaseVertexARB(GLenum mode,
                                                ^~~~
main/dlist.c:1772:49: warning: unused parameter ‘count’ [-Wunused-parameter]
                                         GLsizei count,
                                                 ^~~~~
main/dlist.c:1773:48: warning: unused parameter ‘type’ [-Wunused-parameter]
                                         GLenum type,
                                                ^~~~
main/dlist.c:1774:55: warning: unused parameter ‘indices’ [-Wunused-parameter]
                                         const GLvoid *indices,
                                                       ^~~~~~~
main/dlist.c:1775:49: warning: unused parameter ‘primcount’ [-Wunused-parameter]
                                         GLsizei primcount,
                                                 ^~~~~~~~~
main/dlist.c:1776:47: warning: unused parameter ‘basevertex’ [-Wunused-parameter]
                                         GLint basevertex)
                                               ^~~~~~~~~~
main/dlist.c: In function ‘save_DrawArraysInstancedBaseInstance’:
main/dlist.c:1785:45: warning: unused parameter ‘mode’ [-Wunused-parameter]
 save_DrawArraysInstancedBaseInstance(GLenum mode,
                                             ^~~~
main/dlist.c:1786:44: warning: unused parameter ‘first’ [-Wunused-parameter]
                                      GLint first,
                                            ^~~~~
main/dlist.c:1787:46: warning: unused parameter ‘count’ [-Wunused-parameter]
                                      GLsizei count,
                                              ^~~~~
main/dlist.c:1788:46: warning: unused parameter ‘primcount’ [-Wunused-parameter]
                                      GLsizei primcount,
                                              ^~~~~~~~~
main/dlist.c:1789:45: warning: unused parameter ‘baseinstance’ [-Wunused-parameter]
                                      GLuint baseinstance)
                                             ^~~~~~~~~~~~
main/dlist.c: In function ‘save_DrawElementsInstancedBaseInstance’:
main/dlist.c:1797:47: warning: unused parameter ‘mode’ [-Wunused-parameter]
 save_DrawElementsInstancedBaseInstance(GLenum mode,
                                               ^~~~
main/dlist.c:1798:48: warning: unused parameter ‘count’ [-Wunused-parameter]
                                        GLsizei count,
                                                ^~~~~
main/dlist.c:1799:47: warning: unused parameter ‘type’ [-Wunused-parameter]
                                        GLenum type,
                                               ^~~~
main/dlist.c:1800:52: warning: unused parameter ‘indices’ [-Wunused-parameter]
                                        const void *indices,
                                                    ^~~~~~~
main/dlist.c:1801:48: warning: unused parameter ‘primcount’ [-Wunused-parameter]
                                        GLsizei primcount,
                                                ^~~~~~~~~
main/dlist.c:1802:47: warning: unused parameter ‘baseinstance’ [-Wunused-parameter]
                                        GLuint baseinstance)
                                               ^~~~~~~~~~~~
main/dlist.c: In function ‘save_DrawElementsInstancedBaseVertexBaseInstance’:
main/dlist.c:1810:57: warning: unused parameter ‘mode’ [-Wunused-parameter]
 save_DrawElementsInstancedBaseVertexBaseInstance(GLenum mode,
                                                         ^~~~
main/dlist.c:1811:58: warning: unused parameter ‘count’ [-Wunused-parameter]
                                                  GLsizei count,
                                                          ^~~~~
main/dlist.c:1812:57: warning: unused parameter ‘type’ [-Wunused-parameter]
                                                  GLenum type,
                                                         ^~~~
main/dlist.c:1813:62: warning: unused parameter ‘indices’ [-Wunused-parameter]
                                                  const void *indices,
                                                              ^~~~~~~
main/dlist.c:1814:58: warning: unused parameter ‘primcount’ [-Wunused-parameter]
                                                  GLsizei primcount,
                                                          ^~~~~~~~~
main/dlist.c:1815:56: warning: unused parameter ‘basevertex’ [-Wunused-parameter]
                                                  GLint basevertex,
                                                        ^~~~~~~~~~
main/dlist.c:1816:57: warning: unused parameter ‘baseinstance’ [-Wunused-parameter]
                                                  GLuint baseinstance)
                                                         ^~~~~~~~~~~~

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agomesa: Fix all the whitespace errors in dlist.c
Ian Romanick [Fri, 9 Dec 2016 21:38:44 +0000 (13:38 -0800)]
mesa: Fix all the whitespace errors in dlist.c

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agolinker: Accurately mark a uniform block instance array element as used in a stage
Ian Romanick [Wed, 14 Dec 2016 04:33:08 +0000 (20:33 -0800)]
linker: Accurately mark a uniform block instance array element as used in a stage

Now that information about which array-of-arrays elements are accessed
is tracked, use that information to only mark an instance array element
as used-by-stage if, in fact, it is.

Fixes GL45-CTS.program_interface_query.uniform-block-types.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Walk a list of ir_dereference_array to mark array elements as accessed
Ian Romanick [Thu, 15 Dec 2016 22:26:24 +0000 (14:26 -0800)]
glsl: Walk a list of ir_dereference_array to mark array elements as accessed

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Mark a set of array elements as accessed using a list of array_deref_range
Ian Romanick [Thu, 15 Dec 2016 22:01:28 +0000 (14:01 -0800)]
glsl: Mark a set of array elements as accessed using a list of array_deref_range

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Add structures to track accessed elements of a single array
Ian Romanick [Wed, 14 Dec 2016 04:31:19 +0000 (20:31 -0800)]
glsl: Add structures to track accessed elements of a single array

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Add tracking for elements of an array-of-arrays that have been accessed
Ian Romanick [Wed, 14 Dec 2016 04:29:23 +0000 (20:29 -0800)]
glsl: Add tracking for elements of an array-of-arrays that have been accessed

If there's a better way to provide access to ir_array_refcount_entry
private members to the test functions, I am very interested to know
about it.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: Francisco Jerez <currojerez@riseup.net>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Use simpler visitor to determine which UBO and SSBO blocks are used
Ian Romanick [Mon, 12 Dec 2016 21:37:46 +0000 (13:37 -0800)]
glsl: Use simpler visitor to determine which UBO and SSBO blocks are used

Very soon this visitor will get more complicated.  The users of the
existing ir_variable_refcount visitor won't need the coming
functionality, and this use doesn't need much of the functionality of
ir_variable_refcount.

v2: ir_array_refcount_visitor::get_variable_entry cannot return NULL, so
don't check it.  Suggested by Timothy.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Track the linearized array index for each UBO instance array element
Ian Romanick [Mon, 12 Dec 2016 20:24:18 +0000 (12:24 -0800)]
glsl: Track the linearized array index for each UBO instance array element

v2: Set linearizer_array_index in process_block_array_leaf.  Suggested
by Timothy.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Fix wonkey indentation left from previous commit
Ian Romanick [Mon, 12 Dec 2016 19:58:06 +0000 (11:58 -0800)]
glsl: Fix wonkey indentation left from previous commit

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Split process_block_array into two functions
Ian Romanick [Mon, 12 Dec 2016 19:54:41 +0000 (11:54 -0800)]
glsl: Split process_block_array into two functions

One for the array parts and one for the leaf members.  This will
simplify later changes.

The indentation is wonkey after this patch.  This was done to make it
more obvious that the function is just getting split.  The next patch
will fix the indentation.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoglsl: Fix program interface queries relating to interface blocks.
Kenneth Graunke [Sat, 17 Dec 2016 01:03:15 +0000 (17:03 -0800)]
glsl: Fix program interface queries relating to interface blocks.

This fixes 555 dEQP tests (using the nougat-cts-dev branch), Piglit's
arb_program_interface_query/arb_program_interface_query-resource-query,
and GL45-CTS.program_interface_query.separate-programs-{tess-control,
tess-eval,geometry}.  Only one dEQP program interface failure remains.

I would have liked to split this up into several distinct changes, but
I wasn't sure how to do that given thet tangled nature of these issues.

So, the issues:

   * We need to treat interface blocks declared as an array of instances
     as a single block - removing the outer array.  The resource list
     entry's name should not include the array length.  Properties such
     as GL_ARRAY_SIZE should refer to the variable inside the block, not
     the interface block's array properties.

   * We need to do this prefixing even for structure variables.

   * We need to do this for built-ins (such as gl_PerVertex.gl_Position).

   * After interface array unwrapping, any variable which is an array
     should have [0] appended.  It doesn't matter if it's a TCS/TES/GS
     input or TCS output - that looked like an attempt to unwrap for
     per-vertex variables, but that didn't consider per-patch variables,
     and as far as I can tell there's nothing to justify this.

Several Mesa developers have suggested that Issue 16 contradicts the
main specification, but I believe that it doesn't - the main spec just
isn't terribly clear.  The main ARB_program_interface query spec says:

  "* For an active interface block not declared as an array of block
     instances, a single entry will be generated, using the block name from
     the shader source.

   * For an active interface block declared as an array of instances,
     separate entries will be generated for each active instance.  The name
     of the instance is formed by concatenating the block name, the "["
     character, an integer identifying the instance number, and the "]"
     character."

Issue 16 says that built-ins should be named "gl_PerVertex.gl_Position",
but several people suggested the second bullet above means that it
should be named "gl_PerVertex[array length].gl_Position".

There are two important things to note.  Those bullet points say
"an active interface block", while the others say "variable" or "active
shader storage block member".  They also don't mention applying the
rules recursively (unlike the other bullets).  Both suggest that
these rules apply to blocks themselves, not members of blocks.

In fact, for GL_UNIFORM_BLOCK queries, we do have "block[0]",
"block[1]", ... resource list entries - so those rules are real,
and actually used.  So if they don't apply to block members, then how
should members be named?  Unfortunately, I don't see any rules outside
of issue 16 - where the rationale is very unclear.  I hope to clarify
the spec in the future.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
7 years agoglsl: Drop bogus is_vertex_input from add_shader_variable().
Kenneth Graunke [Fri, 16 Dec 2016 12:54:59 +0000 (04:54 -0800)]
glsl: Drop bogus is_vertex_input from add_shader_variable().

stage_mask is a bitmask of shader stages, so the proper comparison would
be (1 << MESA_SHADER_VERTEX), not MESA_SHADER_VERTEX itself.

But we only care for structure types, and VS inputs cannot be structs.
So we can just drop this entirely.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
7 years agomesa/get: Convert stencil values to TYPE_UINT.
Kenneth Graunke [Sun, 18 Dec 2016 04:04:32 +0000 (20:04 -0800)]
mesa/get: Convert stencil values to TYPE_UINT.

These are listed as Z+ in the GL spec, and often have values of
0xFFFFFFFF.  For glGetFloat, we should return 4294967295.0 rather than
-1.0.  Similarly, for glGetInteger64v, we should return 0xFFFFFFFF, not
the sign extended 0xFFFFFFFFFFFFFFFF.

Fixes 6 dEQP tests matching the pattern
dEQP-GLES3.functional.state_query.integers.stencil*value*mask*getfloat
when run in a single process (with state reset code happening between
tests, which makes dEQP set the stencil value mask to 0xFFFFFFFF).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
7 years agomesa/get: Add TYPE_UINT for casting through a GLuint.
Kenneth Graunke [Sun, 18 Dec 2016 03:55:53 +0000 (19:55 -0800)]
mesa/get: Add TYPE_UINT for casting through a GLuint.

The "State Tables" section of the OpenGL specification lists many values
as belonging to Z+ (non-negative integers), not Z (all integers).

For ordinary glGetInteger queries, this doesn't matter.  However, when
accessing Z+ values via glGetFloat or glGetInteger64, we need to treat
the source value as an unsigned value.  Otherwise, we'll produce a
negative number when bit 31 is set.

This commit merely adds the plumbing.  It doesn't convert any values.

v2: Gotta catch 'em all (add missing cases caught by Ilia)

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
7 years agomesa/get: Make GetFloat/GetDouble of TYPE_INT_N not normalize things.
Kenneth Graunke [Sun, 18 Dec 2016 03:47:38 +0000 (19:47 -0800)]
mesa/get: Make GetFloat/GetDouble of TYPE_INT_N not normalize things.

GetFloat of integer valued things is supposed to perform a simple
int -> float conversion.  INT_TO_FLOAT is not that.  Instead, it
converts [-21474836482147483647] to a normalized [-1.0, 1.0] float.

This is only used for COMPRESSED_TEXTURE_FORMATS, which nobody in
their right mind would try and access via glGetFloat(), but we may
as well fix it.

Found by inspection.

v2: Gotta catch 'em all (fix another case of this caught by Ilia)

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
7 years agoRevert "cso: don't release sampler states that are bound"
Michel Dänzer [Fri, 16 Dec 2016 03:11:30 +0000 (12:11 +0900)]
Revert "cso: don't release sampler states that are bound"

This reverts commit 6dc96de303290e8d1fc294da478c4f370be98dea. No longer
necessary with the previous change.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agocso: Make sanitize_hash safe for samplers
Michel Dänzer [Fri, 16 Dec 2016 02:17:44 +0000 (11:17 +0900)]
cso: Make sanitize_hash safe for samplers

Remove currently bound sampler states from the hash table before pruning
entries from the hash table, so they cannot accidentally be deleted by
the pruning.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agocso: Store hash key in struct cso_sampler
Michel Dänzer [Fri, 16 Dec 2016 02:41:54 +0000 (11:41 +0900)]
cso: Store hash key in struct cso_sampler

Preparation for following changes, no functional change intended.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agocso: Optimize cso_save/restore_fragment_samplers
Michel Dänzer [Fri, 16 Dec 2016 09:00:33 +0000 (18:00 +0900)]
cso: Optimize cso_save/restore_fragment_samplers

Only copy/memset the pointers that actually need to be.

v2:
* Cast info->nr_samplers to int for calculating delta (Nicolai)

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agocso: Store pointers to struct cso_sampler in struct sampler_info
Michel Dänzer [Fri, 16 Dec 2016 02:13:26 +0000 (11:13 +0900)]
cso: Store pointers to struct cso_sampler in struct sampler_info

Preparation for following changes, no functional change intended.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agocso: Don't restore nr_samplers in cso_restore_fragment_samplers
Michel Dänzer [Fri, 16 Dec 2016 09:05:58 +0000 (18:05 +0900)]
cso: Don't restore nr_samplers in cso_restore_fragment_samplers

If info->nr_samplers > ctx->nr_fragment_samplers_saved, the assignment
would prevent cso_single_sampler_done from unbinding the no longer used
samplers from the driver, which could result in use-after-free. This is
probably unlikely to happen in practice though.

Cc: "12.0 13.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoEGL/android: Enhance pbuffer implementation
Liu Zhiquan [Fri, 9 Dec 2016 11:29:56 +0000 (19:29 +0800)]
EGL/android: Enhance pbuffer implementation

Some dri drivers will pass multiple bits in buffer_mask parameter
to droid_image_get_buffer(), more than the actual supported buffer
type combination. For such case, will go through all the bits, and
will not return error when unsupported buffer is requested, only
return error when the allocation for supported buffer failed.

v2: coding style and log changes
v3: coding style changes and update patch format

Signed-off-by: Liu Zhiquan <zhiquan.liu@intel.com>
Signed-off-by: Long, Zhifang <zhifang.long@intel.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
7 years agoradv: Use correct workgroup size limits.
Bas Nieuwenhuizen [Sun, 18 Dec 2016 20:09:28 +0000 (21:09 +0100)]
radv: Use correct workgroup size limits.

Not sure where the 16k comes from, but pretty sure 2k is the max.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: expose the compute queue
Dave Airlie [Wed, 30 Nov 2016 04:08:10 +0000 (04:08 +0000)]
radv: expose the compute queue

v2: Don't expose the SDMA queue and use the CIK check also in the
    second if. (Bas)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: Only emit PFP ME syncs for DMA on the GFX queue.
Bas Nieuwenhuizen [Sun, 18 Dec 2016 13:05:19 +0000 (14:05 +0100)]
radv: Only emit PFP ME syncs for DMA on the GFX queue.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Create an empty CS per ring type.
Bas Nieuwenhuizen [Sat, 17 Dec 2016 20:53:38 +0000 (21:53 +0100)]
radv: Create an empty CS per ring type.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Don't enable CMASK on compute queues.
Bas Nieuwenhuizen [Sat, 17 Dec 2016 20:25:32 +0000 (21:25 +0100)]
radv: Don't enable CMASK on compute queues.

We can't fast clear on compute queues.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Use RELEASE_MEM packet for MEC timestamp query.
Bas Nieuwenhuizen [Sat, 17 Dec 2016 12:27:37 +0000 (13:27 +0100)]
radv: Use RELEASE_MEM packet for MEC timestamp query.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Implement indirect dispatch for the MEC.
Bas Nieuwenhuizen [Mon, 12 Dec 2016 07:45:21 +0000 (08:45 +0100)]
radv: Implement indirect dispatch for the MEC.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: update vkCmdUpdateBuffer for the MEC.
Bas Nieuwenhuizen [Mon, 12 Dec 2016 07:42:44 +0000 (08:42 +0100)]
radv: update vkCmdUpdateBuffer for the MEC.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Implement cache flushing for the MEC.
Bas Nieuwenhuizen [Mon, 12 Dec 2016 07:38:00 +0000 (08:38 +0100)]
radv: Implement cache flushing for the MEC.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: add semaphore support
Dave Airlie [Thu, 1 Dec 2016 01:52:31 +0000 (01:52 +0000)]
radv: add semaphore support

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: pass queue index into winsys submission
Dave Airlie [Thu, 1 Dec 2016 01:14:49 +0000 (01:14 +0000)]
radv: pass queue index into winsys submission

This is so we can submit on separate queues if needed

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: init compute queue and avoid initing transfer queues
Dave Airlie [Thu, 1 Dec 2016 00:15:23 +0000 (00:15 +0000)]
radv: init compute queue and avoid initing transfer queues

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv/winsys: Make WaitIdle queue aware.
Bas Nieuwenhuizen [Sat, 17 Dec 2016 18:10:35 +0000 (19:10 +0100)]
radv/winsys: Make WaitIdle queue aware.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv/meta: update header info
Dave Airlie [Wed, 30 Nov 2016 03:10:28 +0000 (03:10 +0000)]
radv/meta: update header info

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: hook compute clears into clear image api.
Dave Airlie [Wed, 30 Nov 2016 03:09:01 +0000 (03:09 +0000)]
radv: hook compute clears into clear image api.

These aren't used yet but we will want to use them when we
implement a separate compute queue.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: clear image implementation for compute queue
Dave Airlie [Wed, 30 Nov 2016 01:45:24 +0000 (01:45 +0000)]
radv: clear image implementation for compute queue

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv/meta: split clear image out into a separate layer clear function
Dave Airlie [Wed, 30 Nov 2016 02:56:59 +0000 (02:56 +0000)]
radv/meta: split clear image out into a separate layer clear function

This will make it easier to add support for clears on compute queues.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: implement image->image copies using compute shader
Dave Airlie [Wed, 30 Nov 2016 00:26:47 +0000 (00:26 +0000)]
radv: implement image->image copies using compute shader

This is required for having a separate compute queue, we
probably can't use this on GFX queue due to DCC.

v2: Set coord_components = 2 for itoi texture fetch. (Bas)

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: add a compute shader implementation for buffer to image
Dave Airlie [Tue, 29 Nov 2016 23:52:55 +0000 (23:52 +0000)]
radv: add a compute shader implementation for buffer to image

This implements the reverse of the current buffer->image path
and can be used when we need to do image transfer on compute queues

This just adds the code turned off as we don't support separate
computes queues yet, and we don't want to use this path on the GFX
queues for DCC reasons.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: Use correct pitch for views with different block size.
Bas Nieuwenhuizen [Sun, 18 Dec 2016 12:28:04 +0000 (13:28 +0100)]
radv: Use correct pitch for views with different block size.

Needed when accessing a comrpessed texture as R32G32B32A32 from a shader. This
was not encountered previously, as we used the CB for the reinterpretation, which
does not use this pitch.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Store queue family in command buffers.
Dave Airlie [Thu, 1 Dec 2016 00:05:29 +0000 (00:05 +0000)]
radv: Store queue family in command buffers.

v2: Added helper (Bas)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: start fixing up queue allocate for multiple queues
Dave Airlie [Wed, 30 Nov 2016 04:30:06 +0000 (04:30 +0000)]
radv: start fixing up queue allocate for multiple queues

v2: Fix error handling  and zero init the device (Bas)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv/winsys: start adding support for DMA/compute queue
Dave Airlie [Fri, 25 Nov 2016 00:15:43 +0000 (00:15 +0000)]
radv/winsys: start adding support for DMA/compute queue

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv/winsys: Expose number of compute/dma rings.
Bas Nieuwenhuizen [Sat, 17 Dec 2016 18:20:08 +0000 (19:20 +0100)]
radv/winsys: Expose number of compute/dma rings.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agofreedreno/a5xx: border color support
Rob Clark [Thu, 15 Dec 2016 22:27:56 +0000 (17:27 -0500)]
freedreno/a5xx: border color support

Not 100% sure it works if you have border color in VS.. but it might be
right.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: use MRT0 to import linear zs
Rob Clark [Mon, 12 Dec 2016 21:06:55 +0000 (16:06 -0500)]
freedreno/a5xx: use MRT0 to import linear zs

A bit of a hack, but we need to do this until we can do tiled zs in
sysmem (and associated tile/until blits for transfer_map).

Fixes xonotic and glmark2 "refract", when reorder wasn't enabled.
(reorder would paper over the issue by avoiding the extra round-
trip to system memory and back to gmem.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno: fdN_gmem_restore_format() is not gen specific
Rob Clark [Mon, 12 Dec 2016 21:59:09 +0000 (16:59 -0500)]
freedreno: fdN_gmem_restore_format() is not gen specific

Refactor out into a common helper, since this is the same across
generations when we need equiv z/s gmem restore format.

Next patch needs this in a5xx, rather than creating yet another
helper push this into core.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: cargo-cult end-batch sequence more faithfully
Rob Clark [Fri, 9 Dec 2016 23:23:10 +0000 (18:23 -0500)]
freedreno/a5xx: cargo-cult end-batch sequence more faithfully

Fixes some issues at least with GMEM bypass mode, where we'd sometimes
end up with some FS quads not hitting memory.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: misc fix
Rob Clark [Fri, 9 Dec 2016 18:12:51 +0000 (13:12 -0500)]
freedreno/a5xx: misc fix

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fix (at least some) vtx formats
Rob Clark [Fri, 9 Dec 2016 18:10:03 +0000 (13:10 -0500)]
freedreno/a5xx: fix (at least some) vtx formats

Swap/component-order doesn't seem to be quite what that is.  At least
blob was always setting it to XYZW ('11') but we weren't.  Causing
problems w/ formats like sint16..  Hard-coding this instead at least
seems to get glamor working.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: more formats
Rob Clark [Thu, 8 Dec 2016 19:39:58 +0000 (14:39 -0500)]
freedreno/a5xx: more formats

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fixup caps
Rob Clark [Wed, 7 Dec 2016 22:15:43 +0000 (17:15 -0500)]
freedreno/a5xx: fixup caps

Might not be 100% accurate, mostly just copy from a4xx to get started.

We are defn lying about occlusion query at this point (not implemented
yet) but need it to expose anything higher than gl1.4 (glamor needs
gl2.1)

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fix random faults on first sysmem draw
Rob Clark [Wed, 7 Dec 2016 20:21:56 +0000 (15:21 -0500)]
freedreno/a5xx: fix random faults on first sysmem draw

Not sure what this event is, but blob writes it.. and it seems to solve
random write faults at mystery address that would sometimes happen on
first BYPASS draw.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno: update generated headers
Rob Clark [Wed, 7 Dec 2016 20:20:48 +0000 (15:20 -0500)]
freedreno: update generated headers

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fix stride/size for mem->gmem blits
Rob Clark [Wed, 7 Dec 2016 15:17:17 +0000 (10:17 -0500)]
freedreno/a5xx: fix stride/size for mem->gmem blits

<brownpaperbag>these should be the in-GMEM dimensions</brownpaperbag>

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agoradv/winsys: consolidate request->fence code
Dave Airlie [Wed, 30 Nov 2016 03:29:05 +0000 (03:29 +0000)]
radv/winsys: consolidate request->fence code

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: handle fence allocation failing
Dave Airlie [Thu, 1 Dec 2016 00:44:45 +0000 (00:44 +0000)]
radv: handle fence allocation failing

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoradv: Don't bail out on pipeline create failure.
Bas Nieuwenhuizen [Fri, 16 Dec 2016 22:10:31 +0000 (23:10 +0100)]
radv: Don't bail out on pipeline create failure.

The spec says we have to try to create all, and only set failed
pipelines to VK_NULL_HANDLE. If one of them fails, we have to return
an error, but as far as I can see, the spec does not care which of
the suberrors.

Fixes
dEQP-VK.api.object_management.alloc_callback_fail_multiple.compute_pipeline
dEQP-VK.api.object_management.alloc_callback_fail_multiple.graphics_pipeline

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agospirv/nir: add support for ImageGatherExtended
Ilia Mirkin [Sun, 27 Nov 2016 21:09:55 +0000 (16:09 -0500)]
spirv/nir: add support for ImageGatherExtended

The strategy is to do the same thing that the GLSL lower_offset_arrays
pass does - create 4 separate texture gather ops, one per offset, and
read in the results from each gather's w component to recreate the
desired result.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoanv: Fix uniform and storage buffer offset alignment limits.
Francisco Jerez [Thu, 15 Dec 2016 21:34:02 +0000 (13:34 -0800)]
anv: Fix uniform and storage buffer offset alignment limits.

This fixes a regression in a bunch of image store vulkan CTS tests
from commit ad38ba113491869ab0dffed937f7b3dd50e8a735, which started
using OWORD block read messages to implement UBO loads.  The reason
for the failure is that we were giving bogus buffer alignment limits
to the application (1B), so the CTS would happily come back with
descriptor sets pointing at not even word-aligned uniform buffer
addresses.

Surprisingly the sampler messages used to fetch pull constants before
that commit were able to cope with the non-texel aligned addresses,
but the dataport messages used to fetch pull constants after that
commit and the ones used to access storage buffers (before and after
the same commit) aren't as permissive with unaligned addresses.

Cc: <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99097
Reported-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: Remove nir_array from lower_locals_to_regs
Thomas Helland [Thu, 15 Dec 2016 22:36:54 +0000 (23:36 +0100)]
nir: Remove nir_array from lower_locals_to_regs

We do nothing but initialize it, add to it, and delete it.
This is a fallout from removing constant initializer support.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoswr: Implement fence attached work queues for deferred deletion.
Bruce Cherniak [Tue, 13 Dec 2016 01:24:59 +0000 (19:24 -0600)]
swr: Implement fence attached work queues for deferred deletion.

Work can now be added to fences and triggered by fence completion. This
allows for deferred resource deletion, and other asynchronous tasks.

Reviewed-by: George Kyriazis <george.kyriazis@intel.com>
7 years agonir: Turn imov/fmov of undef into undef
Timothy Arceri [Thu, 15 Dec 2016 23:17:33 +0000 (10:17 +1100)]
nir: Turn imov/fmov of undef into undef

Reverting the previous attempt at this a5502a721fd30fd resulted in
the following Vulkan test failing.

dEQP-VK.glsl.return.return_in_dynamic_loop_dynamic_vertex

This time we use the num_components from the alu dest rather than
num_inputs to the op to determine the size of the undef.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99100

7 years agoegl/x11: cleanup init code
Eric Engestrom [Thu, 8 Dec 2016 00:30:34 +0000 (00:30 +0000)]
egl/x11: cleanup init code

No functional change, just rewriting it in an easier-to-understand way.

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
7 years agonir/lower_tex: fix number of components in replace_gradient_with_lod()
Iago Toral Quiroga [Wed, 14 Dec 2016 09:05:44 +0000 (10:05 +0100)]
nir/lower_tex: fix number of components in replace_gradient_with_lod()

We should make the dest in the textureLod() operation have the same number
of components as the destination in the original textureGrad()

Fixes regression in ES3-CTS.gtf.GL3Tests.shadow

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99072
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoRevert "nir: Turn imov/fmov of undef into undef."
Timothy Arceri [Thu, 15 Dec 2016 05:51:13 +0000 (16:51 +1100)]
Revert "nir: Turn imov/fmov of undef into undef."

This reverts commit 6aa730000fea84a14b49828a4bb30761d43903bf.

This was changing the size of the undef to always be 1 (the number of inputs
to imov and fmov) which is wrong, we could be moving a vec4 for example.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
7 years agoi965/vec4: Fix TCS output reads with non-zero component qualifiers.
Kenneth Graunke [Wed, 14 Dec 2016 07:47:05 +0000 (23:47 -0800)]
i965/vec4: Fix TCS output reads with non-zero component qualifiers.

We want to perform the URB read to a vec4 temporary, with no writemask,
then issue a MOV to swizzle the data and store it to the actual
destination, using the final writemask.

We were doing this wrong.  For example, let's say we wanted to read
a vec2 stored in components 2-3 of a vec4.  We would generate a URB
read message of:

   SEND <actual destination>.XY <header with mask set to XY>
   MOV <actual destination>.XY <actual destination>.ZW

This doesn't work, because the URB message reads the .XY components
of the vec4, rather than the ZW.  It writes to the right place, but
with the wrong data.  Then the MOV comes along and overwrites it
with data that didn't even come from the URB at all.

Instead we want to do:

   SEND <temporary> <header with mask set to ZW>
   MOV <actual destination>.XY <temporary>.ZW

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965/disasm: Decode dataport constant cache control fields.
Francisco Jerez [Fri, 9 Dec 2016 06:14:59 +0000 (22:14 -0800)]
i965/disasm: Decode dataport constant cache control fields.

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