Instead of using the copy of shader_info stored in gl_program, it now
uses the one in nir_shader. This is needed for SPIR-V because the
info.tess.tcs_vertices_out is filled in via _mesa_spirv_to_nir which
happens much later than with a GLSL shader. The copy of shader_data in
gl_program is only updated later via brw_shader_gather_info but that
is too late.

For GLSL this shouldn't create any problems because the nir copy of
the shader_info is immediately copied from the gl_program in
glsl_to_nir.

v2: updated after commit "i965: Combine both gl_PatchVerticesIn
    lowering passes." (488972) (Alejandro Piñeiro)

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

The value is copied from the gl_program. If we don’t do this then it
will get reset back to zero in brw_shader_gather_info. This isn’t a
problem for GLSL because in that case the nir_shader is initialised
with a copy of the shader_info from the gl_program.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

This is the same we do for vulkan drivers

This is needed to pass the following CTS test:
KHR-GL45.gl_spirv.spirv_modules_shader_binary_multiple_shader_objects_test

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

input locations used by input attributes are not handled in the same
way in OpenGL vs Vulkan. There is a detailed explanation of such
differences on the following commit:

c2acf97fcc

So with this commit, the same adjustment that is done after
glsl_to_nir, is being done after spirv_to_nir, when it is used on
OpenGL (ARB_gl_spirv).

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

As we plan to reuse it for ARB_gl_spirv implementation.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

After commit "nir: Use derefs in nir_lower_samplers"
(75286c2d08) assumes one deref for both
the texture and the sampler. However there are cases (on OpenGL, using
ARB_gl_spirv) where SPIR-V is not providing a sampler, like for
texture query levels ops. Although we could make spirv_to_nir to
provide a sampler deref for those cases, it is not really needed, and
wrong from the Vulkan point of view.

This patch fixes the following (borrowed) tests run on SPIR-V mode:
  arb_compute_shader/execution/basic-texelFetch.shader_test
  arb_gpu_shader5/execution/sampler_array_indexing/fs-simple-texture-size.shader_test
  arb_texture_query_levels/execution/fs-baselevel.shader_test
  arb_texture_query_levels/execution/fs-maxlevel.shader_test
  arb_texture_query_levels/execution/fs-miptree.shader_test
  arb_texture_query_levels/execution/fs-nomips.shader_test
  arb_texture_query_levels/execution/vs-baselevel.shader_test
  arb_texture_query_levels/execution/vs-maxlevel.shader_test
  arb_texture_query_levels/execution/vs-miptree.shader_test
  arb_texture_query_levels/execution/vs-nomips.shader_test
  glsl-1.30/execution/fs-textureSize-compare.shader_test

v2: merge lower_tex_src_to_offset and calc_sampler_offsets together,
    update texture/sampler index and texture_array_size directly on
    lower_tex_src_to_offset (Jason)
v3: clarify one comment (Jason)

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

So they are not exposed through the introspection API.

It is worth to note that the number of hidden uniforms of GLSL linking
vs SPIR-V linking would be somewhat different due the differen order
of the nir lowerings/optimizations.

For example: gl_FbWposYTransform. This is introduced as part of
nir_lower_wpos_ytransform. On GLSL that is executed after the IR-based
linking. So that means that on GLSL the UniformStorage will not
include this uniform. With the SPIR-V linking, that uniform is already
present, but marked as hidden. So it will be included on the
UniformStorage, but as hidden.

One alternative would create a special how_declared for that case, but
seemed an overkill. Using hidden should be ok as far as it is used
properly.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

Equivalent to the already existing how_declared at GLSL IR. The only
difference is that we are not adding all the declaration_type
available on GLSL, only the one that we will use on the short term. We
would add more mode if needed on the future.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

GLSL has gl_VertexID which is supposed to be non-zero-based.

SPIR-V has both VertexIndex and VertexId builtins whose meanings are
defined by the APIs.

Vulkan defines VertexIndex as being non-zero-based. In Vulkan VertexId
and InstanceId have no meaning and are pretty much just reserved for
OpenGL at this point.

GL_ARB_spirv removes VertexIndex and defines VertexId to be the same
as gl_VertexId (which is also non-zero-based).

Previously in Mesa it was treating VertexIndex as non-zero-based and
VertexId as zero-based, so it was breaking for GL. This behaviour was
apparently based on Khronos bug 14255. However that bug doesn’t seem
to have made a final decision for VertexId.

Assuming there really is no other definition for VertexId for Vulkan
it seems better to just make them both have the same value.

v2: update comment and commit descriptions, based on Jason Ekstrand
    explanation of the meaning/rationale behind all those builtins
    (Jason)

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

info.gs.output_primitive was already being filled. Not sure why this
is not needed on Vulkan, but we found to be needed for
ARB_gl_spirv. Specifically, this is needed to get the following test
passing:

KHR-GL45.gl_spirv.spirv_validation_builtin_variable_decorations_test

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

Patch sets additional formats renderable and enables the extension
when OpenGL ES 3.1 is supported.

v2: instead of dummy_true, have a separate toggle for extension
    (Eric Anholt)

v3: add missing checks, simplify some existing checks and fix
    glCopyTexImage2D check (Nanley Chery)

    add SHORT and BYTE support in read_pixels_es3_error_check

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

One of the reasons we didn't notice that R24_UNORM_X8_TYPELESS
destinations were broken was that an earlier layer was swapping it
out for B8G8R8A8_UNORM.  That made Z24X8 -> Z24X8 blits work.

However, R32_FLOAT -> R24_UNORM_X8_TYPELESS was still totally broken.
The old code only considered one format at a time, without thinking
that format conversion may need to occur.

This patch moves the translation out to a place where it can consider
both formats.  If both are Z24X8, we continue using B8G8R8A8_UNORM to
avoid having to do shader math workarounds.  If we have a Z24X8
destination, but a non-matching source, we use our shader hacks to
actually render to it properly.

Fixes: 804856fa57 (intel/blorp: Handle more exotic destination formats)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

The hardware doesn't support rendering to R24_UNORM_X8_TYPELESS, so
Jason decided to fake it with a bit of shader math and R32_UNORM RTs.

The only problem is that R32_UNORM isn't renderable either...so we've
just traded one bad format for another.

This patch makes us use R32_UINT instead.

Fixes: 804856fa57 (intel/blorp: Handle more exotic destination formats)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

The Vulkan 1.1.82 spec flipped the order to better match D3D.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

This patch ties in the array split, merge, and interleave code.

shader-db changes in the TGSI code are:

              original code  |  array-merge  |       change
              mean      max  |  mean    max  | best  mean %  worst
      -----------------------------------------------------------
      arrays   0.05       2  |   0.00     0  |  -2   -100      0
total temps    5.05      21  |   4.92    20  | -15   -2.59     1
      instr   55.33     988  |  55.20   988  | -15   -0.24     0

Evaluation:

Run shader-db in single thread mode (otherwise the output is
not ordered and the best and worst column don't make sense) to
get results pre-stats.txt and post-stats.txt. Then using
python pandas:

 import pandas as pd
 old_stats = pd.read_csv('pre-stats.txt')
 new_stats = pd.read_csv('post-stats.txt')
 omean = old_stats.mean()
 omax = old_stats.max()
 nmean = new_stats.mean()
 nmax = new_stats.max()
 delta =  new_stats - old_stats
 pd.concat([omean, omax, nmean, nmax, delta.min(),
            delta.mean()/old_stats.mean()*100, delta.max()],
            axis=1, keys=['mean', 'max', 'mean', 'max', 'best',
            'avg change %', 'worst'])

v4: - Correct typo and add bugs that are fixed by this series.
    - Update stats and describe stats evaluation

Bugzilla:
  https://bugs.freedesktop.org/show_bug.cgi?id=105371
  https://bugs.freedesktop.org/show_bug.cgi?id=100200
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

v4: Also track the register given in inst->resource. (thanks: Benedikt Schemmer
    for testing the patches on radeonsi, which revealed that I was missing
    tracking this)
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

Because of the indirect access it is impossible to obtain an accurate per
component and array element tracking. Therefore, the tracking is simplified
to only track whether any element was accessed, whether this happend
conditionally in a loop. In addition, while tracking of temporaries requires
a per-componet tracking that is later fused, for arrays only the components
access mask is neede. The resulting tracking code and evaluation of the array
live range is sufficiently different from the evaluation of the live range of
temporaries to justify implementing this in a different class instead of
adding more complexity to the already existing code for temporary life
range evaluation.

v4: Update commit message to make it clearer why this class is seperate from
    the tracking of temporaries.
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

In preparation of the array live range tracking the evaluation of the read
mask is moved out the register live range tracking to the enclosing call
of the generalized read access tracking.

Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

In preparartion of adding the tracking of the live range the classes that refer
to temporary registers are renamed.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

v2: - Define tests also in the meson.build file.
v4: - Check no-op mapping of all bits.
    - Convert tests to the new class layout used in the merge evaulation.
    - remove dependency on llvm in meson build (Thanks Dylan Baker for pointing
       out that this might not needed)
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

v4: - Update the code to use the new merge logic.
    - Use a cleaner, class-based approach for the evaluation of merges.
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

v4: - Remove logic for evaluation of swizzles and merges since this
        was moved to array_live_range. This class now only handles the
        actual remapping.

Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

This class holds the array length, live range, and accessed components, and
it implements the logic for evaluating how arrays are merged and interleaved.

v4: - Add logic to evaluate merge and interleave of a pair of arrays to
      the class array_live_range.
    - document class
    - update commit message

Thanks Nicolai Hähnle for the pointers given.

Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

On one hand "live range" is the term used in the literature, and on the
other hand a distinction is needed from the array live ranges.

v4: Fix indentions and white spaces

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v3)
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

in constructs like below, currently the live range estimation extends the live range
of t unecessarily to the whole loop because it was not detected that t is
unconditional written and later read only in the "if (a)" scope.

  while (foo)  {
    ...
    if (a) {
       ...
       if (b)
         t = ...
       else
         t = ...
       x = t;
       ...
    }
     ...
  }

This patch adds a unit test for this case and corrects the minimal live range estimation
accordingly.

v4: update comments
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

Array whose elements are only accessed directly are replaced by the
according number of temporary registers. By doing so the otherwise
reserved register range becomes subject to further optimizations like
copy propagation and register merging.

Thanks to the resulting reduced register pressure this patch makes
the piglits

  spec/glsl-1.50/execution -
      variable-indexing/vs-output-array-vec3-index-wr-before-gs
      geometry/max-input-components

pass on r600 (barts) where they would fail before with a "GPR limit exceeded"
error (even with the spilling that was recently added).

v2: * rename method dissolve_arrays to split_arrays
    * unify the tracking and remapping methods for src and dst registers
    * also track access to arrays via reladdr*

v3: * enable this optimization only if the driver requests register merge

v4: * Correct comments
    * Also update inst->resource if it is an array element
      (thanks: Benedikt Schemmer for testing the patches on radeonsi, which
       revealed that I was missing tracking this)

Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

When mesa is compiled in debug mode then this adds the possibility
to print out some statistics about the translated and optimized TGSI
shaders to a file.

The functionality is enabled by setting the environment variable

   GLSL_TO_TGSI_PRINT_STATS

to the file name where the statistics should be collected. The file is
opened in append mode so that statistics from various runs will be
accumulated.

v4: Make accress to log file thread save (thanks for pointing this out Nicolai
    Hähnle)
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Acked-by: Dave Airlie <airlied@redhat.com>

The GLSL operations findLSB, findMSB, and countBits always return
a signed integer type. Let TGSI reflect this.

v2: Properly set values in infer_(src|dst)_type   (Thanks Roland
    Schneidegger for pointing out problems with my 1st approach)
v2: Set values in the common infer_type code path, and only add
    the correct source type for UMSB (Roland Schneidegger)

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

Now that all the build scripts are compatible with both Python 2 and 3,
we can flip the switch and tell Meson to use the latter.

Since Meson already depends on Python 3 anyway, this means we don't need
two different Python stacks to build Mesa.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

In both Python 2 and 3, opening a file without specifying the mode will
open it for reading in text mode ('r').

On Python 2, the read() method of a file object opened in mode 'r' will
return byte strings, while on Python 3 it will return unicode strings.

Explicitly specifying the binary mode ('rb') then decoding the byte
string means we always handle unicode strings on both Python 2 and 3.

Which in turns means all re.match(line) will return unicode strings as
well.

If we also make expandCString return unicode strings, we don't need the
call to the unicode() constructor any more.

We were using the ugettext() method because it always returns unicode
strings in Python 2, contrarily to the gettext() one which returns
byte strings. The ugettext() method doesn't exist on Python 3, so we
must use the right method on each version of Python.

The last hurdles are that Python 3 doesn't let us concatenate unicode
and byte strings directly, and that Python 2's stdout wants encoded byte
strings while Python 3's want unicode strings.

With these changes, the script gives the same output on both Python 2
and 3.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

On Python 3, executing `foo != bar` will first try to call
foo.__ne__(bar), and fallback on the opposite result of foo.__eq__(bar).

Python 2 does not do that.

As a result, those __eq__ methods were never called, when we were
testing for inequality.

Expliclty adding the __ne__ methods fixes this issue, in a way that is
compatible with both Python 2 and 3.

However, this means the __eq__ methods are now called when testing for
`foo != None`, so they need to be guarded correctly.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

The check for ETC2 compatibility was not updated when the fallback
format was changed.

Fixes: 71867a0a61
   st/mesa: Fall back to R8G8B8A8_SRGB for ETC2

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

Instead of copying the list, then sorting the copy in-place, we can just
get a new sorted copy directly.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

In Python 2, the traditional way to sort containers was to use a
comparison function (which returned either -1, 0 or 1 when passed two
objects) and pass that as the "cmp" argument to the container's sort()
method.

Python 2.4 introduced key-functions, which instead only operate on a
given item, and return a sorting key for this item.

In general, this runs faster, because the cmp-function has to get run
multiple times for each item of the container.

Python 3 removed the cmp-function, enforcing usage of key-functions
instead.

This change makes the script compatible with Python 2 and Python 3.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

Python 3 lost the long type: now everything is an int, with the right
size.

This commit makes the script compatible with Python 2 (where we check
for both int and long) and Python 3 (where we only check for int).

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

Mixing the two is a long-standing recipe for errors in Python 2, so much
so that Python 3 now completely separates them.

This commit stops treating both as if they were the same, and in the
process makes the script compatible with both Python 2 and 3.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

On Python 2, the builtin functions filter() returns a list.

On Python 3, it returns an iterator.

Since we want to use those objects in contexts where we need lists, we
need to explicitly turn them into lists.

This makes the code compatible with both Python 2 and Python 3.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

The code was just reimplementing itertools.combinations_with_replacement
in a less efficient way.

This does change the order of the results slightly, but it should be ok.

Signed-off-by: Mathieu Bridon <bochecha@daitauha.fr>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

This is basically copied from the DRI2 destroy path.  Without this,
Raspberry Pi would quickly run out of CMA during the EGL tests in the CTS
due to all the pixmaps laying around.

Fixes: f35198bade ("egl/x11: Implement dri3 support with loader's dri3 helper")
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>

When the SIMD16 Gen4-5 fragment shader payload contains source depth
(g2-3), destination stencil (g4), and destination depth (g5-6), the
single register of stencil makes the destination depth unaligned.

We were generating this instruction in the RT write payload setup:

   mov(16)   m14<1>F   g5<8,8,1>F   { align1 compr };

which is illegal, instructions with a source region spanning more than
one register need to be aligned to even registers.  This is because the
hardware implicitly does (nr | 1) instead of (nr + 1) when splitting the
compressed instruction into two mov(8)'s.

I believe this would cause the hardware to load g5 twice, replicating
subspan 0-1's destination depth to subspan 2-3.  This showed up as 2x2
artifact blocks in both TIS-100 and Reicast.

Normally, we rely on the register allocator to even-align our virtual
GRFs.  But we don't control the payload, so we need to lower SIMD widths
to make it work.  To fix this, we teach lower_simd_width about the
restriction, and then call it again after lower_load_payload (which is
what generates the offending MOV).

Fixes: 8aee87fe4c (i965: Use SIMD16 instead of SIMD8 on Gen4 when possible.)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107212
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=13728
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Diego Viola <diego.viola@gmail.com>

According to the G45 PRM Volume 2 Page 265 we're supposed to only set
these signals when there is an actual depth buffer.  Note that we
already do this for the stencil buffer by virtue of brw->stencil_enabled
invoking _mesa_is_stencil_enabled(ctx) which checks whether the current
drawbuffer's visual has stencil bits (which is updated based on what
buffers are bound).  We just need to do it for depth as well.

Not observed to fix anything.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

This extension is not defined for indirect contexts. Marking it as
"client only", as the old code did here, would make the extension
available in indirect contexts, even though the server would certainly
not have it in its extension list.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Fixes: 5b196f39bd "anv/pipeline: Compile to NIR in compile_graphics"
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

Fixes: 6a60beba40 "intel/tools: Add an error state to aub translator"
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

This fixes both Metro 2033 Redux and Metro Last Light Redux

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99730
Signed-off-by: Eero Tamminen <eero.t.tamminen@intel.com>
Signed-off-by: Vadym Shovkoplias <vadym.shovkoplias@globallogic.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

Return ir_rvalue::error_value with ast_post_inc, ast_post_dec if
parser error was emitted previously. This way process_array_size
won't see bogus IR generated like with commit 9c676a6427.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98699
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>