radv: display an error message if the winsys init failed

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

radv: use a linked list for physical devices

Instead of a static array inside the instance object.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

radv: don't report error with other vendor DRM devices

Enumeration should just skip unsupported DRM devices.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

radv: report INITIALIZATION_FAILED when the amdgpu winsys init failed

The driver should be capable if it reaches the winsys initialization.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

radv: fix a memleak if the physical device initialization failed

The disk cache object should be freed.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

radv: rename radv_devices() to radv_enumerate_physical_devices()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

radv: cleanup radv_CreateInstance()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4806>

llvmpipe: enable ARB_sample_shading

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add min samples support to the fragment shader.

This isn't enabled yet until the state gets hooked up

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: enable GL_ARB_shader_texture_image_samples

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/nir: hooks up texture samples queries

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/sample: add num samples query for txqs (v2)

v2: add false to the existing users (Roland)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: enable 4x sample MSAA + texture multisample

This enables proper support for 4xMSAA and for texture mulitsample
extension.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

drisw: add multisample support to sw dri layer.

This allocates the msaa resources like the dri2 layer
and adds the flushes

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: don't choose pixel centers for multisample

Don't pick the pixel centers for multisample rendering, fix
the setup program.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: choose correct position for multisample

For multisample we don't want pixel centers at this stage,
so don't add them in for that case.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: choose multisample rasterizer functions per triangle (v2)

This just picks the correct cmds to add to the scene.

v2: drop using 32-bit ms (Roland)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: generate multisample triangle rasterizer functions (v2)

This uses the templating to generate multisample version of the
tri plane raster functions

This doesn't generate any optimised version for lower plane numbers,
maybe this is worth doing in the future.

v2: drop generating 32-bit msaa (Roland)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: fixup multisample coverage masks for covered tiles

For fully covered tiles just pass in the filled out mask.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: build 64-bit coverage mask in rasterizer

This adds the logic to build the per-sample masks at the lowest
level of the rasterizer block hierarchy

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add fixed point sample positions to scene.

These will be used in the rasterizer to generate the coverage masks

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add new rast api to pass full 64-bit mask.

The 64-bit mask is a 16-bit mask per sample for up to 4 samples.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: disable opaque variant for multisample

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: fix multisample occlusion queries.

This needs to check the per-sample mask inside the loop if
multisample is enabled.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: move color storing earlier in frag shader

Move the color storage before the late Z test as for sample
shading it needs to be inside a loop with the fragment shader.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: pass mask store into interp for centroid interpolation

This enables centroid interpolation to work, using the current
coverage masks.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: don't allow branch to end for early Z with multisample

Don't allow the branching optimisation with multisample enabled
as we have to check all samples.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: handle gl_SampleMask writing.

This is using a load/store to make it easier to add sample shading later.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add multisample alpha to one support

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add multisample alpha to coverage support.

Converts alpha into coverage mask.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: hook up sample position system value

This creates a global static with the current sample positions,
and passes it to the fragment shader which uses it for interpolation
and sample position support.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: handle multisample color stores.

Extract the final per-sample masks and store to the multisample
color buffers using them.

This retypes the pointer to a uint8_t at entry to make the
GEP simpler, then recasts to the blend type.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: interpolate Z at sample points for early depth test.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: handle multisample early depth test/late depth write

A set of values have to be passed from the early depth test to the
late depth write, when multisampling is enabled, a range of those
values have to be stored between stages, so create storage for them
and pass the values through the storage.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: multisample sample mask + early/late depth pass

Start adding support for multisample masks and the depth passes

The depth passes have to run per-sample, this isn't complete support
it adds the loops, and handles the execution masks.

One mask is stored per sample, they are combined post the early Z
pass into a single shader execution mask, and then the resulting
shader execution mask is anded back in for the late Z pass.

Init the vars to NULL to avoid gcc warnings

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: move some fs code around

this just moves the num_fs loop around for follow on refactors

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add per-sample depth/stencil test

The current depth stencil test code has some optimisations using
the mask when there is only one depth value, multisample requires
per-sample zstencil testing, and for that case just pass in the
mask that needs updating.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: move getting mask value out of depth code. (v2)

In order to add per-sample support to this code, the mask
value is needed not the value from the exec mask.

v2: update comment

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add per-sample interpolation.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add centroid interpolation support.

This just adds the implementation and API to the interpolation builders.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: pass interp location into interpolation code.

This just tracks the attribute interpolation location into the
interp code.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

nir/tgsi: translate the interp location

translate sample and centroid locations.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm: add mask api to force mask

For per-sample shading the mask needs to be forced for each
iteration of the fragment shader.

Just adds the API for now.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm: add sample id/pos intrinsic support

The sample position is looked up in an incoming array using the
sample id.

(These are mostly for ARB_sample_shading support)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add cbuf/zsbuf + coverage samples to the fragment shader key.

These will cause different fragment shaders to be generated.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: change mask input to fragment shader to 64-bit.

In order to handle a 4xMSAA mask (16-bits per sample) increase
the fragment shader API to be 64-bit.

v2: drop pointless if (Roland)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add multisample bit to fragment shader key.

The fragment shader needs to be regenerated when multisample changes.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: plumb multisample state bit into setup code.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe/rast: fix tile clearing for multisample color and depth tiles

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: record sample info for color/depth buffers in scene

This adds the nr_samples + sample_stride to the scene records
for cbufs and zsbuf.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: pass color and depth sample strides into fragment shader.

This just adds the interface and passes the depth and sample strides
into the fragment shader, nothing uses them yet.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

draw: disable point/line smoothing for multisample (v2)

When MSAA is enabled smoothing is ignored

v2: As pointed out by Roland I got this completely wrong,
fix this to work the other way

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: handle multisample render target clears

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add clear texture support for multisample textures.

This adds the clear paths for multisample textures.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add multisample resource copy region support.

This allows direct copies of all samples between two resources.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add internal multisample texture mapping path.

For clearing and copying textures llvmpipe needs to internally
access the per-sample data.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: pass incoming sample_mask into fragment shader context.

This links up the api changing the sample mask to passing it into
the fragment shader.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe/jit: pass fragment sample mask via jit context.

The incoming sample mask for the fragment shader can be passed
via the jit context

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add get_sample_position support (v2)

This just adds the sample values for 4xmsaa, and hooks them
up to the get_sample_position API

v2: move to vulkan standard sample positions

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: fix race between draw and setting fragment shader.

There is a race with u_blitter shaders + pipeline shaders (aaline/aapoint)
where the draw bind can cause a pipeline flush which can use bind_fs_state to
 be reenters and llvmpipe->fs gets the wrong value. Fix this by only
setting the llvmpipe->fs value after the draw binding is complete.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallium/util: split out zstencil clearing code.

llvmpipe will want to reuse this for it's multisample clears.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add a max samples define set to 4.

I doubt I'll care about much higher MSAA levels, so 4 it is.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add multisample support to texture allocator.

This adds a sample stride field and allocates enough memory for
each sample storage. Hook up the sample_stride field to draw
and jit textures and images

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

util: add a resource wrapper to get resource samples

This return 1 as a baseline and should be used in allocator paths.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add samples support to image jit

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

llvmpipe: add num_samples/sample_stride support to jit textures

This adds the support for num_samples/sample_stride retrieval to the
jit texture infrastructure.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

draw: add support for num_samples + sample_stride to the image paths

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

draw: introduce sampler num samples + stride members

This adds the num samples + sampler stride into the texture mapping paths,
currently drivers just pass 0 for now.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/nir: add multisample image operations

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/nir: add multisample support to image size

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/nir/tgsi: add multisample texture sampling.

Both paths are required as u_blitter needs the TGSI path.

This just hooks the instructions up to the sampling code.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/sample: add multisample image operation support

Just adds in the sample stride.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

gallivm/sample: add multisample support for texel fetch

This adds a new callback to get the stride between the per-sample
images, adds a new value for the per-sample index to lookup,
and a flag to use multisampling.

gallivm/sample: add num samples interface for dynamic samplers

This will be used for getting number of samples in jit code.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4122>

virgl: Properly check for encode_stride when encoding transfers

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4763>

llvmpipe: simple texture barrier implementation.

Just flush.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4774>

llvmpipo/nir: free compute shader NIR

I forgot this in the last round.

Fixes: 18f896e55d96 (llvmpipe: add initial nir support)
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4899>

draw/tess: free tessellation control shader i/o memory.

Fixes: 0d02a7b8ca79 (draw: add main tessellation code)
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4899>

nir: add missing group_memory_barrier handling

Totals from 2 (0.00% of 127638) affected shaders:
VGPRs: 164 -> 168 (+2.44%)
CodeSize: 18420 -> 18756 (+1.82%)
Instrs: 3658 -> 3700 (+1.15%)
Cycles: 82912 -> 83080 (+0.20%)
VMEM: 70 -> 69 (-1.43%)
PreVGPRs: 155 -> 168 (+8.39%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
CC: <mesa-stable@lists.freedesktop.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4889>

freedreno/ir3: Disable sin/cos range reduction for mediump.

robclark noted that the blob wasn't doing range reduction in the mediump
case, and I confirmed it on
dEQP-GLES3.functional.shaders.operator.angle_and_trigonometry.sin.mediump_float_fragment
vs
dEQP-GLES3.functional.shaders.operator.angle_and_trigonometry.sin.highp_float_fragment.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4893>

st/nine: Set correctly blend max_rt

Currently nine_convert_blend_state has no way
of knowing the number of rts.
For now set to an upper bound.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4891>

radeonsi: enable TC-compatible HTILE on demand for best Z/S performance

I haven't measured this, but it can only help.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4866>

radeonsi: allow tc_compatible_htile to be mutable

Move the relevant code from si_init_depth_surface to
si_emit_framebuffer_state, so that it can be changed after a pipe_surface
is initialized.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4866>

radeonsi/gfx9: always use IMG_DATA_FORMAT_S8_32 for 8-bit stencil

I wanna remove dependency on tc_compatible_htile from non-dynamic states.
This should be the same as 8_UINT if HTILE is disabled.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4866>

ac/surface: unset RADEON_SURF_TC_COMPATIBLE_HTILE if HTILE hasn't been computed

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4866>

radeonsi: don't wait for idle at the end of gfx IBs

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4894>

ac/nir: export some undef as zero

NIR already optimizes undef usage.
If undef reaches llvm, it's probably because of a broken shader.

In this situation, rather than letting llvm use the undef values
to do more optimization and probably produce incorrect results,
we replace undef values by 0.

"undef" values that are directly used in exports are kept as undef,
because this allows llvm to optimize them away.

This is only enabled for radeonsi.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2689
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

gallium: add a new cap PIPE_CAP_GLSL_ZERO_INIT

Allows driver to select a zero init mode between the 3 possible values.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

mesa: extend GLSLZeroInit semantics

This commit introduces a new way to zero-init variables but keep the
old one to not break any existing behavior.

With this change GLSLZeroInit becomes an integer, with the following
possible values:
 - 0: no 0 init
 - 1: current behavior
 - 2: new behavior. Similar to 1, except ir_var_function_out type are
      0 initialized but ir_var_shader_out.

The rationale behind 2 is: zero initializing ir_var_shader_out can
prevent some optimization where out variables are completely eliminated
when not written to.

On the other hand, zero initializing "ir_var_function_out" has no
effect on correct shaders but typically helps shadertoy since the main
function is:

   void mainImage(out vec4 fragColor) { ... }

So with this change we're sure that fragColor will always get a value.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

glsl: add a is_implicit_initializer flag

Shared globals and glsl_zero_init can cause linker errors if the
variable is only initialized in 1 place.

This commit adds a flag to variables that have been implicitely
initialized to be able in this situation to keep the explicit
initialization value.

Without this change the global-single-initializer-2-shaders piglit
test fails when using glsl_zero_init.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

glsl: rework zero initialization

This commit makes zero_init a bitfield of types of variables to zeroinit.

This will allow some flexibility that will be used in the next commit.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

glsl: init gl_FragColor if zero_init=true

This fixes shaders doing "gl_FragColor += ..." and doesn't hurt correct
shaders, because the zero init is discarded.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

radeonsi: don't print gs_copy_shader stats for shaderdb

Fixes: dbc86fa3de6 ("radeonsi: dump shader stats when hitting the live cache")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4607>

radv: enable shaderInt16 unconditionally with LLVM and only GFX8+ with ACO

The Vulkan spec says:

"shaderInt16 specifies whether 16-bit integers (signed and unsigned)
are supported in shader code. If this feature is not enabled, 16-bit
integer types must not be used in shader code."

I think it's just safe to enable it because 16-bit integers should
be fully supported with LLVM and also with ACO and GFX8+. On GFX8
and earlier generations, throughput of 16-bit int is same as 32-bit
but that should't change anything.

For GFX6-GFX7 ACO support, we have to implement conversions without
SDWA.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4874>

radeonsi: add workaround for issue 2647

For unknown reasons pixel shaders in KSP game get executed with
infinite interpolation coefficients and this causes an infinite
loop in the shader.

This commit adds a hacky workaround that kills pixel shaders if
invalid interp coeffs are detected and enables it for KSP.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2174
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2647
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4700>

zink: use nir_lower_uniforms_to_ubo

Instead of open-coding uniform -> UBO lowering, let's instead use the
one that already exists. This should make things a bit simpler going
forward.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4734>

nir: Always create UBO variable when lowering uniforms to ubo

Zink needs to know the sizes of UBOs, and for normal UBOs we get this
from the nir_var_mem_ubo variables. This allows us to treat all of these
the same way.

We're about to need the same information for the in-progress D3D12
driver, so let's do this in a central location instead of in the driver.

This version is also a bit more careful than the Zink version. In
particular, for two reasons:
1. We increase the variable bindings when we adjust the pre-existing
   UBOs.
2. We increase shader->info.num_ubos when we insert a new UBO variable.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4734>

mesa/st: consider NumUniformBlocks instead of num_ubos when binding

This is the number of uniform blocks at linking time, not after
finalizing shaders.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4734>

compiler/glsl: explicitly store NumUniformBlocks

It's not great to use shader_info for this information, because it
might have gone through lowering of uniforms to UBOs, which can change
the number of UBOs. So let's make sure we know the size of the
UniformBlocks array from when the shader was linked instead.

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4734>

glsl: rename has_implicit_uint_to_int_conversion to *_int_to_uint_*

There is no uint to int implicit conversion in glsl, this is just
a typo in the name of this function. The correct one would be:
has_implicit_int_to_uint_conversion.

Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4884>

driconf: add force_integer_tex_nearest option

And enable it for "GRID Autosport" and "DIRT: Showdown" games.

CC: 20.1 <mesa-stable@lists.freedesktop.org>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1258
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4647>