radv/gfx9: fix buffer to image for 3d images on compute queues

This fixes some of the broken:
dEQP-VK.synchronization.op.multi_queue.*64x64x8* tests.

Fixes: e38685cc62e 'Revert "radv: disable support for VEGA for now."'
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/gfx9: fix 3d image clears on compute queues

This fixes some of the broken:
dEQP-VK.synchronization.op.multi_queue.*64x64x8* tests.

Fixes: e38685cc62e 'Revert "radv: disable support for VEGA for now."'
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/gfx9: fix 3d image to image transfers on compute queues.

This fixes some of the broken:
dEQP-VK.synchronization.op.multi_queue.*64x64x8* tests.

Fixes: e38685cc62e 'Revert "radv: disable support for VEGA for now."'
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

anv/device: Mark all state buffers as needing capture

Previously, we were flagging the instruction state buffer for capture
but not surface state or dynamic state.  We want those captured too.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

intel/aubinator: Gracefully handle dynamic state not being available

Some older versions of the Vulkan driver didn't properly tag dynamic
state as needing to be captured.  Also, this prevents crashes when
looking at dumps on older kernels.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

intel/aubinator: Free section data last

We were walking the sections, printing the batches, and then freeing
them in one pass.  If the batch happens to reference any earlier
sections (which it almost certainly will since it's at the end), we will
access freed memory.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

spirv: consider bitsize when handling OpSwitch cases

This reverts commit 7665383a33f9ce9256aa121cbe4d3bd948dff145 and is
squashed together with https://patchwork.freedesktop.org/patch/194610/
(spirv: avoid infinite loop / freeze in vtn_cfg_walk_blocks()) which
fixes https://bugs.freedesktop.org/show_bug.cgi?id=104359 properly.

Fixes: 9702fac68e (spirv: consider bitsize when handling OpSwitch cases)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104359
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

svga: check for null fs pointer in update_samplers()

This can happen when there's no active fragment shader, such as
when using transform feedback.  This wasn't hit by any Piglit test
but is hit by Daniel Rákos' Nature demo.  VMware bug 2026189.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

st/mesa: increase size of glsl_base_type bitfields

Change 59f458cd8703b97b31b8 added more enums to glsl_base_type.  We
have to bump up the size of the bitfields for fields of this type
for MSVC.  Also, add another assertion to catch another place where
this enum bitfield is used.

Reviewed-by: Neha Bhende <bhenden@vmware.com>

radv: fix pipeline statistics end query on compute queue

It's legal to a pipeline stat query on a compute queue,
but we'd emit the wrong packet here. This should fix it to emit
the correct packet.

Noticed while inspecting the mpv hang.

Fixes: ad61eac250 (radv: factor out eop event writing code. (v2))
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: fix events on compute queues.

The event emission wasn't sending the correct packet for gfx8 compute
queues, which explains why it works on vega fine.

This fixes the mpv vulkan hang.

Fixes: ad61eac250 (radv: factor out eop event writing code. (v2))
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: move local bos usage to a perftest flag.

These seem mildly unstable on vega, crashing CTS in various fun ways,
and looks like leaking memory.

Disable for now, but leave the option to enable them.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

vulkan/wsi: free cmd pools

We destroy the pools but don't free the container.

This fixes:
dEQP-VK.wsi.xlib.swapchain.simulate_oom*

Fixes: d50937f137 (vulkan/wsi: Implement prime in a completely generic way)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: Always use fragment resolve if dest uses DCC.

HW resolve does not support it either.

Fixes: 2a04f5481df "radv/meta: select resolve paths"
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Use correct framebuffer size for partial FS resolves.

Framebuffer is from 0,0, not (dst.x, dst.y).

Fixes: 69136f4e633 "radv/meta: add resolve pass using fragment/vertex shaders"
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Fix fragment resolve destination offset.

The position start at (dst.x, dst.y), so if we want the source to
start at (src.x, src.y), we have to offset by (src.x-dst.x,src.y-dst.y).

Haven't tested that this fixed anything yet, but found by inspection.

Fixes: 69136f4e633 "radv/meta: add resolve pass using fragment/vertex shaders"
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Don't handle DCC in compute resolve.

If the destination has DCC, we will use the FS resolve.

Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Flush caches before subpass resolve.

Fixes: f4e499ec791 "radv: add initial non-conformant radv vulkan driver"
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Invert condition for all samples identical during resolve.

the samples_identical instruction returns 0 if they are differet, so
we have to do the extra work if the result is 0, not if it is != 0.

Fixes: f4e499ec791 "radv: add initial non-conformant radv vulkan driver"
Reviewed-by: Dave Airlie <airlied@redhat.com>

egl: don't try the software path twice

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reported-by: Brendan King <Brendan.King@imgtec.com>

egl: rename LIBGL_ALWAYS_SOFTWARE variable from UseFallback to ForceSoftware

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

egl: let each platform decided how to handle LIBGL_ALWAYS_SOFTWARE

My refactor in 47273d7312cb5b5b6b0b9 missed this early return; because
of it, setting UseFallback one layer above actually prevented the
software path from being used.

Remove this early return and let each platform's dri2_initialize_*()
decide what it can do with the LIBGL_ALWAYS_SOFTWARE restriction.

platform_{surfaceless,x11,wayland} were already handling it themselves.

Fixes: 47273d7312cb5b5b6b0b9 "egl: set UseFallback if LIBGL_ALWAYS_SOFTWARE is set"
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reported-by: Brendan King <Brendan.King@imgtec.com>

egl: link libEGL against the dynamic version of libglapi

Note: the following happens only when using slibtool.
Since this is a very serious breakage, we will keep the workaround until
a better solution is available.

DRI modules store the address of the dispatch table in a TLS variable,
_glapi_tls_Dispatch.

Changes to the way libEGL is built in d884d8d0077c16d459b1 resulted in
it being statically linked against libglapi, and thus containing its own
copy of _glapi_tls_Dispatch. The result was that some applications would
fail to work (e.g. deqp-egl, which dynamically loads libEGL), due to the
DRI module storing the dispatch table address in one copy of
_glapi_tls_Dispatch, and libEGL obtaining the address from another copy
of the variable.

Fixes: d884d8d0077c16d459b1 "egl/dri: link directly to libglapi.so"
Signed-off-by: Brendan King <Brendan.King@imgtec.com>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

radv: don't do format replacement on tc compat htile surfaces.

For copies the texture unit needs to know the depth format so
it can read the htile data properly.

This fixes:
dEQP-VK.renderpass.suballocation.formats.d32_sfloat_s8_uint.load.clear

Fixes: ad3d98da9f (radv: enable tc compatible htile for d32s8 also.)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/gfx9: use correct stencil format for tc compat htile.

This needs to correspond to the bit depth of the Z plane.

noticed in passing reading amdvlk.

Fixes: fc6c77e162df3 (radv: fix TC-compat HTILE with VK_FORMAT_D32_SFLOAT_S8_UINT on Vega)
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

svga: move variant->fs_shadow_compare_units assignment

Fixes a crash since the variant object isn't allocated until later
in the function.  Not sure how this got through.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

amd/common: rework set_userdata_location() and rename to set_loc()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: rename set_userdata_location_shader() to set_loc_shader()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: replace set_userdata_location_indirect() by set_loc_desc()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: rename radv_define_vs_user_sgprs_phase2()

... to set_vs_specific_input_locs().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: rename radv_define_common_user_sgprs_phase2()

... to set_global_input_locs().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: rename add_user_sgpr_array_argument() to add_array_arg()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: replace add_sgpr_argument() by add_arg()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: replace add_user_sgpr_argument() by add_arg()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: replace add_vgpr_argument() by add_arg()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: add new add_arg() helper for SGPRs/VGPRs arguments

The idea is to clean up the add arguments logic.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: rename radv_define_common_user_sgprs_phase1()

... to declare_global_input_sgprs().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: rename radv_define_vs_user_sgprs_phase1()

... to declare_vs_specific_inputs_sgprs().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: do not try to declare input VS SGPRs for GS

It's a no-op anyway but it looked strange to me, remove it.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: add declare_vs_input_vgprs() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: add declare_tes_input_vgprs() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: remove unnecessary num_user_sgprs_used

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: remove unnecessary user_sgpr_count

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

radeonsi: make use of ac_init_exec_full_mask()

Similar to si_init_exec_full_mask().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

svga: use tgsi_util_get_shadow_ref_src_index() in a couple place

No piglit changes.

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>

tgsi: improve comment on tgsi_util_get_shadow_ref_src_index()

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>

svga: fix TGSI_TEXTURE_SHADOW1D coordinate selection

Fixes about 24 Piglit tex-miplevel-selection tests.

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>

svga: fix shadow comparison failures

In some cases, We do shadow comparison cases in the fragment shader
instead of with texture sampler state.  But when we do so, we must
disable the shadow comparison test in the sampler state.  As it
was, we were doing the comparison twice, which resulted in nonsense.
Also, we had the texcoord and texel value swapped in the comparison
instruction.

Fixes about 38 Piglit tex-miplevel-selection tests.

Reviewed-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>

util: add trivial comment on u_upload_create()

r600: fix atomic counter index mode getting emitted on pre-cayman

This is a regression since I added cayman atomic support, not sure
it fixes anything, but the shader dumps look better.

Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: set some dcc parameters depending on if texture will be sampled

This is ported from amdvlk which sets the independent 64b blocks
only for image which will sample dcc.

I'm not sure how to port this to radeonsi.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/radeonsi: set dcc min uncompressed properly for APUs.

This is ported from amdvlk.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

amd/common/radv/radeonsi: use register defines for dcc block sizes.

These are just taken from amdvlk, we probably knew these already,
but may as well port them now.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

st/glsl_to_nir: add patch support to st_nir_assign_var_locations()

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/glsl_to_nir: call post opt functions after opts have finished

We need to move this to a separate loop because
nir_compact_varyings() can alter the IR of a previous stage.

Fixes: 6648bd68fd27 "st/glsl_to_nir: enable NIR link time opts"
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/st_glsl_to_nir: call nir_lower_64bit_pack

Fixes 56 crashes in the radeonsi nir backend.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

docs/features: show es3.1 compat done on r600.

This was already being reported, just missed the docs.

Reported-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa: always compare optype with symbolic name in ATI_fs

Signed-off-by: Miklós Máté <mtmkls@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

mesa: document ati_fragment_shader::cur_pass and swizzlerq

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

mesa: move ATI_fs state compile changes after the error checks

Both in setup and arithmetic instructions. Also, remove the useless
new_*_inst() functions, and refactor check_arith_arg(), because it did
two completely different things.

Piglit: spec/ati_fragment_shader/error04-endshader

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

tnl: fix not having texture coords in ATI_fs in swrast

ATI_fs in swrast only had access to texture coordinates if there was a
valid texture bound and texturing was enabled.

Piglit: spec/ati_fragment_shader/render-sources and render-notexture

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

mesa: fix not having secondary color in ATI_fs in swrast

ATI_fs in swrast only had secondary color if GL_COLOR_SUM was enabled.
This patch probably fixes the same issue in r200.

Piglit: spec/ati_fragment_shader/render-sources and render-precedence

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

mesa: fix validate for secondary interpolator

This patch fixes multiple problems:
- the interpolator check was duplicated
- both had arg instead of argRep
- I split it into color and alpha for better readability and error msg
- the DOT4 check only applies to color instruction according to the spec
- made the DOT4 check fatal, and improved the error msg

Piglit: spec/ati_fragment_shader/error08-secondary

v2: fixed formatting, added spec quotations

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

mesa: fix typo in ATI_fs dstMod error checking

Piglit: spec/ati_fragment_shader/error14-invalidmod

Signed-off-by: Miklós Máté <mtmkls@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: fix crash when an ATI_fs pass begins with an alpha inst

This fixes crash when:
- first pass begins with alpha inst
- first pass ends with color inst, second pass begins with alpha inst
Also, use the symbolic name instead of a number.

Piglit: spec/ati_fragment_shader/api-alphafirst

v2: fixed formatting

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

mesa: add fallback texture for SampleMapATI if there is nothing

This fixes crash in the state tracker.

Piglit: spec/ati_fragment_shader/render-notexture

v2: fixed formatting, moved stuff inside the loop,
    moved the fallback later to fix more cases

Signed-off-by: Miklós Máté <mtmkls@gmail.com>

radeonsi: don't use fast color clear for small images even on APUs

Increase the limit and handle non-square images better.

This makes glxgears 20% faster on APUs, and a little more on dGPUs.
We all use and love glxgears.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

radeonsi: set PNT_SPRITE_ENA = point_quad_rasterization

This is based on how nvc0 translates the state.

gallium/util: add util_num_layers helper

radv: Fix DCC compatible formats.

DCC was disabled when the image format is !!supported, which is one ! too many.

Ironically the commit that introduced it was supposed to lead to more DCC use ...

Fixes: 969537d9358 "radv: Add support for more DCC compression with VK_KHR_image_format_list."
Reviewed-by: Dave Airlie <airlied@redhat.com>

Revert "i965/fs: Use align1 mode on ternary instructions on Gen10+"

This reverts commit 9cd60fce9c22737000a8f8dc711141f8a523fe75.
Above commit caused 2000+ piglit tests to assert fail. Disabling
the align1 mode on gen10 for now to avoid failures.

Cc: Matt Turner <mattst88@gmail.com>
Cc: Rafael Antognolli <rafael.antognolli@intel.com>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Rafael Antognolli <rafael.antognolli@intel.com>

docs: update calendar, add news item and link release notes for 17.2.8

Signed-off-by: Andres Gomez <agomez@igalia.com>

docs: add sha256 checksums for 17.2.8

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit 3281775ab9993d700a0a01a2823b6e7c72fce150)

docs: add release notes for 17.2.8

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit 3482790712e92d660706952f9ff282d904415941)

freedreno: set missing internal_format when importing texture

Fixes running piglits without -fbo. Probably lots of other stuff too.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

amd/common: add ac_export_mrt_z() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd/common: pass the family to ac_llvm_context_init()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

radv: reduce the number of small surfaces that need CMASK or DCC

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

gm107/ir: use lane 0 for manual textureGrad handling

This is parallel to the pre-SM50 change which does this. Adjusts the
shuffles / quadops to make the values correct relative to lane 0, and
then splat the results to all lanes for the final move into the target
register.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-By: Karol Herbst <kherbst@redhat.com>

radv/meta: fix blit paths for depth/stencil (v2.1)

This fixes the layout issue for the blit path as well.

This fixes:
dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.depth_stencil.d32_sfloat_s8_uint_d32_sfloat_s8_uint*

v2: use compatible render passes.
v2.1: use enum

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Cc: "17.2 17.3" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: handle depth/stencil image copy with layouts better. (v3.1)

If we are doing a general->general transfer with HIZ enabled,
we want to hit the tile surface disable bits in radv_emit_fb_ds_state,
however we never get the current layout to know we are in general
and meta hardcoded the transfer layout which is always tile enabled.

This fixes:
dEQP-VK.api.copy_and_blit.core.image_to_image.all_formats.depth_stencil.d32_sfloat_s8_uint_d32_sfloat_s8_uint.optimal_general
dEQP-VK.api.copy_and_blit.core.image_to_image.all_formats.depth_stencil.d32_sfloat_s8_uint_d32_sfloat_s8_uint.general_general

v2: refactor some shared helpers for blit patches
v3: we only need multiple render passes as they should be compatible.
v3.1: use enum (Bas)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Cc: "17.2 17.3" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: refactor blit2d pipeline creation

This just refactors the gfx9 blit2d pipeline creation
to be less lines of code.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/gfx9: add support for 3d images to blit 2d paths

This add support for a 3D image reading path to the blit 2d paths,
like I did for the clear paths.

Fixes: e38685cc62e 'Revert "radv: disable support for VEGA for now."'
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Alex Smith <asmith@feralinteractive.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/gfx9: add 3d sampler image->buffer copy shader. (v3)

On GFX9 we must access 3D textures with 3D samplers AFAICS.

This fixes:
dEQP-VK.api.image_clearing.core.clear_color_image.3d.single_layer

on GFX9 for me.

v1.1: fix tex->sampler_dim to dim
v2: send layer in from outside
v3: don't regress on pre-gfx9

Fixes: e38685cc62e 'Revert "radv: disable support for VEGA for now."'
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Alex Smith <asmith@feralinteractive.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: fix surface max layer count (v2)

looking at traces I noticed we'd set slice_max too large sometimes.

This should fix it.

v2: fix missing - 1

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

intel/fs: Initialize fs_visitor::grf_used on construction.

This should shut up some Valgrind errors during pre-regalloc
scheduling.  The errors were harmless since they could only have led
to the estimation of the bank conflict penalty of an instruction
pre-regalloc, which is inaccurate at that point of the program
compilation, but no less accurate than the intended "return 0"
fall-back path.  The scheduling pass is normally re-run after regalloc
with a well-defined grf_used value and accurate bank conflict
information.

Fixes: acf98ff933d "intel/fs: Teach instruction scheduler about GRF bank conflict cycles."
Reported-by: Eero Tamminen <eero.t.tamminen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

intel/fs/bank_conflicts: Use posix_memalign() instead of overaligned new to obtain vector storage.

The weight_vector_type constructor was inadvertently assuming C++17
semantics of the new operator applied on a type with alignment
requirement greater than the largest fundamental alignment.
Unfortunately on earlier C++ dialects the implementation was allowed
to raise an allocation failure when the alignment requirement of the
allocated type was unsupported, in an implementation-defined fashion.
It's expected that a C++ implementation recent enough to implement
P0035R4 would have honored allocation requests for such over-aligned
types even if the C++17 dialect wasn't active, which is likely the
reason why this problem wasn't caught by our CI system.

A more elegant fix would involve wrapping the __SSE2__ block in a
'__cpp_aligned_new >= 201606' preprocessor conditional and continue
taking advantage of the language feature, but that would yield lower
compile-time performance on old compilers not implementing it
(e.g. GCC versions older than 7.0).

Fixes: af2c320190f3c731 "intel/fs: Implement GRF bank conflict mitigation pass."
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104226
Reported-by: Józef Kucia <joseph.kucia@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

Revert "spirv: consider bitsize when handling OpSwitch cases"

This reverts commit 9702fac68e8bd07be8871f7925d7f9fb98da3699, which
hangs vulkancts and crucible on all platforms.

The patch is being reverted because it disables continuous integration
testing.  The patch from bug 104359 does not apply to master.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104359

radv: fix issue with multisample positions and interp_var_at_sample.

This fixes vmfaults seen on vega with:
dEQP-VK.pipeline.multisample_interpolation.sample_interpolate_at_single_sample_.128_128_1.samples_1

These were caused by the don't allocate cmask but it was just accidental.

The actual problem was the shader was trying to get the sample positions from
a buffer, but the buffer was never getting configured to contain them, as the
previous shader never needed them.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Fixes: 1171b304f3 (radv: overhaul fragment shader sample positions.)
Signed-off-by: Dave Airlie <airlied@redhat.com>

docs: update calendar, add news item and link release notes for 17.3.1

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

docs: add sha256 checksums for 17.3.1

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit f66496d291881f1eaca2ee5d326367fb73537541)

docs: add release notes for 17.3.1

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 4f5e85e9e97de4ae6e3d779ff42bf392c4739234)

radv/gfx9: fix primitive topology when adjacency is used

Found by inspection.

Cc: 17.3 <mesa-stable@lists.freedesktop.org>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

glsl: disable vec3 packing/splitting in tfb separate mode

This fixes a varying packing issue when using transform feedback in
GL_SEPARATE_ATTRIBS mode.  By time we get to linking, we already
know that the number of feedback attributes is under the
GL_MAX_TRANSFORM_FEEDBACK_SEPARATE_ATTRIBS limit so packing isn't
as critical.  In fact, packing/splitting vec3 attributes can cause
trouble because splitting effectively creates another TFB output
which can exceed device limits.  So, disable vec3 packing when it's
not needed to avoid that issue.

Fixes the Piglit ext_transform_feedback-separate test on VMware
driver.

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

glsl: simply packing class comparison

Handle comparing the packing class using the same method as we do
for var->data.is_xfb_only

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

glsl: document varying_matches::assign_locations() params and return value

And change *components to components[] as a reminder that it's an array.

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

glsl: remove some continue statements

In some cases, I think loop code is easier to read without continue
statements.

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

glsl: use bitwise operators in varying_matches::compute_packing_class()

The mix of bitwise operators with * and + to compute the packing_class
values was a little weird.  Just use bitwise ops instead.

v2: add assertion to make sure interpolation bits fit without collision,
per Timothy.  Basically, rewrite function to be simpler.

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

glsl: simplify loop in varying_matches::assign_locations()

The use of break/continue was kind of weird/confusing.

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

glsl: minor simplification in assign_varying_locations()

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