radeonsi/gfx9: proper workaround for LS/HS VGPR initialization bug

When the HS wave is empty, the hardware writes the LS VGPRs starting at
v0 instead of v2. Workaround by shifting them back into place when
necessary. For simplicity, this is always done in the LS prolog.

According to the hardware team, this will be fixed in future chips,
so take that into account already.

Note that this is not a bug fix, as the bug was already worked
around by commit 166823bfd26 ("radeonsi/gfx9: add a temporary workaround
for a tessellation driver bug"). This change merely replaces the
workaround by one that should be better.

v2: add workaround code to shader only when necessary
v3: clarify the prefer_mono comment

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

ac/debug: take ASIC generation into account when printing registers

There were some overlapping changes in gfx9 especially in the CB/DB
blocks which made register dumps rather misleading.

The split is along the lines of the header files, so we'll print VI-only
fields on SI and CI, for example, but we won't print GFX9 fields on
SI/CI/VI, and we won't print SI/CI/VI fields on GFX9.

Acked-by: Marek Olšák <marek.olsak@amd.com>

amd/common: pass chip_class to ac_dump_reg

Acked-by: Marek Olšák <marek.olsak@amd.com>

ac/sid_tables: add FieldTable object

Automatically re-use table entries like StringTable and IntTable do.
This allows us to get rid of the "fields_owner" logic, and simplifies
the next change.

Acked-by: Marek Olšák <marek.olsak@amd.com>

ac/sid_tables: remove unused variable varname_values

Acked-by: Marek Olšák <marek.olsak@amd.com>

radeonsi/gfx9: always flush DB metadata on framebuffer changes

This fixes GL45-CTS.shader_image_load_store.basic-glsl-earlyFragTests.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

util/ralloc: set prev-pointers correctly in ralloc_adopt

Found by inspection.

I'm not aware of any actual failures caused by this, but a precise
sequence of ralloc_adopt and ralloc_free should be able to cause
problems.

v2: make the code slightly clearer (Eric)

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

mesa/main: Fix GetTextureImage error reporting

GetTex*Image should return INVALID_ENUM if target is not valid, however,
GetTextureImage does not receive a target, and instead should return
INVALID_OPERATION if the effective target is not valid. From the
OpenGL 4.6 core profile spec, section 8.11 Texture Queries:

"An INVALID_OPERATION error is generated by GetTextureImage if the effective
 target is not one of TEXTURE_1D, TEXTURE_2D, TEXTURE_3D, TEXTURE_1D_ARRAY,
 TEXTURE_2D_ARRAY, TEXTURE_CUBE_MAP_ARRAY, TEXTURE_RECTANGLE, or
 TEXTURE_CUBE_MAP (for GetTextureImage only)."

Fixes:
KHR-GL45.direct_state_access.textures_image_query_errors

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

egl: remove unused 'Screens' array from _egl_display

This was used by EGL_MESA_screen_surface that has been removed
in commit 7a58262e58d8edac3308777def0950032628edee.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <eml.velikov@collabora.com>

Revert "radv: disable support for VEGA for now."

This reverts commit 611076a41aac3095a82dff2432943d7f8d429822.

With the two previous commits, vega shouldn't be unstable,
doesn't pass CTS, but can do a complete run, and games shouldn't
hang anymore, so bring it back online.

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

radv/gfx9: set descriptor up for base_mip to level range.

This is required on GFX9, fixes a bug in Talos where all the
mipmaps overlay each other.

Just pushing this as well as it fixes Talos.

Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: disable 1d/2d linear optimisation on gfx9.

This causes hangs in some of the CTS tests with a 2d
1536x2 texture.

This fixes hangs with:
dEQP-VK.pipeline.image.suballocation.sampling_type.combined.iew_type.1d_aray.format.r4g4b4a4_unorm_pack16.count_1.size.512x1_array_of_3
if we reenable it, make sure these don't regress.

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

radv/gfx9: fix buffer size on gfx9.

The VI sizing only applies to VI.

This fixes:
dEQP-VK.image.image_size.buffer.*

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

radv: Fix vkCopyImage with both depth and stencil aspects.

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

mesa/mtypes: repack gl_sampler_object.

160->152.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/mtypes: repack gl_texture_object.

reduces size from 1144 to 1128.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/mtypes: repack gl_shader_program_data.

This reduces the size from 144 bytes to 128 bytes.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/mtypes: reorganise gl_shader

This reduces this from 200->182 bytes.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/mtypes: repack display list structs.

This reduces each of these by 8 bytes.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/mtypes: reduce size of gl_sync_object.

Drops from 40->32 bytes.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/mtypes: reorg vertex/fragment program state.

reduces both of these by 8 bytes.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/bindless: reorder gl_bindless_image gl_bindless_sampler.

This makes these use 16-bytes instead of 24-bytes.

Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: fix a memleak when compiling the GS copy shader

Found by inspection.

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

svga: move index buffer bind flag assertion

The buffer bind flags can be promoted in svga_buffer_handle(), so
move the assertion after it. This has already been done for
vertex buffer in commit 6b4bf7e8be, but it misses the one for
index buffer.

Fixes assertion running WarThunder.

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

svga: avoid emitting redundant SetShaderResources and SetVertexBuffers

Minor performance improvement in avoiding binding the same shader resource
or the same vertex buffer for the same slot.

Tested with MTT glretrace.

v2: Per Brian's suggestion, add a helper function to do vertex buffer
    comparision.
v3: Change the helper function to vertex_buffers_equal().

Reviewed-by: Brian Paul <brianp@vmware.com>

spirv: Add support for the HelperInvocation builtin

I have no idea how this got missed but it's been missing since forever.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

loader/dri3: Use client local back to front blit in copySubBuffer if available

The copySubBuffer functionality always attempted a server side blit from
back to fake front if a fake front was present, and we weren't displaying
on a remote GPU.

Now that we always have local blit capability on modern drivers, first
attempt a local blit, and only if that fails, try the server blit.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Axel Davy <axel.davy@normalesup.org>

radeonsi/gfx9: implement primitive binning

This increases performance, but it was tuned for Raven, not Vega.
We don't know yet how Vega will perform, hopefully not worse.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: add more state flags into si_state_dsa

3 flags for primitive binning, 2 flags for out-of-order rasterization
(but that will be done some other time)

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi/gfx9: don't use BREAK_BATCH and FLUSH_DFSM if DFSM is disabled

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

vbo: fix build errors on android

incompatible pointer to integer conversion assigning to 'GLintptr' (aka 'int')
from 'const char *' [-Werror,-Wint-conversion]

      offset = indices;
             ^ ~~~~~~~

Fixes: 2d93b462b4d ("vbo: fix offset in minmax cache key")
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

docs: add news item and link release notes for 17.2.0

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

docs: add sha256 checksums for 17.2.0

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit b4473dd5191878249ccb53f40407206f1e57fa6f)

docs: Update 17.2.0 release notes

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit f5925b2897308530c64e1abf44ebc1ee0e017ada)

radeonsi: eliminate PS color outputs when colormask kills them

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: sort DBG shader flags according to pipe_shader_type

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: ensure cache flushes happen before SET_PREDICATION packets

The data is read when the render_cond_atom is emitted, so we must
delay emitting the atom until after the flush.

Fixes: 0fe0320dc074 ("radeonsi: use optimal packet order when doing a pipeline sync")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: fix ARB_transform_feedback_overflow_query on <= VI

The result written by the shader workaround needs to be written back, or
the CP may read stale data.

Fixes: 78476cfe071a ("radeonsi: enable ARB_transform_feedback_overflow_query")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: fix compute shader state dumping

Fixes: 420c438589c8 ("radeonsi: log draw and compute state into log context")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: add an assertion that only two-dimensional constant references are used

v2: remove some redundant checks

Acked-by: Roland Scheidegger <sroland@vmware.com> (v1)
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de> (v1)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

gallium/radeon: always use two-dimensional constant references

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

gallium/tests: always use two-dimensional constant references

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

pp: always use two-dimensional constant references

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

gallium/hud: always use two-dimensional constant references

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

nine: always generate two-dimensional constant file accesses

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

st/glsl_to_tgsi: inline src_register into translate_src

src_register has no meaningful standalone use, it only makes sense when
called from translate_src.

v2: fix input array handling

Acked-by: Roland Scheidegger <sroland@vmware.com> (v1)
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

st/glsl_to_tgsi: ir_load_ubo always has a second index

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

st/drawpixels: always use two-dimensional constant references

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

tgsi/build: always generate two-dimensional constant file accesses

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

tgsi/ureg: always emit constants (and their decls) as 2D

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

gallium: all drivers should accept two-dimensional constant buffer indexing

Most older drivers seem to just ignore the Dimension setting, so virtually
no changes should be needed.

Acked-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

anv: fix off by one in array check

`anv_formats[ARRAY_SIZE(anv_formats)]` is already one too far.
Spotted by Coverity.

CovID: 1417259
Fixes: 242211933a0682696170 "anv/formats: Nicely handle unknown VkFormat enums"
Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

ac: reorg ac_shader_binary struct to take less space.

This reduces the size from 96 to 80 bytes but putting all the
32-bit sizes at the start.

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

radv: drop emit2d_dst_type.

This is completely unused now.

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

radv/meta: missing initialisations in create_pass().

Otherwise radv_cmd_state_setup_attachments() will complain it has no clearvalues,
when called via radv_process_depth_image_inplace().

v2: use LOAD/STORE instead of DONT_CARE, to preserve stencil values.

Signed-off-by: Xavier Bouchoux <xavierb@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

radv: Enable command buffer chaining by default.

For approx 5-10% performance improvement in dota2.

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

radv: Put semaphore waits in preamble cs.

The separate flush cs gets in the way of batchchain.

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

radv: Actually set the cmd_buffer usage_flags.

Otherwise, the simultaneous uage bit doesn't get set from the begin
info, which we need for batchchaining.

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

util: improve compiler guard

Glibc 2.26 has dropped xlocale.h, but the functions needed (strtod_l()
and strdof_l()) can be found in stdlib.h.
Improve the detection method to allow newer builds to still make use of
the locale-setting.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102454
Cc: Laurent Carlier <lordheavym@gmail.com>
Cc: Emil Velikov <emil.l.velikov@gmail.com>
Cc: Rob Herring <robh@kernel.org>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Laurent Carlier <lordheavym@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

radeon/uvd: add Define Restart Interval to MJPEG bitstream reconstruction

It adds the capacity to decode MJPEG stream with DRI marker

Signed-off-by: Leo Liu <leo.liu@amd.com>

radeon/uvd: fix MJPEG quantization table index

Fixes: 130d1f456b8 ("radeon/uvd: reconstruct MJPEG bitstream")
Signed-off-by: Leo Liu <leo.liu@amd.com>

st/mesa: fix view template initialization in try_pbo_readpixels

I think this is what the code was meant to do, albeit as far as I can tell
the redundant initialization some analyzers complain about should work as
well just fine (only the first layer will be used, if the view contains one
or more layers doesn't really matter).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102467
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: mesa-stable@lists.freedesktop.org

genxml: Make Border Color Pointer an address on Gen4-5, not an offset.

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

i965: Inline emit_reloc in __genx_combine_address

One less layer of baklava.

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

i965: Fix crash in fallback GTT mapping.

We can't perf_debug without a context.

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

i965: Fix state flagging of Gen6 SOL programs.

It doesn't seem like the old code could possibly work.

1. brw_gs_state_dirty made us bail unless one of these flags were set:
   _NEW_TEXTURE, BRW_NEW_GEOMETRY_PROGRAM, BRW_NEW_TRANSFORM_FEEDBACK
2. If there was no geometry program, we called brw_upload_ff_gs_prog()3
3. That checked brw_ff_gs_state_dirty and bailed unless these were set:
   _NEW_LIGHT, BRW_NEW_PRIMITIVE, BRW_NEW_TRANSFORM_FEEDBACK,
   BRW_NEW_VS_PROG_DATA.
4. brw_ff_gs_prog_key pv_first and attr fields were set based on data
   depending on _NEW_LIGHT and BRW_NEW_VS_PROG_DATA.

This means that if we needed a FF GS program, and changed the VS
outputs or provoking vertex mode, we'd fail to notice that we needed
to emit a new program.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Drop useless gen6_brw_upload_ff_gs_prog() wrapper.

gen6...brw?  Drop some baklava layers.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

freedreno: skip batch-cache for compute shaders

It is kind of pointless for compute, and avoids issues with apps kicking
off more than 32 compute shaders at once.

Signed-off-by: Rob Clark <robdclark@gmail.com>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>

m4: Use older autoconf 2.63 compatible ax_check_compile_flag.

CentOS 6 and RHEL 6 have autoconf 2.63.

Fixes: e4b2b69e828c ("configure: Add and use AX_CHECK_COMPILE_FLAG")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965: Move BATCH_SZ define into intel_batchbuffer.c.

It's only used in one file.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

i965: Drop batch_size argument from brw_bufmgr_init().

This is dead code and hasn't been used in a long time.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

i965: Rename brw_bo::offset64 to gtt_offset.

We can drop the meaningless "64" suffix - libdrm_intel originally had
an "offset" field that was an "unsigned long" which was the wrong size,
and we couldn't remove/alter that field without breaking ABI, so we had
to add a uint64_t "offset64" field.

"gtt_offset" is also more descriptive than "offset".

(Patch originally written by Ken, but Chris suggested a better name and
supplied the giant comment making up the bulk of the patch, so I changed
the authorship to him.)

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

i965: Drop the BRW_BATCH_STRUCT macro.

It's used in exactly one place these days, and not much simpler than
just calling intel_batchbuffer_data directly.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

i965: Don't double count the batch in aperture_space.

intel_batchbuffer_reset calls add_exec_bo on the batch right away,
which adds in the batch BO size.

Fixes: 29ba502a4e28 ("i965: Use I915_EXEC_BATCH_FIRST when available.")
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

swr: Report format max_samples=1 to maintain support for "fake" msaa.

Accompanying patch "st/mesa: only try to create 1x msaa surfaces for
'fake' msaa" requires driver to report max_samples=1 to enable "fake"
msaa. Previously, 0 and 1 were treated equivalently in st_init_extensions()
and either could enable "fake" msaa.

This patch raises the swr default msaa_max_count from 0 to 1, so that
swr_is_format_supported will report max_samples=1.

Real msaa can still be enabled by exporting SWR_MSAA_MAX_COUNT with a
pow2 value between 2 and 16.

This patch is necessary to prevent an OpenSWR regression resulting from
the st/mesa patch.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102038
Acked-by: Brian Paul <brianp@vmware.com>
Reviewed-By: George Kyriazis <george.kyriazis@intel.com>

aubinator: remove duplicate initialisation

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

radv: report VM faults if detected

It's fairly simple for now, but this might be quite useful.

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

radeonsi: move si_vm_fault_occured() to AMD common code

For radv, in order to report VM faults when detected.

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

radv: add radv_check_gpu_hangs() helper function

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

radv: disassemble SPIR-V binaries with RADV_DEBUG=spirv

This introduces a new separate option because the output can
be quite verbose. If spirv-dis is not found in the path, this
debug option is useless.

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

radv: move RADV_TRACE_FILE functions to radv_debug.c

At the moment, debugging radv is not really easy because the
driver doesn't report enough information when it hangs. This
new file will be the main location for all debug tools.

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

radv: silent a compiler warning in radv_emit_framebuffer_state()

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

radv: compute correct maximum wave count per SIMD

Ported from RadeonSI (original patch by Marek).

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

st/mesa: only try to create 1x msaa surfaces for "fake" msaa drivers

For software drivers where we want "fake" msaa support for GL 3.x, we
treat 1 sample as being msaa.

For drivers with real msaa support, start format probing at 2x msaa.
For drivers with fake msaa support, start format probing at 1x msaa.

This also tweaks the MaxSamples code in st_init_extensions() so that
we use MaxSamples=1 for fake msaa.  This allows the format proble loops
to run at least one iteration.

This fixes a llvmpipe/VTK regression from commit 6839d3369905eb02151.
And for drivers with fake msaa support, calls such as
glTexImage2DMultisample(samples=1) will now succeed.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102038
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102125
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

nvc0/ir: propagate immediates to CALL input MOVs

On using builtin functions we have to move the input to registers $0 and $1, if
one of the input value is an immediate, we fail to propagate the immediate:

...
mov u32 $r477 0x00000003 (0)
...
mov u32 $r0 %r473 (0)
mov u32 $r1 $r477 (0)
call abs BUILTIN:0 (0)
mov u32 %r495 $r1 (0)
...

With this patch the immediate is propagated, potentially causing the first MOV
to be superfluous, which we'd remove in that case:

...

mov u32 $r0 %r473 (0)
mov u32 $r1 0x00000003 (0)
call abs BUILTIN:0 (0)
mov u32 %r495 $r1 (0)
...

Shaderdb stats:
total instructions in shared programs : 4893460 -> 4893324 (-0.00%)
total gprs used in shared programs    : 582972 -> 582881 (-0.02%)
total local used in shared programs   : 17960 -> 17960 (0.00%)

                local        gpr       inst      bytes
    helped           0          91         112         112
      hurt           0           0           0           0

v2:
 implement some changes proposed by imirkin, the manual deletion of the dead
 mov is necessary after ea22ac23e0 ("nvc0/ir: unlink values pre- and post-call
 to division function") as the potentially dead mov is unlinked properly,
 causing later passes to not notice the mov op at all and thus not cleaning it
 up. That makes up a big chunk of the regression the above commit caused.
 Keep the deletion of the op where it is, deleting it later unnecessarily blows
 up size of the change.

Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0: write 0 to pipeline_statistics.cs_invocations

cs_invocations are currently unsupported, but leaving the field uninitialized
is even worse.

fixes on nvc0:
 * KHR-GL45.pipeline_statistics_query_tests_ARB.functional_default_qo_values
 * KHR-GL45.pipeline_statistics_query_tests_ARB.functional_non_rendering_commands_do_not_affect_queries

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org

llvmpipe: lp_build_gather_elem_vec BE fix for 3x16 load

Fix loading of a 3x16 vector as a single 48-bit load
on big-endian systems (PPC64, S390).

Roland Scheidegger's commit e827d9175675aaa6cfc0b981e2a80685fb7b3a74
plus Ray Strode's patch reduce pre-Roland Piglit failures from ~4000 to ~2000.  This patch fixes
three of the four regressions observed by Ray:

- draw-vertices
- draw-vertices-half-float
- draw-vertices-half-float_gles2

One regression remains:
- draw-vertices-2101010

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100613
Cc: "17.2" "17.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Ben Crocker <bcrocker@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

gallivm: correct channel shift logic on big endian

lp_build_fetch_rgba_soa fetches a texel from a texture.
Part of that process involves first gathering the element
together from memory into a packed format, and then breaking
out the individual color channels into separate, parallel
arrays.

The code fails to account for endianess when reading the packed
values.

This commit attempts to correct the problem by reversing the order
the packed values are read on big endian systems.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100613
Cc: "17.2" "17.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Ray Strode <rstrode@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

util: only use SCHED_IDLE in pthread_setschedparam() when it's defined

Fixes build error when it's not.

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

anv/formats: Nicely handle unknown VkFormat enums

This fixes some crashes in the dEQP-VK.memory.requirements.core.* tests.
I'm not sure whether or not passing out-of-bound formats into the query
is supposed to be allowed but there's no harm in protecting ourselves
from it.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Bugzilla: https://bugs.freedesktop.org/101956
Cc: mesa-stable@lists.freedesktop.org

vbo: fix offset in minmax cache key

Instead of saving primitive offset in the minmax cache key,
save the actual buffer offset which is used in the cache lookup.

Fixes rendering artifact seen with GoogleEarth when run with
VMware driver.

v2: Per Brian's comment, initialize offset to avoid compiler warning.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Brian Paul <brianp@vmware.com>

anv: fix build errors on android

error: incompatible pointer to integer conversion initializing 'VkFence'
   (aka 'unsigned long long') with an expression of type 'void *' [-Werror,-Wint-conversion]

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

winsys/amdgpu: set AMDGPU_GEM_CREATE_VM_ALWAYS_VALID if possible v2

When the kernel supports it set the local flag and
stop adding those BOs to the BO list.

Can probably be optimized much more.

v2: rename new flag to AMDGPU_GEM_CREATE_VM_ALWAYS_VALID

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: set a per-buffer flag that disables inter-process sharing (v4)

For lower overhead in the CS ioctl.
Winsys allocators are not used with interprocess-sharable resources.

v2: It shouldn't crash anymore, but the kernel will reject the new flag.
v3 (christian): Rename the flag, avoid sending those buffers in the BO list.
v4 (christian): Remove setting the kernel flag for now

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

i965: Use BLORP for buffer object stall avoidance blits instead of BLT.

Improves performance of GFXBench4 tests at 1024x768 on a Kabylake GT2:
- Manhattan 3.1 by 1.32134% +/- 0.322734% (n=8).
- Car Chase by 1.25607% +/- 0.291262% (n=5).

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

i965: Always flush caches after blitting to a GL buffer object.

When we blit data into a buffer object, we may need to invalidate any
caches that might contain stale data, so the new data becomes visible.

For example, if the buffer object is bound as a vertex buffer, we need
to invalidate the vertex fetch cache.

While this flushing was missing, it usually happened implicitly for
non-obvious reasons: we're usually on the render ring, and calling
intel_emit_linear_blit() would require switching to the BLT ring,
causing an implicit flush.  This likely provoked the kernel to do
PIPE_CONTROLs on our behalf.  Although, Gen4-5 wouldn't have this
behavior.  At any rate, we should do it ourselves.

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

i965: Add PIPE_CONTRTOL_DATA_CACHE flush to brw_emit_mi_flush().

Although we're phasing out brw_emit_mi_flush(), we still use it in some
places in order to "flush everything".  In a number of those places, we
write data to a buffer that we may then bind as an image surface, SSBO,
or atomic buffer.  Those usages require us to flush the data cache.

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

i965: Add a brw_blorp_copy_buffers() command.

This exposes the new blorp_copy_buffer() functionality to i965.
It should be a drop-in replacement for intel_emit_linear_blit()
(other than the arguments being backwards, for consistency with BLORP).

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

blorp: Make blorp_buffer_copy work on Gen4-6.

Gen4-6 can only handle surfaces up to 8192.  Only Gen7+ can do 16384.

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

blorp: Turn anv_CmdCopyBuffer into a blorp_buffer_copy() helper.

I want to be able to copy between buffer objects using BLORP in the i965
driver.  Anvil already had code to do this, in a reasonably efficient
manner - first using large bpp copies, then smaller bpp copies.

This patch moves that logic into BLORP as blorp_buffer_copy(), so we
can use it in both drivers.

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