vbo/dlist: rework vertex_store management

Simplify the whole thing:
- keep a single function (grow_vertex_storage) that extends the array
- call it when the array is full (in ATTR_UNION) or at the beginning of
  a primitive when the vertex count is known to avoid multiple reallocations.
- allocate save->vertex_store in vbo_save_NewList if needed

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: rework primitive store handling

Simplify the logic by simply growing the primitive array when needed and
remove remove the call to compile_vertex_list in _save_End if the
array is full (it'll be resized later if needed instead).

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: rework buffer sizes

Use 20Mb as the default bo size and use the same limit when
allocating the CPU buffers in realloc_storage.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove unused functions

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: add documentation

Add a bit of documentation to describe the logic of vbo_save_api.

while at it improve the packing of vbo_save_context.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove vbo_save_context::vert_count

The same property can be deduced directly from the vertex_store,
so remove this one to avoid sync problems between the two and
simplify the code.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove vbo_save_copied_vtx

The copy_vertices() function is rarely called and removing
this decreases vbo_save_context's size by 20kb.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: fix max_index_count value

We just need an upper bound anyway.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: rework out of memory

We don't need to store/init a GLvertexformat (3kb) only to handle the
out of memory situation.
Instead, we can call _mesa_noop_vtxfmt_init when needed.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: don't force list compilation if out of vertex space

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: limit allocation sizes

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove vbo_save_context::max_vert

max_vert is easy to deduce from vertex_store so remove it to simplify
the code.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: realloc vertex stores

Same logic that for the primitives: since we're using an intermediate
RAM storage, we don't need to artificially split lists when the buffer
is full.

Instead, we can simply realloc-ate it.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove vbo_save_context::buffer_map

buffer_map pointed to the first vertex in the store for the current list.

With the previous commit, the first vertex is always at index 0, so we
can drop buffer_map.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: reset vertex_store::used in reset_counters

This means that each list will start writing its vertices at index 0.

The previous mode (append) was useful in the past when the vertices
were written directly to a mapped-buffer object.

Now that there's an intermediate storage in RAM, the added complexity
of the append-mode has no value so drop it.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove vbo_save_context::buffer_ptr

buffer_ptr is always "buffer_in_ram + used" so compute it when needed
and drop it from the struct to avoid out-of-sync errors.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: don't force list compilation if out of prim space

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: realloc prims array instead of free/malloc

Now that we don't have links to the prim_store->prims pointers, we
can replace the free/malloc by a realloc version to grow the prim
storage as needed by a list.

This will be used in the next commit to avoid splitting lists in
multiple vbo_save_vertex_list instances because the prim_store was full.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: use prim_store directly

prim_store won't be shared anymore between multiple list, so use it
directly instead of proxies.

This simplifies the code and avoid to keep them in sync.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: don't store prim_store

This is not used anywhere, so we can drop it from vbo_save_context.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: remove vbo_save_vertex_store::bufferobj

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: use a single buffer object

Instead of using 2 bo, 1 for the indices, one for the vertices store
everything in the same buffer.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: use buffer_in_ram_size

The next commit will refactor buffer upload so we need to decouple
the size from the CPU buffer holding the vertices from the size of
the GPU buffer where we're uploading them.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: move VAO update at the end

No functional changes but this is needed for future commits.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: apply start_offset after indices construction

This will make our life easier for the next commit.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

vbo/dlist: simplify add_vertex function

Use the correct pointer directly instead of adding/removing the same
offset.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12646>

freedreno/a6xx: Fix VS primid with tess + GS.

Analogous to the previous commit.

Fixes: 8115cde3ba6 ("tu, freedreno/a6xx, ir3: Rewrite tess PrimID handling")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12768>

tu: Fix VS primid with tess + GS

If tess is enabled, then VS chains to the HS and we have to use the HS
value for the PrimID sysval. This is still true if GS is also enabled,
so we need to use the HS instead of the GS if both are enabled.  In
particular if the HS or DS uses gl_PrimitiveID but the GS doesn't then
we still need to setup the register.

This fixes a bunch of dEQP-VK.tessellation.invariance.primitive_set.*
cases on a650, although it seems they managed to still pass on a630 with
the broken PrimID handling.

Fixes: 8115cde3ba6 ("tu, freedreno/a6xx, ir3: Rewrite tess PrimID handling")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12768>

nir: fix serialization of loop/if control

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fixes: e76ae39ae22 ("nir: add support for user defined select control")
Fixes: b56451f82c1 ("nir: add support for user defined loop control")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12778>

ci: Use FDO_DISTRIBUTION_TAG where possible

This allows for de duplicating some variables.

Signed-off-by: Rohan Garg <rohan.garg@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12750>

radeonsi: enable nir option pack_varying_options

flat interpolation type is not enabled because we need to set
    SPI_PS_INPUT_CNTL_x.FLAT_SHADE=1
which is per vec4.

16 bit varying also need to set
    SPI_PS_INPUT_CNTL_x.FP16_INTERP_MODE
but varying packing only support 32 bit, so we don't need to
worry about this for now.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12537>

nir/linker: pack varyings with different interpolation qualifier

Driver like radeonsi load varying in a scalar manner, so prefer to pack
varying with different interpolation qualifier into same slot to save
space.

But driver like panfrost/bifrost can load varying in vector manner,
so prefer to pack varying with same interpolation qualifier.

Driver can add interpolation qualifiers which are able to be
packed into same varying slot to pack_varying_options nir option.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12537>

nir/lower_io_to_vector: check centroid & sample when merge variable

These qualifiers should be respected for different varying load code
generation.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12537>

mesa: fix glthread deadlock when EGL multi thread shared context

This fixes dEQP EGL tests when glthread is enabled:
dEQP-EGL.functional.sharing.gles2.multithread.random.images.copyteximage2d.*
dEQP-EGL.functional.sharing.gles2.multithread.random.images.texsubimage2d.*
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_server_sync.images.teximage2d.*
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_server_sync.images.texsubimage2d.*
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_sync.images.copyteximage2d.*
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_sync.images.texsubimage2d.*

Deadlock happens when:

  Thread A: call EGL context functions which will involve _mesa_glthread_finish(), like
  eglMakeCurrent() and eglDestroyContext(). It will hold the EGLDisplay.Mutex and wait
  on glthread job queue empty (util_queue_fence_wait(&last->fence)). glthread job thread
  executes batch holding gl_context.Shared.TexMutex (glthread_unmarshal_batch()).

  Thread B: call EGLImage import functions like EGLImageTargetTexture2DOES() which will
  hold gl_context.Shared.TexMutex (egl_image_target_texture()) then validate EGLImage
  and hold EGLDisplay.Mutex (dri2_lookup_egl_image()).

This fixes the deadlock by moving the EGLImage validation out of gl_context.Shared.TexMutex
lock area.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12336>

mesa: add ValidateEGLImage driver callback

Check EGLImage imported.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12336>

gallium/api: add validate_egl_image interface

For being used by mesa state tracker.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12336>

gallium/dri: add dri_screen egl image validate hooks

For being used by EGLImage import in the following commits.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12336>

gbm/dri: implement image lookup extension version 2

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12336>

egl/dri2: seperate EGLImage validate and lookup

Version 2 of DRIImageLookupExtension add two interface for EGLImage
validate and lookup. This is for resolving deak lock in the following
commits.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12336>

docs: update calendar and link releases notes for 21.1.8

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12780>

docs: add release notes for 21.1.8

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12780>

gl.h: Remove dllimport

__declspec(dllimport) isn't needed for linking to functions, extern
is sufficient. The preprocessor define used to determine when to
set it (_DLL) was incorrect, since that preprocessor define is set
anytime you're linking against dynamic CRT as opposed to static CRT.

Reviewed-by: Bill Kristiansen <billkris@Microsoft.com>
Reviewed-by: Marek Ol\9aák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12776>

aux/trace: dump more pipe_vertex_element members

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12754>

mesa/st: use uint for instance_divisor instead of int

this is confusing

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12755>

venus: raise the ring buffer size to 64KB

When the renderer supports largeRing, raise the ring buffer size to 64KB
and raise the direct submit threshold to 4KB.  The numbers are randomly
picked.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: update venus-protocol headers

There are some decoder changes.  But we mainly want to gain largeRing
support.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: make ring buffer size configurable

Until we can assume large ring support.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: rework vn_instance_submission

The goal is to remove local_cs_data.  To that end, vn_ring_submit is
modified to take a cs than a pointer.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: add and use VN_CS_ENCODER_INITIALIZER

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: init experimental features before the ring

We can no longer use
vn_call_vkGetVenusExperimentalFeatureData100000MESA which requires the
ring to be initialized.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: support reply shmem without ring

Let vn_instance_get_reply_shmem_locked use the renderer if it is called
before the ring is initialized.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: add vn_renderer_submit_simple_sync

It submits a command stream to the renderer and waits.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: init roundtrip fields in vn_instance later

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

venus: reorder version fields in vn_instance

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12568>

docs: Update Windows llvmpipe doc for driver split

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

wgl: Refactor drivers to a libgallium_wgl.dll

I'd like to support EGL on Windows, using the same architecture as Linux.
On Linux, libgallium_dri.so is a "megadriver," containing the Mesa Gallium
state tracker, plus the actual driver implementations.

Now, on Windows, libgallium_wgl.dll is a proper OpenGL ICD, and OpenGL32.dll
is just a stub DLL which implements the wgl* APIs in terms of that ICD.

This is the more "architecturally clean" way to share the state tracker
between EGL and WGL.

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

wgl: Make overridden entrypoints local to stw_ext_context

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

wgl: Move wgl* non-extension definitions to libgl-gdi

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

wgl: Move opengl32.def to target instead of frontend

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

wgl: Don't use BUILD_GL32 for wgl frontend

The set of GL exports from OpenGL32.dll is controlled by a .def file,
there's no need to also use __declspec(dllexport) for all of them.

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

static-glapi: Fix MSVC preprocessor definitions

Previously, any DLL which links any function from libglapi-static would
end up exporting the entire GL API surface area, due to the dllexport
introduced by BUILD_GL32. This fix allows DLLs to internally include
definitions of gl* APIs without having to export the entire surface.

Also, remove unreachable with_shared_glapi branch already in an else block.

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

wgl: Fix unit test when using shared glapi

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

mapi: Fix shared-glapi build with MSVC

Reviewed By: Bill Kristiansen <billkris@microsoft.com>

Reviewed-by: Charmaine Lee >charmainel@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12677>

glx: Prevent crashes when an extension isn't found

Avoid a NULL-pointer dereference.

Fixes: 2c8a85b712e ("glx: Stash a copy of the XExtCodes in the glx_display")
Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12751>

freedreno/ir3: Update physical_predecessors for streamout block

Caught by newly added ir3_validate checks.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12742>

freedreno/ir3: Improve error msg for block level validation

Printing whatever happened to be the last instruction is misleading.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12742>

freedreno/ir3: Validate physical successors

Each block's physical successors should be a superset of the logical
successors.

Also validate that the successors are sane (ie. we shouldn't have the
2nd one if we don't have the first)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12742>

freedreno/ir3: Fix double printing of branch suffix

print_instr_name() already prints the suffix.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12742>

freedreno/ir3: Fix physical successors for break out of loop

We can't just clobber the existing successor[0] which is based on
logical successor[0]

Suggested-by: Connor Abbott <cwabbott0@gmail.com>
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12742>

freedreno/ir3: Update physical_successors after retargetting jumps

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12742>

glsl: break out early if compound assignment's operand errored out

Fixes compiler crashes on:

struct Foo
{
  float does_exist_member;
};

in vec2 tex;
out vec4 color;

void
main(void)
{
  Foo foo;

  foo.does_not_exist_member %= 3; /* or any of: <<=, >>=, &=, |=, ^= */
  color = vec4(tex.xy, tex.xy);
}

Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
CC: mesa-stable
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12717>

glsl: propagate errors from *=, /=, +=, -= operators

Fixes compiler crash on:

void main()
{
    gl_FragColor = a += 1;
}

(a is not declared anywhere)

Found with AFL++.

Fixes: d1fa69ed61d ("glsl: do not attempt assignment if operand type not parsed correctly")
Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12717>

zink: fix some pipe caps for max instructions

all shaders are supported now

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12760>

zink: use dispatch table for (almost) all vulkan calls

try to avoid hitting the loader when possible

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12757>

lima: Improve error messages for unsupported GP operations

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12738>

lima: Fix crashes for GPUs with more than four cores

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12738>

lima: Enable PIPE_CAP_VERTEX_COLOR_UNCLAMPED

Fixes lighting being too bright in Neverball.

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12738>

glsl: fix variable scope for do-while loops

Without this the following code was successfully compiling.

   void main()
   {
      do
         int var_1 = 0;
      while (false);

      var_1++;
   }

Fixes: a87ac255cf7e ("Initial commit.  lol")

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12465>

glsl: handle scope correctly when inlining loop expression

We need to clone the previously evaluated loop expression when
inlining otherwise we will have conflicts with shadow variables
defined in deeper scopes.

Fixes: 5c02e2e2de75 ("glsl: Generate IR for switch statements")

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5255

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12465>

glsl: fix variable scope for loop-expression

We need to evaluate the loop-expression of a for loop before
we evaluate the loop body otherwise we will find the wrong
variable for the following loop.

   int var_0 = 0;
   for(; var_0 < 10; var_0++) {
      const float var_0 = 1.0;
      gl_FragColor = vec4(0.0, var_0, 0.0, 0.0);
   }

Fixes: a87ac255cf7e ("Initial commit.  lol")

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5262

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12465>

mesa: fix mesa_problem() call in _mesa_program_state_flags()

Previously STATE_INTERNAL returned 0 for the unhandled states
but the outer switch throws the error for unhandled states.

Fixes: b4f3497786ef ("mesa: remove STATE_INTERNAL")

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5316

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12746>

zink: use EXT_primitive_topology_list_restart where available

this also fixes an error where PATCHES was reported as supported

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12626>

zink: hook up VK_EXT_primitive_topology_list_restart

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12626>

lavapipe: fix primitive restart with indexed indirect draws

restart index needs to be set

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12761>

util/primconvert: force restart rewrites if original primtype wasn't supported

if we're doing a primtype change, the restart needs to be eliminated to
avoid losing restart data after the rewrite

Fixes: 583070748cb ("util/primconvert: handle rewriting of prim-restart draws with unsupported primtype")

Acked-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12625>

microsoft/spirv_to_dxil: Add `install : true` to spirv_to_dxil library.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12758>

ci: Add osmesa to Windows GitLab CI

osmesa CI coverage on Windows was lost when Windows CI with
Meson build was migrated from AppVeyor to GitLab CI

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8672>

main: don't always clamp pixels read from snorm buffers

From the OpenGL 4.6 spec, section 18.2.8:

"For a signed normalized fixed-point color buffer, each component is
clamped to [0,1] if read color clamping is enabled, or if type
represents un- signed integer components; otherwise type represents
signed integer components, and each component is clamped to [−1,1]."

Signed-off-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12749>

zink: clamp query results to 500 per qbo on 32bit

this makes 4000byte qbos instead of 40000, which avoids bypassing slab
allocation and makes the buffers more reusable to avoid exploding 32bit
address space

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12752>

llvmpipe: memcpy user_buffers at set_constant_buffer time.

The data in the user buffer is only valid for a short period of time, and
we could use-after-free it if rendering hadn't been flushed by shader
deletion time.

Fixes: #5254
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12724>

llvmpipe: Fix leak of CS local memory with 0 threads.

Fixes: bde08ce4d7bf ("llvmpipe: handle compute shader launch with 0 threads")
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12724>

llvmpipe: Free CS shader images on context destroy.

Cc: mesa-stable
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12724>

iris: signal the syncobj after a failed batch

When a batch fails we either recreate our context (in case we got -EIO
or -ENOMEM) or we abort() (every other error). If we don't abort and a
later batch has a dependency on the batch that failed, then this newer
batch will fail with -EINVAL since it requires a syncobj that was
never submitted, which means we'll abort().

To avoid this problem, in this patch we simply signal syncobjs of
failed batches. This means we may be breaking our dependency tracking,
but IMHO it's better than simply letting it abort() later.

In other words, this moves the situation for some apps from "app
causes a GPU hang and then aborts" to "app causes a GPU hang but keeps
running".

Note: on some older Kernels (like today's Debian 5.10 Kernel) I see X
simply freezing after the GPU hang when the app doesn't decide to
abort(). Switching to a more recent Kernel fixes this issue for me, so
in case it happens to you make sure you have the most recent stable
trees.

v2:
  - Fix coding style (Ken).
  - Use the big comment block provided by Ken (Ken).
  - Adjust the commit message so avoid saying we retry (Ken).
  - Rebase after the syncobj ownership changes.
  - Drive-by add a missing white space in the header.

Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12657>

ci/freedreno: Skip slow SizedDeclarationsPrimitive in CI.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12753>

ci/freedreno: Mark a630 basic-glsl-misc-fs as flaky.

My branch mostly fixed it, but apparently it's not stable yet.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12753>

ci/freedreno: Fix typo in glx-tfp flake annotation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12753>

freedreno/a6xx: Emit a WFI after event writes flushing CCU.

Following on from !12475, do the same CCU workaround for freedreno as
turnip is doing, so that we flush CCU correctly for when color/depth is
next read under a different cache domain from where it was written.

Fixes instability (particularly if the GPU is clocked up) on
dEQP-GLES3.functional.fbo.blit.depth_stencil.depth32f_stencil8* in bypass
mode on a630.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11481>

ac,radv: remove unused inputs array and VS input code

The previous commit stopped using "inputs".

"load_layer_id" has always been broken and it was probably unused anyway.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12570>

ac,radeonsi: load VS inputs at the call site of nir_intrinsic_load_input

to match ACO

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12570>

ac/llvm: implement nir_intrinsic_elect

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12570>