gallium: move pipe_draw_info::index_bias to pipe_draw_start_count_bias

this moves index_bias into the multidraw struct, enabling draws where the value
changes to be merged; the draw_info struct member is renamed and moved to the end
of the struct for tc use

u_vbuf still has some checks to split draws if index_bias changes, maybe
this can be removed at some point?

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10166>

gallium: rename pipe_draw_start_count -> pipe_draw_start_count_bias

and add an index_bias member

no functional changes yet, just the rename and unused struct member

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10166>

mesa: fix glShaderSource() error handling

Section 7.1 (SHADER OBJECTS) of the OpenGL 4.6 spec says:

   "An INVALID_VALUE error is generated if count is negative."

However a count of 0 is not an error. Previously it would cause a
GL_OUT_OF_MEMORY error.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10477>

microsoft/compiler: Maintain sorting of resource type in the context

This change moves the SRVs associated with read-only SSBOs to be emitted
before any other UAV. We do this because the validator expects resources
to be emitted in a specific order, as noted by `emit_module`.

Previously, we emitted SSBOs as SRVs (read-only) or UAVs (read-write)
after other UAVs.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10514>

v3d: DRM_FORMAT_MOD_BROADCOM_SAND128 only available for NV12 format.

We were exposing as available DRM_FORMAT_MOD_BROADCOM_SAND128 for
any format.

Fixes: 95c4f0f91098 "v3d: Enables DRM_FORMAT_MOD_BROADCOM_SAND128 support"
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10524>

v3d: YUV formats at query_dmabuf_modifiers are external_only

This fixes Issue https://github.com/Igalia/meta-webkit/issues/185
"Issue Raspberry 4-64 + Mesa VC4 driver + Gstreamer = red Label on video"

Fixes: 95c4f0f91098 "v3d: Enables DRM_FORMAT_MOD_BROADCOM_SAND128 support"
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10524>

v3d: YUV formats at is_dmabuf_modifier_supported are external_only

This fixes Issue https://github.com/Igalia/meta-webkit/issues/185
"Issue Raspberry 4-64 + Mesa VC4 driver + Gstreamer = red Label on video"

Fixes: 6ee10ab3de86 "gallium: Add pipe_screen::is_dmabuf_modifier_supported"
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10524>

glx: init __GLXvendorInfo to NULL

Since 01ba8a8d02b the dd variable isn't necessarly initialized,
so the compiler complains:

   warning: ‘dd’ may be used uninitialized in this function

Initialize dd to NULL to fix this.

Fixes: 01ba8a8d02b ("glx: Implement GLX_EXT_no_config_context")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Remove some dead declarations from glxclient.h

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: s/Display */struct glx_display */ over internal API

We'd like to avoid __glXInitialize as much as possible since it involves
taking a global lock. This means converting internal APIs to operate as
much as possible in terms of something other than a Display *, since if
that's all you have then you're forced to call __glXInitialize to get to
the glx_display.

The contortions in DRI2 displease me, but DRI2 displeases me, so.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Move server GLX vendor and version strings to glx_screen

These can in fact vary between screens.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Simplify some overuse of GetGLXScreenConfigs

If you call this on your current display and screen you're just going to
look up the same value we already stored in your context when we created
it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Stash a copy of the XExtCodes in the glx_display

Instead of a pointer into xlib's state for it. Mostly because it lets us
remove our copy of majorOpcode from the display without taking another
pointer indirection.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Remove unused opcode argument to __glX{Get,QueryServer}String

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Remove warn-once-ery around GLX 1.3 functions

Like, yes, this is an app bug, but we kinda emulate 1.3 anyway already
and are moving further in that direction, and at this point the app is
not going to get fixed. Just hush.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Simplify glXIsDirect

This answer is cached for us at context creation, we never really need
to generate protocol here.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Stop force-enabling extensions "implied" by GLX 1.3

1.3 has been ubiquitous since xserver 1.2 in early 2007, and has always
been supported with DRI2.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

glx: Stop pretending the GLX major number isn't 1

Clarify the comments and logic accordingly.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10418>

venus: handle wsi image queue ownership transfer for Android

1. pre-allocate command pools at device creation if anb enabled
2. force VK_SHARING_MODE_CONCURRENT for wsi image if necessary
3. pre-allocate and record command buffers at android wsi image creation
4. transfer in the ownership at vkAcquireImageANDROID
5. transfer out the ownership at vkQueueSignalReleaseImageANDROID

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10335>

venus: handle VK_IMAGE_LAYOUT_PRESENT_SRC_KHR transfer

Whenver VK_IMAGE_LAYOUT_PRESENT_SRC_KHR is used, replace it with
VK_IMAGE_LAYOUT_GENERAL as required for proper layout and ownership
transfer for external memory backed swapchain images.

This will be Android only until common WSI is fixed.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10335>

venus: enable VK_EXT_queue_family_foreign

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10335>

venus: update venus-protocol headers

Advertise VK_EXT_queue_family_foreign

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10335>

docs: mark off GL_ARB_shader_clock for zink

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10521>

radv: fix possible use-after-free when inserting GS copy shader from cache

If we give radv_pipeline_cache_insert_shaders() a variant which was
retrived from the cache, it will free it.

I haven't encountered this issue, but it seems possible.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10433>

radv: fix use-after-free upon GS copy shader cache hits

If radv_pipeline_cache_insert_shaders() finds a GS copy shader in the
cache, it will free the variant in gs_variants and replace it with the one
in the cache.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10433>

vulkan: fix use-after-free in vk_common_DestroyDebugReportCallbackEXT

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Fixes: bd1705a4808 ("vulkan: Make vk_debug_report_callback derive from vk_object_base")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10433>

radv: fix computation of the number of user SGPRS for NGG GS state

The NGG GS state uses one user SGPR.

Cc: 21.1 mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10485>

zink: hook up EXT_image_drm_format_modifier

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10520>

zink: hook up VK_EXT_shader_subgroup_ballot

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10520>

zink: hook up VK_EXT_conservative_rasterization

just the ext bits

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10520>

zink: hook up VK_EXT_sample_locations

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10520>

driconf: add workaround for Golf With Friends

The game has a shader that uses texture functions that rely on implicit
derivatives after a discard.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4547

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10278>

util/hash_table: do not leak u64 struct key

For non 64bit devices the key stored in hash_table_u64 is wrapped in
hash_key_u64 structure, which is never free.

This commit fixes this issue by just removing the user-defined
`delete_function` parameter in hash_table_u64_{destroy,clear} (which
nobody is using) and using instead a delete function to free this
structure.

Fixes: 608257cf82f ("i965: Fix INTEL_DEBUG=bat")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10480>

v3dv: check dest bitsize in color blit

Otherwise, if src_bit_size > 0 and dst_bit_size == 0, we end up doing a
bad shift in `1 << (dst_bit_size - 1)`, as `dst_bit_size - 1` is a
negative value (in this case would be MAX_UINT32).

Fixes CID#1468134 "Bad bit shift operation (BAD_SHIFT)":
   "large_shift: In expression 1 << dst_bit_size - 1U, left shifting by
    more than 31 bits has undefined behavior. The shift amount,
    dst_bit_size - 1U, is 4294967295."

v2:
 - Use an assertion instead (Iago)

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10251>

v3dv: rename VC5 to V3D

As we are not using anymore references to the old VC5, let's rename
definitions from VC5 to V3D in the Vulkan driver.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10402>

broadcom/simulator: change references to VC5

We are referring the driver as V3D instead old VC5; so let's update the
references.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10402>

broadcom/qpu: rename from VC5 to V3D

Get rid of old references to VC5.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10402>

v3d: rename VC5 enums and definitions

As the driver was renamed in the past from VC5 to V3D, let's rename also
the definitions and enumerations to keep it consistent across the code.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10402>

v3d: rename header include guards

Long time ago VC5 was renamed to V3D, but the include guards `VC5_FOO_H`
were not.

In order to keep consistency, let's rename these guards from `VC5_FOO_H`
to `V3D_FOO_H`.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10402>

spirv: Don't replicate patch bool in vtn_variable

When we originally added patch variable handling to spirv_to_nir, we
were splitting I/O block variables in spirv_to_nir, so we weren't
guaranteed to have a nir_variable early enough in processing.

Since b0c643d8f57 ("spirv: Use NIR per-member splitting"), we've been
using NIR per-member splitting where we have a nir_variable which has
a separate nir_variable_data per member.  With this, we can drop
vtn_variable::patch and use the patch boolean on the nir_variable
instead.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10469>

clover: Add constructor for constant_argument.

Fix defects reported by Coverity Scan.

member_not_init_in_gen_ctor: The compiler-generated constructor for this class does not initialize buf.
member_not_init_in_gen_ctor: The compiler-generated constructor for this class does not initialize st.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10399>

intel/isl: There are seven aux states

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10515>

zink: use first-created shader variant as the default

it's not really expected that most apps are going to be switching variants
all the time, so having the "default" one be more dynamic allows skipping
shader lookups more frequently

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10513>

zink: add fastpath for getting default shader variants

no need for hashing or lookups if this is the default variant

also have an alt variant for vertex stages to account for streamout and
halfz lowering

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10513>

zink: create entrypoints for descriptor variables with spirv 1.5

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10512>

zink: generate spirv 1.5 from ntv when using vk >= 1.2

this has an ntv option so further tuning here should be easy enough

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10512>

zink: export PIPE_CAP_TGSI_CLOCK

ARB_shader_clock

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10510>

zink: support nir_intrinsic_shader_clock

simple const unop

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10510>

zink: add spirv builder for unops with a const operand

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10510>

zink: add conversion util for nir_scope -> SpvScope

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10510>

zink: hook up VK_KHR_shader_clock

more extensions wooo

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10510>

ci: Uprev piglit to b3a9fa345 ("framework/replay: Quote resource names before signing")

The PIGLIT_BUILD_CL_TESTS related changes were shamelessly stolen from
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6441/diffs?commit_id=5742be00e6b493b5289ed8120eb06f10d341ea3a.

v2: Remove glslang-tools from arm_build.sh.  Suggested by Michel.

Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10492>

venus: enable external memory support

This enables VK_KHR_external_memory_fd and
VK_EXT_external_memory_dma_buf support.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: rework external memory capability queries

The idea is to allow the renderer to use a completely different external
memory handle type (e.g., OPAQUE_WIN32!?) than the driver (always
OPAQUE_FD and DMA_BUF).  It hides the mismatch by doing translations in
vkGetPhysicalDevice*.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: update venus-protocol for external memory

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: make sure gem_handle and vn_renderer_bo are 1:1

When two vn_renderer_bo's share the same gem_handle, destroying one of
them would invalidate the other.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: use sparse array to manage vn_renderer_bo

It should be faster.  More importantly, we want to use it to keep track
of all BOs for correct dmabuf import.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: move some common members to vn_renderer_bo

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: merge bo create and init ops

There is no good reason to seprate them.  I also plan to adopt
util_sparse_array, which requires the kernel BO to be created first (to
use its gem_handle/res_id as the key).

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: move vn_renderer_bo_ops to vn_renderer

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: pass vn_renderer in vn_renderer_bo functions

We will move vn_renderer_bo_ops to vn_renderer in the following commit.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: add dev->renderer pointer

dev->instance->renderer is slower to type and to follow.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: use vn_renderer_shmem

vn_renderer_bo_create_cpu becomes unused and is removed.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: add vn_renderer_shmem

CPU BOs and GPU BOs are used different enough that it makes sense to
treat them as different objects.  This commit adds vn_renderer_shmem to
represent CPU BOs.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: rename VN_CS_ENCODER_INITIALIZER

Rename it to VN_CS_ENCODER_INITIALIZER_LOCAL to make it clear that it is
only for local variables.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: update venus-protocol headers to use accessors

Switch to the newly added vn_instance_submit_command accessors.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

venus: provide accessors for vn_instance_submit_command

This will allow us to change vn_instance_submit_command without
regenerating the headers.

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10437>

docs: update calendar for 21.1.0-rc3

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10503>

docs: update calendar for 21.1.0-rc2

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10503>

docs: update calendar for 21.1.0-rc1

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10503>

freedreno/ci: Skip texsubmimage cube_map_array

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/ci: Disable counterstrike trace on a306 for now

The combination of removing bottlenecks in userspace (userspace fences,
etc) and slow GPU results in hitting full ringbuffer on a306.  Haven't
figured out a reasonable way to work around that in userspace until a
kernel fix is in place, so disable this one for now.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Assume explicit fences if in_fence_fd

If we ever see explicit fencing used, then we can disable implicit
fencing, even for internal or unfenced batches.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Async submit support

Move the submit ioctl to it's own thread to unblock the driver thread
and let it move on to the next frame.

Note that I did experiment with doing the append_bo() parts
synchronously on the theory that we should be more likely to hit the
fast path if we did that part of submit merging before the bo was
potentially re-used in the next batch/submit.  It helped some things
by a couple percent, but hurt more things.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: pipe should hold reference to device

A more direct solution would be for bo's to have a reference to the
device.  But bo's are ref/unrefd more frequently.

This avoids async submits unrefing a bo after the device handle-
table is freed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: fd_submit should hold ref to fd_pipe

Also, move this into the base class, no reason for it to be in backend.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Avoid flushing deferred submits for u_trace

This is a bit ugly, but with userspace fences and deferred submits it is
better to poll until the timestamp buffer is ready.  Otherwise we could
be triggering deferred submits to flush sooner than they normally would,
changing the behavior of what we are trying to measure.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm/sp: Implement deferred submit merging

For submits flushed with (a) no required fence, and (b) no externally
visible effects (ie. imported/exported bo), we can defer flushing the
submit and merge it into a later submit.

This is a bit more work in userspace, but it cuts down the number of
submit ioctls.  And a common case is that later submits overlap in the
bo's used (for example, blit upload to a buffer, which is then used in
the following draw pass), so it reduces the net amount of work needed
to be done in the kernel to handle the submit ioctl.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/19
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm/sp: Split submit prep and finish

For deferred submits, we still need to do the prep steps immediately,
but the ioctl construction can be deferred.. prepare for that by
splitting the prep out.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Add pipe tracking for deferred submits

Now that we have some bo state tracking for userspace fences, we can
build on this to add a way for the pipe implementation to defer a submit
flush in order to merge submits into a single ioctl.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Re-work fd_submit fence interface

Move everything into a struct assocated with the pipe_fence_handle, so
that the drm layer can fill in the seqn/fd fences directly.

This will give us a comvenient place to insert a util_queue_fence in the
next commit.

While we're at it, extract the uint32_t fence (previously called
'timestamp' in place, a kgsl legacy) into a struct that encapsulates
both the kernel fence and the userspace fence.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Reference count submits

To merge submits, we'll need drm to internally hold an extra reference
to the submit.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Drop unused create_fence() arg

It was always zero.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Move fence struct to header

Just a bit of code-motion to prep for later patches.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: last_fence optimization for TC async flushes

Without deferred flushes, we would track the last_fence any time when
there had be no rendering/etc since the last flush, to avoid creating
empty batches simply for the purpose of getting a fence.  We can do the
same thing with deferred flushes, where the fence is created from the
frontend thread before we know what will be flushed, by repopulating
the pre-created fence.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/batch: Don't create fences for every batch

We only need fences for batches flushed via pctx->flush().  This will
serve as a useful signal about submits that can be merged.

v2. disable this optimization pre-a6xx until I can debug issues that
    submit merging exposes

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Inline the fence-table

In the common case, a bo will have no more than a single fence attached.
We can inline the storage to avoid a separate allocation in this case.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Userspace fences

Add a per-fd_pipe fence "timeline" so we can detect cases where we don't
need to call into the kernel to determine if a fd_bo is still busy.

This reuses table_lock, rather than introducing a per-bo lock to protect
fence state updates because (a) the common / hotpath pattern is to
update fences on a lot of objects, but checking the fence state of a
single object is less common, and (b) because we already hold the table
lock in common spots where we need to check the bo's fence state (ie.
allocations from the bo-cache).

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Add locked version fd_{bo,pipe}_del()

This will be needed in the next patch, so we can reuse the bo table_lock
for fence related locking.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Move the growable array helper

We'll need this to track userspace fences attached to a fd_bo.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Add FD_BO_PREP_FLUSH

There are a couple cases where we want to use _NOSYNC, but at the same
time we want to ensure that rendering related to a bo is actually
flushed.

This doesn't do anything yet, but when we start deferring/merging
submits we'll need a way to trigger anything deferred to flush.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Cleanup bo cpu_prep flags

Also add some STATIC_ASSERT()

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Cleanup bo allocation flags

Most of them were actually unused.  The memory type (KMEM vs SMI) only
applied to very old a2xx era devices that had a small/fast stacked
memory (SMI) vs normal memory (KMEM).  And the cache flags are ignored
(ie. everything is writecombine), but we can add new cache flags later
when they actually do something.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno/drm: Move submit->primary to base class

Gets rid of a bit of duplication between the two current
implementations, and will be needed in next patch.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Allow resource shadowing for TC

TC_TRANSFER_MAP_NO_INVALIDATE just means we can't invalidate.  It
doesn't preclude the resource shadow path.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Add a couple debug traces

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Make sure we actually flush if we need a fence

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Avoid staging blits with stencil on older gens

This avoids infinite recursion with fallback to
util_resource_copy_region()

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

freedreno: Small indent fix

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10444>

ci/broadcom: update expected results

Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10501>