The current approach of explicitly saving/restoring some states is
unnecessarily complicated and inefficient. For example, some meta OPs
that use memory fills/copies will have nested save/restores. This patch
is the first step towards avoiding unnecessary state re-emits around
meta OPs.
The changes are:
- Move radv_meta_saved_state to radv_cmd_buffer::state
- Add radv_meta_begin/end helpers that initialize radv_meta_saved_state
and restore states used by the meta OP
- Remove all explicit saves/restores, use the new helpers
radv_meta_begin/end is called inside the entrypoint and not some nested
helper function which means that state is only restored once per meta
OP.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39774>
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39720>
The Vulkan spec says:
"Store and resolve operations are only performed at the end of a
render pass instance that does not specify the
VK_RENDERING_SUSPENDING_BIT_KHR flag."
VK_RENDERING_SUSPENDING_BIT is also illegal with custom resolves.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39782>
This is mostly for not calling CmdBeginRendering() while rendering
is already active in order to catch potential driver issues. This
requires a small refactoring of how the rendering info is passed for
resolves though.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39782>
Right after "normal" layout transitions and just before the rendering
state is set, mostly because it doesn't need to be saved/restored
either.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39729>
This was an oversight of VK_KHR_dynamic_rendering_local_read which has
been addressed by VK_KHR_maintenance10 which introduced new flags to
give more information to implementations.
The Vulkan spec says:
"VK_RENDERING_ATTACHMENT_INPUT_ATTACHMENT_FEEDBACK_BIT_KHR is
intended to give implementations similar information as a subpass
where an attachment could be used as both a color attachment and
input attachment. Some implementations require extra work to make
this scenario work beyond just considering the image layouts.
Implementations which have no such considerations may treat this
flag as a noop. The primary use case for this flag is to enable
feedback loops inside a single shader."
"Applications are encouraged to use
VK_RENDERING_LOCAL_READ_CONCURRENT_ACCESS_CONTROL_BIT_KHR if
maintenance10 is available and they use feedback loops with
VK_KHR_dynamic_rendering_local_read. Feedback loops are still
allowed when not using the rendering flag, but the performance
implication was an oversight in the original definition of
VK_KHR_dynamic_rendering_local_read."
Because it's clearly defined by the Vulkan spec, let's just pessimize
always to avoid relying on some shaders state which require to do very
late decompression passes. This will allow us to do more cleanups and
optimizations related to the framebuffer. Also note that DRLR is still
a niche feature.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39538>
If the rendering state is inherited in the secondary, otherwise nothing
wait for the pending flushes after a decompression pass. One more
argument to stop delaying this.
Fixes
dEQP-VK.renderpasses.dynamic_rendering.partial_secondary_cmd_buff.local_read.*
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39678>
This is just wrong if the secondary uses ESO because the emitted
pipelines would be NULL in the secondary, but if the app re-binds
the same pipeline in the primary it would consider it as already
emitted. A sequence like this would break:
CmdBindPipeline(compute)
CmdDispatch()
CmdExecuteCommands() --> with ESO compute
CmdBindPipeline(compute)
CmdDispatch()
This tracking is probably useless anyways because it's unlikely that
apps will rebind the same pipeline right after CmdExecuteCommands() but
let's keep it because this is a bugfix.
Fixes
dEQP-VK.api.command_buffers.pipeline_shader_object_mix_with_secondaries.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39587>
PAL always set WD_SWITCH_ON_EOP for pre gfx10 when primitive
restart is enabled to prevent gpu hang.
It only happens when specific index stream with primitive
restart. Since we don't know what's the exact problem,
just follow PAL to disable 4x primitive rate when primitive
restart is enabled.
GFX10+ does not use this function.
Cc: mesa-stable
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39292>
If the main CS is SDMA and the gang CS is ACE, this would emit a
SDMA_FENCE packet on ACE which just hangs.
Fixes: b1938901d0 ("radv: Use SDMA fence packet when flushing gang semaphores")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39211>
For transfer queue operations that aren't supported by SDMA,
implement them with ACE (Async Compute Engine) using the pre-
existing compute copy functions.
Add a helper radv_get_pm4_cs that returns the ACE gang CS for
transfer command buffers and the main CS for graphics/compute
command buffers. Use radv_get_pm4_cs to make sure to emit the
compute commands to the correct command stream.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25594>
They will be called from the transfer copy functions.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39057>
We need to use gang semaphores in the following two scenarios:
1. Leader to follower semaphore:
Increment the leader to follower semaphore when the leader wants
to block the follower: a transfer operation on ACE needs to wait
for a previous operation on SDMA.
2. Follower to leader semaphore:
Increment the follower to leader semaphore when the follower wants
to block the leader: a transfer operation on SDMA needs to wait
for a previous operation on ACE.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39057>
Change the explanation to use "leader" and "follower" terminology.
Explain better how it is used with GFX/ACE and SDMA/ACE.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39057>
