With this patch, VM binds remain synchronous in relation to vm_bind()
KMD backend calls. However, the syscalls required for VM bind is
reduce in 2(in the optimal cases), the syncobj create and destroy
syscall are replaced by he usage a timeline syncobj.
Next step will be make this completely asynchronous.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26805>
With this patch, VM binds remain synchronous in relation to vm_bind()
KMD backend calls. However, the syscalls required for VM bind is
reduce in 2(in the optimal cases), the syncobj create and destroy
syscall are replaced by he usage a timeline syncobj.
Next step will be make this completely asynchronous.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26805>
mesa-overlay with control arg fails to setup socket properly when multiple
adapters launched. First adapter listens to socket, blocking all remaining
adapters. This is a common occurrence with games using lauchers where the
launcher is also a vulkan app. Fixed by deferring socket listening to first
frame rendered, which should be the game.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27538>
This avoids the surprising behaviour where `--target jobname` works for
some jobs but not others, because gitlab adds `X/N` at the end of these
job names.
If the user does specify something like `jobname 1/.*` to only run the
first, the extra `\d+/\d+` is ignored, just like if the job isn't
`parallel:` and therefore doesn't end with `X/N`.
If the user really wants to fail to match parallel jobs (previous
behaviour), they can simply add a `$` at the end of the job name/regex
(but also, I don't see why someone would want that behaviour).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27530>
This is really the right thing to do; it makes libva pick up the driver
we just built in a meson devenv. Not doing so can be kinda fatal;
otherwise we can end up loading different gallium-drivers for OpenGL and
VA-API, which can end up being pretty fatal.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27514>
geometry shaders don't specify the input topology, only the class of topology.
normalize when generating a passthrough gs.
asahi will be more picky about this in the future.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Antonino Maniscalco <antonino.maniscalco@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27457>
I don't remember why I implemented it like this in !13959, but
AFAICT there's no need to manually remove this queue from vk_device's
queues list.
On the other hand, this hack causes issues if syncobj timeline isn't
supported; amdgpu always support timeline, but amdgpu over virtio-gpu
doesn't.
The issue is as follow: the sequence in vk_queue.c:
case VK_QUEUE_SUBMIT_MODE_DEFERRED:
vk_queue_push_submit(queue, submit);
return vk_device_flush(queue->base.device);
Would fail to produce the expected result, because vk_device_flush would
fail to realize that the blit queue has some pending work because
"vk_foreach_queue(queue, device)" would never process the queue.
Then, the call to vk_drm_syncobj_export_sync_file() would fail, because
the syncobj handle was never used in a submit.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27412>
Persona 3 Reload is largely affected by the way amplification works with
NGG GS and disabling it drastically improve performance.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27518>
NGG GS doesn't perform well in some cases and having an option to
disable it for performance experiments is very useful.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27518>
If otherwise no helper lanes are required by the shader, then demote
behaves like discard and immediately terminates the invocations.
With maximal reconvergence, however, we need to ensure that helper lanes
are not terminated unless the entire quad was demoted.
In order to fix this, generally enable helper lanes in this unlikely
corner case and avoid a major refactor.
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27277>
With shader object, when VS+GS or TES+GS are compiled separately and
the VS has written (but unused) outputs, the ESGS vertex stride
must be passed through an user SGPR. This is because when the GS is
compiled we can't know the number of ES outputs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27388>
Apparently, Midgard GPUs don't like when the last 2 words of
compute/vertex jobs contain garbage. Extend the compute job definition
to include a padding section thus aligning the job on a 64-byte boundary,
and add the according pan_section_pack() calls where we have a
compute job filled.
Fixes: b76420be1f ("panfrost: Split command stream descriptor definitions per-gen")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/10558
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Tested-by: Anton Bambura <jenneron@postmarketos.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27515>
Drop the Blend descriptor definition in v4.xml, and adjust
pan_blitter accordingly.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27515>
