Rather than doing a synchronous round trip to the host to query fault
count, use the shmem->global_faults field if available. The kernel is
already only reporting faults associated with the same drm_file (which
in this case maps 1:1 to guest process), so the only thing we really
miss is, if the app is using multiple contexts, _which_ context the
fault was in. But vulkan doesn't even use that, which sounds like a
good enough argument that per-submitqueue granularity isn't really
needed.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23533>
A remnant from a very early version, prior to refactoring out
fd_ringbuffer_sp to share more code btwn msm and virtgpu backends.
Somehow this never got removed.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23533>
Sync with commit f16c04291100 ("drm/xe: Rename engine to exec_queue").
With that Iris and ANV had some major renames that were done manually
as there is to many "engine" in non-related code.
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/24476>
Use the genxml helpers to pack/unpack midgard and bifrost surface
descriptors. Also changed how we describe midgard's descriptors, to make
it more straightforward. We currently only use SURFACE_WITH_STRIDE for
midgard, but pandecode should still be able to decode other surface
descriptor types.
This commit shouldn't change panfrost/pandecode behavior.
Consequenctly, this refactor also prepares pandecode to handle the
SURFACE_YUV descriptor for bifrost in the following patch.
Signed-off-by: Italo Nicola <italonicola@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21109>
We don't support any ASTC formats yet, but the textureCompressionASTC_LDR
feature was incorrectly set to true. Fix this by setting it to false and
don't advertise ASTC support for texture compression.
Fixes dEQP-VK.api.info.format_properties.compressed_formats
Fixes: 8991e646 ("pvr: Add a Vulkan driver for Imagination Technologies PowerVR Rogue GPUs")
Signed-off-by: Luigi Santivetti <luigi.santivetti@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24448>
We really want to avoid cpu jobs as much as possible. Also,
the texel buffer path we have for this should be able to
handle most cases.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24466>
this otherwise underflows the array and provides a (probably huge) garbage
value for the binding id, which then causes the driver to massively overallocate
both the layout and set/pool/buffer
the main result of this is that on radv any simple test that should be near-instant
takes 2-3 seconds to execute, which somehow nobody noticed
Fixes: e3b746e3a3 ("zink: use GPL to handle (simple) separate shader objects")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24501>
When we switch the channels by re-creating vec4 values we have to
take into account that the source values may be used in an ALU op,
and with that we have to take read-port limitations into account.
Fixes: 18a8d148d8
r600/sfn: Cleanup copy-prop into vec4 source values
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24519>
The new UAPI allows for an EXEC with zero pushes just fine so we no
longer need the no-op push just for synchronization. This lets us drop
the whole empty push we allocate per-queue as well.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>
