vk_framebuffer provides exactly the same thing, so let the core
implement the VkFramebuffer entry points for us.
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/28104>
The vk runtime provides a default implementation, drop the stubs we
had.
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/28104>
Make sure the object name reflects the fact this object is related
to vkCmdEvent not vkEvent.
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/28104>
The texture descriptor is stored in a CPU-only template that gets copied
to the descriptor set. No need to reserve space in the texture payload
BO.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28104>
Doesn't make a difference in practice, because we are allocating with
zalloc, but let's explicitly call util_dynarray_init() for correctness.
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/28104>
SCOPE_COMMAND means the allocation last only for the vkXxx call. Both
batches and clear values live longer than that.
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/28104>
No panvk_device_set_lost() users yet, and if we start reporting
device-lost events, we will use vk_device_set_lost() directly.
Replace the only call to panvk_device_is_lost() by a vk_device_is_lost()
call.
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/28104>
VkDeviceMemory objects passed to BindImageMemory2() are not allowed to
bu VK_NULL_HANDLE, so let's drop the if (mem) test and assert()
instead.
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/28104>
We recently stopped enforcing clang-format in CI, but we still want to
be able to use it on new changes without pulling unrelated formatting
changes. Let's run it one more time to fix the existing formatting
issues.
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/28104>
Destroy the decode context when all private BOs attached to the device
have been destroyed, otherwise we a have use-after-free situation when
panvk_priv_bo_destroy() calls pandecode_inject_mmap().
Fixes: 211aa20194 ("panvk: Move away from panfrost_{bo,device}")
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/28104>
Summary:
- ensure headers used outside runtime are included in dependency source
- drop redundant idep_vulkan_common_entrypoints_h
- drop redundant icd side tricks for the order of header gen
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28066>
Some GPU gens have variants impacting the GPU capabilities. Let the
kmod backend report a variant ID through a new
pan_kmod_dev_props::gpu_variant field, and patch the panfrost_model
logic to match both the gpu_id and the gpu_variant.
All existing entries are assumed to have no variant, hence the
gpu_variant field assigned to zero.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Antonino Maniscalco <antonino.maniscalco@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26358>
Now that the rest of the code has been designed to not use
panfrost_{bo,device}, we can transition the remaining wrappers to
pan_kmod_{bo,vm,dev}.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
We wrap panfrost_device's BOs with a panvk_priv_bo object and store
the result in panvk_device. This way we'll be able to transition to
explicit BO allocation without changing BO users code.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
We don't need the panfrost_bo object which contains both the BO and
its CPU/GPU mappings. We store the GPU address at bind time. We also
have a hack for index buffers which are currently walked by the CPU
to extract the min/max index. Add a comment to make sure this field
goes away when indirect draw is fixed.
While at it, keep a ref to the buffer object so we don't end up with
a invalid deref (UAF) if the vulkan user does something silly like
freeing the VkDeviceMemory object while the VkBuffer is still active.
Flag this with a TODO to make sure we don't forget about it.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
We don't need the panfrost_bo object which contains both the BO and
its CPU/GPU mappings. We store the GPU address at bind time and store
it in the pimage object.
While at it, keep a ref to the buffer object so we don't end up with
a invalid deref (UAF) if the vulkan user does something silly like
freeing the VkDeviceMemory object while the VkImage is still active.
Flag this with a TODO to make sure we don't forget about it.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
Those are all the BOs that are allocated for internal use (descriptor
allocation, mostly). Such BOs get allocated and immediately mapped in
CPU and GPU VMs.
The new panvk_priv_bo abstraction uses a pan_kmod_bo as we're trying
to get away from panfrost_bo in panvk.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
Will be needed for when we will transition away from panfrost_{bo,device}.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
If we want to get rid of panfrost_device in panvk, we need to instantiate
our own blend shader and blitter caches.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
For now it points to panfrost_device::decode_ctx, but it will be
explicitly instantiated when we move away from panfrost_{device,bo}.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
Each logical device should come with its own GPU VM space, and given the
GPU VM space is attached the FD in panfrost, we need to move the
panfrost_device to panvk_device. As a result, we also need to move the
meta stuff there, because the meta logic is allocating GPU buffers.
Note that we instantiate a new kmod_dev at the physical device level
so we can query device properties. This device shouldn't be used for
any buffer allocation or GPU submission though.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
There's no reason to dereference the physical device if we can use
the value stored in vk_device, plus it gets rid of a few
panfrost_device_fd() call sites, which will help us transition away
from panfrost_device.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
This allows us to get rid of some panvk_physical_device::pdev
accesses.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
We're better off passing the panvk_physical_device device directly,
as it contains both the meta cache and the panfrost_device object.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
All the helpers use pan_image_mem::bo just to get the base address.
So let's just store this information here, and let the user keep track
of the BO that's attached to the image resource.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
In order to do that, we create a pan_sample.h header, and make it so
the buffer allocation is left to the caller.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Constantine Shablya <constantine.shablya@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26698>
