Needed for timeline semaphores to work.
Value taken from the toolkit documentation v11.5.0.

Signed-off-by: leozhang <nowerzt@gmail.com>

When building VLC for UWP ARM64 with aarch64-w64-mingw32 from the llvm-mingw toolchain,
this patch fixes the error: incompatible pointer types passing 'CUdeviceptr *' (aka 'unsigned int *')
to parameter of type 'unsigned long long *' [-Werror,-Wincompatible-pointer-types]

These functions can be used to create HW cuda device using primary
contexts - which is the preferred approach.
Signed-off-by: Oleg Dobkin <olegd@anyvision.co>
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>

Use of void * can lead to silent errors if you accidentally pass
the address of a pointer instead of the pointer.

I got the struct names from the API docs, although having those
match is not actually important.

This now matches the pattern used in the official headers.

It can be useful to determine if the decoder context is the same as the display
context.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>

NVIDIA driver implementation defines and expects ‘CUdeviceptr’ to
be 64-bit wide on the 64-bit platforms.
FFMPEG build on PPC64 architecture on the other hand, falls back to
an incorrect (32-bit wide data type) definition of this data type.
This patch adds a new build time check to pick up the correct definition
on PPC64 architecture.
Signed-off-by: BtbN <btbn@btbn.de>

This function is needed by the vf_thumbnail_cuda filter in ffmpeg.

This is needed to clean up the mipmapped array returned by the CUDA 10
vulkan interop.

Signed-off-by: Seungha Yang <seungha.yang@navercorp.com>

Specifically, two sets of functions:

1) Loading a compiled CUDA module and launching kernels from it.

2) Texture object creation and destruction.

This part of the interop API allows cuda to work with external
semaphores to synchronise work with Vulkan.

Signed-off-by: softworkz <softworkz@hotmail.com>
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>

This is needed to match a device selected from Vulkan when doing
interop.

This adds support for the CUDA/Vulkan interop API, which requires
very new driver versions. As such, we allow the symbols to be
optionally loaded. I expect that when the next Video SDK release
comes out, it will also require the very new drivers and we'll make
these symbols compulsory.

Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>

This will allow mpv to use this loader, instead of having to maintain
its own code.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>