If you like my work and want to support what I do, support me on Ko-Fi 💜!

amffi is a wrapper of Advanced Media Framework's C++ API.

This works as a cross platform COM style API, in a similar way to AMDs own examples.

32-bit environments are currently supported through the usage of macros to change the calling convention between x86_64 and x86, and any breakage is considered a bug.

Please note, while this compiles on any OS and hardware, testing it on Linux requires installing the driver itself. For RDNA 3 and above users, AMD has a guide here, for RDNA 2 and 1 users I have written a guide here.

#include "public/common/AMFFactory.h"

int main() {
    AMF_RESULT res = AMF_OK;
    res = g_AMFFactory.Init();

    amf::AMFContextPtr context;
    res = g_AMFFactory.GetFactory()->CreateContext(&context);
}

use amf_bindings::amf_init;
use amf_bindings::version::AMF_VERSION;
use amf_bindings::factory::AMFFactory;
use amf_bindings::context::AMFContext;

fn main() {
    let library = amf_init().unwrap();

    let factory: AMFFactory = library.init_factory(AMF_VERSION).unwrap();
    let context: AMFContext = factory.create_context().unwrap();
}

Examples are WIP, and are direct ports of AMF C++ examples, and are not nessicarily the best way to use the wrapper. For example:

    if (memoryTypeIn == amf::AMF_MEMORY_VULKAN)
    {
        res = amf::AMFContext1Ptr(context)->InitVulkan(NULL);
        AMF_RETURN_IF_FAILED(res, L"InitVulkan(NULL) failed");
        PrepareFillFromHost(context, memoryTypeIn, formatIn, widthIn, heightIn, false);
    }
#ifdef _WIN32
    if (memoryTypeIn == amf::AMF_MEMORY_DX11)
    {
        res = context->InitDX11(NULL); // can be DX11 device
        AMF_RETURN_IF_FAILED(res, L"InitDX11(NULL) failed");
        PrepareFillFromHost(context, memoryTypeIn, formatIn, widthIn, heightIn, false);
    }
#endif

    #[cfg(windows)]
    if MEMORY_TYPE == amffi::core::data::AMFMemoryType::DX11 {
        context
            .init_dx11(
                None,
                amffi::core::data::AMFDXVersion::DX11_0,
            )
            .unwrap()
    }
    if MEMORY_TYPE == amffi::core::data::AMFMemoryType::Vulkan {
        context.init_vulkan(None).unwrap();
    }

Examples are meant to display how to use functionality, and make it easy to compare code to the C++ code.

Certain APIs, such as buffer observers, have been wrapped to create a more idiomatic feeling API, such as

fn import_dx_buffer(context: &AMFContext, tex: &ID3D11Texture2D) {
    let surface: AMFSurface = context.create_surface_from_dx11_native(tex);
    // Observers are unsafe, as they create a cyclical lifetime
    let observer_handle = unsafe {
        surface.add_observer(|surface: AMFSurface| {
            println!("Releasing!");
        });
    }
}

This is done by creating trait wrappers for the equivalent C++ interface, such as:

// This means both adding traits, and implementing Observer is unsafe
pub unsafe trait SurfaceObserver {
    fn on_surface_data_release(&mut self, surface: AMFSurface);
}

And wrapping them in a struct equivalent to how the AMD driver expects them, like so:

#[repr(C)]
pub(crate) struct InternalSurfaceObserver<T: SurfaceObserver> {
    vtbl: &'static AMFSurfaceObserverVtbl,
    this: T,
}

impl<T: SurfaceObserver> InternalSurfaceObserver<T> {
    pub(crate) fn new(observer: T) -> Self {
        Self {
            vtbl: &AMFSurfaceObserverVtbl {
                on_surface_data_release: internal_observer::<T>,
            },
            this: observer,
        }
    }
}

stdcall! {
    fn internal_observer<T: SurfaceObserver>(this: *mut *const AMFSurfaceObserverVtbl, surface: ManuallyDrop<AMFSurface>) {
        let this = unsafe { &mut *(this as *mut InternalSurfaceObserver<T>) };
        this.this.on_surface_data_release(^surface);
    }
}

class AMF_NO_VTABLE AMFSurfaceObserver
{
public:
    virtual void AMF_STD_CALL OnSurfaceDataRelease(AMFSurface* pSurface) = 0;
};

If a feature required for your work is missing, please open an issue.