Skip to main content

Crate wgpu_external_frame

Crate wgpu_external_frame 

Source
Expand description

Zero-copy import of externally produced GPU frames into wgpu textures.

Browser engines, media decoders, capture pipelines, and compositors all hand their output over as a platform handle to memory the GPU already holds: a DMA-BUF file descriptor on Linux, an AHardwareBuffer on Android, an IOSurface on Apple platforms, a shared ID3D12Resource handle on Windows. This crate turns each of those into a wgpu::Texture on a device the caller already owns, without ever routing a pixel through the CPU.

Each platform gets its own module, because the handles genuinely have nothing in common: one is a file descriptor with a DRM format modifier and an explicit fence, one is a reference-counted NDK object, one is a CoreFoundation object, one is a Win32 HANDLE. There is no unifying trait, and inventing one would only hide the differences a caller has to know about anyway. A module exists only on its own platform, so these docs link the one built for the target they were generated for and name the others.

  • dma_buf — Linux DMA-BUF, imported through Vulkan external memory (VK_EXT_external_memory_dma_buf) or EGL (EGL_LINUX_DMA_BUF_EXT plus glEGLImageTargetTexture2DOES), whichever backend the wgpu device runs on.
  • ahardware_buffer — Android AHardwareBuffer, imported through Vulkan (VK_ANDROID_external_memory_android_hardware_buffer) and adopted by wgpu without a copy, except for YCbCr buffers the driver describes only through an external format, which a GPU pass converts into planes wgpu owns.
  • io_surface — macOS and iOS IOSurface, packed RGBA/BGRA or biplanar 4:2:0 YCbCr one plane per texture, imported through MTLDevice::newTextureWithDescriptor:iosurface:plane:.
  • shared_handle — Windows shared texture handles, imported through ID3D12Device::OpenSharedHandle.

Every import reaches under wgpu to its hal layer, so the caller must run on the backend that platform’s import requires; each entry point says which, and asserts it rather than guessing.

Modules§

dma_buf
Linux DMA-BUF import.

Structs§

YcbcrEncoding
How a YCbCr frame’s samples map to non-linear R’G’B’: the matrix and the code range a consumer needs to convert them.

Enums§

YcbcrMatrix
YCbCr to R’G’B’ matrix coefficients.
YcbcrRange
The code range YCbCr samples span.