HostPinnedRing.hpp File Reference

Vendor-neutral N-slot sysmem ring for DVP / AMD-pinned / host-copy transfers. More...

Detailed Description

Vendor-neutral N-slot sysmem ring for DVP / AMD-pinned / host-copy transfers.

Counterpart to ImportedGpuBufferRing (which holds GPU-side buffers for true direct P2P). HostPinnedRing holds host-side page-locked buffers that a video card's DMA engine writes into, and a per-backend transfer primitive that moves slot contents to/from a QRhi texture.

The class is consumed by every vendor strategy that goes through host RAM on the hot path: DeckLink (all variants), AJA's DVP path, Magewell, Datapath, Osprey, USB capture, GenICam, etc. The vendor supplies the page-locked buffer (via its DMABufferLock, MWPinVideoBuffer, etc) and the slot index it just filled; the ring runs the QRhi-side ingest via the configured backend.

Backend selection is automatic:

1. DVP (NVIDIA Quadro/Tesla, Win+Linux) — fastest sysmem→texture mechanism on NVIDIA HW. Uses nv_dvp_copy_buffer_to_texture.
2. AMD-pinned (AMD, Win+Linux) — GL_AMD_pinned_memory-class buffer bound as GL_PIXEL_UNPACK_BUFFER + glTexSubImage2D. Only when QRhi is on the OpenGL backend.
3. cudaHostRegister (NVIDIA, any backend) — page-locks the slot then cuMemcpy2D into a CUDA-imported QRhi texture. Fallback when DVP isn't loaded but CUDA is.
4. CPU staging — plain QRhi uploadTexture from the slot's host pointer. Always available.

The picker walks the list in order, choosing the first backend compatible with the live QRhi backend + the GpuCapabilities probe.

Go to the source code of this file.

Classes
struct	score::gfx::interop::HostPinnedSlot
struct	score::gfx::interop::HostPinnedRingConfig
class	score::gfx::interop::HostPinnedRing

Namespaces
namespace	score
	Base toolkit upon which the software is built.
namespace	score::gfx
	Graphics rendering pipeline for ossia score.

Enumerations
enum class	score::gfx::interop::HostPinnedRingBackend : uint8_t {   None = 0 , Dvp , AmdPinned , CudaHostReg ,   CpuStaging }
enum class	score::gfx::interop::HostPinnedDirection : uint8_t { CaptureToTexture = 0 , TextureToBuffer }

Functions
const char *	score::gfx::interop::hostPinnedBackendName (HostPinnedRingBackend) noexcept
	Short name for logs and harness rows.

Enumeration Type Documentation

HostPinnedDirection

enum class score::gfx::interop::HostPinnedDirection : uint8_t

strong

Direction of the per-slot transfer.

Enumerator
CaptureToTexture	Vendor wrote into slot, ring uploads to texture.
TextureToBuffer	Ring downloads texture to slot, vendor reads.

HostPinnedRingBackend

enum class score::gfx::interop::HostPinnedRingBackend : uint8_t

strong

Which backend a HostPinnedRing instance is using. Set by create() and queryable via backend().

Enumerator
Dvp	NVIDIA DVP — fastest sysmem path.
AmdPinned	AMD GL pinned-memory or virtual-memory.
CudaHostReg	cudaHostRegister + cuMemcpy2D.
CpuStaging	QRhi uploadTexture / readbackTexture fallback.