|
VLink
2.3.0
A high-performance communication middleware
|
Vendor-independent allocation and sharing contract for zerocopy::FastBuffer.
More...
Go to the source code of this file.
Classes | |
| class | vlink::FastBufferPluginInterface |
| Supplies platform buffers without exposing a vendor runtime to VLink. More... | |
Namespaces | |
| vlink | |
Vendor-independent allocation and sharing contract for zerocopy::FastBuffer.
FastBufferPluginInterface separates native allocation and host access from the platform that allocates the storage. Implementations may use ordinary or pinned CPU memory, device memory, shared memory, DMA-BUF or virtual allocations. Unsupported allocation domains and optional operations report failure.
| Operation | Methods | Contract |
|---|---|---|
| Allocation and ownership | create(), release() | One owned reference per result |
| CPU mapping | map(), unmap() | Explicit access/cache scope |
| Native API integration | native_handle(), import_native() | Provider-defined local object |
| Completion | synchronize() | Wait for tracked work |
| Payload copy | copy(), copy_to_host(), copy_from_host() | Synchronous completion |
| Descriptor transport | export_handle(), import_handle() | No local pointer on the wire |
| Descriptor size | get_handle_size() | Per-allocation support |
| Dead-owner cleanup | destroy_handle() | Named OS resources only |
A zerocopy::FastBuffer::Buffer address belongs to its local process and address space. CPU callers must use map() or copy_to_host(); a non-null device address is not proof of CPU accessibility. Native objects and numeric file descriptors likewise require provider-specific import or OS handle transfer.
VLINK_FASTBUFFER_PLUGIN before the process creates its first buffer. The interface ABI and the descriptor protocol ID are separate compatibility contracts; the latter identifies interoperable native descriptors within the core envelope.