tile_io

Trait and utilities for copying data between TileTensors.

Structs

• ​GenericToLocalTileCopier: A TileCopier that moves a tile from generic memory into registers.
• ​GenericToSharedAsyncTileCopier: An AsyncTileCopier that asynchronously moves a tile from generic memory into shared memory using NVIDIA's cp.async instruction.
• ​GenericToSharedTileCopier: A TileCopier that moves a tile from generic memory into shared memory.
• ​LocalToGenericTileCopier: A TileCopier that moves a tile from registers into generic memory.
• ​LocalToSharedTileCopier: A TileCopier that moves a tile from registers into shared memory.
• ​SharedToGenericTileCopier: A TileCopier that moves a tile from shared memory into generic memory.
• ​SharedToLocalTileCopier: A TileCopier that moves a tile from shared memory into registers.

Traits

• ​AsyncTileCopier: Trait for asynchronously copying a TileTensor between address spaces.
• ​TileCopier: Trait for copying a TileTensor from one address space to another.

Functions

• ​copy_dram_to_local: Synchronously copies a tile from DRAM (generic memory) to registers.
• ​copy_dram_to_sram: Synchronously copies a tile from DRAM (generic memory) to SRAM (shared).
• ​copy_dram_to_sram_async: Asynchronously copies a tile from DRAM (generic memory) to SRAM (shared).
• ​copy_local_to_dram: Synchronously copies a tile from registers (LOCAL) to DRAM (generic).
• ​copy_local_to_shared: Synchronously copies a tile from registers (LOCAL) to SRAM (shared).
• ​copy_sram_to_dram: Synchronously copies a tile from SRAM (shared memory) to DRAM (generic).
• ​copy_sram_to_local: Synchronously copies a tile from SRAM (shared memory) to registers.