Glossary#

This glossary is to help users understand the basic concepts or terminologies used in the rocPRIM library.

Terminologies

Warp#: Refers to a group of threads that execute in SIMT (Single Instruction, Multiple Thread) fashion. Also known as wavefronts on AMD GPUs.
Hardware Warp Size#: Refers to the number of threads in a warp defined by the hardware. On Nvidia GPUs, a warp size is 32 while on AMD GPUs, a warp size is 64.
Logical Warp Size#: Refers to the number of threads in a warp defined by the user, which can be equal to or less than the size of the hardware warp size.
Lane ID#: Refers to the thread identifier within the warp. A logical lane ID refers to the thread identifier in a “logical warp”, which can be smaller than a hardware warp size (And can be defined as lane_id() % WarpSize).
Warp ID#: Refers to the identifier of the hardware/logical warp in a block. Warp ID is guaranteed to be unique among warps.
Block#: Refers to a group of threads that are executed on the same compute unit (streaming multiprocessor). These threads can n be indexed using 1 Dimension {X}, 2 Dimensions {X, Y} or 3 Dimensions {X, Y, Z}. A block consists of multiple warps.
Tile#: Refers to a block, but in the C++AMP/HCC nomenclature.
Flat ID#: Refers to a flattened identifier of a block (tile) or a thread identifier. Flat ID is a 1D value created from 2D or 3D n identifier. Example: flat id of thread id (X, Y) in 2D thread block 128x4 (XxY) is Y * 128 + X.