CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
|
Structure describing the tiled structure of a GEMM-like computation.
#include <library.h>
Public Attributes | |
cutlass::gemm::GemmCoord | threadblock_shape |
Describes the shape of a threadblock (in elements) More... | |
int | threadblock_stages |
Describes the number of pipeline stages in the threadblock-scoped mainloop. More... | |
cutlass::gemm::GemmCoord | warp_count |
Number of warps in each logical dimension. More... | |
MathInstructionDescription | math_instruction |
Core math instruction. More... | |
int | minimum_compute_capability |
Minimum compute capability (e.g. 70, 75) of a device eligible to run the operation. More... | |
int | maximum_compute_capability |
Minimum compute capability (e.g. 70, 75) of a device eligible to run the operation. More... | |
|
inline |
MathInstructionDescription cutlass::library::TileDescription::math_instruction |
int cutlass::library::TileDescription::maximum_compute_capability |
int cutlass::library::TileDescription::minimum_compute_capability |
cutlass::gemm::GemmCoord cutlass::library::TileDescription::threadblock_shape |
int cutlass::library::TileDescription::threadblock_stages |
cutlass::gemm::GemmCoord cutlass::library::TileDescription::warp_count |