MLIR
20.0.0git
|
Base struct to hold GPU mapping information for a given operation. More...
#include "mlir/Dialect/Linalg/TransformOps/GPUHeuristics.h"
Public Attributes | |
SmallVector< int64_t > | numThreads |
Number of threads to use for the mapping. More... | |
SmallVector< Attribute > | threadMapping |
Thread mapping attributes, one per entry of numThreads . More... | |
Base struct to hold GPU mapping information for a given operation.
Definition at line 20 of file GPUHeuristics.h.
SmallVector<int64_t> mlir::transform::gpu::MappingInfo::numThreads |
Number of threads to use for the mapping.
Note: When the number of threads used is smaller than the total number of available threads, predication ensues. It is often useful to use more threads and saturate memory bandwidth for some operations, even if others end up being predicated.
Definition at line 26 of file GPUHeuristics.h.
SmallVector<Attribute> mlir::transform::gpu::MappingInfo::threadMapping |
Thread mapping attributes, one per entry of numThreads
.
Definition at line 29 of file GPUHeuristics.h.