MLIR
22.0.0git
|
#include "mlir/Dialect/GPU/TransformOps/Utils.h"
Public Member Functions | |
GpuLaneIdBuilder (MLIRContext *ctx, int64_t warpSize, bool unused, DeviceMaskingAttrInterface mask=nullptr) | |
![]() | |
GpuIdBuilder ()=default | |
GpuIdBuilder (MLIRContext *ctx, bool useLinearMapping, const MappingIdBuilderFnType &builder) | |
Public Attributes | |
int64_t | warpSize = 32 |
![]() | |
SmallVector< DeviceMappingAttrInterface > | mappingAttributes |
The mapping attributes targeted by this generator. More... | |
GpuIdBuilderFnType | idBuilder |
The constructor that builds the concrete IR for mapping ids. More... | |
Additional Inherited Members | |
![]() | |
using | MappingIdBuilderFnType = std::function< DeviceMappingAttrInterface(MLIRContext *, mlir::gpu::MappingId)> |
Builder for lane id.
The idBuilder
method returns nD values used for indexing rewrites as well as 1D sizes for predicate generation. This useLinearMapping
case is the only supported case. If mask
is provided, it will be used to filter the active lanes.
mlir::transform::gpu::GpuLaneIdBuilder::GpuLaneIdBuilder | ( | MLIRContext * | ctx, |
int64_t | warpSize, | ||
bool | unused, | ||
DeviceMaskingAttrInterface | mask = nullptr |
||
) |
Definition at line 349 of file Utils.cpp.
References mlir::get().
int64_t mlir::transform::gpu::GpuLaneIdBuilder::warpSize = 32 |