IndexedAccessOpInterfaceImpl.cpp File Reference

#include "mlir/Dialect/GPU/Transforms/IndexedAccessOpInterfaceImpl.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "mlir/Dialect/MemRef/IR/MemoryAccessOpInterfaces.h"
#include "mlir/IR/Dialect.h"
#include "mlir/IR/Operation.h"
#include "mlir/IR/PatternMatch.h"

Go to the source code of this file.

Functions
static int64_t	get1DAccessSize (MMAMatrixType matrixType, int64_t leadingDim, bool transpose)
	Given a GPU matrix type that will be loaded or stored, the leading dimension of the matrix in memory, and whether or not the matrix is transposed, compute the size of the linear memory that the load/store spans as dC + leadingDim * (dR - 1) where dR and dC are the non-contiguous and contiguous matrix dimensions, respectively (we get to the dX-1th row and then access the first dY elements of it).

Function Documentation

get1DAccessSize()

int64_t get1DAccessSize ( MMAMatrixType matrixType, int64_t leadingDim, bool transpose )

static

Given a GPU matrix type that will be loaded or stored, the leading dimension of the matrix in memory, and whether or not the matrix is transposed, compute the size of the linear memory that the load/store spans as dC + leadingDim * (dR - 1) where dR and dC are the non-contiguous and contiguous matrix dimensions, respectively (we get to the dX-1th row and then access the first dY elements of it).

Definition at line 31 of file IndexedAccessOpInterfaceImpl.cpp.

References mlir::gpu::MMAMatrixType::getShape().