tensor_tiled_copy(Prefetch)

Specify the prefetch copy operation for matrices. This interface differs from tensor_tiled_copy(Trans) in its input parameters, and it uses C++ function overloading for its implementation.

The TiledCopy policy and tensor src matrix input are involved. The shape and stride of the tensor src matrix must be consistent with those defined in the TiledCopy policy.

Interface Definition

template<typename TiledCopy,

typename dtypeS, typename LayoutS>

void tensor_tiled_copy(TiledCopy copy, Tensor<dtypeS, LayoutS> src);

Template Parameters

**Table 1** Template parameter definition
Parameter	Type	Description
TiledStore	typename	Copy tiling policy type.
dtypeS	typename	Src precision type.
LayoutS	typename	Src layout type.

Parameters

**Table 2** Parameter definition
Parameter	Type	Description	Input/Output
copy	TiledCopy	Copy tiling policy.	Input
src	Tensor<dtypeS, LayoutS>	Source matrix for transposition.	Input

Return Value

void

Examples

#include <stdio.h>
#include <stdlib.h>
#include "kupl_mma.h"
using namespace kupl::tensor;

int main()
{
    double *src = (double*)malloc(sizeof(double) * 8);
    auto shape_s = make_shape(Int<8>{});
    auto stride_s = make_stride(Int<1>{});
    auto layout_s = make_layout(shape_s, stride_s);
    auto tensor_s = make_tensor(src, layout_s);

    auto atom_copy_shape = make_shape(Int<1>{});
    auto tiled_copy = make_tiled_copy(Ops<KP36_PREFETCH_L1>, atom_copy_shape);

    tensor_tiled_copy(tiled_copy, tensor_s);

    free(src);
    return 0;
}

The preceding example demonstrates an L1 prefetch operation on a double-precision buffer. A single prefetch instruction processes eight double-precision data records.

Parent topic: Matrix Programming API Functions