heurexport.h File Reference

---

Detailed Description

Matrix type-dependent part of the heuristic management module.

#include <oski/matrix.h>

Go to the source code of this file.

Heuristic type module implementations.
#define	oski_HeurIsApplicable   MANGLE_MOD_(oski_HeurIsApplicable)
#define	oski_HeurEstimateCost   MANGLE_MOD_(oski_HeurEstimateCost)
#define	oski_HeurEvaluate   MANGLE_MOD_(oski_HeurEvaluate)
#define	oski_HeurApplyResults   MANGLE_MOD_(oski_HeurApplyResults)
Defines
#define	INC_OSKI_HEUREXPORT_H
	heurexport.h included.
Functions
int	oski_HeurIsApplicable (const oski_matrix_t A)
	Heuristic module-specific implementation of oski_IsHeurApplicable().
double	oski_HeurEstimateCost (const oski_matrix_t A)
	Heuristic module-specific implementation of oski_EstimateHeurCost().
void *	oski_HeurEvaluate (const oski_matrix_t A)
	Heuristic module-specific implementation of oski_EvaluateHeur().
int	oski_HeurApplyResults (const void *results, oski_matrix_t A)
	Heuristic module-specific implementation of oski_ApplyHeurResults().

---

Function Documentation

double oski_HeurEstimateCost (  const oski_matrix_t  A  )

Heuristic module-specific implementation of oski_EstimateHeurCost(). Todo: This implementation currently assumes our pessimistic upper-bound on the heuristic evaluation cost of 40 x (the time to stream through the matrix), and should be changed to use a build-time benchmark.

void* oski_HeurEvaluate (  const oski_matrix_t  A  )

Heuristic module-specific implementation of oski_EvaluateHeur(). Returns: Returns a oski_symmrb_t structure if the heuristic could be evaluated and returned a non-trivial block size (i.e., block size not equal to 1x1). Note: This heuristic uses the same approximate fill ratio used in the BCSR-heuristic case.

int oski_HeurIsApplicable (  const oski_matrix_t  A  )

Heuristic module-specific implementation of oski_IsHeurApplicable(). This implementation says register blocking is 'applicable' (i.e., could be considered) for the given matrix if all of the following conditions hold: The matrix has a CSR or CSC representation. AND at least one of the following conditions holds: The user explicitly provided a hint that the structure has blocks (HINT_SINGLE_BLOCKSIZE, HINT_MULTIPLE_BLOCKSIZES), and that such blocks are aligned (HINT_ALIGNED_BLOCKS). The execution time is 'dominated' by sparse matrix-vector multiply. For the definnition of 'dominated', see IsTimeDominatedByMatMult().

---