/ani/mrses

#ifndef _MRSES_IMPL_H

#define _MRSES_IMPL_H

#include <stdlib.h>

#include <math.h>

#include "mrses.h"

#include "hw_sched.h"

#include "tools.h"

#define HW_ALIGN 16		// For SPE support minimum 16 is required (should be multiple of MRSESDataType anyway)

#define SPE_BLOCK 16		// This indicates minimal block of processing by SPE code

#define SIMD_BLOCK 16		// In bytes

#define HW_ITERATE_BLOCKS 16

//#define TRACE_TIMINGS 1

//#define USE_FAST_RANDOM 1		//* default random/rand somehow sequentalize threads

//#define FIX_RANDOM 1

//#define HW_USE_BLOCKED_MULTIPLY	//* Alternative outdated way of processing in SPU (don't use)

#define blas_asum cblas_sasum

#define blas_scal cblas_sscal

#define blas_axpy cblas_saxpy

#define blas_gemv cblas_sgemv

#define blas_gemm cblas_sgemm

#define blas_syrk cblas_ssyrk

#define blas_trsm cblas_strsm

#define blas_trsv cblas_strsv

#define blas_dot  cblas_sdot

#define lapack_potrf spotrf_

struct MRSESContextT {

    MRSESDistance dist;

    int properties;		/**< Number of properties */

    int alloc;			/**< Size of vectors in A and B matrixes for alignment */

    int nA, nB;			/**< Number of samples */

    MRSESDataType *A;		/**< Actually: (A - mean(A))/sqrt(nA) */

    MRSESDataType *B;		/**< Actually: (B - mean(B))/sqrt(nB) */

    MRSESDataType *mean;	/**< Actually: mean(A) - mean(B) */

    HWSched sched;

    int iterate_size;		/**< Number of blocks per block group */

    int max_block_size;

    int block_size;

    int cur_chunk;

    int width; 

    int iterations;   

    MRSESIntType *index;	/**< In compute: provided by client app, no allocation, in iteration: allocated, indicates that iterations were started before */

    MRSESIntType *ires;		/**< Array of resulting indeces */

    MRSESDataType *result;	/**< Provided by client app, no allocation */

};

typedef struct MRSESContextT MRSESContextS;

#endif /* _MRSES_IMPL_H */