1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
|
#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//#define HW_USE_BLOCKED_MULTIPLY //* Alternative outdated way of processing in SPU (don't use)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 */
|