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
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
|
#include "float_mat.h"
#define ADD +=
#define MUL *
void REAL_mat_mul1(unsigned m, unsigned n, unsigned p,
REAL * restrict c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
c[i*stride_c+k] = 0;
}
}
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
for(unsigned j=0; j<n; j++) {
c[i*stride_c+k] ADD (a[i*stride_a+j] MUL b[j*stride_b+k]);
}
}
}
}
void REAL_mat_mul2(unsigned m, unsigned n, unsigned p,
REAL * restrict c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
REAL total = 0;
for(unsigned j=0; j<n; j++) {
total ADD (a[i*stride_a + j] MUL b[j*stride_b + k]);
}
c[i*stride_c+k] = total;
}
}
}
void REAL_mat_mul3(unsigned m, unsigned n, unsigned p,
REAL * restrict c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
REAL total0 = 0, total1 = 0;
unsigned j;
for(j=0; j+1<n; j+=2) {
total0 ADD (a[i*stride_a + j] MUL b[j*stride_b + k]);
total1 ADD (a[i*stride_a + (j+1)] MUL b[(j+1)*stride_b + k]);
}
if (j < n) {
total0 ADD a[i*stride_a + j] MUL b[j*stride_b + k];
}
total0 ADD total1;
c[i*stride_c+k] = total0;
}
}
}
void REAL_mat_mul4(unsigned m, unsigned n, unsigned p,
REAL * c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
const REAL *pa_i = a;
REAL * pc_i = c;
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
const REAL *pb_j_k = b+k, *pa_i_j = pa_i;
REAL total = 0;
for(unsigned j=0; j<n; j++) {
total ADD (*pa_i_j MUL *pb_j_k);
pa_i_j ++;
pb_j_k += stride_b;
}
pc_i[k] = total;
}
pa_i += stride_a;
pc_i += stride_c;
}
}
void REAL_mat_mul5(unsigned m, unsigned n, unsigned p,
REAL * c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
const REAL *pa_i = a;
REAL * pc_i = c;
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
const REAL *pb_j_k = b+k, *pa_i_j = pa_i;
REAL total = 0;
for(unsigned j2=0, n2=n/2; j2<n2; j2++) {
REAL p0 = *pa_i_j MUL *pb_j_k;
pa_i_j ++;
pb_j_k += stride_b;
REAL p1 = *pa_i_j MUL *pb_j_k;
pa_i_j ++;
pb_j_k += stride_b;
total ADD p0;
total ADD p1;
}
if (n%2) {
total ADD *pa_i_j MUL *pb_j_k;
}
pc_i[k] = total;
}
pa_i += stride_a;
pc_i += stride_c;
}
}
#define CHUNK \
total ADD (*pa_i_j MUL *pb_j_k); \
pa_i_j ++; \
pb_j_k += stride_b;
void REAL_mat_mul6(unsigned m, unsigned n, unsigned p,
REAL * c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
const REAL *pa_i = a;
REAL * pc_i = c;
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
const REAL *pb_j_k = b+k, *pa_i_j = pa_i;
REAL total = 0;
unsigned j2=0, n2=n/2;
if (n2 > 0) {
do {
CHUNK
CHUNK
j2++;
} while (j2 < n2);
}
if (n%2) {
total ADD (*pa_i_j MUL *pb_j_k);
}
pc_i[k] = total;
}
pa_i += stride_a;
pc_i += stride_c;
}
}
#define UNROLL 4
void REAL_mat_mul7(unsigned m, unsigned n, unsigned p,
REAL * c, unsigned stride_c,
const REAL *a, unsigned stride_a,
const REAL *b, unsigned stride_b) {
const REAL *pa_i = a;
REAL * pc_i = c;
for(unsigned i=0; i<m; i++) {
for(unsigned k=0; k<p; k++) {
const REAL *pb_j_k = b+k, *pa_i_j = pa_i;
REAL total = 0;
{
unsigned j4=0, n4=n/UNROLL;
if (n4 > 0) {
do {
CHUNK
CHUNK
CHUNK
CHUNK
j4++;
} while (j4 < n4);
}
}
{
unsigned j4=0, n4=n%UNROLL;
if (n4 > 0) {
do {
CHUNK
j4++;
} while (j4 < n4);
}
}
pc_i[k] = total;
}
pa_i += stride_a;
pc_i += stride_c;
}
}
|
