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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
|
static const int RBM = 4, RBN = 5, RBK = 6;
static const int CBM = 24, CBN = 25, CBK = 24;
static inline void kloop(size_t p, t* a0, size_t lda, t* b0, size_t ldb, t* c, size_t ldc)
{
t* c_0 = &c[ldc*0];
t* c_1 = &c[ldc*1];
t* c_2 = &c[ldc*2];
t* c_3 = &c[ldc*3];
t c_0_0 = c_0[0];
t c_0_1 = c_0[1];
t c_0_2 = c_0[2];
t c_0_3 = c_0[3];
t c_0_4 = c_0[4];
t c_1_0 = c_1[0];
t c_1_1 = c_1[1];
t c_1_2 = c_1[2];
t c_1_3 = c_1[3];
t c_1_4 = c_1[4];
t c_2_0 = c_2[0];
t c_2_1 = c_2[1];
t c_2_2 = c_2[2];
t c_2_3 = c_2[3];
t c_2_4 = c_2[4];
t c_3_0 = c_3[0];
t c_3_1 = c_3[1];
t c_3_2 = c_3[2];
t c_3_3 = c_3[3];
t c_3_4 = c_3[4];
for (t *a = a0, *b = b0; a < a0 + p/RBK*RBK; a += RBK, b += RBK*ldb)
{
t* a_0 = &a[lda*0];
t* a_1 = &a[lda*1];
t* a_2 = &a[lda*2];
t* a_3 = &a[lda*3];
t* b_0 = &b[ldb*0];
t* b_1 = &b[ldb*1];
t* b_2 = &b[ldb*2];
t* b_3 = &b[ldb*3];
t* b_4 = &b[ldb*4];
t* b_5 = &b[ldb*5];
c_0_0 = fma(a_0[0], b_0[0], c_0_0);
c_0_1 = fma(a_0[0], b_0[1], c_0_1);
c_0_2 = fma(a_0[0], b_0[2], c_0_2);
c_0_3 = fma(a_0[0], b_0[3], c_0_3);
c_0_4 = fma(a_0[0], b_0[4], c_0_4);
c_1_0 = fma(a_1[0], b_0[0], c_1_0);
c_1_1 = fma(a_1[0], b_0[1], c_1_1);
c_1_2 = fma(a_1[0], b_0[2], c_1_2);
c_1_3 = fma(a_1[0], b_0[3], c_1_3);
c_1_4 = fma(a_1[0], b_0[4], c_1_4);
c_2_0 = fma(a_2[0], b_0[0], c_2_0);
c_2_1 = fma(a_2[0], b_0[1], c_2_1);
c_2_2 = fma(a_2[0], b_0[2], c_2_2);
c_2_3 = fma(a_2[0], b_0[3], c_2_3);
c_2_4 = fma(a_2[0], b_0[4], c_2_4);
c_3_0 = fma(a_3[0], b_0[0], c_3_0);
c_3_1 = fma(a_3[0], b_0[1], c_3_1);
c_3_2 = fma(a_3[0], b_0[2], c_3_2);
c_3_3 = fma(a_3[0], b_0[3], c_3_3);
c_3_4 = fma(a_3[0], b_0[4], c_3_4);
c_0_0 = fma(a_0[1], b_1[0], c_0_0);
c_0_1 = fma(a_0[1], b_1[1], c_0_1);
c_0_2 = fma(a_0[1], b_1[2], c_0_2);
c_0_3 = fma(a_0[1], b_1[3], c_0_3);
c_0_4 = fma(a_0[1], b_1[4], c_0_4);
c_1_0 = fma(a_1[1], b_1[0], c_1_0);
c_1_1 = fma(a_1[1], b_1[1], c_1_1);
c_1_2 = fma(a_1[1], b_1[2], c_1_2);
c_1_3 = fma(a_1[1], b_1[3], c_1_3);
c_1_4 = fma(a_1[1], b_1[4], c_1_4);
c_2_0 = fma(a_2[1], b_1[0], c_2_0);
c_2_1 = fma(a_2[1], b_1[1], c_2_1);
c_2_2 = fma(a_2[1], b_1[2], c_2_2);
c_2_3 = fma(a_2[1], b_1[3], c_2_3);
c_2_4 = fma(a_2[1], b_1[4], c_2_4);
c_3_0 = fma(a_3[1], b_1[0], c_3_0);
c_3_1 = fma(a_3[1], b_1[1], c_3_1);
c_3_2 = fma(a_3[1], b_1[2], c_3_2);
c_3_3 = fma(a_3[1], b_1[3], c_3_3);
c_3_4 = fma(a_3[1], b_1[4], c_3_4);
c_0_0 = fma(a_0[2], b_2[0], c_0_0);
c_0_1 = fma(a_0[2], b_2[1], c_0_1);
c_0_2 = fma(a_0[2], b_2[2], c_0_2);
c_0_3 = fma(a_0[2], b_2[3], c_0_3);
c_0_4 = fma(a_0[2], b_2[4], c_0_4);
c_1_0 = fma(a_1[2], b_2[0], c_1_0);
c_1_1 = fma(a_1[2], b_2[1], c_1_1);
c_1_2 = fma(a_1[2], b_2[2], c_1_2);
c_1_3 = fma(a_1[2], b_2[3], c_1_3);
c_1_4 = fma(a_1[2], b_2[4], c_1_4);
c_2_0 = fma(a_2[2], b_2[0], c_2_0);
c_2_1 = fma(a_2[2], b_2[1], c_2_1);
c_2_2 = fma(a_2[2], b_2[2], c_2_2);
c_2_3 = fma(a_2[2], b_2[3], c_2_3);
c_2_4 = fma(a_2[2], b_2[4], c_2_4);
c_3_0 = fma(a_3[2], b_2[0], c_3_0);
c_3_1 = fma(a_3[2], b_2[1], c_3_1);
c_3_2 = fma(a_3[2], b_2[2], c_3_2);
c_3_3 = fma(a_3[2], b_2[3], c_3_3);
c_3_4 = fma(a_3[2], b_2[4], c_3_4);
c_0_0 = fma(a_0[3], b_3[0], c_0_0);
c_0_1 = fma(a_0[3], b_3[1], c_0_1);
c_0_2 = fma(a_0[3], b_3[2], c_0_2);
c_0_3 = fma(a_0[3], b_3[3], c_0_3);
c_0_4 = fma(a_0[3], b_3[4], c_0_4);
c_1_0 = fma(a_1[3], b_3[0], c_1_0);
c_1_1 = fma(a_1[3], b_3[1], c_1_1);
c_1_2 = fma(a_1[3], b_3[2], c_1_2);
c_1_3 = fma(a_1[3], b_3[3], c_1_3);
c_1_4 = fma(a_1[3], b_3[4], c_1_4);
c_2_0 = fma(a_2[3], b_3[0], c_2_0);
c_2_1 = fma(a_2[3], b_3[1], c_2_1);
c_2_2 = fma(a_2[3], b_3[2], c_2_2);
c_2_3 = fma(a_2[3], b_3[3], c_2_3);
c_2_4 = fma(a_2[3], b_3[4], c_2_4);
c_3_0 = fma(a_3[3], b_3[0], c_3_0);
c_3_1 = fma(a_3[3], b_3[1], c_3_1);
c_3_2 = fma(a_3[3], b_3[2], c_3_2);
c_3_3 = fma(a_3[3], b_3[3], c_3_3);
c_3_4 = fma(a_3[3], b_3[4], c_3_4);
c_0_0 = fma(a_0[4], b_4[0], c_0_0);
c_0_1 = fma(a_0[4], b_4[1], c_0_1);
c_0_2 = fma(a_0[4], b_4[2], c_0_2);
c_0_3 = fma(a_0[4], b_4[3], c_0_3);
c_0_4 = fma(a_0[4], b_4[4], c_0_4);
c_1_0 = fma(a_1[4], b_4[0], c_1_0);
c_1_1 = fma(a_1[4], b_4[1], c_1_1);
c_1_2 = fma(a_1[4], b_4[2], c_1_2);
c_1_3 = fma(a_1[4], b_4[3], c_1_3);
c_1_4 = fma(a_1[4], b_4[4], c_1_4);
c_2_0 = fma(a_2[4], b_4[0], c_2_0);
c_2_1 = fma(a_2[4], b_4[1], c_2_1);
c_2_2 = fma(a_2[4], b_4[2], c_2_2);
c_2_3 = fma(a_2[4], b_4[3], c_2_3);
c_2_4 = fma(a_2[4], b_4[4], c_2_4);
c_3_0 = fma(a_3[4], b_4[0], c_3_0);
c_3_1 = fma(a_3[4], b_4[1], c_3_1);
c_3_2 = fma(a_3[4], b_4[2], c_3_2);
c_3_3 = fma(a_3[4], b_4[3], c_3_3);
c_3_4 = fma(a_3[4], b_4[4], c_3_4);
c_0_0 = fma(a_0[5], b_5[0], c_0_0);
c_0_1 = fma(a_0[5], b_5[1], c_0_1);
c_0_2 = fma(a_0[5], b_5[2], c_0_2);
c_0_3 = fma(a_0[5], b_5[3], c_0_3);
c_0_4 = fma(a_0[5], b_5[4], c_0_4);
c_1_0 = fma(a_1[5], b_5[0], c_1_0);
c_1_1 = fma(a_1[5], b_5[1], c_1_1);
c_1_2 = fma(a_1[5], b_5[2], c_1_2);
c_1_3 = fma(a_1[5], b_5[3], c_1_3);
c_1_4 = fma(a_1[5], b_5[4], c_1_4);
c_2_0 = fma(a_2[5], b_5[0], c_2_0);
c_2_1 = fma(a_2[5], b_5[1], c_2_1);
c_2_2 = fma(a_2[5], b_5[2], c_2_2);
c_2_3 = fma(a_2[5], b_5[3], c_2_3);
c_2_4 = fma(a_2[5], b_5[4], c_2_4);
c_3_0 = fma(a_3[5], b_5[0], c_3_0);
c_3_1 = fma(a_3[5], b_5[1], c_3_1);
c_3_2 = fma(a_3[5], b_5[2], c_3_2);
c_3_3 = fma(a_3[5], b_5[3], c_3_3);
c_3_4 = fma(a_3[5], b_5[4], c_3_4);
}
for (t *a = a0 + p/RBK*RBK, *b = b0 + p/RBK*RBK*ldb; a < a0 + p; a++, b += ldb)
{
t* a_0 = &a[lda*0];
t* a_1 = &a[lda*1];
t* a_2 = &a[lda*2];
t* a_3 = &a[lda*3];
t* b_0 = &b[ldb*0];
c_0_0 = fma(a_0[0], b_0[0], c_0_0);
c_0_1 = fma(a_0[0], b_0[1], c_0_1);
c_0_2 = fma(a_0[0], b_0[2], c_0_2);
c_0_3 = fma(a_0[0], b_0[3], c_0_3);
c_0_4 = fma(a_0[0], b_0[4], c_0_4);
c_1_0 = fma(a_1[0], b_0[0], c_1_0);
c_1_1 = fma(a_1[0], b_0[1], c_1_1);
c_1_2 = fma(a_1[0], b_0[2], c_1_2);
c_1_3 = fma(a_1[0], b_0[3], c_1_3);
c_1_4 = fma(a_1[0], b_0[4], c_1_4);
c_2_0 = fma(a_2[0], b_0[0], c_2_0);
c_2_1 = fma(a_2[0], b_0[1], c_2_1);
c_2_2 = fma(a_2[0], b_0[2], c_2_2);
c_2_3 = fma(a_2[0], b_0[3], c_2_3);
c_2_4 = fma(a_2[0], b_0[4], c_2_4);
c_3_0 = fma(a_3[0], b_0[0], c_3_0);
c_3_1 = fma(a_3[0], b_0[1], c_3_1);
c_3_2 = fma(a_3[0], b_0[2], c_3_2);
c_3_3 = fma(a_3[0], b_0[3], c_3_3);
c_3_4 = fma(a_3[0], b_0[4], c_3_4);
}
c_0[0] = c_0_0;
c_0[1] = c_0_1;
c_0[2] = c_0_2;
c_0[3] = c_0_3;
c_0[4] = c_0_4;
c_1[0] = c_1_0;
c_1[1] = c_1_1;
c_1[2] = c_1_2;
c_1[3] = c_1_3;
c_1[4] = c_1_4;
c_2[0] = c_2_0;
c_2[1] = c_2_1;
c_2[2] = c_2_2;
c_2[3] = c_2_3;
c_2[4] = c_2_4;
c_3[0] = c_3_0;
c_3[1] = c_3_1;
c_3[2] = c_3_2;
c_3[3] = c_3_3;
c_3[4] = c_3_4;
}
|
