@@ -24,101 +24,101 @@ python3 dot_product.py
24
24
``` bash
25
25
--------------------------------------------------------------------------------
26
26
S=1024, K=1024
27
- out_f32f32: -670.21264648 , time:0.08947158ms
28
- out_f32x4f32: -670.21435547 , time:0.02821302ms
29
- out_f32f32_th: -670.21374512 , time:0.09709382ms
27
+ out_f32f32: -332.80715942 , time:0.01124835ms
28
+ out_f32x4f32: -332.80645752 , time:0.01134133ms
29
+ out_f32f32_th: -332.80691528 , time:0.01127815ms
30
30
--------------------------------------------------------------------------------
31
- out_f16f32: -670.32208252 , time:0.04000235ms
32
- out_f16x2f32: -670.15814209 , time:0.05491829ms
33
- out_f16x8packf32: -669.90997314 , time:0.01669478ms
34
- out_f16f16_th: -670.50000000 , time:0.02021313ms
31
+ out_f16f32: -333.19879150 , time:0.01110196ms
32
+ out_f16x2f32: -333.44345093 , time:0.01122665ms
33
+ out_f16x8packf32: -333.64193726 , time:0.01099825ms
34
+ out_f16f16_th: -332.75000000 , time:0.01118803ms
35
35
--------------------------------------------------------------------------------
36
36
--------------------------------------------------------------------------------
37
37
S=1024, K=2048
38
- out_f32f32: 1040.51086426 , time:0.04557490ms
39
- out_f32x4f32: 1040.50720215 , time:0.06275582ms
40
- out_f32f32_th: 1040.50842285 , time:0.04762864ms
38
+ out_f32f32: -142.86260986 , time:0.01630998ms
39
+ out_f32x4f32: -142.86064148 , time:0.01116729ms
40
+ out_f32f32_th: -142.86035156 , time:0.01143432ms
41
41
--------------------------------------------------------------------------------
42
- out_f16f32: 1041.44299316 , time:0.03214121ms
43
- out_f16x2f32: 1041.79589844 , time:0.03448486ms
44
- out_f16x8packf32: 1042.22717285 , time:0.02689457ms
45
- out_f16f16_th: 1041 .00000000 , time:0.02859521ms
42
+ out_f16f32: -143.31562805 , time:0.01554394ms
43
+ out_f16x2f32: -142.84217834 , time:0.01099968ms
44
+ out_f16x8packf32: -143.60864258 , time:0.01112890ms
45
+ out_f16f16_th: -143 .00000000 , time:0.01136470ms
46
46
--------------------------------------------------------------------------------
47
47
--------------------------------------------------------------------------------
48
48
S=1024, K=4096
49
- out_f32f32: -1859.81457520 , time:0.08664179ms
50
- out_f32x4f32: -1859.81628418 , time:0.08621526ms
51
- out_f32f32_th: -1859.81933594 , time:0.08647323ms
49
+ out_f32f32: -3116.77270508 , time:0.02791572ms
50
+ out_f32x4f32: -3116.77929688 , time:0.01236105ms
51
+ out_f32f32_th: -3116.77709961 , time:0.01418424ms
52
52
--------------------------------------------------------------------------------
53
- out_f16f32: -1860.23291016 , time:0.05826116ms
54
- out_f16x2f32: -1860.91186523 , time:0.04677963ms
55
- out_f16x8packf32: -1860.25988770 , time:0.04591107ms
56
- out_f16f16_th: -1861 .00000000 , time:0.04904127ms
53
+ out_f16f32: -3118.24951172 , time:0.02777576ms
54
+ out_f16x2f32: -3118.13208008 , time:0.01556611ms
55
+ out_f16x8packf32: -3118.15527344 , time:0.01114249ms
56
+ out_f16f16_th: -3118 .00000000 , time:0.01161337ms
57
57
--------------------------------------------------------------------------------
58
58
--------------------------------------------------------------------------------
59
59
S=2048, K=1024
60
- out_f32f32: 858.98229980 , time:0.04499865ms
61
- out_f32x4f32: 858.98461914 , time:0.04623890ms
62
- out_f32f32_th: 858.98376465 , time:0.06848693ms
60
+ out_f32f32: -1549.67492676 , time:0.01551032ms
61
+ out_f32x4f32: -1549.67419434 , time:0.01115298ms
62
+ out_f32f32_th: -1549.67382812 , time:0.01146293ms
63
63
--------------------------------------------------------------------------------
64
- out_f16f32: 858.85339355 , time:0.03274632ms
65
- out_f16x2f32: 858.94274902 , time:0.02831578ms
66
- out_f16x8packf32: 859.46844482 , time:0.02884459ms
67
- out_f16f16_th: 859 .00000000 , time:0.03692698ms
64
+ out_f16f32: -1549.45434570 , time:0.01545978ms
65
+ out_f16x2f32: -1549.04064941 , time:0.01100898ms
66
+ out_f16x8packf32: -1549.04748535 , time:0.01111746ms
67
+ out_f16f16_th: -1550 .00000000 , time:0.01136041ms
68
68
--------------------------------------------------------------------------------
69
69
--------------------------------------------------------------------------------
70
70
S=2048, K=2048
71
- out_f32f32: -1205.77990723 , time:0.08356524ms
72
- out_f32x4f32: -1205.77624512 , time:0.08583307ms
73
- out_f32f32_th: -1205.77807617 , time:0.08613133ms
71
+ out_f32f32: -4219.10205078 , time:0.02766943ms
72
+ out_f32x4f32: -4219.10009766 , time:0.01223850ms
73
+ out_f32f32_th: -4219.10693359 , time:0.01404524ms
74
74
--------------------------------------------------------------------------------
75
- out_f16f32: -1205.40588379 , time:0.06001544ms
76
- out_f16x2f32: -1205.29028320 , time:0.04738235ms
77
- out_f16x8packf32: -1205.72924805 , time:0.04624581ms
78
- out_f16f16_th: -1205 .00000000 , time:0.04907203ms
75
+ out_f16f32: -4218.69335938 , time:0.02764416ms
76
+ out_f16x2f32: -4219.42822266 , time:0.01547956ms
77
+ out_f16x8packf32: -4219.27929688 , time:0.01113629ms
78
+ out_f16f16_th: -4220 .00000000 , time:0.01157045ms
79
79
--------------------------------------------------------------------------------
80
80
--------------------------------------------------------------------------------
81
81
S=2048, K=4096
82
- out_f32f32: -893.49169922 , time:0.16136765ms
83
- out_f32x4f32: -893.48596191 , time:0.16174912ms
84
- out_f32f32_th: -893.48901367 , time:0.16518927ms
82
+ out_f32f32: -2869.79296875 , time:0.05231595ms
83
+ out_f32x4f32: -2869.78149414 , time:0.02043509ms
84
+ out_f32f32_th: -2869.78759766 , time:0.02305937ms
85
85
--------------------------------------------------------------------------------
86
- out_f16f32: -894.42169189 , time:0.11468077ms
87
- out_f16x2f32: -894.61779785 , time:0.08950567ms
88
- out_f16x8packf32: -895.26538086 , time:0.08448958ms
89
- out_f16f16_th: -894 .00000000 , time:0.09156108ms
86
+ out_f16f32: -2870.39965820 , time:0.05218816ms
87
+ out_f16x2f32: -2871.60571289 , time:0.02775407ms
88
+ out_f16x8packf32: -2870.28857422 , time:0.01228762ms
89
+ out_f16f16_th: -2870 .00000000 , time:0.01509762ms
90
90
--------------------------------------------------------------------------------
91
91
--------------------------------------------------------------------------------
92
92
S=4096, K=1024
93
- out_f32f32: 141.78890991 , time:0.08385873ms
94
- out_f32x4f32: 141.78639221 , time:0.08500123ms
95
- out_f32f32_th: 141.78683472 , time:0.08647728ms
93
+ out_f32f32: -1801.87890625 , time:0.02767515ms
94
+ out_f32x4f32: -1801.88061523 , time:0.01203156ms
95
+ out_f32f32_th: -1801.88317871 , time:0.01396847ms
96
96
--------------------------------------------------------------------------------
97
- out_f16f32: 141.80113220 , time:0.05876780ms
98
- out_f16x2f32: 141.62113953 , time:0.04708385ms
99
- out_f16x8packf32: 141.15240479 , time:0.04586506ms
100
- out_f16f16_th: 141.50000000 , time:0.04933500ms
97
+ out_f16f32: -1801.71777344 , time:0.02766609ms
98
+ out_f16x2f32: -1801.05224609 , time:0.01547670ms
99
+ out_f16x8packf32: -1799.91137695 , time:0.01112270ms
100
+ out_f16f16_th: -1801.00000000 , time:0.01154137ms
101
101
--------------------------------------------------------------------------------
102
102
--------------------------------------------------------------------------------
103
103
S=4096, K=2048
104
- out_f32f32: -1238.80456543 , time:0.16236329ms
105
- out_f32x4f32: -1238.80737305 , time:0.16246724ms
106
- out_f32f32_th: -1238.80859375 , time:0.16496468ms
104
+ out_f32f32: 643.72991943 , time:0.05231857ms
105
+ out_f32x4f32: 643.72863770 , time:0.02044320ms
106
+ out_f32f32_th: 643.73022461 , time:0.02305865ms
107
107
--------------------------------------------------------------------------------
108
- out_f16f32: -1238.78466797 , time:0.11416745ms
109
- out_f16x2f32: -1239.28540039 , time:0.08488607ms
110
- out_f16x8packf32: -1238.85302734 , time:0.08867455ms
111
- out_f16f16_th: -1239 .00000000 , time:0.09029007ms
108
+ out_f16f32: 644.73352051 , time:0.05214262ms
109
+ out_f16x2f32: 644.69067383 , time:0.02766657ms
110
+ out_f16x8packf32: 644.65740967 , time:0.01228309ms
111
+ out_f16f16_th: 644 .00000000 , time:0.01508307ms
112
112
--------------------------------------------------------------------------------
113
113
--------------------------------------------------------------------------------
114
114
S=4096, K=4096
115
- out_f32f32: 556.32690430 , time:0.31692672ms
116
- out_f32x4f32: 556.33087158 , time:0.31752276ms
117
- out_f32f32_th: 556.32879639 , time:0.32040811ms
118
- --------------------------------------------------------------------------------
119
- out_f16f32: 554.45031738 , time:0.23417449ms
120
- out_f16x2f32: 553.61444092 , time:0.16469955ms
121
- out_f16x8packf32: 554.04040527 , time:0.16465998ms
122
- out_f16f16_th: 554.50000000 , time:0.17046404ms
115
+ out_f32f32: 7372.59375000 , time:0.17362595ms
116
+ out_f32x4f32: 7372.59960938 , time:0.18044138ms
117
+ out_f32f32_th: 7372.58251953 , time:0.18282819ms
118
+ --------------------------------------------------------------------------------
119
+ out_f16f32: 7371.09033203 , time:0.10100150ms
120
+ out_f16x2f32: 7371.48632812 , time:0.05214143ms
121
+ out_f16x8packf32: 7369.69873047 , time:0.02043009ms
122
+ out_f16f16_th: 7372.00000000 , time:0.02451396ms
123
123
--------------------------------------------------------------------------------
124
124
```
0 commit comments