forked from alibaba/MNN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathperformance.cl
93 lines (85 loc) · 2.39 KB
/
performance.cl
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
#define MAD_V4(x, y) \
x = mad(y, x, y); \
y = mad(x, y, x); \
x = mad(y, x, y); \
y = mad(x, y, x);
#define MAD_V16(x, y) \
MAD_V4(x, y); \
MAD_V4(x, y); \
MAD_V4(x, y); \
MAD_V4(x, y);
#define MAD_V64(x, y) \
MAD_V16(x, y); \
MAD_V16(x, y); \
MAD_V16(x, y); \
MAD_V16(x, y);
#define MAD_V128(x, y) \
MAD_V64(x, y); \
MAD_V64(x, y); \
MAD_V64(x, y); \
MAD_V64(x, y);
#define MAD_V256(x, y) \
MAD_V128(x, y); \
MAD_V128(x, y); \
MAD_V128(x, y); \
MAD_V128(x, y);
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
__kernel void float_precision(__global float* output_ptr, float mul_value) {
float mul_x = mul_value;
float mul_y = (float)get_local_id(0);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
output_ptr[get_global_id(0)] = mul_y;
}
__kernel void half4_precision(__global half* output_ptr, float mul_value) {
half mul = (half)mul_value;
half4 mul_x = (half4)(mul);
half4 mul_y = (half4)get_local_id(0);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
MAD_V256(mul_x, mul_y);
output_ptr[get_global_id(0)] = (mul_y.S0) + (mul_y.S1) + (mul_y.S2) + (mul_y.S3);
}