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
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2017-2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#include "../include/BiF_Definitions.cl"
#include "../../Headers/spirv.h"
#include "../include/exp_for_hyper.cl"
#include "../IMF/FP32/tanh_s_la_noLUT.cl"
#if defined(cl_khr_fp64)
#include "../IMF/FP64/tanh_d_la.cl"
#endif // defined(cl_khr_fp64)
float SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(tanh, _f32, )( float x )
{
float result;
if( __intel_relaxed_isnan(x) )
{
result = SPIRV_OCL_BUILTIN(nan, _i32, )(0);
}
else if(__UseHighAccuracyMath)
{
result = __ocl_svml_tanhf_noLUT(x);
}
else if(SPIRV_OCL_BUILTIN(fabs, _f32, )(x) < as_float(0x3A71E7A0)) // 0.00092279352247715
{
result = x;
}
else if( SPIRV_OCL_BUILTIN(fabs, _f32, )(x) < as_float(0x3EACB527) ) // 0.33731958270072937
{
float sinhx, coshx;
{
float x2 = x * x;
float x3 = x * x2;
float x5 = x3 * x2;
sinhx = (as_float(0x3C088889) * x5) + (as_float(0x3E2AAAAB) * x3) + x;
}
{
float pexp = __intel_exp_for_tanh( x, -2.0f);
float nexp = __intel_exp_for_tanh(-x, -2.0f);
coshx = 2.0f * ( pexp + nexp );
}
result = sinhx / coshx;
}
else if (SPIRV_OCL_BUILTIN(fabs, _f32, )(x) < as_float(0x41987E0C)) // 19.061546325683594
{
float exp2x = __intel_exp_for_hyper(2 * x, 0.0f);
result = (exp2x - 1) / (exp2x + 1);
}
else
{
result = (x > 0) ? 1.0f : -1.0f;
}
return result;
}
GENERATE_SPIRV_OCL_VECTOR_FUNCTIONS_1ARG_LOOP( tanh, float, float, f32 )
#if defined(cl_khr_fp64)
INLINE double SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(tanh, _f64, )( double x )
{
return __ocl_svml_tanh(x);
}
GENERATE_SPIRV_OCL_VECTOR_FUNCTIONS_1ARG_LOOP( tanh, double, double, f64 )
#endif // defined(cl_khr_fp64)
#if defined(cl_khr_fp16)
INLINE half SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(tanh, _f16, )( half x )
{
return SPIRV_OCL_BUILTIN(tanh, _f32, )((float)x);
}
GENERATE_SPIRV_OCL_VECTOR_FUNCTIONS_1ARG_LOOP( tanh, half, half, f16 )
#endif // defined(cl_khr_fp16)
|
