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
|
/*========================== 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 "../IMF/FP32/rootn_s_la.cl"
#if defined(cl_khr_fp64)
#include "../IMF/FP64/rootn_d_la.cl"
#endif // defined(cl_khr_fp64)
float SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(rootn, _f32_i32, )( float x, int n )
{
float result;
if(__FastRelaxedMath && (!__APIRS))
{
// Defined for x > 0 and n is nonzero. Derived
// implementations implement this as:
// exp2(log2( x) / n) for x > 0.
// Defined for x < 0 and n is odd. Derived
// implementations implement this as:
// -exp2(log2(-x) / n) for x < 0.
// Defined as +0 for x = +/-0 and y > 0.
// Undefined for all other cases.
float pr = x;
// TBD: Which is faster?
// Note that USC has a pattern match optimization to turn
// log-mul-exp into pow. Additionally, there are some specific
// LLVM optimizations for pow. So, preferring pow for now.
#if 0
pr = SPIRV_OCL_BUILTIN(fabs, _f32, )( pr );
pr = SPIRV_OCL_BUILTIN(log2, _f32, )( pr );
pr = pr * 1.0f / n;
pr = SPIRV_OCL_BUILTIN(exp2, _f32, )( pr );
#else
pr = SPIRV_OCL_BUILTIN(native_powr, _f32_f32, )( pr, 1.0f / n );
#endif
// For rootn(), we'll return the positive result for both +0.0f and -0.0f.
float nr = -pr;
result = ( x >= 0.0f ) ? pr : nr; // positive result for non-negative x, else negative result
}
else
{
result = __ocl_svml_rootnf(x, n);
}
return result;
}
GENERATE_SPIRV_OCL_VECTOR_FUNCTIONS_2ARGS_VV_LOOP( rootn, float, float, int, f32, i32 )
#if defined(cl_khr_fp64)
INLINE double SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(rootn, _f64_i32, )( double y, int x )
{
return __ocl_svml_rootn_v2(y, x);
}
GENERATE_SPIRV_OCL_VECTOR_FUNCTIONS_2ARGS_VV_LOOP( rootn, double, double, int, f64, i32 )
#endif // defined(cl_khr_fp64)
#if defined(cl_khr_fp16)
INLINE half SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(rootn, _f16_i32, )( half y, int x )
{
return SPIRV_OCL_BUILTIN(rootn, _f32_i32, )((float)y, x);
}
GENERATE_SPIRV_OCL_VECTOR_FUNCTIONS_2ARGS_VV_LOOP( rootn, half, half, int, f16, i32 )
#endif // defined(cl_khr_fp16)
|
