File: pow_base_fp64.cl

package info (click to toggle)
pocl 6.0-6
  • links: PTS, VCS
  • area: main
  • in suites: trixie
  • size: 25,304 kB
  • sloc: lisp: 149,513; ansic: 103,778; cpp: 54,947; python: 1,513; sh: 949; ruby: 255; pascal: 226; tcl: 180; makefile: 173; java: 72; xml: 49
file content (198 lines) | stat: -rw-r--r-- 6,491 bytes parent folder | download | duplicates (5)
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
/*===--------------------------------------------------------------------------
 *                   ROCm Device Libraries
 *
 * This file is distributed under the University of Illinois Open Source
 * License. See ROCM_LICENSE.TXT for details.
 *===------------------------------------------------------------------------*/

#include "ocml_helpers.h"
#include "singlevec.h"

extern _CL_OVERLOADABLE v2type MATH_PRIVATE(epln)(vtype);
extern _CL_OVERLOADABLE vtype MATH_PRIVATE(expep)(v2type);

vtype
#if defined(COMPILING_POWR)
MATH_MANGLE(powr)(vtype x, vtype y)
#elif defined(COMPILING_POWN)
MATH_MANGLE(pown)(vtype x, inttype ny)
#elif defined(COMPILING_ROOTN)
MATH_MANGLE(rootn)(vtype x, inttype ny)
#else
MATH_MANGLE(pow)(vtype x, vtype y)
#endif
{

    vtype ax = BUILTIN_ABS_F64(x);

#if defined(COMPILING_POWN)
    vtype y = convert_vtype(ny);
#elif defined(COMPILING_ROOTN)
    v2type y = rcp(convert_vtype(ny));
#else
    vtype ay = BUILTIN_ABS_F64(y);
    // flush denorm y to zero
    itype is_denorm = (as_itype(ay) < (itype)IMPBIT_DP64);
    y = (is_denorm) ? as_vtype(as_itype(y) & (itype)SIGNBIT_DP64) : y;
#endif

    vtype ret = MATH_PRIVATE(expep)(omul(y, MATH_PRIVATE(epln)(ax)));

#if defined(COMPILING_POWN) || defined(COMPILING_ROOTN)
    itype nyi = convert_itype(ny);
    itype is_odd = (nyi << 63);
#else
    vtype tay = BUILTIN_TRUNC_F64(ay);
    itype is_int = (ay == tay);
    vtype unused;
    itype is_odd = is_int ? (fract(tay*0.5, &unused) != vZERO_DP64) : (itype)0;

#ifdef SINGLEVEC
    if (is_odd && (ax > x))
      ret = copysign(ret, -0.0);
#else
    ret = copysign(ret, as_vtype(is_odd & (ax > x)));
#endif

#endif


    // edge cases
#if defined(COMPILING_POWN) || defined(COMPILING_ROOTN)
    vtype ret0 = (nyi > (itype)0) ? (vtype)0.0 : vINFINITY_DP64;
    vtype retI = (nyi > (itype)0) ? vINFINITY_DP64 : (vtype)0.0f;
    ret = (ax > (vtype)0.0) ? ret : ret0;
    ret = (isinf(x)) ? retI : ret;
    ret = (isnan(x)) ? x : ret;

    itype xneg = as_itype(x) & (itype)SIGNBIT_DP64;
    ret = (is_odd & xneg) ? copysign(ret, x) : ret;

#if defined(COMPILING_POWN)
    ret = (nyi == (itype)0) ? (vtype)1.0 : ret;
#elif defined(COMPILING_ROOTN)
    ret = (SV_NOT(is_odd) SV_AND (x < (vtype)0.0)) ? vNAN_DP64 : ret;
    ret = (nyi == (itype)0) ? vNAN_DP64: ret;
#endif

    return ret;
#else  /* POW / POWR */
    itype ax_eq_0 = (as_itype(ax) == as_itype(vZERO_DP64));
    itype ay_eq_0 = (as_itype(ay) == as_itype(vZERO_DP64));

    itype ax_lt_1 = (ax < vONE_DP64);
    itype ax_gt_1 = (ax > vONE_DP64);

    itype ax_eq_nan = (isnan(ax));
    itype ay_eq_nan = (isnan(ay));
    itype ay_eq_pinf = (ay == vINFINITY_DP64);

#ifdef SINGLEVEC
    itype y_pos = (as_itype(y) & (itype)SIGNBIT_DP64) == 0;
    itype x_neg = (as_itype(x) & (itype)SIGNBIT_DP64) != 0;
    itype y_neg = !y_pos;
#else
    itype x_neg = (as_itype(x) & (itype)SIGNBIT_DP64);
    itype y_neg = (as_itype(y) & (itype)SIGNBIT_DP64);
    itype y_pos = (y_neg ^ (itype)SIGNBIT_DP64);
#endif

    itype y_eq_ninf = (y == vNINFINITY_DP64);
    itype y_eq_pinf = (y == vINFINITY_DP64);

#endif /* POW / POWR */


#if defined(COMPILING_POWR)

    itype ax_eq_pinf = (ax == vINFINITY_DP64);
    itype ax_eq_1 = (ax == vONE_DP64);
    itype ay_lt_inf = (ay < vINFINITY_DP64);
    itype ax_lt_pinf = (ax < vINFINITY_DP64);
    itype ax_ne_0 = SV_NOT(ax_eq_0);

    #ifndef FINITE_MATH_ONLY
        ret = (ax_lt_1 & y_eq_ninf) ? vINFINITY_DP64 : ret;
        ret = (ax_lt_1 & y_eq_pinf) ? vZERO_DP64 : ret;
        ret = (ax_eq_1 & ay_lt_inf) ? vONE_DP64 : ret;
        ret = (ax_eq_1 & ay_eq_pinf) ? vNAN_DP64 : ret;
        ret = (ax_gt_1 & y_eq_ninf) ? vZERO_DP64 : ret;
        ret = (ax_gt_1 & y_eq_pinf) ? vINFINITY_DP64 : ret;

        ret = (ax_lt_pinf & ay_eq_0) ? vONE_DP64 : ret;

        ret = (ax_eq_pinf & y_neg) ? vZERO_DP64 : ret;
        ret = (ax_eq_pinf & y_pos) ? vINFINITY_DP64 : ret;
        ret = (ax_eq_pinf & y_eq_pinf) ? vINFINITY_DP64 : ret;
        ret = (ax_eq_pinf & ay_eq_0) ? vNAN_DP64 : ret;

        ret = (ax_eq_0 & y_neg) ? vINFINITY_DP64 : ret;
        ret = (ax_eq_0 & y_pos) ? vZERO_DP64 : ret;
        ret = (ax_eq_0 & ay_eq_0) ? vNAN_DP64 : ret;
        ret = (ax_ne_0 & x_neg) ? vNAN_DP64 : ret;

        ret = (ax_eq_nan) ? x : ret;
        ret = (ay_eq_nan) ? y : ret;
    #else
        ret = (ax_eq_1) ? vONE_DP64 : ret;
        ret = (ay_eq_0) ? vONE_DP64 : ret;
        ret = (ax_eq_0 & y_pos) ? vZERO_DP64 : ret;
    #endif

    return ret;
#endif

#if defined(COMPILING_POW)

    itype is_not_int = SV_NOT(is_int);
    itype is_not_odd = SV_NOT(is_odd);

    itype x_eq_ninf = (x == vNINFINITY_DP64);
    itype x_eq_pinf = (x == vINFINITY_DP64);

    #ifndef FINITE_MATH_ONLY
        vtype xinf = copysign(vINFINITY_DP64, x);
        vtype xzero = copysign(vZERO_DP64, x);

        ret = (x_neg & is_not_int) ? vNAN_DP64 : ret;

        ret = (ax_lt_1 & y_eq_ninf) ? vINFINITY_DP64 : ret;
        ret = (ax_gt_1 & y_eq_ninf) ? vZERO_DP64 : ret;
        ret = (ax_lt_1 & y_eq_pinf) ? vZERO_DP64 : ret;
        ret = (ax_gt_1 & y_eq_pinf) ? vINFINITY_DP64 : ret;

        ret = (ax_eq_0 & y_neg & is_odd) ? xinf : ret;
        ret = (ax_eq_0 & y_neg & is_not_odd) ? vINFINITY_DP64 : ret;
        ret = (ax_eq_0 & y_pos & is_odd) ? xzero : ret;
        ret = (ax_eq_0 & y_pos & is_not_odd) ? vZERO_DP64 : ret;
        ret = (ax_eq_0 & y_eq_ninf) ? vINFINITY_DP64 : ret;

        ret = ((x == (vtype)(-1.0)) & ay_eq_pinf) ? vONE_DP64 : ret;

        ret = (x_eq_ninf & y_neg & is_odd) ? (vtype)(-0.0) : ret;
        ret = (x_eq_ninf & y_neg & is_not_odd) ? vZERO_DP64 : ret;
        ret = (x_eq_ninf & y_pos & is_odd) ? vNINFINITY_DP64 : ret;
        ret = (x_eq_ninf & y_pos & is_not_odd) ? vINFINITY_DP64 : ret;
        ret = (x_eq_pinf & y_neg) ? vZERO_DP64 : ret;
        ret = (x_eq_pinf & y_pos) ? vINFINITY_DP64 : ret;
        ret = (ax_eq_nan) ? x : ret;
        ret = (ay_eq_nan) ? y : ret;

    #else
        // XXX work around conformance test incorrectly checking these cases
        vtype xinf = copysign(vINFINITY_DP64, x);
        ret = (ax_eq_0 & y_neg & is_odd) ? xinf : ret;
        ret = (ax_eq_0 & y_neg & is_not_odd) ? vINFINITY_DP64 : ret;

        vtype xzero = copysign(0.0f, x);
        ret = (ax_eq_0 & y_pos & is_odd) ? xzero : ret;
        ret = (ax_eq_0 & y_pos & is_not_odd) ? vZERO_DP64 : ret;
    #endif

    ret = ay_eq_0 ? vONE_DP64 : ret;
    ret = (x == vONE_DP64) ? vONE_DP64 : ret;
    return ret;

#endif

}