1 /* -------------------------------------------------------------- */
2 /* (C)Copyright 2007,2008, */
3 /* International Business Machines Corporation */
4 /* All Rights Reserved. */
5 /* */
6 /* Redistribution and use in source and binary forms, with or */
7 /* without modification, are permitted provided that the */
8 /* following conditions are met: */
9 /* */
10 /* - Redistributions of source code must retain the above copyright*/
11 /* notice, this list of conditions and the following disclaimer. */
12 /* */
13 /* - Redistributions in binary form must reproduce the above */
14 /* copyright notice, this list of conditions and the following */
15 /* disclaimer in the documentation and/or other materials */
16 /* provided with the distribution. */
17 /* */
18 /* - Neither the name of IBM Corporation nor the names of its */
19 /* contributors may be used to endorse or promote products */
20 /* derived from this software without specific prior written */
21 /* permission. */
22 /* */
23 /* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND */
24 /* CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, */
25 /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
26 /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
27 /* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR */
28 /* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, */
29 /* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT */
30 /* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; */
31 /* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) */
32 /* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN */
33 /* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR */
34 /* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, */
35 /* EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
36 /* -------------------------------------------------------------- */
37 /* PROLOG END TAG zYx */
38 #ifdef __SPU__
39 #ifndef _NEXTAFTERD2_H_
40 #define _NEXTAFTERD2_H_ 1
41
42 #include <spu_intrinsics.h>
43
44 /*
45 * FUNCTION
46 * vector double _nextafterd2(vector double x, vector double y)
47 *
48 * DESCRIPTION
49 * The nextafterf4 function returns a vector containing the next representable
50 * floating-point number after the element of x, in the direction of the
51 * corresponding element y.
52 *
53 * Special Cases:
54 * - nextafter(NaN, y) = NaN
55 * - nextafter(x, NaN) = NaN
56 * - x = largest finite value, y = infinity, result is undefined
57 * - x = largest finite negative value, y = -infinity, result is undefined
58 * - x != y, and result = 0, considered an underflow
59 *
60 */
61
_nextafterd2(vector double x,vector double y)62 static __inline vector double _nextafterd2(vector double x, vector double y)
63 {
64 vec_double2 n1ulp = (vec_double2)spu_splats(0x8000000000000001ull);
65 vector unsigned char mov_carry = {0x80,0x80,0x80, 7, 0x80,0x80,0x80,0x80,
66 0x80,0x80,0x80,15, 0x80,0x80,0x80,0x80};
67 vec_double2 zerod = spu_splats(0.0);
68 vec_llong2 one = spu_splats(1ll);
69 vec_ullong2 xlt0, xgty, xeqy, xeq0;
70 vec_llong2 xllong;
71 vec_int4 carry;
72 vec_llong2 delta, deltap1;
73 vec_double2 result;
74
75 /* Compiler Bug. Replace xtmp/ytmp with x/y when spu_cmpgt(x,y) doesn't change x/y!*/
76 volatile vec_double2 xtmp = x;
77 volatile vec_double2 ytmp = y;
78
79 /*
80 * The idea here is to treat x as a signed long long value, which allows us to
81 * add or subtact one to/from it to get the next representable value.
82 */
83
84 xeq0 = spu_cmpeq(xtmp, zerod);
85 xlt0 = spu_cmpgt(zerod, xtmp);
86 xeqy = spu_cmpeq(xtmp, ytmp);
87 xgty = spu_cmpgt(xtmp, ytmp);
88
89 /* If x = -0.0, set x = 0.0 */
90 x = spu_andc(x, (vec_double2)xeq0);
91
92 xllong = (vec_llong2)x;
93
94 /* Determine value to add to x */
95 delta = (vec_llong2)spu_xor(xgty, xlt0);
96
97 //deltap1 = delta + one;
98 carry = spu_genc((vec_int4)delta, (vec_int4)one);
99 carry = spu_shuffle(carry, carry, mov_carry);
100 deltap1 = (vec_llong2)spu_addx((vec_int4)delta, (vec_int4)one, (vec_int4)carry);
101
102 delta = spu_sel(deltap1, delta, (vec_ullong2)delta);
103
104 //xllong = xllong + delta;
105 carry = spu_genc((vec_int4)xllong, (vec_int4)delta);
106 carry = spu_shuffle(carry, carry, mov_carry);
107 xllong = (vec_llong2)spu_addx((vec_int4)xllong, (vec_int4)delta, (vec_int4)carry);
108
109 /* Fix the case of x = 0, and answer should be -1 ulp */
110 result = spu_sel((vec_double2)xllong, n1ulp, spu_and((vec_ullong2)delta, xeq0));
111
112 /*
113 * Special Cases
114 */
115
116 /* x = y */
117 result = spu_sel(result, y, xeqy);
118
119 return result;
120 }
121
122 #endif /* _NEXTAFTERD2_H_ */
123 #endif /* __SPU__ */
124