1 /*
2 (C) Copyright 2001,2006,
3 International Business Machines Corporation,
4 Sony Computer Entertainment, Incorporated,
5 Toshiba Corporation,
6
7 All rights reserved.
8
9 Redistribution and use in source and binary forms, with or without
10 modification, are permitted provided that the following conditions are met:
11
12 * Redistributions of source code must retain the above copyright notice,
13 this list of conditions and the following disclaimer.
14 * Redistributions in binary form must reproduce the above copyright
15 notice, this list of conditions and the following disclaimer in the
16 documentation and/or other materials provided with the distribution.
17 * Neither the names of the copyright holders nor the names of their
18 contributors may be used to endorse or promote products derived from this
19 software without specific prior written permission.
20
21 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
22 IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23 TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
24 PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
25 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
26 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
27 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
28 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33 #ifndef _FMOD_H_
34 #define _FMOD_H_ 1
35
36 #include <spu_intrinsics.h>
37 #include <errno.h>
38 #include "headers/vec_literal.h"
39
40 /* This implementation returns zero if y is a denorm or zero.
41 */
_fmod(double x,double y)42 static __inline double _fmod(double x, double y)
43 {
44 int n, shift;
45 vec_uchar16 swap_words = VEC_LITERAL(vec_uchar16, 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11);
46 vec_uchar16 propagate = VEC_LITERAL(vec_uchar16, 4,5,6,7, 192,192,192,192, 12,13,14,15, 192,192,192,192);
47 vec_uchar16 splat_hi = VEC_LITERAL(vec_uchar16, 0,1,2,3,0,1,2,3, 8,9,10,11, 8,9,10,11);
48 vec_uint4 vx, vy, z;
49 vec_uint4 x_hi, y_hi;
50 vec_uint4 abs_x, abs_y;
51 vec_uint4 exp_x, exp_y;
52 vec_uint4 zero_x, zero_y;
53 vec_uint4 logb_x, logb_y;
54 vec_uint4 mant_x, mant_y;
55 vec_uint4 normal, norm, denorm;
56 vec_uint4 result, result0, resultx, cnt, sign, borrow;
57 vec_uint4 lsb = (vec_uint4)(VEC_SPLAT_U64(0x0000000000000001ULL));
58 vec_uint4 sign_mask = (vec_uint4)(VEC_SPLAT_U64(0x8000000000000000ULL));
59 vec_uint4 implied_1 = (vec_uint4)(VEC_SPLAT_U64(0x0010000000000000ULL));
60 vec_uint4 mant_mask = (vec_uint4)(VEC_SPLAT_U64(0x000FFFFFFFFFFFFFULL));
61 vec_ullong2 domain;
62 vec_int4 verrno;
63 vec_double2 vc = { 0.0, 0.0 };
64 vec_int4 fail = { EDOM, EDOM, EDOM, EDOM };
65
66 vx = (vec_uint4)spu_promote(x, 0);
67 vy = (vec_uint4)spu_promote(y, 0);
68
69 abs_x = spu_andc(vx, sign_mask);
70 abs_y = spu_andc(vy, sign_mask);
71
72 sign = spu_and(vx, sign_mask);
73
74 x_hi = spu_shuffle(abs_x, abs_x, splat_hi);
75 y_hi = spu_shuffle(abs_y, abs_y, splat_hi);
76
77 /* Determine ilogb of abs_x and abs_y and
78 * extract the mantissas (mant_x, mant_y)
79 */
80 exp_x = spu_rlmask(x_hi, -20);
81 exp_y = spu_rlmask(y_hi, -20);
82
83 resultx = spu_cmpgt(y_hi, x_hi);
84
85 zero_x = spu_cmpeq(exp_x, 0);
86 zero_y = spu_cmpeq(exp_y, 0);
87
88 logb_x = spu_add(exp_x, -1023);
89 logb_y = spu_add(exp_y, -1023);
90
91 mant_x = spu_andc(spu_sel(implied_1, abs_x, mant_mask), zero_x);
92 mant_y = spu_andc(spu_sel(implied_1, abs_y, mant_mask), zero_y);
93
94 /* Compute fixed point fmod of mant_x and mant_y. Set the flag,
95 * result0, to all ones if we detect that the final result is
96 * ever 0.
97 */
98 result0 = spu_or(zero_x, zero_y);
99
100 n = spu_extract(spu_sub(logb_x, logb_y), 0);
101
102 while (n-- > 0) {
103 borrow = spu_genb(mant_x, mant_y);
104 borrow = spu_shuffle(borrow, borrow, propagate);
105 z = spu_subx(mant_x, mant_y, borrow);
106
107 result0 = spu_or(spu_cmpeq(spu_or(z, spu_shuffle(z, z, swap_words)), 0), result0);
108
109 mant_x = spu_sel(spu_slqw(mant_x, 1), spu_andc(spu_slqw(z, 1), lsb),
110 spu_cmpgt((vec_int4)spu_shuffle(z, z, splat_hi), -1));
111 }
112
113 borrow = spu_genb(mant_x, mant_y);
114 borrow = spu_shuffle(borrow, borrow, propagate);
115 z = spu_subx(mant_x, mant_y, borrow);
116
117 mant_x = spu_sel(mant_x, z,
118 spu_cmpgt((vec_int4)spu_shuffle(z, z, splat_hi), -1));
119 mant_x = spu_andc(mant_x, VEC_LITERAL(vec_uint4, 0,0,-1,-1));
120
121 result0 = spu_or(spu_cmpeq(spu_or(mant_x, spu_shuffle(mant_x, mant_x, swap_words)), 0), result0);
122
123 /* Convert the result back to floating point and restore
124 * the sign. If we flagged the result to be zero (result0),
125 * zero it. If we flagged the result to equal its input x,
126 * (resultx) then return x.
127 *
128 * Double precision generates a denorm for an output.
129 */
130 cnt = spu_cntlz(mant_x);
131 cnt = spu_add(cnt, spu_and(spu_rlqwbyte(cnt, 4), spu_cmpeq(cnt, 32)));
132 cnt = spu_add(spu_shuffle(cnt, cnt, splat_hi), -11);
133
134 shift = spu_extract(exp_y, 0) - 1;
135 denorm = spu_slqwbytebc(spu_slqw(mant_x, shift), shift);
136
137 exp_y = spu_sub(exp_y, cnt);
138
139 normal = spu_cmpgt((vec_int4)exp_y, 0);
140
141 /* Normalize normal results, denormalize denorm results.
142 */
143 shift = spu_extract(cnt, 0);
144 norm = spu_slqwbytebc(spu_slqw(spu_andc(mant_x, VEC_LITERAL(vec_uint4, 0x00100000, 0, -1, -1)), shift), shift);
145
146 mant_x = spu_sel(denorm, norm, normal);
147
148 exp_y = spu_and(spu_rl(exp_y, 20), normal);
149
150 result = spu_sel(exp_y, spu_or(sign, mant_x), VEC_LITERAL(vec_uint4, 0x800FFFFF, -1, 0x800FFFFF, -1));
151
152 result = spu_sel(spu_andc(result, spu_rlmask(result0, -1)), vx,
153 resultx);
154
155 #ifndef _IEEE_LIBM
156 /*
157 * If y is zero, set errno to EDOM
158 */
159 domain = spu_cmpeq(vc, (vec_double2) vy);
160 verrno = spu_splats(errno);
161 errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0);
162 #endif
163
164 return (spu_extract((vec_double2)result, 0));
165 }
166 #endif /* _FMOD_H_ */
167