1 | /* -------------------------------------------------------------- */ |
---|
2 | /* (C)Copyright 2001,2008, */ |
---|
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 */ |
---|
10 | /* without modification, are permitted provided that the */ |
---|
11 | /* following conditions are met: */ |
---|
12 | /* */ |
---|
13 | /* - Redistributions of source code must retain the above copyright*/ |
---|
14 | /* notice, this list of conditions and the following disclaimer. */ |
---|
15 | /* */ |
---|
16 | /* - Redistributions in binary form must reproduce the above */ |
---|
17 | /* copyright notice, this list of conditions and the following */ |
---|
18 | /* disclaimer in the documentation and/or other materials */ |
---|
19 | /* provided with the distribution. */ |
---|
20 | /* */ |
---|
21 | /* - Neither the name of IBM Corporation nor the names of its */ |
---|
22 | /* contributors may be used to endorse or promote products */ |
---|
23 | /* derived from this software without specific prior written */ |
---|
24 | /* permission. */ |
---|
25 | /* */ |
---|
26 | /* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND */ |
---|
27 | /* CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, */ |
---|
28 | /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */ |
---|
29 | /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */ |
---|
30 | /* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR */ |
---|
31 | /* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, */ |
---|
32 | /* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT */ |
---|
33 | /* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; */ |
---|
34 | /* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) */ |
---|
35 | /* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN */ |
---|
36 | /* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR */ |
---|
37 | /* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, */ |
---|
38 | /* EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ |
---|
39 | /* -------------------------------------------------------------- */ |
---|
40 | /* PROLOG END TAG zYx */ |
---|
41 | #ifdef __SPU__ |
---|
42 | #ifndef _EXPD2_H_ |
---|
43 | #define _EXPD2_H_ 1 |
---|
44 | |
---|
45 | #include <spu_intrinsics.h> |
---|
46 | #include "floord2.h" |
---|
47 | |
---|
48 | #define LOG2E 1.4426950408889634073599 // 1/log(2) |
---|
49 | |
---|
50 | /* |
---|
51 | * FUNCTION |
---|
52 | * vector double _expd2(vector double x) |
---|
53 | * |
---|
54 | * DESCRIPTION |
---|
55 | * _expd2 computes e raised to the input x for |
---|
56 | * each of the element of the double word vector. |
---|
57 | * |
---|
58 | * Calculation is performed by reducing the input argument |
---|
59 | * to within a managable range, and then computing the power |
---|
60 | * series to the 11th degree. |
---|
61 | * |
---|
62 | * Range reduction is performed using the property: |
---|
63 | * |
---|
64 | * exp(x) = 2^n * exp(r) |
---|
65 | * |
---|
66 | * Values for "n" and "r" are determined such that: |
---|
67 | * |
---|
68 | * x = n * ln(2) + r, |r| <= ln(2)/2 |
---|
69 | * |
---|
70 | * n = floor( (x/ln(2)) + 1/2 ) |
---|
71 | * r = x - (n * ln(2)) |
---|
72 | * |
---|
73 | * To enhance the precision for "r", computation is performed |
---|
74 | * using a two part representation of ln(2). |
---|
75 | * |
---|
76 | * Once the input is reduced, the power series is computed: |
---|
77 | * |
---|
78 | * __12_ |
---|
79 | * \ |
---|
80 | * exp(x) = 1 + \ (x^i)/i! |
---|
81 | * / |
---|
82 | * /____ |
---|
83 | * i=2 |
---|
84 | * |
---|
85 | * The resulting value is scaled by 2^n and returned. |
---|
86 | * |
---|
87 | */ |
---|
88 | |
---|
89 | static __inline vector double _expd2(vector double x) |
---|
90 | { |
---|
91 | // log(2) in extended machine representable precision |
---|
92 | vec_double2 ln2_hi = spu_splats(6.9314575195312500E-1); // 3FE62E4000000000 |
---|
93 | vec_double2 ln2_lo = spu_splats(1.4286068203094172E-6); // 3EB7F7D1CF79ABCA |
---|
94 | |
---|
95 | // coefficients for the power series |
---|
96 | // vec_double2 f01 = spu_splats(1.00000000000000000000E0); // 1/(1!) |
---|
97 | vec_double2 f02 = spu_splats(5.00000000000000000000E-1); // 1/(2!) |
---|
98 | vec_double2 f03 = spu_splats(1.66666666666666666667E-1); // 1/(3!) |
---|
99 | vec_double2 f04 = spu_splats(4.16666666666666666667E-2); // 1/(4!) |
---|
100 | vec_double2 f05 = spu_splats(8.33333333333333333333E-3); // 1/(5!) |
---|
101 | vec_double2 f06 = spu_splats(1.38888888888888888889E-3); // 1/(6!) |
---|
102 | vec_double2 f07 = spu_splats(1.98412698412698412698E-4); // 1/(7!) |
---|
103 | vec_double2 f08 = spu_splats(2.48015873015873015873E-5); // 1/(8!) |
---|
104 | vec_double2 f09 = spu_splats(2.75573192239858906526E-6); // 1/(9!) |
---|
105 | vec_double2 f10 = spu_splats(2.75573192239858906526E-7); // 1/(10!) |
---|
106 | vec_double2 f11 = spu_splats(2.50521083854417187751E-8); // 1/(11!) |
---|
107 | vec_double2 f12 = spu_splats(2.08767569878680989792E-9); // 1/(12!) |
---|
108 | |
---|
109 | // rx = floor(1/2 + x/log(2)) |
---|
110 | vec_double2 rx = _floord2(spu_madd(x,spu_splats(LOG2E),spu_splats(0.5))); |
---|
111 | |
---|
112 | // extract the exponent of reduction |
---|
113 | vec_int4 exp = spu_convts(spu_roundtf(rx),0); |
---|
114 | |
---|
115 | // reduce the input to within [ -ln(2)/2 ... ln(2)/2 ] |
---|
116 | vec_double2 r; |
---|
117 | r = spu_nmsub(rx,ln2_hi,x); |
---|
118 | r = spu_nmsub(rx,ln2_lo,r); |
---|
119 | |
---|
120 | vec_double2 result; |
---|
121 | vec_double2 r2 = spu_mul(r,r); |
---|
122 | |
---|
123 | // Use Horner's method on the power series |
---|
124 | /* result = ((((c12*x + c11)*x + c10)*x + c9)*x + c8)*x + c7)*x + c6)*x^6 + |
---|
125 | ((((((c5*x + c4)*x + c3)*x + c2)*x + c1)*x + c0 |
---|
126 | */ |
---|
127 | |
---|
128 | #ifdef __SPU_EDP__ |
---|
129 | vec_double2 p1, p2, r4, r6; |
---|
130 | |
---|
131 | p1 = spu_madd(f12, r, f11); |
---|
132 | p2 = spu_madd(f05, r, f04); |
---|
133 | r4 = spu_mul(r2, r2); |
---|
134 | p1 = spu_madd(p1, r, f10); |
---|
135 | p2 = spu_madd(p2, r, f03); |
---|
136 | p1 = spu_madd(p1, r, f09); |
---|
137 | p2 = spu_madd(p2, r, f02); |
---|
138 | p1 = spu_madd(p1, r, f08); |
---|
139 | r6 = spu_mul(r2, r4); |
---|
140 | p1 = spu_madd(p1, r, f07); |
---|
141 | p2 = spu_madd(p2, r2, r); |
---|
142 | p1 = spu_madd(p1, r, f06); |
---|
143 | |
---|
144 | result = spu_madd(r6, p1, p2); |
---|
145 | result = spu_add(result, spu_splats(1.0)); |
---|
146 | |
---|
147 | #else |
---|
148 | |
---|
149 | result = spu_madd(r,f12,f11); |
---|
150 | result = spu_madd(result,r,f10); |
---|
151 | result = spu_madd(result,r,f09); |
---|
152 | result = spu_madd(result,r,f08); |
---|
153 | result = spu_madd(result,r,f07); |
---|
154 | result = spu_madd(result,r,f06); |
---|
155 | result = spu_madd(result,r,f05); |
---|
156 | result = spu_madd(result,r,f04); |
---|
157 | result = spu_madd(result,r,f03); |
---|
158 | result = spu_madd(result,r,f02); |
---|
159 | result = spu_madd(result,r2,r); |
---|
160 | result = spu_add(result,spu_splats(1.0)); |
---|
161 | |
---|
162 | #endif /* __SPU_EDP__ */ |
---|
163 | |
---|
164 | |
---|
165 | // Scale the result - basically a call to ldexpd2() |
---|
166 | vec_int4 e1, e2; |
---|
167 | vec_int4 min = spu_splats(-2044); |
---|
168 | vec_int4 max = spu_splats(2046); |
---|
169 | vec_uint4 cmp_min, cmp_max; |
---|
170 | vec_uint4 shift = (vec_uint4) { 20, 32, 20, 32 }; |
---|
171 | vec_double2 f1, f2; |
---|
172 | |
---|
173 | /* Clamp the specified exponent to the range -2044 to 2046. |
---|
174 | */ |
---|
175 | cmp_min = spu_cmpgt(exp, min); |
---|
176 | cmp_max = spu_cmpgt(exp, max); |
---|
177 | exp = spu_sel(min, exp, cmp_min); |
---|
178 | exp = spu_sel(exp, max, cmp_max); |
---|
179 | |
---|
180 | /* Generate the factors f1 = 2^e1 and f2 = 2^e2 |
---|
181 | */ |
---|
182 | e1 = spu_rlmaska(exp, -1); |
---|
183 | e2 = spu_sub(exp, e1); |
---|
184 | |
---|
185 | f1 = (vec_double2)spu_sl(spu_add(e1, 1023), shift); |
---|
186 | f2 = (vec_double2)spu_sl(spu_add(e2, 1023), shift); |
---|
187 | |
---|
188 | /* Compute the product x * 2^e1 * 2^e2 |
---|
189 | */ |
---|
190 | result = spu_mul(spu_mul(result, f1), f2); |
---|
191 | |
---|
192 | return result; |
---|
193 | } |
---|
194 | |
---|
195 | #endif /* _EXPD2_H_ */ |
---|
196 | #endif /* __SPU__ */ |
---|