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 _SQRTF4_H_
43 #define _SQRTF4_H_ 1
44
45 #include <spu_intrinsics.h>
46
47 /*
48 * FUNCTION
49 * vector float _sqrtf4(vector float in)
50 *
51 * DESCRIPTION
52 * The _sqrtf4 function computes the square root of the vector input "in"
53 *and returns the result.
54 *
55 */
_sqrtf4(vector float in)56 static __inline vector float _sqrtf4(vector float in)
57 {
58 vec_uint4 exp, valid;
59 vec_uint4 mask = spu_splats((unsigned int)0xFF000000);
60 vec_uint4 half = spu_splats((unsigned int)0x00800000);
61 vec_float4 one = spu_splats(1.0f);
62 vec_float4 three = spu_splats(3.0f);
63 vec_float4 x, y0, y1, y1_n1, y1_p1, y1_p2, y1_p3;
64 vec_float4 mant, err, err_p1, err_p2, err_p3;
65 vec_float4 out;
66
67 /* Compute the mantissa of the result seperately from
68 * the exponent to assure complete accuracy over the allowable
69 * input range. The mantissa is computed for inputs in the
70 * range [0.5, 2.0).
71 */
72 x = spu_sel(in, one, mask);
73 y0 = spu_rsqrte(x);
74
75 /* Perform one iteration of the Newton-Raphsom method in single precision
76 * arithmetic.
77 */
78 y1 = spu_mul(spu_nmsub(x, spu_mul(y0, y0), three),
79 spu_mul(y0, (vec_float4)(spu_sub((vec_uint4)(x), half))));
80
81 /* Correct the result for possible error. The range of error is -3 to +1.
82 * Identify the extent of the error and correct for it.
83 */
84 y1_p3 = (vec_float4)spu_add((vec_uint4)(y1), 3);
85 y1_p2 = (vec_float4)spu_add((vec_uint4)(y1), 2);
86 y1_p1 = (vec_float4)spu_add((vec_uint4)(y1), 1);
87 y1_n1 = (vec_float4)spu_add((vec_uint4)(y1), -1);
88
89 err = spu_nmsub(y1, y1, x);
90 err_p1 = spu_nmsub(y1_p1, y1_p1, x);
91 err_p2 = spu_nmsub(y1_p2, y1_p2, x);
92 err_p3 = spu_nmsub(y1_p3, y1_p3, x);
93
94 mant = spu_sel(y1_n1, y1, spu_cmpgt((vec_int4)(err), -1));
95 mant = spu_sel(mant, y1_p1, spu_cmpgt((vec_int4)(err_p1), -1));
96 mant = spu_sel(mant, y1_p2, spu_cmpgt((vec_int4)(err_p2), -1));
97 mant = spu_sel(mant, y1_p3, spu_cmpgt((vec_int4)(err_p3), -1));
98
99 /* Compute the expected exponent. If the exponent is zero or the input is
100 * negative, then set the result to zero.
101 */
102 exp = spu_rlmask(spu_add((vec_uint4)(in), (vec_uint4)(one)), -1);
103
104 valid = spu_cmpgt(spu_and((vec_int4)(in), (vec_int4)(mask)), 0);
105
106 /* Merge the computed exponent and mantissa.
107 */
108 out = spu_and(spu_sel(mant, (vec_float4)(exp), spu_splats(0xFF800000)), (vec_float4)(valid));
109
110
111 return (out);
112
113 }
114
115 #endif /* _SQRTF4_H_ */
116 #endif /* __SPU__ */
117
