2 * Copyright 2023 Siemens
4 * The authors hereby grant permission to use, copy, modify, distribute,
5 * and license this software and its documentation for any purpose, provided
6 * that existing copyright notices are retained in all copies and that this
7 * notice is included verbatim in any distributions. No written agreement,
8 * license, or royalty fee is required for any of the authorized uses.
9 * Modifications to this software may be copyrighted by their authors
10 * and need not follow the licensing terms described here, provided that
11 * the new terms are clearly indicated on the first page of each file where
16 * Copyright (c) 1994-2009 Red Hat, Inc. All rights reserved.
18 * This copyrighted material is made available to anyone wishing to use,
19 * modify, copy, or redistribute it subject to the terms and conditions
20 * of the BSD License. This program is distributed in the hope that
21 * it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
22 * including the implied warranties of MERCHANTABILITY or FITNESS FOR
23 * A PARTICULAR PURPOSE. A copy of this license is available at
24 * http://www.opensource.org/licenses. Any Red Hat trademarks that are
25 * incorporated in the source code or documentation are not subject to
26 * the BSD License and may only be used or replicated with the express
27 * permission of Red Hat, Inc.
30 /*****************************************************************
31 * The following routines are coded directly from the algorithms
32 * and coefficients given in "Software Manual for the Elementary
33 * Functions" by William J. Cody, Jr. and William Waite, Prentice
35 *****************************************************************/
37 /* Based on newlib/libm/mathfp/s_sqrt.c in Newlib. */
39 #include "amdgcnmach.h"
41 v64si
v64df_numtest (v64df
);
42 v64si
v64df_ispos (v64df
);
44 #if defined (__has_builtin) \
45 && __has_builtin (__builtin_gcn_frexpv_mant) \
46 && __has_builtin (__builtin_gcn_frexpv_exp) \
47 && __has_builtin (__builtin_gcn_ldexpv)
49 DEF_VD_MATH_FUNC (v64df
, sqrt
, v64df x
)
51 FUNCTION_INIT (v64df
);
53 /* Check for special values. */
54 v64si num_type
= v64df_numtest (x
);
55 VECTOR_IF (num_type
== NAN
, cond
)
57 VECTOR_RETURN (x
, cond
);
58 VECTOR_ELSEIF (num_type
== INF
, cond
)
59 VECTOR_IF2 (v64df_ispos (x
), cond2
, cond
)
61 VECTOR_RETURN (VECTOR_INIT (z_notanum
.d
), cond2
);
62 VECTOR_ELSE2 (cond2
,cond
)
64 VECTOR_RETURN (VECTOR_INIT (z_infinity
.d
), cond
);
68 /* Initial checks are performed here. */
69 VECTOR_IF (x
== 0.0, cond
)
70 VECTOR_RETURN (VECTOR_INIT (0.0), cond
);
72 VECTOR_IF (x
< 0.0, cond
)
74 VECTOR_RETURN (VECTOR_INIT (z_notanum
.d
), cond
);
77 /* Find the exponent and mantissa for the form x = f * 2^exp. */
78 v64df f
= __builtin_gcn_frexpv_mant (x
);
79 v64si exp
= __builtin_gcn_frexpv_exp (x
);
80 v64si odd
= (exp
& 1) != 0;
82 /* Get the initial approximation. */
83 v64df y
= 0.41731 + 0.59016 * f
;
86 /* Calculate the remaining iterations. */
91 /* Calculate the final value. */
92 VECTOR_COND_MOVE (y
, y
* __SQRT_HALF
, odd
);
93 VECTOR_COND_MOVE (exp
, exp
+ 1, odd
);
95 y
= __builtin_gcn_ldexpv (y
, exp
);
97 VECTOR_RETURN (y
, NO_COND
);
102 DEF_VARIANTS (sqrt
, df
, df
)