4
0
mirror of git://sourceware.org/git/newlib-cygwin.git synced 2025-01-23 23:47:22 +08:00
Kwok Cheung Yeung e18743072b amdgcn: Add vectorized math routines
This implements a set of vectorized math routines to be used by the
compiler auto-vectorizer.  Versions for vectors with 2 lanes up to
64 lanes (in powers of 2) are provided.

These routines are based on the scalar versions of the math routines in
libm/common, libm/math and libm/mathfp.  They make extensive use of the GCC
C vector extensions and GCN-specific builtins in GCC.
2023-01-18 13:22:58 -05:00

104 lines
3.4 KiB
C

/*
* Copyright 2023 Siemens
*
* The authors hereby grant permission to use, copy, modify, distribute,
* and license this software and its documentation for any purpose, provided
* that existing copyright notices are retained in all copies and that this
* notice is included verbatim in any distributions. No written agreement,
* license, or royalty fee is required for any of the authorized uses.
* Modifications to this software may be copyrighted by their authors
* and need not follow the licensing terms described here, provided that
* the new terms are clearly indicated on the first page of each file where
* they apply.
*/
/*
* Copyright (c) 1994-2009 Red Hat, Inc. All rights reserved.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the BSD License. This program is distributed in the hope that
* it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
* including the implied warranties of MERCHANTABILITY or FITNESS FOR
* A PARTICULAR PURPOSE. A copy of this license is available at
* http://www.opensource.org/licenses. Any Red Hat trademarks that are
* incorporated in the source code or documentation are not subject to
* the BSD License and may only be used or replicated with the express
* permission of Red Hat, Inc.
*/
/*****************************************************************
* The following routines are coded directly from the algorithms
* and coefficients given in "Software Manual for the Elementary
* Functions" by William J. Cody, Jr. and William Waite, Prentice
* Hall, 1980.
*****************************************************************/
/* Based on newlib/libm/mathfp/sf_sqrt.c in Newlib. */
#include "amdgcnmach.h"
v64si v64sf_numtestf (v64sf);
v64si v64sf_isposf (v64sf);
#if defined (__has_builtin) \
&& __has_builtin (__builtin_gcn_frexpvf_mant) \
&& __has_builtin (__builtin_gcn_frexpvf_exp) \
&& __has_builtin (__builtin_gcn_ldexpvf)
DEF_VS_MATH_FUNC (v64sf, sqrtf, v64sf x)
{
FUNCTION_INIT (v64sf);
/* Check for special values. */
v64si num_type = v64sf_numtestf (x);
VECTOR_IF (num_type == NAN, cond)
errno = EDOM;
VECTOR_RETURN (x, cond);
VECTOR_ELSEIF (num_type == INF, cond)
VECTOR_IF2 (v64sf_isposf (x), cond2, cond)
errno = EDOM;
VECTOR_RETURN (VECTOR_INIT (z_notanum_f.f), cond2);
VECTOR_ELSE2 (cond2,cond)
errno = ERANGE;
VECTOR_RETURN (VECTOR_INIT (z_infinity_f.f), cond);
VECTOR_ENDIF
VECTOR_ENDIF
/* Initial checks are performed here. */
VECTOR_IF (x == 0.0f, cond)
VECTOR_RETURN (VECTOR_INIT (0.0f), cond);
VECTOR_ENDIF
VECTOR_IF (x < 0.0f, cond)
errno = EDOM;
VECTOR_RETURN (VECTOR_INIT (z_notanum_f.f), cond);
VECTOR_ENDIF
/* Find the exponent and mantissa for the form x = f * 2^exp. */
v64sf f = __builtin_gcn_frexpvf_mant (x);
v64si exp = __builtin_gcn_frexpvf_exp (x);
v64si odd = (exp & 1) != 0;
/* Get the initial approximation. */
v64sf y = 0.41731f + 0.59016f * f;
f *= 0.5f;
/* Calculate the remaining iterations. */
y = y * 0.5f + f / y;
y = y * 0.5f + f / y;
/* Calculate the final value. */
VECTOR_COND_MOVE (y, y * (float) __SQRT_HALF, odd);
VECTOR_COND_MOVE (exp, exp + 1, odd);
exp >>= 1;
y = __builtin_gcn_ldexpvf (y, exp);
VECTOR_RETURN (y, NO_COND);
FUNCTION_RETURN;
}
DEF_VARIANTS (sqrtf, sf, sf)
#endif