mirror of
git://sourceware.org/git/newlib-cygwin.git
synced 2025-02-24 09:39:33 +08:00
125e39bfea
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.
76 lines
2.0 KiB
C
76 lines
2.0 KiB
C
/*****************************************************************
|
|
* 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/s_sqrt.c in Newlib. */
|
|
|
|
#include "amdgcnmach.h"
|
|
|
|
v64si v64df_numtest (v64df);
|
|
v64si v64df_ispos (v64df);
|
|
|
|
#if defined (__has_builtin) \
|
|
&& __has_builtin (__builtin_gcn_frexpv_mant) \
|
|
&& __has_builtin (__builtin_gcn_frexpv_exp) \
|
|
&& __has_builtin (__builtin_gcn_ldexpv)
|
|
|
|
DEF_VD_MATH_FUNC (v64df, sqrt, v64df x)
|
|
{
|
|
FUNCTION_INIT (v64df);
|
|
|
|
/* Check for special values. */
|
|
v64si num_type = v64df_numtest (x);
|
|
VECTOR_IF (num_type == NAN, cond)
|
|
errno = EDOM;
|
|
VECTOR_RETURN (x, cond);
|
|
VECTOR_ELSEIF (num_type == INF, cond)
|
|
VECTOR_IF2 (v64df_ispos (x), cond2, cond)
|
|
errno = EDOM;
|
|
VECTOR_RETURN (VECTOR_INIT (z_notanum.d), cond2);
|
|
VECTOR_ELSE2 (cond2,cond)
|
|
errno = ERANGE;
|
|
VECTOR_RETURN (VECTOR_INIT (z_infinity.d), cond);
|
|
VECTOR_ENDIF
|
|
VECTOR_ENDIF
|
|
|
|
/* Initial checks are performed here. */
|
|
VECTOR_IF (x == 0.0, cond)
|
|
VECTOR_RETURN (VECTOR_INIT (0.0), cond);
|
|
VECTOR_ENDIF
|
|
VECTOR_IF (x < 0.0, cond)
|
|
errno = EDOM;
|
|
VECTOR_RETURN (VECTOR_INIT (z_notanum.d), cond);
|
|
VECTOR_ENDIF
|
|
|
|
/* Find the exponent and mantissa for the form x = f * 2^exp. */
|
|
v64df f = __builtin_gcn_frexpv_mant (x);
|
|
v64si exp = __builtin_gcn_frexpv_exp (x);
|
|
v64si odd = (exp & 1) != 0;
|
|
|
|
/* Get the initial approximation. */
|
|
v64df y = 0.41731 + 0.59016 * f;
|
|
|
|
f *= 0.5f;
|
|
/* Calculate the remaining iterations. */
|
|
y = y * 0.5f + f / y;
|
|
y = y * 0.5f + f / y;
|
|
y = y * 0.5f + f / y;
|
|
|
|
/* Calculate the final value. */
|
|
VECTOR_COND_MOVE (y, y * __SQRT_HALF, odd);
|
|
VECTOR_COND_MOVE (exp, exp + 1, odd);
|
|
exp >>= 1;
|
|
y = __builtin_gcn_ldexpv (y, exp);
|
|
|
|
VECTOR_RETURN (y, NO_COND);
|
|
|
|
FUNCTION_RETURN;
|
|
}
|
|
|
|
DEF_VARIANTS (sqrt, df, df)
|
|
|
|
#endif
|