mirror of
git://sourceware.org/git/newlib-cygwin.git
synced 2025-02-04 21:40:34 +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.
83 lines
2.5 KiB
C
83 lines
2.5 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_logarithm.c in Newlib. */
|
|
|
|
#include "amdgcnmach.h"
|
|
|
|
v64si v64df_finite (v64df);
|
|
v64si v64df_isnan (v64df);
|
|
|
|
static const double a[] = { -0.64124943423745581147e+02,
|
|
0.16383943563021534222e+02,
|
|
-0.78956112887481257267 };
|
|
static const double b[] = { -0.76949932108494879777e+03,
|
|
0.31203222091924532844e+03,
|
|
-0.35667977739034646171e+02 };
|
|
static const double C1 = 22713.0 / 32768.0;
|
|
static const double C2 = 1.428606820309417232e-06;
|
|
|
|
#if defined (__has_builtin) \
|
|
&& __has_builtin (__builtin_gcn_frexpv_mant) \
|
|
&& __has_builtin (__builtin_gcn_frexpv_exp) \
|
|
|
|
DEF_VD_MATH_FUNC (v64df, log, v64df x)
|
|
{
|
|
FUNCTION_INIT (v64df);
|
|
|
|
/* Check for domain/range errors here. */
|
|
VECTOR_IF (x == 0.0, cond)
|
|
errno = ERANGE;
|
|
VECTOR_RETURN (VECTOR_INIT (-z_infinity.d), cond);
|
|
VECTOR_ELSEIF (x < 0.0, cond)
|
|
errno = EDOM;
|
|
VECTOR_RETURN (VECTOR_INIT (z_notanum.d), cond);
|
|
VECTOR_ELSEIF (__builtin_convertvector (~v64df_finite (x), v64di), cond)
|
|
VECTOR_RETURN (VECTOR_MERGE (VECTOR_INIT (z_notanum.d),
|
|
VECTOR_INIT (z_infinity.d),
|
|
v64df_isnan (x)),
|
|
cond);
|
|
VECTOR_ENDIF
|
|
|
|
/* Get the exponent and mantissa where x = f * 2^N. */
|
|
v64df f = __builtin_gcn_frexpv_mant (x);
|
|
v64si N = __builtin_gcn_frexpv_exp (x);
|
|
|
|
v64df z = f - 0.5;
|
|
|
|
VECTOR_IF (f > __SQRT_HALF, cond)
|
|
VECTOR_COND_MOVE (z, (z - 0.5) / (f * 0.5 + 0.5), cond);
|
|
VECTOR_ELSE (cond)
|
|
VECTOR_COND_MOVE (N, N - 1, cond);
|
|
VECTOR_COND_MOVE (z, z / (z * 0.5 + 0.5), cond);
|
|
VECTOR_ENDIF
|
|
|
|
v64df w = z * z;
|
|
|
|
/* Use Newton's method with 4 terms. */
|
|
z += z * w * ((a[2] * w + a[1]) * w + a[0]) / (((w + b[2]) * w + b[1]) * w + b[0]);
|
|
|
|
v64df Nf = __builtin_convertvector (N, v64df);
|
|
VECTOR_COND_MOVE (z, (Nf * C2 + z) + Nf * C1, N != 0);
|
|
|
|
VECTOR_RETURN (z, NO_COND);
|
|
|
|
FUNCTION_RETURN;
|
|
}
|
|
|
|
DEF_VARIANTS (log, df, df)
|
|
|
|
DEF_VD_MATH_FUNC (v64df, log1p, v64df x)
|
|
{
|
|
/* TODO: Implement algorithm with better precision. */
|
|
return v64df_log_aux (1 + x, __mask);
|
|
}
|
|
|
|
DEF_VARIANTS (log1p, df, df)
|
|
|
|
#endif
|