xref: /freebsd/contrib/arm-optimized-routines/math/exp2f.c (revision 072a4ba82a01476eaee33781ccd241033eefcf0b)
131914882SAlex Richardson /*
231914882SAlex Richardson  * Single-precision 2^x function.
331914882SAlex Richardson  *
431914882SAlex Richardson  * Copyright (c) 2017-2018, Arm Limited.
5*072a4ba8SAndrew Turner  * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
631914882SAlex Richardson  */
731914882SAlex Richardson 
831914882SAlex Richardson #include <math.h>
931914882SAlex Richardson #include <stdint.h>
1031914882SAlex Richardson #include "math_config.h"
1131914882SAlex Richardson 
1231914882SAlex Richardson /*
1331914882SAlex Richardson EXP2F_TABLE_BITS = 5
1431914882SAlex Richardson EXP2F_POLY_ORDER = 3
1531914882SAlex Richardson 
1631914882SAlex Richardson ULP error: 0.502 (nearest rounding.)
1731914882SAlex Richardson Relative error: 1.69 * 2^-34 in [-1/64, 1/64] (before rounding.)
1831914882SAlex Richardson Wrong count: 168353 (all nearest rounding wrong results with fma.)
1931914882SAlex Richardson Non-nearest ULP error: 1 (rounded ULP error)
2031914882SAlex Richardson */
2131914882SAlex Richardson 
2231914882SAlex Richardson #define N (1 << EXP2F_TABLE_BITS)
2331914882SAlex Richardson #define T __exp2f_data.tab
2431914882SAlex Richardson #define C __exp2f_data.poly
2531914882SAlex Richardson #define SHIFT __exp2f_data.shift_scaled
2631914882SAlex Richardson 
2731914882SAlex Richardson static inline uint32_t
2831914882SAlex Richardson top12 (float x)
2931914882SAlex Richardson {
3031914882SAlex Richardson   return asuint (x) >> 20;
3131914882SAlex Richardson }
3231914882SAlex Richardson 
3331914882SAlex Richardson float
3431914882SAlex Richardson exp2f (float x)
3531914882SAlex Richardson {
3631914882SAlex Richardson   uint32_t abstop;
3731914882SAlex Richardson   uint64_t ki, t;
3831914882SAlex Richardson   /* double_t for better performance on targets with FLT_EVAL_METHOD==2.  */
3931914882SAlex Richardson   double_t kd, xd, z, r, r2, y, s;
4031914882SAlex Richardson 
4131914882SAlex Richardson   xd = (double_t) x;
4231914882SAlex Richardson   abstop = top12 (x) & 0x7ff;
4331914882SAlex Richardson   if (unlikely (abstop >= top12 (128.0f)))
4431914882SAlex Richardson     {
4531914882SAlex Richardson       /* |x| >= 128 or x is nan.  */
4631914882SAlex Richardson       if (asuint (x) == asuint (-INFINITY))
4731914882SAlex Richardson 	return 0.0f;
4831914882SAlex Richardson       if (abstop >= top12 (INFINITY))
4931914882SAlex Richardson 	return x + x;
5031914882SAlex Richardson       if (x > 0.0f)
5131914882SAlex Richardson 	return __math_oflowf (0);
5231914882SAlex Richardson       if (x <= -150.0f)
5331914882SAlex Richardson 	return __math_uflowf (0);
5431914882SAlex Richardson #if WANT_ERRNO_UFLOW
5531914882SAlex Richardson       if (x < -149.0f)
5631914882SAlex Richardson 	return __math_may_uflowf (0);
5731914882SAlex Richardson #endif
5831914882SAlex Richardson     }
5931914882SAlex Richardson 
6031914882SAlex Richardson   /* x = k/N + r with r in [-1/(2N), 1/(2N)] and int k.  */
6131914882SAlex Richardson   kd = eval_as_double (xd + SHIFT);
6231914882SAlex Richardson   ki = asuint64 (kd);
6331914882SAlex Richardson   kd -= SHIFT; /* k/N for int k.  */
6431914882SAlex Richardson   r = xd - kd;
6531914882SAlex Richardson 
6631914882SAlex Richardson   /* exp2(x) = 2^(k/N) * 2^r ~= s * (C0*r^3 + C1*r^2 + C2*r + 1) */
6731914882SAlex Richardson   t = T[ki % N];
6831914882SAlex Richardson   t += ki << (52 - EXP2F_TABLE_BITS);
6931914882SAlex Richardson   s = asdouble (t);
7031914882SAlex Richardson   z = C[0] * r + C[1];
7131914882SAlex Richardson   r2 = r * r;
7231914882SAlex Richardson   y = C[2] * r + 1;
7331914882SAlex Richardson   y = z * r2 + y;
7431914882SAlex Richardson   y = y * s;
7531914882SAlex Richardson   return eval_as_float (y);
7631914882SAlex Richardson }
7731914882SAlex Richardson #if USE_GLIBC_ABI
7831914882SAlex Richardson strong_alias (exp2f, __exp2f_finite)
7931914882SAlex Richardson hidden_alias (exp2f, __ieee754_exp2f)
8031914882SAlex Richardson #endif
81