glibc/sysdeps/ieee754/flt-32/math_config.h
Paul Zimmermann 6e98983c09 math: Optimized generic exp10f with wrappers
It is inspired by expf and reuses its tables and internal functions.
The error checks are inlined and errno setting is in separate tail
called functions, but the wrappers are kept in this patch to handle
the _LIB_VERSION==_SVID_ case.

Double precision arithmetics is used which is expected to be faster on
most targets (including soft-float) than using single precision and it
is easier to get good precision result with it.

Result for x86_64 (i7-4790K CPU @ 4.00GHz) are:

Before new code:
  "exp10f": {
   "workload-spec2017.wrf (adapted)": {
    "duration": 4.0414e+09,
    "iterations": 1.00128e+08,
    "reciprocal-throughput": 26.6818,
    "latency": 54.043,
    "max-throughput": 3.74787e+07,
    "min-throughput": 1.85038e+07
   }

With new code:
  "exp10f": {
   "workload-spec2017.wrf (adapted)": {
    "duration": 4.11951e+09,
    "iterations": 1.23968e+08,
    "reciprocal-throughput": 21.0581,
    "latency": 45.4028,
    "max-throughput": 4.74876e+07,
    "min-throughput": 2.20251e+07
   }

Result for aarch64 (A72 @ 2GHz) are:

Before new code:
  "exp10f": {
   "workload-spec2017.wrf (adapted)": {
    "duration": 4.62362e+09,
    "iterations": 3.3376e+07,
    "reciprocal-throughput": 127.698,
    "latency": 149.365,
    "max-throughput": 7.831e+06,
    "min-throughput": 6.69501e+06
   }

With new code:
  "exp10f": {
   "workload-spec2017.wrf (adapted)": {
    "duration": 4.29108e+09,
    "iterations": 6.6752e+07,
    "reciprocal-throughput": 51.2111,
    "latency": 77.3568,
    "max-throughput": 1.9527e+07,
    "min-throughput": 1.29271e+07
   }

Checked on x86_64-linux-gnu, powerpc64le-linux-gnu, aarch64-linux-gnu,
and sparc64-linux-gnu.
2020-06-19 10:48:15 -03:00

166 lines
4.0 KiB
C

/* Configuration for math routines.
Copyright (C) 2017-2020 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
#ifndef _MATH_CONFIG_H
#define _MATH_CONFIG_H
#include <math.h>
#include <math_private.h>
#include <nan-high-order-bit.h>
#include <stdint.h>
#ifndef WANT_ROUNDING
/* Correct special case results in non-nearest rounding modes. */
# define WANT_ROUNDING 1
#endif
#ifndef WANT_ERRNO
/* Set errno according to ISO C with (math_errhandling & MATH_ERRNO) != 0. */
# define WANT_ERRNO 1
#endif
#ifndef WANT_ERRNO_UFLOW
/* Set errno to ERANGE if result underflows to 0 (in all rounding modes). */
# define WANT_ERRNO_UFLOW (WANT_ROUNDING && WANT_ERRNO)
#endif
#ifndef TOINT_INTRINSICS
/* When set, the roundtoint and converttoint functions are provided with
the semantics documented below. */
# define TOINT_INTRINSICS 0
#endif
#if TOINT_INTRINSICS
/* Round x to nearest int in all rounding modes, ties have to be rounded
consistently with converttoint so the results match. If the result
would be outside of [-2^31, 2^31-1] then the semantics is unspecified. */
static inline double_t
roundtoint (double_t x);
/* Convert x to nearest int in all rounding modes, ties have to be rounded
consistently with roundtoint. If the result is not representible in an
int32_t then the semantics is unspecified. */
static inline int32_t
converttoint (double_t x);
#endif
static inline uint32_t
asuint (float f)
{
union
{
float f;
uint32_t i;
} u = {f};
return u.i;
}
static inline float
asfloat (uint32_t i)
{
union
{
uint32_t i;
float f;
} u = {i};
return u.f;
}
static inline uint64_t
asuint64 (double f)
{
union
{
double f;
uint64_t i;
} u = {f};
return u.i;
}
static inline double
asdouble (uint64_t i)
{
union
{
uint64_t i;
double f;
} u = {i};
return u.f;
}
#define NOINLINE __attribute__ ((noinline))
attribute_hidden float __math_oflowf (uint32_t);
attribute_hidden float __math_uflowf (uint32_t);
attribute_hidden float __math_may_uflowf (uint32_t);
attribute_hidden float __math_divzerof (uint32_t);
attribute_hidden float __math_invalidf (float);
/* Shared between expf, exp2f, exp10f, and powf. */
#define EXP2F_TABLE_BITS 5
#define EXP2F_POLY_ORDER 3
extern const struct exp2f_data
{
uint64_t tab[1 << EXP2F_TABLE_BITS];
double shift_scaled;
double poly[EXP2F_POLY_ORDER];
double shift;
double invln2_scaled;
double poly_scaled[EXP2F_POLY_ORDER];
} __exp2f_data attribute_hidden;
#define LOGF_TABLE_BITS 4
#define LOGF_POLY_ORDER 4
extern const struct logf_data
{
struct
{
double invc, logc;
} tab[1 << LOGF_TABLE_BITS];
double ln2;
double poly[LOGF_POLY_ORDER - 1]; /* First order coefficient is 1. */
} __logf_data attribute_hidden;
#define LOG2F_TABLE_BITS 4
#define LOG2F_POLY_ORDER 4
extern const struct log2f_data
{
struct
{
double invc, logc;
} tab[1 << LOG2F_TABLE_BITS];
double poly[LOG2F_POLY_ORDER];
} __log2f_data attribute_hidden;
#define POWF_LOG2_TABLE_BITS 4
#define POWF_LOG2_POLY_ORDER 5
#if TOINT_INTRINSICS
# define POWF_SCALE_BITS EXP2F_TABLE_BITS
#else
# define POWF_SCALE_BITS 0
#endif
#define POWF_SCALE ((double) (1 << POWF_SCALE_BITS))
extern const struct powf_log2_data
{
struct
{
double invc, logc;
} tab[1 << POWF_LOG2_TABLE_BITS];
double poly[POWF_LOG2_POLY_ORDER];
} __powf_log2_data attribute_hidden;
#endif