New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
2017-05-25 23:41:38 +08:00
|
|
|
/* Configuration for math routines.
|
Improve performance of sinf/cosf/sincosf
Here is the correct patch with both filenames and int cast fixed:
This patch is a complete rewrite of sinf, cosf and sincosf. The new version
is significantly faster, as well as simple and accurate.
The worst-case ULP is 0.56072, maximum relative error is 0.5303p-23 over all
4 billion inputs. In non-nearest rounding modes the error is 1ULP.
The algorithm uses 3 main cases: small inputs which don't need argument
reduction, small inputs which need a simple range reduction and large inputs
requiring complex range reduction. The code uses approximate integer
comparisons to quickly decide between these cases - on some targets this may
be slow, so this can be configured to use floating point comparisons.
The small range reducer uses a single reduction step to handle values up to
120.0. It is fastest on targets which support inlined round instructions.
The large range reducer uses integer arithmetic for simplicity. It does a
32x96 bit multiply to compute a 64-bit modulo result. This is more than
accurate enough to handle the worst-case cancellation for values close to
an integer multiple of PI/4. It could be further optimized, however it is
already much faster than necessary.
Simple benchmark showing speedup factor on AArch64 for various ranges:
range 0.7853982 sinf 1.7 cosf 2.2 sincosf 2.8
range 1.570796 sinf 1.9 cosf 1.9 sincosf 2.7
range 3.141593 sinf 2.0 cosf 2.0 sincosf 3.5
range 6.283185 sinf 2.3 cosf 2.3 sincosf 4.2
range 125.6637 sinf 2.9 cosf 3.0 sincosf 5.1
range 1.1259e15 sinf 26.8 cosf 26.8 sincosf 45.2
ChangeLog:
2018-05-18 Wilco Dijkstra <wdijkstr@arm.com>
* newlib/libm/common/Makefile.in: Regenerated.
* newlib/libm/common/Makefile.am: Add sinf.c, cosf.c, sincosf.c
sincosf.h, sincosf_data.c. Add -fbuiltin -fno-math-errno to CFLAGS.
* newlib/libm/common/math_config.h: Add HAVE_FAST_ROUND, HAVE_FAST_LROUND,
roundtoint, converttoint, force_eval_float, force_eval_double, eval_as_float,
eval_as_double, likely, unlikely.
* newlib/libm/common/cosf.c: New file.
* newlib/libm/common/sinf.c: Likewise.
* newlib/libm/common/sincosf.h: Likewise.
* newlib/libm/common/sincosf.c: Likewise.
* newlib/libm/common/sincosf_data.c: Likewise.
* newlib/libm/math/sf_cos.c: Add #if to build conditionally.
* newlib/libm/math/sf_sin.c: Likewise.
* newlib/libm/math/wf_sincos.c: Likewise.
--
2018-06-20 20:07:22 +08:00
|
|
|
Copyright (c) 2017-2018 Arm Ltd. All rights reserved.
|
New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
2017-05-25 23:41:38 +08:00
|
|
|
|
|
|
|
Redistribution and use in source and binary forms, with or without
|
|
|
|
modification, are permitted provided that the following conditions
|
|
|
|
are met:
|
|
|
|
1. Redistributions of source code must retain the above copyright
|
|
|
|
notice, this list of conditions and the following disclaimer.
|
|
|
|
2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
notice, this list of conditions and the following disclaimer in the
|
|
|
|
documentation and/or other materials provided with the distribution.
|
|
|
|
3. The name of the company may not be used to endorse or promote
|
|
|
|
products derived from this software without specific prior written
|
|
|
|
permission.
|
|
|
|
|
|
|
|
THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS AND ANY EXPRESS OR IMPLIED
|
|
|
|
WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
|
|
|
|
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
|
|
|
|
IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
|
|
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
|
|
|
|
TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
|
|
|
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
|
|
|
|
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
|
|
|
|
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
|
|
|
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
|
|
|
|
|
|
|
|
#ifndef _MATH_CONFIG_H
|
|
|
|
#define _MATH_CONFIG_H
|
|
|
|
|
|
|
|
#include <math.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
|
|
|
|
|
Improve performance of sinf/cosf/sincosf
Here is the correct patch with both filenames and int cast fixed:
This patch is a complete rewrite of sinf, cosf and sincosf. The new version
is significantly faster, as well as simple and accurate.
The worst-case ULP is 0.56072, maximum relative error is 0.5303p-23 over all
4 billion inputs. In non-nearest rounding modes the error is 1ULP.
The algorithm uses 3 main cases: small inputs which don't need argument
reduction, small inputs which need a simple range reduction and large inputs
requiring complex range reduction. The code uses approximate integer
comparisons to quickly decide between these cases - on some targets this may
be slow, so this can be configured to use floating point comparisons.
The small range reducer uses a single reduction step to handle values up to
120.0. It is fastest on targets which support inlined round instructions.
The large range reducer uses integer arithmetic for simplicity. It does a
32x96 bit multiply to compute a 64-bit modulo result. This is more than
accurate enough to handle the worst-case cancellation for values close to
an integer multiple of PI/4. It could be further optimized, however it is
already much faster than necessary.
Simple benchmark showing speedup factor on AArch64 for various ranges:
range 0.7853982 sinf 1.7 cosf 2.2 sincosf 2.8
range 1.570796 sinf 1.9 cosf 1.9 sincosf 2.7
range 3.141593 sinf 2.0 cosf 2.0 sincosf 3.5
range 6.283185 sinf 2.3 cosf 2.3 sincosf 4.2
range 125.6637 sinf 2.9 cosf 3.0 sincosf 5.1
range 1.1259e15 sinf 26.8 cosf 26.8 sincosf 45.2
ChangeLog:
2018-05-18 Wilco Dijkstra <wdijkstr@arm.com>
* newlib/libm/common/Makefile.in: Regenerated.
* newlib/libm/common/Makefile.am: Add sinf.c, cosf.c, sincosf.c
sincosf.h, sincosf_data.c. Add -fbuiltin -fno-math-errno to CFLAGS.
* newlib/libm/common/math_config.h: Add HAVE_FAST_ROUND, HAVE_FAST_LROUND,
roundtoint, converttoint, force_eval_float, force_eval_double, eval_as_float,
eval_as_double, likely, unlikely.
* newlib/libm/common/cosf.c: New file.
* newlib/libm/common/sinf.c: Likewise.
* newlib/libm/common/sincosf.h: Likewise.
* newlib/libm/common/sincosf.c: Likewise.
* newlib/libm/common/sincosf_data.c: Likewise.
* newlib/libm/math/sf_cos.c: Add #if to build conditionally.
* newlib/libm/math/sf_sin.c: Likewise.
* newlib/libm/math/wf_sincos.c: Likewise.
--
2018-06-20 20:07:22 +08:00
|
|
|
/* Compiler can inline round as a single instruction. */
|
|
|
|
#ifndef HAVE_FAST_ROUND
|
|
|
|
# if __aarch64__
|
|
|
|
# define HAVE_FAST_ROUND 1
|
|
|
|
# else
|
|
|
|
# define HAVE_FAST_ROUND 0
|
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* Compiler can inline lround, but not (long)round(x). */
|
|
|
|
#ifndef HAVE_FAST_LROUND
|
|
|
|
# if __aarch64__ && (100*__GNUC__ + __GNUC_MINOR__) >= 408 && __NO_MATH_ERRNO__
|
|
|
|
# define HAVE_FAST_LROUND 1
|
|
|
|
# else
|
|
|
|
# define HAVE_FAST_LROUND 0
|
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#if HAVE_FAST_ROUND
|
|
|
|
# define TOINT_INTRINSICS 1
|
|
|
|
|
|
|
|
static inline double_t
|
|
|
|
roundtoint (double_t x)
|
|
|
|
{
|
|
|
|
return round (x);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline uint64_t
|
|
|
|
converttoint (double_t x)
|
|
|
|
{
|
|
|
|
# if HAVE_FAST_LROUND
|
|
|
|
return lround (x);
|
|
|
|
# else
|
|
|
|
return (long) round (x);
|
|
|
|
# endif
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
2017-05-25 23:41:38 +08:00
|
|
|
#ifndef TOINT_INTRINSICS
|
|
|
|
# define TOINT_INTRINSICS 0
|
|
|
|
#endif
|
|
|
|
#ifndef TOINT_RINT
|
|
|
|
# define TOINT_RINT 0
|
|
|
|
#endif
|
|
|
|
#ifndef TOINT_SHIFT
|
|
|
|
# define TOINT_SHIFT 1
|
|
|
|
#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;
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifndef IEEE_754_2008_SNAN
|
|
|
|
# define IEEE_754_2008_SNAN 1
|
|
|
|
#endif
|
|
|
|
static inline int
|
|
|
|
issignalingf_inline (float x)
|
|
|
|
{
|
|
|
|
uint32_t ix = asuint (x);
|
|
|
|
if (!IEEE_754_2008_SNAN)
|
|
|
|
return (ix & 0x7fc00000) == 0x7fc00000;
|
|
|
|
return 2 * (ix ^ 0x00400000) > 2u * 0x7fc00000;
|
|
|
|
}
|
|
|
|
|
Improve performance of sinf/cosf/sincosf
Here is the correct patch with both filenames and int cast fixed:
This patch is a complete rewrite of sinf, cosf and sincosf. The new version
is significantly faster, as well as simple and accurate.
The worst-case ULP is 0.56072, maximum relative error is 0.5303p-23 over all
4 billion inputs. In non-nearest rounding modes the error is 1ULP.
The algorithm uses 3 main cases: small inputs which don't need argument
reduction, small inputs which need a simple range reduction and large inputs
requiring complex range reduction. The code uses approximate integer
comparisons to quickly decide between these cases - on some targets this may
be slow, so this can be configured to use floating point comparisons.
The small range reducer uses a single reduction step to handle values up to
120.0. It is fastest on targets which support inlined round instructions.
The large range reducer uses integer arithmetic for simplicity. It does a
32x96 bit multiply to compute a 64-bit modulo result. This is more than
accurate enough to handle the worst-case cancellation for values close to
an integer multiple of PI/4. It could be further optimized, however it is
already much faster than necessary.
Simple benchmark showing speedup factor on AArch64 for various ranges:
range 0.7853982 sinf 1.7 cosf 2.2 sincosf 2.8
range 1.570796 sinf 1.9 cosf 1.9 sincosf 2.7
range 3.141593 sinf 2.0 cosf 2.0 sincosf 3.5
range 6.283185 sinf 2.3 cosf 2.3 sincosf 4.2
range 125.6637 sinf 2.9 cosf 3.0 sincosf 5.1
range 1.1259e15 sinf 26.8 cosf 26.8 sincosf 45.2
ChangeLog:
2018-05-18 Wilco Dijkstra <wdijkstr@arm.com>
* newlib/libm/common/Makefile.in: Regenerated.
* newlib/libm/common/Makefile.am: Add sinf.c, cosf.c, sincosf.c
sincosf.h, sincosf_data.c. Add -fbuiltin -fno-math-errno to CFLAGS.
* newlib/libm/common/math_config.h: Add HAVE_FAST_ROUND, HAVE_FAST_LROUND,
roundtoint, converttoint, force_eval_float, force_eval_double, eval_as_float,
eval_as_double, likely, unlikely.
* newlib/libm/common/cosf.c: New file.
* newlib/libm/common/sinf.c: Likewise.
* newlib/libm/common/sincosf.h: Likewise.
* newlib/libm/common/sincosf.c: Likewise.
* newlib/libm/common/sincosf_data.c: Likewise.
* newlib/libm/math/sf_cos.c: Add #if to build conditionally.
* newlib/libm/math/sf_sin.c: Likewise.
* newlib/libm/math/wf_sincos.c: Likewise.
--
2018-06-20 20:07:22 +08:00
|
|
|
/* Force the evaluation of a floating-point expression for its side-effect. */
|
|
|
|
#if __aarch64__ && __GNUC__
|
|
|
|
static inline void
|
|
|
|
force_eval_float (float x)
|
|
|
|
{
|
|
|
|
__asm__ __volatile__ ("" : "+w" (x));
|
|
|
|
}
|
|
|
|
static inline void
|
|
|
|
force_eval_double (double x)
|
|
|
|
{
|
|
|
|
__asm__ __volatile__ ("" : "+w" (x));
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
static inline void
|
|
|
|
force_eval_float (float x)
|
|
|
|
{
|
|
|
|
volatile float y = x;
|
|
|
|
}
|
|
|
|
static inline void
|
|
|
|
force_eval_double (double x)
|
|
|
|
{
|
|
|
|
volatile double y = x;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* Evaluate an expression as the specified type, normally a type
|
|
|
|
cast should be enough, but compilers implement non-standard
|
|
|
|
excess-precision handling, so when FLT_EVAL_METHOD != 0 then
|
|
|
|
these functions may need to be customized. */
|
|
|
|
static inline float
|
|
|
|
eval_as_float (float x)
|
|
|
|
{
|
|
|
|
return x;
|
|
|
|
}
|
|
|
|
static inline double
|
|
|
|
eval_as_double (double x)
|
|
|
|
{
|
|
|
|
return x;
|
|
|
|
}
|
|
|
|
|
New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
2017-05-25 23:41:38 +08:00
|
|
|
#ifdef __GNUC__
|
|
|
|
# define HIDDEN __attribute__ ((__visibility__ ("hidden")))
|
|
|
|
# define NOINLINE __attribute__ ((noinline))
|
Improve performance of sinf/cosf/sincosf
Here is the correct patch with both filenames and int cast fixed:
This patch is a complete rewrite of sinf, cosf and sincosf. The new version
is significantly faster, as well as simple and accurate.
The worst-case ULP is 0.56072, maximum relative error is 0.5303p-23 over all
4 billion inputs. In non-nearest rounding modes the error is 1ULP.
The algorithm uses 3 main cases: small inputs which don't need argument
reduction, small inputs which need a simple range reduction and large inputs
requiring complex range reduction. The code uses approximate integer
comparisons to quickly decide between these cases - on some targets this may
be slow, so this can be configured to use floating point comparisons.
The small range reducer uses a single reduction step to handle values up to
120.0. It is fastest on targets which support inlined round instructions.
The large range reducer uses integer arithmetic for simplicity. It does a
32x96 bit multiply to compute a 64-bit modulo result. This is more than
accurate enough to handle the worst-case cancellation for values close to
an integer multiple of PI/4. It could be further optimized, however it is
already much faster than necessary.
Simple benchmark showing speedup factor on AArch64 for various ranges:
range 0.7853982 sinf 1.7 cosf 2.2 sincosf 2.8
range 1.570796 sinf 1.9 cosf 1.9 sincosf 2.7
range 3.141593 sinf 2.0 cosf 2.0 sincosf 3.5
range 6.283185 sinf 2.3 cosf 2.3 sincosf 4.2
range 125.6637 sinf 2.9 cosf 3.0 sincosf 5.1
range 1.1259e15 sinf 26.8 cosf 26.8 sincosf 45.2
ChangeLog:
2018-05-18 Wilco Dijkstra <wdijkstr@arm.com>
* newlib/libm/common/Makefile.in: Regenerated.
* newlib/libm/common/Makefile.am: Add sinf.c, cosf.c, sincosf.c
sincosf.h, sincosf_data.c. Add -fbuiltin -fno-math-errno to CFLAGS.
* newlib/libm/common/math_config.h: Add HAVE_FAST_ROUND, HAVE_FAST_LROUND,
roundtoint, converttoint, force_eval_float, force_eval_double, eval_as_float,
eval_as_double, likely, unlikely.
* newlib/libm/common/cosf.c: New file.
* newlib/libm/common/sinf.c: Likewise.
* newlib/libm/common/sincosf.h: Likewise.
* newlib/libm/common/sincosf.c: Likewise.
* newlib/libm/common/sincosf_data.c: Likewise.
* newlib/libm/math/sf_cos.c: Add #if to build conditionally.
* newlib/libm/math/sf_sin.c: Likewise.
* newlib/libm/math/wf_sincos.c: Likewise.
--
2018-06-20 20:07:22 +08:00
|
|
|
# define likely(x) __builtin_expect (!!(x), 1)
|
|
|
|
# define unlikely(x) __builtin_expect (x, 0)
|
New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
2017-05-25 23:41:38 +08:00
|
|
|
#else
|
|
|
|
# define HIDDEN
|
|
|
|
# define NOINLINE
|
Improve performance of sinf/cosf/sincosf
Here is the correct patch with both filenames and int cast fixed:
This patch is a complete rewrite of sinf, cosf and sincosf. The new version
is significantly faster, as well as simple and accurate.
The worst-case ULP is 0.56072, maximum relative error is 0.5303p-23 over all
4 billion inputs. In non-nearest rounding modes the error is 1ULP.
The algorithm uses 3 main cases: small inputs which don't need argument
reduction, small inputs which need a simple range reduction and large inputs
requiring complex range reduction. The code uses approximate integer
comparisons to quickly decide between these cases - on some targets this may
be slow, so this can be configured to use floating point comparisons.
The small range reducer uses a single reduction step to handle values up to
120.0. It is fastest on targets which support inlined round instructions.
The large range reducer uses integer arithmetic for simplicity. It does a
32x96 bit multiply to compute a 64-bit modulo result. This is more than
accurate enough to handle the worst-case cancellation for values close to
an integer multiple of PI/4. It could be further optimized, however it is
already much faster than necessary.
Simple benchmark showing speedup factor on AArch64 for various ranges:
range 0.7853982 sinf 1.7 cosf 2.2 sincosf 2.8
range 1.570796 sinf 1.9 cosf 1.9 sincosf 2.7
range 3.141593 sinf 2.0 cosf 2.0 sincosf 3.5
range 6.283185 sinf 2.3 cosf 2.3 sincosf 4.2
range 125.6637 sinf 2.9 cosf 3.0 sincosf 5.1
range 1.1259e15 sinf 26.8 cosf 26.8 sincosf 45.2
ChangeLog:
2018-05-18 Wilco Dijkstra <wdijkstr@arm.com>
* newlib/libm/common/Makefile.in: Regenerated.
* newlib/libm/common/Makefile.am: Add sinf.c, cosf.c, sincosf.c
sincosf.h, sincosf_data.c. Add -fbuiltin -fno-math-errno to CFLAGS.
* newlib/libm/common/math_config.h: Add HAVE_FAST_ROUND, HAVE_FAST_LROUND,
roundtoint, converttoint, force_eval_float, force_eval_double, eval_as_float,
eval_as_double, likely, unlikely.
* newlib/libm/common/cosf.c: New file.
* newlib/libm/common/sinf.c: Likewise.
* newlib/libm/common/sincosf.h: Likewise.
* newlib/libm/common/sincosf.c: Likewise.
* newlib/libm/common/sincosf_data.c: Likewise.
* newlib/libm/math/sf_cos.c: Add #if to build conditionally.
* newlib/libm/math/sf_sin.c: Likewise.
* newlib/libm/math/wf_sincos.c: Likewise.
--
2018-06-20 20:07:22 +08:00
|
|
|
# define likely(x) (x)
|
|
|
|
# define unlikely(x) (x)
|
New expf, exp2f, logf, log2f and powf implementations
Based on code from https://github.com/ARM-software/optimized-routines/
This patch adds a highly optimized generic implementation of expf,
exp2f, logf, log2f and powf. The new functions are not only
faster (6x for powf!), but are also smaller and more accurate.
In order to achieve this, the algorithm uses double precision
arithmetic for accuracy, avoids divisions and uses small table
lookups to minimize the polynomials. Special cases are handled
inline to avoid the unnecessary overhead of wrapper functions and
set errno to POSIX requirements.
The new functions are added under newlib/libm/common, but the old
implementations are kept (in newlib/libm/math) for non-IEEE or
pre-C99 systems. Targets can enable the new math code by defining
__OBSOLETE_MATH_DEFAULT to 0 in newlib/libc/include/machine/ieeefp.h,
users can override the default by defining __OBSOLETE_MATH.
Currently the new code is enabled for AArch64 and AArch32 with VFP.
Targets with a single precision FPU may still prefer the old
implementation.
libm.a size changes:
arm: -1692
arm/thumb/v7-a/nofp: -878
arm/thumb/v7-a+fp/hard: -864
arm/thumb/v7-a+fp/softfp: -908
aarch64: -1476
2017-05-25 23:41:38 +08:00
|
|
|
#endif
|
|
|
|
|
|
|
|
HIDDEN float __math_oflowf (unsigned long);
|
|
|
|
HIDDEN float __math_uflowf (unsigned long);
|
|
|
|
HIDDEN float __math_may_uflowf (unsigned long);
|
|
|
|
HIDDEN float __math_divzerof (unsigned long);
|
|
|
|
HIDDEN float __math_invalidf (float);
|
|
|
|
|
|
|
|
/* Shared between expf, exp2f 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 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 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 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 HIDDEN;
|
|
|
|
|
|
|
|
#endif
|