Skip to content
fast log and exp functions for x86/x64 SSE
Branch: master
Clone or download
herumi Merge pull request #11 from jschueller/patch-1
Dont assume expd_v size argument is a 2/4 multiple
Latest commit 595f39f Feb 26, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
include/cybozu avoid xbyak at default Jun 15, 2017
Makefile avoid xbyak at default Jun 15, 2017
avx2.cpp
bench.cpp avoid xbyak at default Jun 15, 2017
bench.sln update sln to Visual Studio 14 Mar 21, 2017
bench.vcproj first commit Apr 16, 2010
ck.cpp remove unused code and rename vec_expd Aug 26, 2011
fastexp.cpp avoid xbyak at default Jun 15, 2017
fmath.hpp Dont assume expd_v size argument is a 2/4 multiple Feb 25, 2019
readme.md change old url to new url Aug 29, 2016
readme.txt change old url to new url Aug 29, 2016

readme.md

fast approximate function of exponential function exp and log

How to use

include fmath.hpp and use fmath::log, fmath::exp, fmath::expd.

fmath::PowGenerator is a class to generate a function to compute pow(x, y) of x >= 0 for a given fixed y > 0.

eg. fmath::PowGenerator f(1.234); f.get(x) returns pow(x, 1.234);

Prototpye of function

  • float fmath::exp(float);
  • float fmath::log(float);
  • double fmath::logd(double);
  • __m128 fmath::exp_ps(__m128);
  • __m128 fmath::log_ps(__m128);
  • void fmath::expv_d(double *p, size_t n); // for double p[n];

Experimental

If you install xbyak and define FMATH_USE_XBYAK before including fmath.hpp, then fmath::exp() and fmath::exp_ps() will be about 10~20 % faster. Xbyak version uses SSE4.1 if available.

AVX version of fmath::exp is experimental

Remark

gcc puts warnings such as "dereferencing type-punned pointer will break strict-aliasing rules." It is no problem. Please change #if 1 in fmath.hpp:423 if you worry about it. But it causes a little slower.

-ffast-math option of gcc may generate bad code for fmath::expd.

License

modified new BSD License http://opensource.org/licenses/BSD-3-Clause

History

  • 2012/Oct/30 fix fmath::expd for small value
  • 2011/Aug/26 add fmath::expd_v
  • 2011/Mar/25 exp supports AVX
  • 2011/Mar/25 exp, exp_ps support avx
  • 2010/Feb/16 add fmath::exp_ps, log_ps and optimize functions
  • 2010/Jan/10 add fmath::PowGenerator
  • 2009/Dec/28 add fmath::log()
  • 2009/Dec/09 support cygwin
  • 2009/Dec/08 first version

Author

MITSUNARI Shigeo(herumi@nifty.com) http://herumi.in.coocan.jp/

Benchmark

compiler

  • Visual Studio 2010RC
  • icc 11.1
  • gcc 4.3.2 on cygwin
  • gcc 4.4.1 on 64bit Linux

option

  • cl(icl):

/Ox /Ob2 /GS- /Zi /D_SECURE_SCL=0 /MD /Oy /arch:SSE2 /fp:fast /DNOMINMAX

  • gcc:

-O3 -fomit-frame-pointer -DNDEBUG -fno-operator-names -msse2 -mfpmath=sse -march=native

see fastexp.cpp

You can’t perform that action at this time.