[Bug math/28472] pow(10, i) accuracy

["newbie-02 at gmx dot de" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28472

--- Comment #5 from b. <newbie-02 at gmx dot de> ---
if ... I'm right in testing ... Intel 'oneAPI' compiler(s) 'icpx' and 'dpcpp'
produce clean powers of 10 ( matching E-string evaluation ) for long double
figures with 'exp10l( x )'.  

since they do not compile themselves but use clang/clang++ for this purpose -
with a huge bunch of options and includes - it should be possible to have a
look at it and gather some inspiration ... ??? hot spots seem to be calling
clang with '-cc1' and '-x c++'.  

( for the bug itself i propose testing and setting to new, besides the
deviation it is a consistency problem which is difficult to explain to users:
that one get's different results in one system for different formulations of
the same  mathematical task. ('pow( 10, x )' != '1Ex' != 'pow10( x )' !=
'exp10( x )' ).  

( and of course gcc shouldn't be second in 'best compilers' ) 

reg. the proposal of using exp10( x ) instead of pow( 10, x ):  
'exp10( x )' compiled with gcc -O2: 230 fails between x = -324 and x = 308, 
'pow( 10, x )': only 2, 
as well with 'exp10( x )' 
( complaining " incompatible implicit declaration of built-in function ‘exp10’
" ) 
as for '__builtin_exp10( x )'.  
as well for gcc, g++, clang++, 
'clang' additional complaints and producing even more fails.

-- 
You are receiving this mail because:
You are on the CC list for the bug.