[Bug math/19869] floor and fmod not consistent with each other

[joseph at codesourcery dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=19869

--- Comment #2 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
As you're using i686, your program may be subject to excess precision 
issues.  Thus, to demonstrate a problem in glibc you need to show specific 
double values - preferably as hex floats - that, as inputs to specific 
glibc functions (make sure to use -fno-builtin so the values do go to 
glibc functions, not built-in versions), produce a specific incorrect 
result; a C program doing double arithmetic to compute values may produce 
apparent inconsistencies not because of a glibc bug but because of the 
same value sometimes being used with excess precision and sometimes 
without.

Does your problem appear with -mfpmath=sse -fno-builtin?  If not, that 
indicates there is no glibc issue.  If it does appear with -mfpmath=sse 
-fno-builtin, you should be able to identify either a single double value 
(represented as a hex float) on which floor produces incorrect results, or 
a pair of two such values (represented as hex floats) on which fmod 
produces incorrect results.

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