[gs-bugs] [Bug 689794] gs gets /rangecheck in --initgraphics-- when compiled with -ffast-math

bugs.ghostscript.com-bugzilla-daemon at ghostscript.com bugs.ghostscript.com-bugzilla-daemon at ghostscript.com
Fri Apr 18 16:50:24 PDT 2008


http://bugs.ghostscript.com/show_bug.cgi?id=689794





------- Additional Comments From williambader at hotmail.com  2008-04-18 16:50 -------
Does it matter if it isn't transitive?
Isn't the worst that can happen is that rotation by an almost multiple of 90
followed by the reverse rotation by an almost multiple of 90 might now do one
exact and the other by the more general case instead of doing both by the more
general case?  Any small rotation by an almost multiple of 90 can cause the
rangecheck, so in the cases where it makes a difference, the result would
probably be a rangecheck, and even if gs didn't get a rangecheck, the trig
functions would probably not end up canceling exactly, so the result might still
be a very very small rotation.
My guess with the optimization is that trig functions are usually implemented
based on radians as series or tables, and if the function implementation itself
is an approximation and then if the argument to the function is an approximation
of a multiple of pi/2, getting multiples of 90 to return exact values could be
tricky, and very small values can cause problems.



------- You are receiving this mail because: -------
You are the QA contact for the bug, or are watching the QA contact.



More information about the gs-bugs mailing list