Martin Ling <martin-uf...@earth.li> writes: > -ffast-math: > > "This option is not turned on by any -O option besides -Ofast, since it > can result in incorrect output for programs that depend on an exact > implementation of IEEE or ISO rules/specifications for math functions. > It may, however, yield faster code for programs that do not require the > guarantees of these specifications." > > Or more succintly from an older version of that documentation: > > "Might allow some programs designed to not be too dependent on IEEE > behavior for floating-point to run faster, or die trying." > > Since neither UFRaw nor Lensfun are deliberately designed this way > (and doing so would require also tracking GCC changes to see what else > might break with each new compiler version), I think it's reasonable to > say it's not a bug for UFRaw to fail when compiled with that option. > > -O3 should always be fine though.
This doesn't seem like a bug in ufraw. IEEE 754 requires a specific answer for each computation, with the tricky bits being in rounding and intermediate precision. i386 has extra-precision registers, and reordering computation can produce different results. So I think the kinds of things fast-math *should* be getting into trouble with is computations where it's considered a bug to get an answer which is different in the lowest-order bits, or when the thing to be computed is unstable. Getting 0 instead of 800 for the line in question is not a case like that, and I think is likely to be an outright optimizer bug rather than the case being warned about. Given all that, I think it makes sense for ufraw not to ask for fast-math.
pgpkMzkhIp9Tv.pgp
Description: PGP signature
------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_feb
_______________________________________________ ufraw-devel mailing list ufraw-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ufraw-devel
