hi:
have floating point intensive application crashes randomly when processing floats. might caused denormalized numbers.
flush-to-zero mode on arm set default on rpi? (precission can ignored).
in other case, how can set up?
use gcc cross-toolchain linaro, rpi 2 b+.
thanks.
have floating point intensive application crashes randomly when processing floats. might caused denormalized numbers.
flush-to-zero mode on arm set default on rpi? (precission can ignored).
in other case, how can set up?
use gcc cross-toolchain linaro, rpi 2 b+.
thanks.
setting ftz (flush 0 mode).
looks bit 24 in fpscr (just google arm fpscr).
there better way can suggest, try:-or similar.
think fpscr set 0 default on linux.
should work pi's, using vfp or neon.
looks bit 24 in fpscr (just google arm fpscr).
there better way can suggest, try:-
code: select all
asm( "vmsr fpscr,%0" :: "r" (1 << 24) )
think fpscr set 0 default on linux.
should work pi's, using vfp or neon.
raspberrypi
Comments
Post a Comment