Tim Olson wrote:
On Jan 8, 2009, at 9:30 AM, Hans-Martin Mosner wrote:
AFAIK, the Croquet project needs to have exactly reproducible float arithmetics on all platforms.
Slightly off-topic, but since we have such a knowledgeable crowd...
As noted by John, Croquet uses fdlibm for bit-identical floating point math. Does anyone have a feeling for how difficult (or impossible) it will be to achieve identical computation on OpenCL-compliant devices? For example, how difficult would it be to port, say, fdlibm, so that trancedentals use the exact same code? Any other show-stoppers that might not occur to the naive mind :-) ?
Thanks, Josh
Are you using the exact same code on all platforms for trancendentals and other floating-point functions? If not, that is going to be the greatest source of different results. They tend to vary widely on accuracy, since there is no 1/2 ULP accuracy requirement for those functions, unlike for the defined IEEE operations (+,-,*,/,sqrt)
-- Tim Olson