mode = guimath.feround()
lastmode = guimath.feround(mode)
Set the floating point rounding mode. Mode 1, 2, 3, 4 refers to FE_DOWNWARD, FE_TONEAREST, FE_TOWARDZERO, FE_UPWARD respectively. The default, and most accurate, mode is FE_TONEAREST. The mode is changed only if the argument is 1-4. If there is no support for this function the return value is 0.
This function is useful to determine if a simulation depends unduly on double precision round-off error.
This affects calculations performed in both Python and HOC.
Example:
from neuron import h gm = h.GUIMath() print 'default rounding mode %d' % gm.feround() def test_round(mode): gm = h.GUIMath() old = gm.feround(mode) x = 0 for i in xrange(1, 1000001): x += 0.1 print 'round mode %d x=%25.17lf' % (mode, x) gm.feround(old) for i in xrange(1, 5): test_round(i)Output:
default rounding mode 2 round mode 1 x= 99999.99999613071850035 round mode 2 x= 100000.00000133288267534 round mode 3 x= 99999.99999613071850035 round mode 4 x= 100000.00000432481465396