Closed johguenther closed 1 week ago
Will be fixed in v1.25.0.
Be aware that vminnm/vmaxnm
will not be generated by default for --arch=arm
. In addition you need to specify --cpu=cortex-a35
or newer.
fminnm/fmaxnm
will be generated by default when --arch=aarch64
is selected.
For min (analog for max) ISPC issues
fmin
orvmin
instructions, which are not IEEE 754 compliant: in case any operand is NaN then NaN is the result. I suggest to better choose the alternative instructionsfminnm
/vminnm
: those follow IEEE 754 and select the numerical value instead if NaN is one operand and are thus consistent to x86 behavior. OSPRay uses min/max to filter out NaNs.