Must I use -use_fast_math to generate ptx files?

What I’ve found is yes. But I want to try whether avoiding fast_math helps improve accuracy.

If I don’t use -use_fast_math while generating .ptx files, I’ll have runtime error:
OptiX Error: Invalid value (Details: Function “RTresult _rtProgramCreateFromPTXFile(RTcontext, const char*, const char*, RTprogram_api**)” caught exception: defs/uses not defined for PTX instruction (Try --use_fast_math): madc.hi, [1310866])
(sample2.cxx:219)

If I use -ftz -prec-div -prec-sqrt -fmad instead, I got the same error as using nothing.
Even if I use -ftz=true -prec-div=false -prec-sqrt=false -fmad=true to get the same effect as -use_fast_math, I got the same error.

Any ideas?
Thanks!

Is that with OptiX 3.6.3?
If not, please try again with that version because there have been fixes for the madc.hi case.

See https://devtalk.nvidia.com/default/topic/779816/optix/optix-3-6-3-released/
“Bug fix Double-precision cosine and other transcendentals now work properly, although ray tracing internals are still single-precision.”