[BUG] About the rounding mode of TF32

[beginlner]

https://github.com/NVIDIA/cutlass/blob/main/include/cutlass/tfloat32.h#L80 Why does the code represent "round toward nearest even", but not "round to nearest, ties away from zero"?

[hwu36]

I don't fully understand your questions. Exactly as this rounding method says, it "round_half_ulp_truncate", so it is not "to the nearest even"

we do this because it only takes two instructions to do the conversion.

[github-actions[bot]]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[mnicely]

Closing due to inactivity