David Majnemer cc13f3ba45 Correctly round FP -> BF16 when SDAG expands such nodes (#82399) ...

We did something pretty naive:
- round FP64 -> BF16 by first rounding to FP32
- skip FP32 -> BF16 rounding entirely
- taking the top 16 bits of a FP32 which will turn some NaNs into
infinities

Let's do this in a more principled way by rounding types with more
precision than FP32 to FP32 using round-inexact-to-odd which will negate
double rounding issues.

2024-02-21 12:37:02 -05:00

67 KiB

Raw Blame History

View Raw