Open jeffsetter opened 5 years ago
Cool. I assume bfloat and 32 bit are sufficient for now?
Yes. BFloat and Float32 would be sufficient.
Btw, if you're interested in looking at the code that Halide uses to implement bfloat16, these two files that Andrew wrote are useful: https://github.com/StanfordAHA/Halide-to-Hardware/blob/master/src/Float16.cpp https://github.com/StanfordAHA/Halide-to-Hardware/blob/master/src/EmulateFloat16Math.cpp
Add support for all floating operators: add, mul, sub le, lt, ge, gt, neq, neq, neg, sqr, flr, ceil, abs, min, max div, rem log, exp, pow, sqrt sin, cos, tan, asin, acos, atan2, tanh
These are listed in order from most important to least important.