This is supported on x86 and target feature 
avx only.Expand description
Rounds packed single-precision (32-bit) floating point elements in a
according to the flag ROUNDING. The value of ROUNDING may be as follows:
- 0x00: Round to the nearest whole number.
- 0x01: Round down, toward negative infinity.
- 0x02: Round up, toward positive infinity.
- 0x03: Truncate the values.
For a complete list of options, check the LLVM docs.