Misc CL fixes
This series fixes a few issues getting us further in the CL CTS:
- When
vec8
/vec16
instructions were added,nir_lower_bit_size
wasn't updated to be able to support lowering them. - Fix a bad rebase around CL kernel sampling.
- Remove undefs before converting to DXIL. DXIL ALU ops don't support undefs, so just flush them to 0s.
- Add a test which validates the prior change.
The vec8
/vec16
change I'll probably push upstream immediately.
Edited by Jesse Natalie