NIR: Optimize some cases of shifts and immediates
The idea for the first patch occurred to me while I was looking at some shaders that were hurt but a different patch series. The idea for the second patch occurred to me while I was looking at a shader that was helped by the first patch.
There was a third patch that tried to optimize cases like (#a << b) << #c
, but that didn't affect anything in shader-db on any Intel platforms.