17:46 KungFuJesus: Alright, so it's in both the fragment and the vertex shader parsers. It seems that the saturation bit fields manage to get mangled with store merging in GCC. I'm seeing it in at least the dot product 2 operand instruction but I have to dig a bit further to see exactly why this is. That instruction emits a constant "0" for the saturation flag
17:58 KungFuJesus: yeah wow, it fixes it if I use a variable that is set to 0 before the giant switch/case loop at the very top
17:58 KungFuJesus: this is interesting, it could be a GCC bug with big endian