when is the gcn assembler coming?
opencl is driving me nuts again.
it thinks that precalculating 20 offsets into LDS in the outer loop is more important than staying
below 128 registers.
and i didnt even enable optimization.
please can i at least disable all "optimizations"?
