For example, in the final test from trigonometry.shader_test, we have:
sincos r1.x <v4:float>, null <float>, l(3.00000000e+01) <s:float>
which currently gets turned into
mov sr0.xyzw <v4:float>, l(3.00000000e+01) <s:float>
sin r1.x <v4:float>, sr0.xyzw <v4:float>
That doesn't seem ideal, and should possibly be rejected by the
validator, but ends up working out well enough for the SPIR-V backend.
It's a bit more problematic to deal with for the GLSL backend.
043f6f21b0