[Symmetry]

VLIWs tend to do well for programs with predictable-at-compile-time memory access patterns which tends to correlate with floating point heavy code.