This would allow mixing armel and softvfp ABIs, but not hard float ABIs, at least across compilation unit boundaries (that said, GCC never seems to optimize ABI

This would allow mixing armel and softvfp ABIs, but not hard float ABIs, at least across compilation unit boundaries (that said, GCC never seems to optimize ABI bottlenecks within a compilation unit anyway)