ghash-x86_64.pl: minimize stack frame usage.
ghash-x86.pl: modulo-scheduling MMX loop in respect to input vector
results in up to 10% performance improvement.