4-operation form is preferred over 3-operation because it breaks a long
dependency chain, thus allowing a superscalar processor to execute more
operations in parallel.
The idea was taken from: http://www.zorinaq.com/papers/md5-amd64.html

AMD Athlon(tm) II X3 450 Processor, x86_64

$ for i in $(seq 1 4); do ./avutil_md5_test2; done
size: 1048576  runs: 1024  time:    5.821 +- 0.019
size: 1048576  runs: 1024  time:    5.822 +- 0.019
size: 1048576  runs: 1024  time:    5.841 +- 0.018
size: 1048576  runs: 1024  time:    5.821 +- 0.018

$ for i in $(seq 1 4); do ./avutil_md5_test2; done
size: 1048576  runs: 1024  time:    5.646 +- 0.019
size: 1048576  runs: 1024  time:    5.646 +- 0.018
size: 1048576  runs: 1024  time:    5.642 +- 0.019
size: 1048576  runs: 1024  time:    5.641 +- 0.019
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>