* bs_transpose_dst only
** original
==> test.ccomp.host.out <==
cycles: 3080223

==> test.ccomp.k1c.out <==
cycles: 10145951

==> test.gcc.host.out <==
cycles: 1485887

==> test.gcc.k1c.out <==
cycles: 4078535

** neg and
==> test.ccomp.host.out <==
cycles: 2905049

==> test.ccomp.k1c.out <==
cycles: 7995063

==> test.gcc.host.out <==
cycles: 1858263

==> test.gcc.k1c.out <==
cycles: 5255763

** cmove mais mauvais scheduling de registres
==> test.ccomp.host.out <==
cycles: 4363682

==> test.ccomp.k1c.out <==
cycles: 7208629

==> test.gcc.host.out <==
cycles: 2916854

==> test.gcc.k1c.out <==
cycles: 5646730

** cmove via match du and
==> test.ccomp.host.out <==
cycles: 2553732

==> test.ccomp.k1c.out <==
cycles: 7208629

==> test.gcc.host.out <==
cycles: 1849125

==> test.gcc.k1c.out <==
cycles: 5255763

** hand optimized loads
cycles: 6027072

* both bs_transpose_dst and bs_transpose_rev
** with both cmove
6890902