Using this trick, the dependency in hrp2 is removed from the
jrl-walkgen.pc file.

Elie Moussy authored 11 years ago and olivier stasse committed 11 years ago

Conflicts:

	src/PreviewControl/PreviewControl.cpp
	src/ZMPRefTrajectoryGeneration/OnlineStepPositionTrajectoryGeneration.cpp
	src/ZMPRefTrajectoryGeneration/OrientationsPreview.cpp
	src/ZMPRefTrajectoryGeneration/ZMPVelocityReferencedQP.cpp

This test covers more widely some functionnalities of the Herdt PG.
The results should be checked again. Some asymmetries arise and
are too important to be explained only by the constraints.

The fix proposed by http://gcc.gnu.org/bugzilla/show_bug.cgi?id=323
is not working when recompiling jrl-walkgen and jrl-dynamic
with -msse and -mfpmath=sse.

However the two other algorithms do not suffer from the transition
between x86_32 and x86_64. Thus it might still come from a x87 problem
(aka an intermediate high resolution representation in CPU which is
lost after memory transfer) specific to Morisawa.

The trajectories are however very similar between each other. Thus
this differenciation is to maintain a working unitary test.

Morisawa 2007 is failing on 12.04 x86_64,
and working on 10.04 x86_32.
For the latter reason, the test is still not activated.

With the use of spatial algebra in JRL-dynamics ZMP needs
backward dynamics computation for proper evaluation.
The internal setting of the WPG used an approximation of
the momentum derivatives previously implemented in JRL-dynamics.
This commit fix this problem and the references for the unitary
test.

This should be also validate with the simple-humanoid and not
HRP-2.