[Quantlib-dev] new linear algebra library and memory management improvement

[<fra...@ya...>]

Hi all,
=20
I have carried out some code profiling on the new LFM implementation. (ratc=
het cap pricing using 3 factors)  Results showed that about 38% of the comp=
utation time was spend for dynamic memory allocation (the disposable trick =
saves some of them but not all of them, because a time step computation req=
uires 41 dynamic allocations !).  I have removed all these memory operation=
s by passing a reference toward the result instead of returning it eg: =20
=20
Existing code:  Disposable<Array> LiborForwardModelProcess::drift(Time t, c=
onst Array& x)   =20
My code:     void LiborForwardModelProcess::drift(Time t, const ublas::vect=
or<Real>& x, ublas::vector<Real>& result )
=20
This had improved speed by the expected extent with VC++ 2003 but only by 2=
2% with gcc.      The remaining time consuming operations being linear alge=
bra operations, I have also replaced the existing Matrix/Vector implementat=
ion by the boost uBlas library. This gave pretty good results, indeed, all =
operations but matrix*scalar operation are greatly accelerated. I removed t=
hese operations by embedding them in procedures as follows:
=20
Existing code:      lfmParam_->covariance(result, t, x);        result*=3Dd=
t;   =20
My code:      lfmParam_->covariance(result, t, x, dt);  // dt product is do=
ne inside
=20
Finally the overall gain of speed due to these changes is around 50% with b=
oth VC2003 and gcc.    I'm considering to implement these changes for the w=
hole QuantLib, but before doing so I would like to know your opinion on the=
 subject. I do not see any other way to improve memory management, but I'm =
hesitating between two solutions for the uBlas migration :
Embedding all uBlas code in the existing Array and Matrix classes. The main=
 advantage is that no other file would be altered. However it seems really =
tedious to expose all the nice features of uBlas using this architecture.
Replacing completely all Array and Matrix in the QuantLib code.
I tend to prefer the second solution because it is much neater in my opinio=
n even if it is not the simplest one in the short run. I'm aware that uBlas=
 is not the most efficient linear algebra library available, however profil=
ing shows that it is sufficient for the LFM case (sheer linera algebra oper=
ations accounts for a small part of the computation time). More complex ope=
rations can be performed using ATLAS through uBlas bindings.
Any suggestion/advice are more than  welcome.  =20
 Thanks for your attention,=20
best regards, =20
Fran=E7ois