[petsc-dev] simplification of solvers that utilize KSPNASH STCG or GLTR

Munson, Todd tmunson at mcs.anl.gov
Sun Aug 21 23:33:58 CDT 2016


I don't see any reason not to make the change.

Todd.

> On Aug 21, 2016, at 9:40 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
> 
>   Todd,
> 
>    The TAO solvers that utilize KSPNASH STCG or GLTR have all this horribly redundant unneeded code 
> 
>      /* Solve the trust region subproblem */
>      if (NTR_KSP_NASH == tr->ksp_type) {
>        ierr = KSPNASHSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
>        ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
>        ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
>        tao->ksp_its+=its;
>        tao->ksp_tot_its+=its;
>        ierr = KSPNASHGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
>      } else if (NTR_KSP_STCG == tr->ksp_type) {
>        ierr = KSPSTCGSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
>        ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
>        ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
>        tao->ksp_its+=its;
>        tao->ksp_tot_its+=its;
>        ierr = KSPSTCGGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
>      } else { /* NTR_KSP_GLTR */
>        ierr = KSPGLTRSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
>        ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
>        ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
>        tao->ksp_its+=its;
>        tao->ksp_tot_its+=its;
>        ierr = KSPGLTRGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
>      }
> 
> This can be simplified a great deal by making the names 
> 
> KSPCGSetRadius(), KSPGGGetNormD() etc you only need to change the registration from 
> 
>  ierr = PetscObjectComposeFunction((PetscObject)ksp,"KSPNASHSetRadius_C",KSPNASHSetRadius_NASH);CHKERRQ(ierr);
>  ierr = PetscObjectComposeFunction((PetscObject)ksp,"KSPNASHGetNormD_C",KSPNASHGetNormD_NASH);CHKERRQ(ierr);
>  ierr = PetscObjectComposeFunction((PetscObject)ksp,"KSPNASHGetObjFcn_C",KSPNASHGetObjFcn_NASH);CHKERRQ(ierr);
> 
> to 
> 
> 
>  ierr = PetscObjectComposeFunction((PetscObject)ksp,"KSPCGSetRadius_C",KSPCGSetRadius_NASH);CHKERRQ(ierr);
>  ierr = PetscObjectComposeFunction((PetscObject)ksp,"KSPCGGetNormD_C",KSPCGGetNormD_NASH);CHKERRQ(ierr);
>  ierr = PetscObjectComposeFunction((PetscObject)ksp,"KSPCGGetObjFcn_C",KSPCGGetObjFcn_NASH);CHKERRQ(ierr);
> 
> Is there any reason we shouldn't do this? Essentially these three KSPs are a special subclass of KSPCG that supports these additional methods.  Anything specific to an individual one like the Lanczo stuff still remains with it.
> 
>  Barry
> 
> 
> 
> 




More information about the petsc-dev mailing list