Changeset [5ea812358c091a21ff35f15f5f89adf9c0c956c8] by Jed Brown

November 16th, 2008 @ 11:32 PM

Replace explicit passing of work arrays with VLA.

Start with plumbing for element mappings. The first implementation will just use nodal coordinates.

Also, replace some hairy indexing with VLA pointer types. External tests show that GCC produces better inner-loop code for

double (A)[N]; / A[i][j] in inner loop */

than

double A; / A[i*N+j] in inner loop */

In fact, the assembly produced at -O3 with the VLA-pointer seems roughly equivalent to hoisting

double restrict a = &A[i*N]; / a[j] in inner loop */

out of the innermost loop. GCC -Wcast-qual produces a warning for

const double a; const double (b)[2] = (const double (*)[2])a;

claiming that cast discards qualifiers from pointer target type' but it is not clear to me that this is actually the case.

Signed-off-by: Jed Brown jed@59A2.org http://github.com/jedbrown/dohp/...

Committed by Jed Brown

  • M CMakeLists.txt
  • M include/dohpjacobi.h
  • M include/dohptype.h
  • M include/private/jacimpl.h
  • M src/jacobi/impls/tensor/efstopo.c
  • M src/jacobi/impls/tensor/tensor.c
  • M src/jacobi/impls/tensor/tensor.h
  • M src/jacobi/interface/jacobi.c
  • M src/jacobi/tests/ex1.c
New-ticket Create new ticket

Create your profile

Help contribute to this project by taking a few moments to create your personal profile. Create your profile »

An implementation of the ``dual order hp'' version of the finite element method. This project targets parallel domain-decomposition methods for strongly coupled nonlinear problems with PDE constraints.