lp:~florian-rathgeber/dolfin/gpu-backend

Branch merges

Propose for merging

No branches dependent on this one.

Related source package recipes

1 recipe using this branch. (?)

Create packaging recipe

Related snap packages

Related bugs

Related blueprints

5513. By Florian Rathgeber on 2013-06-03

Add CMake build script for debugging builds

5512. By Florian Rathgeber on 2011-02-03

CUDAKrylovSolver: compute convergence i.t.o. residual norm (instead of sqrt(<r,d>)) for better comparability to Cusp and PETSc solvers; bugfix: changed GRID_DIM to power of 2 (otherwise vecdot kernels fail)

5511. By Florian Rathgeber on 2011-02-02

added positive definite Helmholtz demo for both CPU and GPU; STATUS: Cusp GPU solver works, CUDA GPU solver produces wrong results for all but very small problems

5510. By Florian Rathgeber on 2011-02-02

CuspKrylovSolver: const arguments to solve functions (as in GenericLinearSolver)
moved parameter "fixed_iteration_number" to KrylovSolver in order to be able to set it from global parameters

5509. By Florian Rathgeber on 2011-01-27

CUDAKrylovSolver: convergence check uses residual instead of squared residual (squared residual might easily get below floating point precision and hence the check could become inaccurate)

5508. By Florian Rathgeber on 2011-01-27

added advection-diffusion GPU demos using CUDA backend with both assembled matrix and local-matrix-approach

5507. By Florian Rathgeber on 2011-01-27

refactored GPUSolver:
* renamed to CUDAKrylovSolver for consistency
* derives from GenericLinearSolver for compatibility with DefaultFactory, but does not implement full interface (set_operator etc.)
* inherits default_parameters from KrylovSolver
* residual is reset (r = b - Ax) every 50 iterations, added kernels to compute this for A both GPUMatrix and GPULocalMatrix
CuspKrylovSolver: output final residual after convergence
STATUS: CUDAKrylovSolver seems to produce wrong results

5506. By Florian Rathgeber on 2011-01-26

added check for CUDA-capable GPUs present and their compute capability and set code generation accordingly in FindCUDA.cmake

5505. By Florian Rathgeber on 2011-01-25

updated dolfin-config.cmake.in, seems to correctly pass compiler flags now

5504. By Florian Rathgeber on 2011-01-24

added GPUAssembler parameters to GlobalParameters, updated parameters in CuspKrylovSolver to be in sync with KrylovSolver