- 05 Oct, 2016 8 commits
-
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
PerfTest compiles all FLL implementations configurations and then allows to choosej one at run-time. The lookup phase was split between algorithm and kernel. Both lookups code, which was almost identical, has been factorized. It is also easier to add a new lookup phase by creating a parameter matcher.
-
Quentin Khan authored
-
Quentin Khan authored
The methods made a copy of the functor when called. This prevented retrieving the functor's internal state after the call. A uniform reference is now used to get a reference when possible and an rvalue reference otherwise.
-
- 03 Oct, 2016 1 commit
-
-
Quentin Khan authored
-
- 23 Sep, 2016 2 commits
-
-
Quentin Khan authored
-
Quentin Khan authored
-
- 22 Sep, 2016 14 commits
-
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
Add the symbolic data to P2M, P2L, M2P, L2P operators.
-
Quentin Khan authored
- Add multipole_t and local_expansion_t structures to split corresponding data in FUnifCell. - Add multipole and local_expansion attributes in FUnifCell. - Make FUnifKernel compatible with StarPU interface.
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
- The symbolic data is given to allow real kernels to find a cell's depth / box. - The P2P did not compute the correct indices for the U-list items
-
Quentin Khan authored
This is done to make constructors consistent accross the algorithms.
-
- 16 Sep, 2016 3 commits
-
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
- 14 Sep, 2016 10 commits
-
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
The Lagrange and Chebyshev kernels are disabled until they are made compatible with the StarPU algorithm.
-
Quentin Khan authored
-
Quentin Khan authored
- Add FMM operator overloads that take specific node symbolic data, multipole and local expansion instead of the whole node data. - Add a mutex to allow concurrent execution, this implies the explicit copy constructor/operator because the mutex cannot be copied/moved. - Add a fuse_results operation to gather the information from the kernel copies when a multi-threaded algorithms uses them (eg. tasks, StarPU). - Make the multipole and local expansion types explicit.
-
Quentin Khan authored
Group some of the node symbolic data (depth, morton index) in a symbolic_data_t object. This will allow easy manipulation.
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
Quentin Khan authored
-
- 08 Sep, 2016 1 commit
-
-
Berenger Bramas authored
-
- 05 Sep, 2016 1 commit
-
-
BLANCHARD Pierre authored
-