| Commit message (Collapse) | Author | Age |
| |
|
| |
|
|
|
|
| |
to keep accuracy of smallest eigenvalues.
|
|
|
|
| |
zero coefficients.
|
| |
|
|
|
|
| |
install(DIRECTORY ...) command.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
products, and enable +=/-= assignement for such products.
This changeset also improves the performance by working on column of the result at once.
|
| |
|
| |
|
| |
|
|
|
|
| |
sizes, remove numerous useless temporaries.
|
|
|
|
| |
in GeneralizedEigenSolver
|
| |
|
| |
|
| |
|
|
|
|
| |
vectorized path when FMA is available.
|
|
|
|
| |
with array::tanh, enable fast tanh in fast-math mode only.
|
| |
|
| |
|
|
|
|
| |
variant for very old compilers (enum, template<typename C> for both function definitions) fixes the problem.
|
| |
|
| |
|
|
|
|
| |
corresponds to clang 3.7)
|
| |
|
| |
|
|
|
|
| |
Also moved assertion to finished() method and adapted unit-test
|
| |
|
|\
| |
| |
| |
| |
| | |
request PR-215)
Fix_tanh_inconsistent_for_tensorflow
|
| |
| |
| |
| |
| |
| | |
bug #1242).
The unit test tests all combinations of 2x2 block-sizes from 0 to 3.
|
|/ |
|
| |
|
| |
|
|
|
|
| |
functions to the Eigen::half_impl namespace together with ADL
|
| |
|
| |
|
| |
|
|
|
|
| |
allowed. For instance, this permits to vectorize 5x5 matrices (including product)
|
|
|
|
| |
logic decides on the sizes that are OK for vectorization.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|