| 1 |
This program tests TSQR::SequentialTsqr, which implements the |
| 2 |
sequential cache-blocked version of TSQR. TSQR (Tall Skinny QR) |
| 3 |
computes the QR factorization of a tall and skinny matrix (with many |
| 4 |
more rows than columns) distributed in a block row layout across one |
| 5 |
or more processes. SequentialTsqr is built on the computational |
| 6 |
kernels implemented in TSQR::Combine. These tests use the default |
| 7 |
back-end implementation of TSQR::Combine (which you can see as the |
| 8 |
default template parameter value in Tsqr_Combine.hpp). |
| 9 |
|
| 10 |
By default, TSQR::SequentialTsqr will only be tested with real |
| 11 |
arithmetic Scalar types (currently, float and double, corresponding to |
| 12 |
LAPACK's "S" resp. "D" data types). If you build Trilinos with |
| 13 |
complex arithmetic support (Teuchos_ENABLE_COMPLEX), and invoke this |
| 14 |
program with the "--complex" option, complex arithmetic Scalar types |
| 15 |
will also be tested (currently, std::complex<float> and |
| 16 |
std::complex<double>, corresponding to LAPACK's "C" resp. "Z" data |
| 17 |
types). |
| 18 |
|
| 19 |
This program can test accuracy ("--verify") or performance |
| 20 |
("--benchmark"). For accuracy tests, it computes both the |
| 21 |
orthogonality $\| I - Q^* Q \|_F$ and the residual $\| A - Q R \|_F$. |
| 22 |
For performance tests, it repeats the test with the same data for a |
| 23 |
number of trials (specified by the "--ntrials=<n>" command-line |
| 24 |
option). The typical use case for SequentialTsqr is data sets too |
| 25 |
large to fit in cache; it will work otherwise, and efficiently too, |
| 26 |
but in that case you would be better off just testing TSQR::Combine. |