Add dlasr

[btracey]

No description provided.

[btracey]

Better documentation and more checks coming.

I'm not sure how to test this other than to rewrite the code again. However, this code will be called by a bunch of routines so it should be effectively tested then.

[btracey]

Good news: Just added tests for Dlasr
Bad news: It exposed a mismatch between the condition number computation for Dgecon and Dpocon that's significant (0.021 to 0.024 while all of the others match to 1e-14)
Worse news: Cgo and native give the same wrong answer.

Suggestions? Log a bug with OpenBLAS, and comment out the tests I've added for that case. What about the dpocon script itself? It's generating random cases at present.

[kortschak]

Worse news: Cgo and native give the same wrong answer.

This is not good, but it's not terrible. At least the issue is consistent and code that work elsewhere works just as well here.

Suggestions? Log a bug with OpenBLAS, and comment out the tests I've added for that case.

I think so.

What about the dpocon script itself? It's generating random cases at present.

What do you mean?

[btracey]

The added case arises when generating the random test cases (at least with the random number stream on my machine). In order to pass the tests this case shouldn't be generated. I could, for example, make the loop shorter, but that dousn't feel like the right solution.

[kortschak]

Maybe log cases >1e-14 and fail case >1e-2?

[btracey]

Filed OpenMathLib/OpenBLAS#664

[btracey]

Filed http://icl.cs.utk.edu/lapack-forum/viewtopic.php?f=13&t=4817

[btracey]

Implemented suggestions.

[kortschak]

The docs in LAPACK are very informative. Could this be expanded?

[btracey]

Done.

[kortschak]

LGTM after test comment improved. I'll leave the doc comment up to you.

[kortschak]

Why is this here?