Fix blas mat-vec multiplication on array with only 1 nontrivial dimension

[sebasv]

First attempt at fixing #584

[sebasv]

The failing tests are behavior intended by this fix, needs discussion if this is desired behavior.

[jturner314]

I'm sorry that it's taken me so long to review this PR. Thanks for your investigative work on this.

The only change I'd like to make is to add a comment explaining why those .max() calls are necessary.

[jturner314]

It would be nice to add a comment, e.g.

                        // Determine stride between rows or columns. Note that the stride is
                        // adjusted to at least `k` or `m` to handle the case of a matrix with a
                        // trivial (length 1) dimension, since the stride for the trivial dimension
                        // may be arbitrary.

[bluss]

Using the maintainer push to PR to update.

I'll add your comment @jturner314 jturner314 but not in that way, because it causes a textual conflict.

[bluss]

@jturner314 @sebasv do you remember what the failing test comment was about? I guess it's obsolete now, because the tests pass and this looks ready to merge to me.

[jturner314]

@jturner314 @sebasv do you remember what the failing test comment was about? I guess it's obsolete now, because the tests pass and this looks ready to merge to me.

I'm not sure. Maybe it's referring to the mat_vec_product_1d test added by the PR? (That would be surprising because the test was added after @sebasv made that comment, but maybe he just hadn't pushed the test yet?) That test fails before the fix in this PR.

[bluss]

ok must be obsolete, thanks :)

Thanks so much @sebasv for reporting and fixing! (It's me who renamed the test and the PR to avoid using the term "dot product" here.)

[sebasv]

Just for completeness, if I recall correctly there were two tests not related to this PR which failed because they called a dot product on a transposed array with a trivial dimension, which messed up the strides. If no tests fail now, this was probably resolved with another bugfix. @bluss @jturner314 thank you for your reviews and for merging the fix!