[query] Fixed bug where hl.nd.qr didn't support 0 elements

[johnc1231]

I'm not sure this is the right fix. The segfault was coming from dgeqrf itself, and it was happening only when N is 0 (M being 0 was fine). The real issue is that when N is 0, LAPACK computes LWORK to be 0, meaning we don't actually allocate any memory for the WORK array and we segfault. To contrast, if I do a workspace size query for shape (0, 10), I get 320 for LWORK. This also seems wrong though, as you shouldn't really need any work space for an empty matrix?

Anyway, the segfault preventing fix was just to make sure LWORK was at least 1. I was hoping you (Patrick) could use Julia interface to play with the LAPACK functions and see what behavior you notice, and if it's consistent with the above.

[patrick-schultz]

LAPACK/MKL documentation says WORK must be an array of length max(1, LWORK). I think the "official" way to fix this is to leave LWORK as whatever the workspace query returns, and allocate the WORK arrray of size max(1, LWORK).

[patrick-schultz]

That's weird about the M=0 case. The LAPACK version (source here) returns LWORK=0 if either M or N is 0. But it doesn't seem worth complicating things over.

[johnc1231]

Ah ok, I'll just change the allocation and leave LWORK alone.