Mark Gates

Changing lwork to be IN/OUT would be a nice solution originally, but it is not backwards compatible. The application would then have to know whether the LAPACK version was 3.10...

The problem seems to be if a user did something like this: ``` // query for lwork double dummy[1]; info = dgeqrf( m, n, ..., dummy, -1 ); // allocate...

Using a single `ar cr` is the right thing to do, rather than multiple `ar cr` updating the same file. In terms of correctness and speed, I advocate moving to...

Here's a take on implementing this in ScaLAPACK, so far just the BLACS library. If this seems like a good direction, can @weslleyspereira or someone do the other directories? Or...

Found this buried in my inbox. @ax3l Yes, BLAS++ and LAPACK++ have C++ macros BLASPP_VERSION and LAPACKPP_VERSION, and also runtime functions blaspp_version() and lapackpp_version(). ``` // Version 2022.05.00 #define LAPACKPP_VERSION...

As I recall, the idea was it would run the LAPACK++ CMake process _after_ building LAPACK. We can test it out.

> 1. We are using an old version of BLAS++ and LAPACK++ (2010.10.02). Is that intentional? 2020.10.02. You had me worried for a minute. I think that was just what...

`gemmt` was introduced in Intel MKL, circa 2015. In cuBLAS, similar functionality is called [`syrkx`](https://docs.nvidia.com/cuda/cublas/index.html#cublas-t-syrkx) and [`herkx`](https://docs.nvidia.com/cuda/cublas/index.html#cublas-t-herkx), which predated the MKL `gemmt` function (not sure exactly what version of CUDA...

Another use case is in the Polar Decomposition. The QDWH iteration converges to a unitary matrix U, and then computes a Hermitian matrix H = U^H A. Again, we have...

The triangular application is rather weird and probably rare. How often is A*B a triangular matrix? So I wouldn't name it based on triangular. Interpreting the output as Hermitian (or...