rebase

Correct INFO value returned by SLASQ2/DLASQ2 (Reference-LAPACK 506)

Fix division by zero in the non-x86 codepath of C/ZROTG

Add LAPACKE fixes from Reference-LAPACK PR 537

Apply fixes from Reference-LAPACK PR468 and 539 for array declarations in ?ORGBR/?UNGBR

Apply lapack-testing fix from Reference-LAPACK PR536

fixes changing back from a single OMP thread for error exit testing to the originally requested number of threads for computational tests

Apply lapack-testing fixes from Reference-LAPACK PR538

Delay creation of the (soft)link until after the library has been built

Replace spurious AVX512 requirement in the Haswell drot microkernel with an AVX2/FMA3 guard

* Add mixed clang/ifort build on OSX to the Azure CI config based on https://github.com/oneapi-src/oneapi-ci
(and remove debugging tools from the clang+gfortran job)

* Remove extraneous libgfortran dependency of ifort builds

* remove FEXTRALIB from link line of shared library as ifort keeps track of dependencies (and they are different for a .dylib than what f_check got for an executable)

rebase

Fix obscure ctest crashes on OSX and add OSX builds to Azure CI

rebase

Restore __volatile__ keyword in ARM64 DYNAMIC_ARCH detection mechanism

POWER10: Improve dgemm performance

This patch uses vector pair pointer for input load operation
which helps to generate power10 lxvp instructions.

POWER10: Optimized zgemv

This patch makes use of Matrix-Multiply Assist (MMA)
feature introduced in POWER ISA v3.1 for zgemv_n and zgemv_t.

Fix unwanted fallback to implicit typing in slanv2/dlanv2

Use "old" compute(24) function with clang due to register limitations

Pass NO_AVX512 macro def when `DYNAMIC_ARCH` is enabled

Fix spillover of host-specific build flags into the shared part of x86 DYNAMIC_ARCH builds

Copy missing fixes from the final revision of Reference-LAPACK PR477

for #3139

(copies fixes made in Reference-LAPACK PR 477 after the initial cherrypick)

rebase

Remove Unnecessary/Erroneous Adds/Reads In sgemm_tcopy_16.S COPY1x8 Macro

POWER10:  Adding check for little endian

This patch makes sure that recent POWER10 patches are used
only for little endian.

There appears to have been some code leak when copying from the COPY2x8
macro above where we're reading 8 bytes into d4-d7 directly after
reading 4 bytes into s4-s7. These 32 bytes in d4-7 are unused and can
possibly overrun the boundary of allocated memory -- Valgrind detected
this which is what dragged my attention to it for a 128,1 copy.

Additionally, there is no need to update the addresses stored in A0-A7
as the only possible paths after running this macro will overwrite A0-7
if looping to the next 8 rows, or overwrite A0-3 if moving to 4 rows --
in which case A4-7 are unused.

Fix compilation of the benchmarks on older OSX versions