Use KERNEL_DEFINITIONS rather than COMMON_OPTS to pass -march=skylake…

Add -march=skylake-avx512 to CFLAGS when the target is Skylake

Revert "Add -march=skylake-avx512 when required"

Should fix 1806 and #1801

Use avx512 workaround with msys2/mingw64 as well

Add a C+intrinsics version of the SGEMM/skylakex kernel

for most sizes this is 1.2x to 1.4x faster than the current code

Update common_mips64.h for the 1st loop of blas_memory_alloc

Improve CMake help output and add SYMBOLPREFIX and -SUFFIX options

Handle conflicting usage of ARCH in at least some BSD environments

Add optimized *copy versions for skylakex

fixes #1796

in case a conflicting setting of ARCH (for architecture) gets pulled in from the environment
(originally suggested by dloghin in #1753)

Add -march=skylake-avx512 when required

fixes #1797

very minor gains since it's not super hot code, but general principles

fix parallel build issues with APFS/HFS+/ext2/3 in netlib-lapack

the compare is more costly than the work

Add optimized n/t copy versions for skylakex; in the patch the
tcopy is also rewritten using intrinsics; the ncopy file
will be worked on in a future commit

merge develop

The problem is that OpenBLAS sets the LAPACKE_LIB and the TMGLIB to the
same object and uses the `ar` feature to update the archive file. If the
underlying filesystem does not have sub-second timestamp resolution and
the system is fast enough (or `ccache` is used), the timestamp of the
builds which should be added to the previously generated archive is the
same as the archive file itself and therefore `make` does not update the
archive.

Since OpenBLAS takes care to not run the different targets updating the
archive in parallel, the easiest solution is to declare the respective
targets `.PHONY`, forcing `make` to always update them.

fixes #1682

update travis alpine chroot with avx512 intrinsics headers

skylake dgemm: Add a 16x8 kernel

Minor gains for small matrixes, but at 512x512 and above the gain
gets more significant.

The next step for the avx512 dgemm code is adding a 16x8 kernel.
In the 8x8 kernel, each FMA has a matching load (the broadcast);
in the 16x8 kernel we can reuse this load for 2 FMAs, which
in turn reduces pressure on the load ports of the CPU and gives
a nice performance boost (in the 25% range).

address #1782 2nd loop

Create a AVX512 enabled version of DGEMM

Fix unknown type name __WAIT_STATUS on RHEL5

With glibc 2.5 one must have #define _XOPEN_SOURCE >= 500 to use wait.
But reading glibc code this is actually needed only if stdlib.h was
included before sys/wait.h. This was the case here through
openblas_utest.h. So changing include fix compilation on RHEL5 and
should ne hurt with more recent distro.

* Problem found when using with gcc 5.5 and 4.7.2 on RHEL5/CENTOS5
* Fix #1519

Check for Immintrin.h presence in the AVX512 compatibility test as well

This patch adds dgemm_kernel_4x8_skylakex.c which is
* dgemm_kernel_4x8_haswell.s converted to C + intrinsics
* 8x8 support added
* 8x8 kernel implemented using AVX512

Performance is a work in progress, but already shows a 10% - 20%
increase for a wide range of matrix sizes.

Convert fldmia/fstmia instructions to UAL syntax for clang7

second part of fix for #1774, containing files missed in #1775