Ensure that the right xerbla gets included in OSX DYNAMIC_ARCH builds

Fixed MSA enabled optimization on Loongson-3A4000

loongarch64: Optimize dgemm_kernel

Merge back from release branch to copy 0.3.19 tag

Update from develop branch for 0.3.19 release

Update Changelog for 0.3.19 release

Update -tp option for recent nvfortran on x86_64

Add feature-based fallback for unknown x86_64 cpus

Enable delayed (re)init of Windows threads beyond Cygwin

fix bug in zscal function

Makefile: also consider -O, -Og and -Os when stripping flags

Fix bracing in cpuid_mips64.c

memset can not be used in zscal because of
the stride parameters.

Signed-off-by: Wu Zhigang <zhigang.wu@starfivetech.com>

gcc also supports -O, -Og and -Os as optimization flags.
They may be given on the make command-line by users.

For the calculation of LAPACK_NOOPT, all such flags should be considered.

Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
[Retrieved from:
https://git.buildroot.net/buildroot/tree/package/openblas/0003-Makefile-also-consider-Os-when-determining-LAPACK_NO.patch]
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>

* Move accesses to the overflow flag into the scope of the blas lock

optimize cgemm on ARM cortex A53 & cortex A55

Add CMake support for Power

Add sgemm part for Arm SVE

Fallback for Power kernels

Roll back SkylakeX DGEMM kernels to 4x8 when compiling for DYNAMIC_ARCH

* Adjust ar commad in ARMV7 Darwin NDK build after homebrew update to NDK 23b

Fix hardcoded library name in cpp_thread_test Makefile