DYNAMIC_ARCH: Improve mapping for future AMD cpus

POWER10: Changing store instructions for Level1 functions

This patch changes 32 bytes stores to two 16 bytes stores
to fix a recent degradation due to 32 bytes stores.

Initial attempt at proper cpu detection on RISCV

Fix riscv64 detect

Automatically downgrade C910V to RISCV64_GENERIC if the compiler lacks vector support

Fix CMAKE generator rules for ?laswp_ncopy and ?neg_tcopy kernels

Makefile fixes related to C_LAPACK, plus Travis CI fixes

Signed-off-by: Han Gao <gaohan@uniontech.com>

When CORE is empty, use -march=loongson3a. Fix it.

Signed-off-by: Han Gao <gaohan@uniontech.com>

build: minor fixes to build on windows with make

This patch contains following fixes

1. Fix to build without PIC flag
2. Define LAPACK_COMPLEX_STRUCTURE for windows. Builds are failing
without it and changes are consistent with the CMake rules defined
in system.cmake (line 576)

* Fix f2c-like support functions to compile with MSVC, and
re-enable C_LAPACK for MSVC in CMAKE

* Add MSVC&flang build to Azure CI in order to check C_LAPACK correctness

Fix undefined PREFETCHSIZEs in PPC440 GEMV kernels

Adapt commands for tests with GNU make.

* Add C equivalents of the Fortran routines from Reference-LAPACK as fallbacks, and C_LAPACK variable to trigger their use

Consolidate actions on GitHub runners.

Fix missing braces from previous commit (PR3600)

Re-organize build matrix for Ubuntu and MacOS runners.
Don't start runners that don't do anything.
Run tests.

* limit flang optimizations to -O2

Fix build targets in Makefile.prebuild

* Prevent addition of a BOM to test input (which would distort the names of output files)