there is no more requirement for ARMv6 instructions,
and VFP on ARMv5 is uncommon

to be used on armv5 and new platforms

otherwise some macros like YIELDING are not defined correctly

- use on arm, arm64 and any new platform
- use faster integer math instead of double
- use similar scale as rdtsc so that timeouts work

should fix issue #597

- was writing 0 to lock variable, so was ineffective
- only exit loop if both lock was 0 and strex was successful

install OpenBLASConfigVersion.cmake

Add POWER7/POWER8 as targets

ppc64le platform support (ELF ABI v2)

make TARGET=GENERIC

?gemm3m is unimplemented on generic target.

Fix test execution when USE_OPENMP=0

optimizations for haswell

The standard way to disable OpenMP support is to set USE_OPENMP=0,
as indicated by other checks to see if USE_OPENMP equals 1. The
problem is obviously then that `ifdef USE_OPENMP` is very much not
what we want to test for. This causes tests to fail when no OpenMP
library is installed.

added benchmark scripts

small modification of gemm.c

added gesv benchmark

bugfix for benchmark Makefile on MAC

bugfixes, to build benchmarks with mingw on Windows OS

bugfix for arm locking

added blas level1 swap  benchmark

e.g.
make HOSTCC=gcc CC=arm-linux-androideabi-gcc NO_LAPACK=1 TARGET=ARMV7

In Android NDK, it uses armeabi-v7a-hard ABI.
TARGET_CFLAGS += -mhard-float -D_NDK_MATH_NO_SOFTFP=1
TARGET_LDFLAGS += -Wl,--no-warn-mismatch -lm_hard
For more information, please check hard-float example at
android_ndk/tests/device/hard-float/jni/.

added blas level1 copy benchmark

Bugfix for armv6 memory barrier

2) If OpenBLAS cannot find Fortran compiler, it will only build BLAS
(without LAPACK).

added optimized cscal and zscal functions for steamroller