The reason was that the invocation of the cpuid_tool through the makefile
was using LD_LIBRARY_PATH=. to force link&use of the latest-build libcpuid
library. This doesn't seem to work, so using LD_PRELOAD to explicitly load
libcpuid.so into cpuid_tool.
The committed approach of course doesn't work on Mac OS X, where
make test-old should be used.
The reason for the apparent slowness was that the "cpuid_tool" that
was being called from run_tests.py is actually a shell script - a
wrapper, installed by libtool. The script was heavy enough to cause
substantial overhead. Just bypassing it (by using the real
cpuid_tool binary) reduced "make test" running time from 4.1 to 0.7s
on my machine.
It is a bit hacky, though, so "make test-old" is retained, which
uses the old invocation.
The last change to flags detection caused a bunch of tests to fail.
The reason is that they are bogus, all recent Intel chips don't have
RDTSCP indicated in the test files, whereas they have it in reality.
I figured it will be easier to add "--fix" option to run_tests.py,
rather than fixing each testfile by hand.
This is also extended in the Makefile:
"make test" runs the tests and reports discrepancies.
"make fix-tests" fixes any offending tests. This blindly assumes that
libcpuid is sane.
The .am files contained obsolete data (to include libcpuid.sln/libcpuid.vcproj
in a dist). These files are now in libcpuid_vc71.* and libcpuid_vc9.* flavours,
thus this change. "make dist" now works.
