Implement XSAVE/XRSTOR for AVX (state components 0, 1 and 2)
Refactor existing FXSAVE / FXRSTOR implementation so as to use
the new code, since these are sub-cases of the general XSAVE/XRSTOR
functionality.
Add a new CPUID level to indicate CPUs which are AVX2 compatible,
and enable it by default on AVX2 compatible hosts.
For both the AVX and AVX2 simulated CPUIDs, claim that XSAVEOPT is not
supported, in an attempt to avoid having to implement it.
Remove CPUID kludgery to do with OSX 10.10 (Yosemite) in order to
persuade it not to use XSAVE/XRSTOR.
libvex_ir.h: add new guarded load conversion "ILGop_IdentV128"
as required by XSAVE/XRSTOR support.
git-svn-id: svn://svn.valgrind.org/vex/trunk@3169 8f6e269a-dfd6-0310-a8e1-e2731360e62c
9 files changed