NEWS

6/24/2015 (Balint Joo/Thorsten Kurth) - Added lots of cleanup for KNL. Added packers for QDP-JIT (OCSRI layout for now)
1/12/2016 (Balint Joo/Aaron Walden) - Added AVX512 output from code generator. Tested on Intel SDE. Single and Double precision work, 16-bit precision is currently broken.

4/9/2015 (Balint Joo): Added Basic Multi-Shift solvers -- but can still add more optimizations. Need to test multi-node etc.
3/5/2015 (Balint Joo): Changed the interface to solvers, to accept isign parameter. For CG this lets one choose the operator either as M^\dagger M or M M^\dagger. This will allow CGNE vs CGNR implementations later. For BiCG it enables one to perform either the M or the M^\dagger solve without creating a new solver object, Also added checkerboard only spinor packers. Solvers which work only on one checkerboard can now just allocate cb-s worth of spinors in the calling wrappers.

2/4/2015 (Balint Joo, Diptorup Deb): Added BlueGene/Q intrinsics (QPX) and the ability to build without QDP++ (but still need QMP)

9/11/2014 (Balint Joo): Initial public release
9/25/2014 (Diptorup Deb, Balint Joo): 
	 *)  Added scalar code from qphix-codegen (FIXME: errant prefetches currently ignored via macro)
         *)  Added fixes to compile with GCC toolchain, compiles on BG/Q with BGClang (clang 3.6.0(trunk)) at ALCF:
             --enable-cean configur option: Enable Cilk Array Notation (enable for use with ICC and GCC-4.9, default: disabled )
             --enable-mm-malloc :  Enable _mm_malloc for allocating aligned memory (default enabled) 
             - clang doesn't like restrict keyword: Make this better with -Drestrict=__restrict__ in CXXFLAGS