Yask Versions Save

Add checks when applying NUMA MPI pinning.

A few small changes:

• Avoid some compiler warnings
• Use NUMA pinning when using Intel MPI

Major changes:

• Remove unmaintained GA tuner script.
• Fix issues with g++ compile.

Remove reliance on compiler-specific feature.

Most significant changes:

• Add -offload knob to yask.sh to simplify benchmarking offload kernels.
• Add detection of multiple GPUs to yask.sh.
• Remove buggy per-stage stats output.

Some small device-offload fixes.

Remove more warnings and simplify build process for non-Intel compilers.

Some small changes to allow building by more compilers, targeting more hardware:

• Remove some errors and warnings flagged by some compilers.
• Add more flexibility in Makefiles to allow easier specification of compilers and options.
• Add more information in the tutorial about making changes and generating code for more offload devices.

Main changes:

• Changed default temporal-blocking angles to be minimal instead of fold-multiples. New option -no-round_up_temporal_angles will restore old behavior.
• Added temporal blocking to 2D wave example code.

• Re-enable printing of option values in YASK compiler and kernel test utility.
• Change default value of early-load option for offloading.