use new ISA extensions to improve instrumentation

_From [bruen...@google.com](https://code.google.com/u/109494838902877177630/) on November 08, 2013 17:20:27_

*\* TODO OP_bextr from AMD TPM, or OP_pext from Intel BMI2, could be useful for Dr. Memory

They extract a contiguous set of bits, which would simplify Dr. Memory when
using 2-bit shadow mappings.

*\* TODO flags-less arithmetic instrs

BMI2 added shifts and multiplies that do not touch eflags.

drmem table lookup is "shr;shr;add w/ scale;movzx;cmp".

Could use lea for add, and flags-less shifts => don't need to save flags?
Though have other cmps: heap ignore (goes away w/ -replace_malloc), whether
regs defined.

*\* TODO AMD XOP adds comparisons that don't touch the flags: OP_vpcomd, etc.

But we'd have to move data in and out of xmm/ymm registers, and spill them.

_Original issue: http://code.google.com/p/drmemory/issues/detail?id=1376_