1. xpow_optimizaton.patch: it uses https://github.com/intel/VML to improve the xpow functions for non complex data type. Test shows that the larger the array size, the more improvement it can get. When array size is 128, the performance of those updated functions can be increased more than 2 times.

2. bsxfun_pow_optimizaton.patch: it is on the basis of the above patch. Originally we didn't plan to improve these bsxfun_pow functions. But VML uses those vectorized math functions from libmvec of which the precision is lower than libm and this will lead bsxfun.cc-tst to fail. So we implement this patch. With this patch, when array size is 128, the performance of those updated functions can be increased more than 20%. The larger the array size, the more improvement it can get. However, when the array size is less than 16, the performance will drop ~10%. So if bsxfun.cc-tst's failure is OK, we recommend the first patch, as it has no side effect.