Use another implementation for PDEP

[yanjiew1]

This approach is from chapter 7 of Hacker's Delight. It eliminates the need of POPCNT and BZHI instructions totally, enhancing portability.

GCC generates less instructions compared to the original implementation.

[yanjiew1]

The performance can be slower when compiled with "-O3" flag because the new implementation cannot take advantage of the LEA instruction.
If we compile this code with "-O3", GCC may inline the function and then optimize the code in a way that the shifts of the ppp are computed once for each mask. Thus, the number of instructions may not be reduced when "-O3" is used.