Optimize CRC intrinisics for targets lacking of CRC extension

[Jim Huang]

The existing functions _mm_crc32_u8, _mm_crc32_u16, and _mm_crc32_u32 are currently executed using generic C programming in the absence of CRC extensions. There is a potential implementation designed for CPUs which are equipped to handle PMULL (Polynomial Multiply Long) instructions but lack support for CRC32 instructions.

Reference:

• libdeflate/lib/arm/crc32_impl.h
• Fastest CRC32 for x86, Intel and AMD, + comprehensive derivation and discussion of various approaches
• Writing better code with help from the compiler