Closed k0ekk0ek closed 4 months ago
disable vectorized parsers
Note that standard libraries use vectorized routines all the time (memcpy, memcmp, are all vectorized).
I am aware. It's not recommended behavior, but some NSD users may not be interested in maximum performance and instead prefer having just one parser kernel for simplicity(?) This is merely to accommodate their specific use case. I've opted to enable all kernels by default. In practice, I suspect very few people will actively disable the better performing kernels.
I've had a comment from an NSD user that maybe users would like to disable vectorized parsers. Add flags to make and CMake based builds to explicitly enable/disable westmere and haswell based parsing.