V. Rancurel

The threaded encoding/decoding functions shall be in the lib itself, and would be what people would use if they wanted to use the lib with the max performance without worrying...

The most mature compiler is clang/llvm

nttec.h now

Here is a simulation: https://github.com/vrancurel/f4mul

Lâm, can you give it a try ?

We can probably further optimize the loop we do to adjust with SIMD operations themselves, e.g. with cmpeq_epu16(), etc Bitfields operation are in fact unlikely so we can process them...

A lot of doxygen modules are missing: e.g. rest.c: REST based routines vfs: Virtual File System abstraction etc

For S3, can be implemented with POST. For sproxyd, can be implemented with GC service (code already there)

have to think if retry might be at application level.

we can imagine both: read-cache might be enhanced by checking version on the backing store. write-behind can be enabled if user is not bothered by asynchronous behavior.