lamphamsy

The purpose is to let compiler optimize/vectorize operations It's expected to obtain a perf comparable to that in SIMD mode

For RS FNT, the current decoding algorithm use FFT of length `2*k` that should be less than `q`. Hence, the acceptable `k` must be smaller than half of `q`. Source...

- [x] `Buffers` with a `meta` part - [x] No-SIMD working - [x] SIMD working

`Buffers` consists of a vector of `n` DATA buffers (array of T) and an OPTIONAL vector of `n` META buffers (array of `uint8_t`). A meta buffer corresponds to a data...

Merging SIMD parts

1

It merges SIMD parts together. It waits for - [x] #262

Try to refactor the `while` loops in `FecCode::decoding_packet` and `FecCode::encoding_packet`

- Branch: master - Mode: Debug or Release - SIMD: off I got following errors on MacOs: ```code /Scality/WorkingSpace/SourceCodes/IronMan/vr/ntl/src/simd/definitions.h:87:22: error: unknown type name '__m128' using RegisterType = __m128; ^ /Scality/WorkingSpace/SourceCodes/IronMan/vr/ntl/src/simd/definitions.h:88:18:...

It focuses on `FNT(q = 2^p + 1)`. Given a block of `n` symbols of `FNT(q)`, a metadata part should indicate indices of all symbols whose value = `2^p`. If...

Indices should be declared as `size_t`

1

instead of `int` or `unsigned` etc.

NF4 pack symbols and write back results directly into input data: https://github.com/scality/nttec/blob/master/src/fec_rs_nf4.h#L107 It should not modify input data as other FECs