Attempt to optimize RLNC Recoder Matrix Multiplication

[itzmeanjan]

The matrix multiplication in RLNC Recoder, which multiples a random sampled coding vector with a matrix, made of coding coefficients of received erasure-coded pieces, can be optimized to benefit from cache locality, by keeping the matrix in transposed form.

rlnc/src/full/recoder.rs

Lines 130 to 143 in 4e6cc41

	// Compute the resulting coding vector for the original source pieces
	// by multiplying the random sampled recoding vector by the matrix of received coding vectors.
	let computed_coding_vector = (0..self.num_pieces_coded_together)
	.map(|coeff_idx| {
	random_recoding_vector
	.iter()
	.enumerate()
	.fold(Gf256::default(), |acc, (recoding_vec_idx, &cur)| {
	let row_begins_at = recoding_vec_idx * self.num_pieces_coded_together;
	acc + Gf256::new(cur) * self.coding_vectors[row_begins_at + coeff_idx]
	})
	.get()
	})
	.collect::<Vec<u8>>();