Global DOUBLE_TO_FLOAT_CROSSOVER

to be benchmarked.

Global FFPACK::buildMatrix (const Field &F, typename Field::ConstElement_ptr E, typename Field::ConstElement_ptr C, const size_t lda, const size_t *B, const size_t *T, const size_t me, const size_t mc, const size_t lambda, const size_t mu)

is this :

Global FFPACK::Invert2 (const Field &F, const size_t M, typename Field::Element_ptr A, const size_t lda, typename Field::Element_ptr X, const size_t ldx, int &nullity)

not tested.

Parameters

	F	the computation domain
	M	order of the matrix
[in,out]	A	input matrix ( $M \times M$ ). On output, `A` is modified and represents a "psycological" factorisation `LU`.
	lda	leading dimension of A
[out]	X	this is the inverse of `A` if `A` is invertible (non `NULL` and $\mathtt{nullity} = 0$ ). It is untouched otherwise.
	ldx	leading dimension of `X`
	nullity	dimension of the kernel of `A`

Returns: pointer to $X = A^{-1}$