Pascal Giorgi
Pascal Giorgi
A lot of work remains to do : - use sparse matrix operation from FFLAS or LinBox - use generic field instead of hardcoded ones (beware of rand() use) -...
in SIMD code for FFT with floating point values, we can gain one multiplication by precomputing the W^i/p as in the integral case.
Working on #159, I found a weird thing about the Compose blackbox in blackbox/compose.h file. There is a specialisation of Compose when the two composed blackboxes share the same type....
The code for online order basis (OPMBasis) is hanging forever when added in the order basis test suite (test-order-basis). The code need to be checked.
The fgemm checkers are designed to check FFLAS matrix multiplication using the Freivalds technique : to check C=AB, one can choose a vector v at random and check Cv =...
Test for fgemm checkers are done only with modular field. All fields must be tested, especially the ones using floating point representation. This is related to #93.
There is a generalised issue with the call to BLAS functions. All FFLAS-FFPACK functions use size_t as the primary type for dimensions (i.e. vector, matrix) while BLAS uses int. Therefore,...
This PR enables new feature for RNS double conversions. In particular, the rns_double_extended structure allows now to choose moduli that can go beyond the 26 bits limit of rns_double. This...