void write_matrix(Givaro::Integer p, size_t m, size_t n, T* C, size_t ldc){
size_t www=(p.bitsize()*log(2.))/log(10.);
for (size_t i=0;i<m;++i){
cout<<"[ ";
cout.width(www+1);
cout<<std::right<<C[i*ldc];
for (size_t j=1;j<n;++j){
cout<<" ";
cout.width(www);
cout<<std::right<<C[i*ldc+j];
}
cout<<"]"<<endl;
}
cout<<endl;
}
int tmain(){
srand( (int)seed);
srand48(seed);
Givaro::Integer::seeding(seed);
typedef Givaro::Modular<Ints> Field;
Givaro::Integer p;
FFLAS::Timer chrono, TimFreivalds;
double time=0.,timev=0.;
#ifdef BENCH_FLINT
double timeFlint=0.;
#endif
for (size_t loop=0;loop<iters;loop++){
Givaro::Integer::random_exact_2exp(p, b);
Givaro::IntPrimeDom IPD;
IPD.nextprimein(p);
Ints ip; Givaro::Caster<Ints,Givaro::Integer>(ip,p);
Givaro::Caster<Givaro::Integer,Ints>(p,ip); // to check consistency
Field F(ip);
size_t lda,ldb,ldc;
lda=k;
ldb=n;
ldc=n;
typename Field::RandIter Rand(F,seed);
typename Field::Element_ptr A,B,C;
A= FFLAS::fflas_new(F,m,lda);
B= FFLAS::fflas_new(F,k,ldb);
C= FFLAS::fflas_new(F,m,ldc);
// for (size_t i=0;i<m;++i)
// for (size_t j=0;j<k;++j)
// Rand.random(A[i*lda+j]);
// for (size_t i=0;i<k;++i)
// for (size_t j=0;j<n;++j)
// Rand.random(B[i*ldb+j]);
// for (size_t i=0;i<m;++i)
// for (size_t j=0;j<n;++j)
// Rand.random(C[i*ldc+j]);
PAR_BLOCK { FFLAS::pfrand(F,Rand, m,k,A,m/size_t(MAX_THREADS)); }
PAR_BLOCK { FFLAS::pfrand(F,Rand, k,n,B,k/MAX_THREADS); }
PAR_BLOCK { FFLAS::pfzero(F, m,n,C,m/MAX_THREADS); }
Ints alpha,beta;
alpha=F.one;
beta=F.zero;
#ifdef BENCH_FLINT
// FLINT MUL //
fmpz_t modp,tmp;
fmpz_init(modp);
fmpz_init(tmp);
fmpz_set_mpz(modp, *(reinterpret_cast<const mpz_t*>(&p)));
fmpz_mat_t AA,BB,CC,DD;
fmpz_mat_init (AA, m, k);
fmpz_mat_init (BB, k, n);
fmpz_mat_init (CC, m, n);
fmpz_mat_init (DD, m, n);
fmpz_t aalpha, bbeta;
fmpz_set_mpz(aalpha,*(reinterpret_cast<const mpz_t*>(&alpha)));
fmpz_set_mpz(bbeta,*(reinterpret_cast<const mpz_t*>(&beta)));
for (size_t i=0;i<m;++i)
for (size_t j=0;j<k;++j)
fmpz_set_mpz(fmpz_mat_entry(AA,i,j),*(reinterpret_cast<const mpz_t*>(A+i*lda+j)));
for (size_t i=0;i<k;++i)
for (size_t j=0;j<n;++j)
fmpz_set_mpz(fmpz_mat_entry(BB,i,j),*(reinterpret_cast<const mpz_t*>(B+i*ldb+j)));
for (size_t i=0;i<m;++i)
for (size_t j=0;j<n;++j)
fmpz_set_mpz(fmpz_mat_entry(CC,i,j),*(reinterpret_cast<const mpz_t*>(C+i*ldc+j)));
chrono.clear();chrono.start();
// DD= A.B
fmpz_mat_mul(DD,AA,BB);
// CC = beta.C
fmpz_mat_scalar_mul_fmpz(CC,CC,bbeta);
// CC = CC + DD.alpha
fmpz_mat_scalar_addmul_fmpz(CC,DD,aalpha);
// CC = CC mod p
for (size_t i=0;i<m;++i)
for (size_t j=0;j<n;++j)
fmpz_mod(fmpz_mat_entry(CC,i,j),fmpz_mat_entry(CC,i,j),modp);
chrono.stop();
timeFlint+=chrono.usertime();
fmpz_mat_clear(AA);
fmpz_mat_clear(BB);
#endif
//END FLINT CODE //
using FFLAS::CuttingStrategy::Recursive;
using FFLAS::StrategyParameter::TwoDAdaptive;
// RNS MUL_LA
chrono.clear();chrono.start();
// PAR_BLOCK{
// FFLAS::fgemm(F,FFLAS::FflasNoTrans,FFLAS::FflasNoTrans,m,n,k,alpha,A,lda,B,ldb,beta,C,ldc, SPLITTER(NUM_THREADS,Recursive,TwoDAdaptive) );
// }
{
FFLAS::fgemm(F,FFLAS::FflasNoTrans,FFLAS::FflasNoTrans,m,n,k,alpha,A,lda,B,ldb,beta,C,ldc,FFLAS::ParSeqHelper::Sequential());
}
chrono.stop();
time+=chrono.realtime();
TimFreivalds.start();
bool pass = FFLAS::freivalds(F, FFLAS::FflasNoTrans, FFLAS::FflasNoTrans, m,n,k, alpha, A, k, B, n, C,n);
TimFreivalds.stop();
timev+=TimFreivalds.usertime();
if (!pass) {
std::cout<<"FAILED"<<std::endl;
std::cout << "p:=" << p << ';'<<std::endl;
write_matrix(std::cout<<"A:=",p,m,k,A,lda)<<';'<<std::endl;
write_matrix(std::cout<<"B:=",p,k,n,B,ldb)<<';'<<std::endl;
write_matrix(std::cout<<"C:=",p,m,n,C,ldc)<<';'<<std::endl;
}
FFLAS::fflas_delete(A);
FFLAS::fflas_delete(B);
FFLAS::fflas_delete(C);
}
double Gflops=(2.*double(m)/1000.*double(n)/1000.*double(k)/1000.0) / time * double(iters);
// Gflops*=p.bitsize()/16.;
cout<<typeid(Ints).name()
<< " | Time: "<< (time/double(iters)) << " (total:" << time <<") Gflops: "<<Gflops<<" | perword: "<< (Gflops*double(p.bitsize()))/64. ;
FFLAS::writeCommandString(std::cout << '|' << p << " (" << p.bitsize()<<")|", as) << " | Freivalds: "<< timev/double(iters) << std::endl;
#ifdef BENCH_FLINT
cout<<"Time FLINT: "<<timeFlint<<endl;
#endif
return 0;
}