double syr2krect(size_t N, size_t iterations = 1){
Eigen::Matrix<value_type, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor> A(N, size_t(1.5*N)), B(N, size_t(1.5*N)), C(N, N);
value_type c = 3, d = 5;
rminit(N, 1.5*N, A);
rminit(N, 1.5*N, B);
minit(N, C);
std::vector<double> times;
for(size_t i = 0; i < iterations; ++i){
auto start = std::chrono::steady_clock::now();
//C = c * B * A.transpose() + c * A * B.transpose() + d * C;
C *= d;
C += c * A * B.transpose();
C += c * B * A.transpose();
auto end = std::chrono::steady_clock::now();
auto diff = end - start;
times.push_back(std::chrono::duration<double, std::milli> (diff).count()); //save time in ms for each iteration
}
double tmin = *(std::min_element(times.begin(), times.end()));
double tavg = average_time(times);
// check to see if nothing happened during run to invalidate the times
if(variance(tavg, times) > max_variance){
std::cerr << "eigen kernel 'syr2krect': Time deviation too large! \n";
}
return tavg;
}
void
dvma_init()
{
vm_offset_t segmap_addr;
/*
* Create phys_map covering the entire DVMA space,
* then allocate the segment pool from that. The
* remainder will be used as the DVMA page pool.
*/
phys_map = uvm_map_create(pmap_kernel(),
DVMA_MAP_BASE, DVMA_MAP_END, 1);
if (phys_map == NULL)
panic("unable to create DVMA map");
/*
* Reserve the DVMA space used for segment remapping.
* The remainder of phys_map is used for DVMA scratch
* memory pages (i.e. driver control blocks, etc.)
*/
segmap_addr = uvm_km_valloc_wait(phys_map, dvma_segmap_size);
if (segmap_addr != DVMA_MAP_BASE)
panic("dvma_init: unable to allocate DVMA segments");
/*
* Create the VM pool used for mapping whole segments
* into DVMA space for the purpose of data transfer.
*/
rminit(dvma_segmap, dvma_segmap_size, segmap_addr,
"dvma_segmap", NUM_DVMA_SEGS);
}
double syr2krect(size_t N, size_t iterations = 1){
typedef mtl::tag::row_major row_major;
typedef mtl::mat::parameters<row_major> parameters;
typedef mtl::dense2D<value_type, parameters> dense2D;
typedef mtl::dense_vector<value_type> dense_vector;
dense2D A(N, 1.5*N), B(N, 1.5*N), C(N, N);
value_type c = 3, d = 5;
rminit(N, 1.5*N, A);
rminit(N, 1.5*N, B);
minit(N, C);
std::vector<double> times;
for(size_t i = 0; i < iterations; ++i){
auto start = std::chrono::steady_clock::now();
//C = c * B * trans(A) + c * A * trans(B) + d * C;
C *= d;
C += c * A * trans(B);
C += c * B * trans(A);
auto end = std::chrono::steady_clock::now();
auto diff = end - start;
times.push_back(std::chrono::duration<double, std::milli> (diff).count()); //save time in ms for each iteration
}
double tmin = *(std::min_element(times.begin(), times.end()));
double tavg = average_time(times);
// check to see if nothing happened during run to invalidate the times
if(variance(tavg, times) > max_variance){
std::cerr << "mtl kernel 'syr2krect': Time deviation too large! \n";
}
return tavg;
}
void
dvma_init()
{
/*
* Create the resource map for DVMA pages.
*/
dvmamap = malloc((sizeof(struct map) * dvma_max_segs),
M_DEVBUF, M_WAITOK);
rminit(dvmamap, btoc(DVMA_MAP_AVAIL), btoc(DVMA_MAP_BASE),
"dvmamap", dvma_max_segs);
/*
* Enable DVMA in the System Enable register.
* Note: This is only necessary for VME slave accesses.
* On-board devices are always capable of DVMA.
*/
*enable_reg |= ENA_SDVMA;
}
