phynexis-deps/armadillo-code
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

faster handling of sparse matrix views with one column

Browse Source
This commit is contained in:
conrad
2024-05-22 14:49:53 +10:00
parent b955116e1f
commit b12eefdae8
2 changed files with 38 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/armadillo_bits/SpSubview_meat.hpp
+18 -8
View File
@@ -44,18 +44,28 @@ SpSubview<eT>::SpSubview(const SpMat<eT>& in_m, const uword in_row1, const uword
m.sync_csc();
// There must be a O(1) way to do this
uword lend = m.col_ptrs[in_col1 + in_n_cols];
uword lend_row = in_row1 + in_n_rows;
uword count = 0;
// count the number of non-zeros in the subview
uword count = 0;
for(uword i = m.col_ptrs[in_col1]; i < lend; ++i)
if( (n_cols == 1) && (n_rows == m.n_rows) )
{
const uword m_row_indices_i = m.row_indices[i];
count = m.col_ptrs[aux_col1 + 1] - m.col_ptrs[aux_col1];
}
else
{
arma_debug_print("counting non-zeros in sparse subview");
const bool condition = (m_row_indices_i >= in_row1) && (m_row_indices_i < lend_row);
uword lend = m.col_ptrs[in_col1 + in_n_cols];
uword lend_row = in_row1 + in_n_rows;
count += condition ? uword(1) : uword(0);
for(uword i = m.col_ptrs[in_col1]; i < lend; ++i)
{
const uword m_row_indices_i = m.row_indices[i];
const bool condition = (m_row_indices_i >= in_row1) && (m_row_indices_i < lend_row);
count += condition ? uword(1) : uword(0);
}
}
access::rw(n_nonzero) = count;
include/armadillo_bits/fn_dot.hpp
+20 -2
View File
@@ -289,13 +289,31 @@ dot
{
arma_debug_sigprint();
typedef typename T1::elem_type eT;
if(is_SpSubview_col<T2>::value)
{
// TODO: refactor to use C++17 "if constexpr" to avoid reinterpret_cast shenanigans
const SpSubview_col<eT>& yy = reinterpret_cast< const SpSubview_col<eT>& >(y);
if(yy.n_rows == yy.m.n_rows)
{
arma_debug_print("using sparse column vector specialisation");
const quasi_unwrap<T1> U(x);
arma_conform_assert_same_size(U.M.n_elem, uword(1), yy.n_elem, uword(1), "dot()");
return dense_sparse_helper::dot(U.M.memptr(), yy.m, yy.aux_col1);
}
}
const Proxy<T1> pa(x);
const SpProxy<T2> pb(y);
arma_conform_assert_same_size(pa.get_n_rows(), pa.get_n_cols(), pb.get_n_rows(), pb.get_n_cols(), "dot()");
typedef typename T1::elem_type eT;
eT result = eT(0);
typename SpProxy<T2>::const_iterator_type it = pb.begin();