static void exec(LHS &lhs, RHS const &rhs)
{
expr::evaluate(rhs);
length_type const size = lhs.size(1, 0);
for (index_type i=0; i<size; ++i)
lhs.put(i, rhs.get(i));
}
static void exec(LHS &lhs, RHS const &rhs)
{
using namespace impl;
typedef typename simd::LValue_access_traits<typename LHS::value_type> WAT;
typedef typename simd::Proxy_factory<RHS, false>::access_traits EAT;
length_type const vec_size =
simd::Simd_traits<typename LHS::value_type>::vec_size;
Ext_data<LHS, layout_type> dda(lhs, SYNC_OUT);
simd::Proxy<WAT,true> lp(dda.data());
simd::Proxy<EAT,false> rp(simd::Proxy_factory<RHS,false>::create(rhs));
length_type const size = dda.size(0);
length_type n = size;
// loop using proxy interface. This generates the best code
// with gcc 3.4 (with gcc 4.1 the difference to the first case
// above is negligible).
while (n >= vec_size)
{
lp.store(rp.load());
n -= vec_size;
lp.increment();
rp.increment();
}
// Process the remainder, using simple loop fusion.
for (index_type i = size - n; i != size; ++i) lhs.put(i, rhs.get(i));
}
static void exec(LHS &lhs, RHS const &rhs)
{
expr::evaluate(rhs);
length_type const rows = lhs.size(2, 0);
length_type const cols = lhs.size(2, 1);
for (index_type j=0; j<cols; ++j)
for (index_type i=0; i<rows; ++i)
lhs.put(i, j, rhs.get(i, j));
}
static void exec(LHS &lhs, RHS const &rhs)
{
expr::evaluate(rhs);
length_type const size0 = lhs.size(3, 0);
length_type const size1 = lhs.size(3, 1);
length_type const size2 = lhs.size(3, 2);
for (index_type i=0; i<size0; ++i)
for (index_type k=0; k<size2; ++k)
for (index_type j=0; j<size1; ++j)
lhs.put(i, j, k, rhs.get(i, j, k));
}