typename Promotion<T1, T2>::type
dotp_ext(
const_Vector<T1, Block1> op1,
const_Vector<T2, Block2> op2)
{
typedef typename Promotion<T1, T2>::type value_type;
test_assert(op1.size() == op2.size());
vsip::dda::Data<Block1, dda::in> raw1(op1.block());
T1 const *p1 = raw1.ptr();
stride_type str1 = raw1.stride(0);
vsip::dda::Data<Block2, dda::in> raw2(op2.block());
T2 const *p2 = raw2.ptr();
stride_type str2 = raw2.stride(0);
value_type sum = value_type();
for (index_type i=0; i<op1.size(); ++i)
{
sum += *p1 * *p2;
p1 += str1;
p2 += str2;
}
return sum;
}
void
vector_add(
const_Vector<T, Block1> res,
const_Vector<T, Block2> op1,
const_Vector<T, Block3> op2)
{
vsip::dda::Data<Block1, vsip::dda::out> raw_res(res.block());
float* p_raw = raw_res.ptr();
stride_type str_raw = raw_res.stride(0);
vsip::dda::Data<Block2, vsip::dda::in> raw1(op1.block());
float const *p1 = raw1.ptr();
stride_type str1 = raw1.stride(0);
vsip::dda::Data<Block3, vsip::dda::in> raw2(op2.block());
float const *p2 = raw2.ptr();
stride_type str2 = raw2.stride(0);
for (index_type i=0; i<res.size(); ++i)
{
*p_raw = *p1 + *p2;
p1 += str1;
p2 += str2;
p_raw += str_raw;
}
}
void operator()(
char const * str,
const_Vector<T, Block1> vin,
Vector <T, Block2> vout)
{
typedef vsip::impl::Layout<1, row1_type,
vsip::impl::Stride_unit, vsip::impl::Cmplx_split_fmt>
LP;
typedef vsip::impl::Ext_data<Block1, LP> layout1;
typedef vsip::impl::Ext_data<Block2, LP> layout2;
// PROFILE: Check sizes, check layout costs.
// Important to check size. If vectors are too large, our
// buffers will overflow.
test_assert(vin.size() == size_);
test_assert(vout.size() == size_);
if (verbose_)
{
cout << "Test_FFT_split: " << str << endl;
cout << "Block_layout<Block1>:\n";
print_layout<typename vsip::impl::Block_layout<Block1>::layout_type>(cout);
cout << endl;
cout << "LP:\n";
print_layout<LP>(cout);
cout << endl;
cout << " access_type<LP>(vin.block()) = "
<< access_type<LP>(vin.block()) << endl;
cout << " access_type<LP>(vout.block()) = "
<< access_type<LP>(vout.block()) << endl;
cout << " mem_required<LP>(vin.block()) = "
<< vsip::impl::mem_required<LP>(vin.block()) << endl;
cout << " mem_required<LP>(vout.block()) = "
<< vsip::impl::mem_required<LP>(vout.block()) << endl;
}
test_assert(vsip::impl::mem_required<LP>(vin.block()) <= sizeof(T)*size_);
test_assert(vsip::impl::mem_required<LP>(vout.block()) <= sizeof(T)*size_);
layout1 rin (vin.block(), vsip::impl::SYNC_IN,
make_pair(buffer_ + 0, buffer_ + size_));
layout2 rout(vout.block(), vsip::impl::SYNC_OUT,
make_pair(buffer_ + 2*size_, buffer_ + 3*size_));
test_assert(rin.stride(0) == 1);
test_assert(rin.size(0) == size_);
test_assert(rout.stride(0) == 1);
test_assert(rout.size(0) == size_);
fft_unit_stride_split(rin.data().first, rin.data().second,
rout.data().first, rout.data().second,
size_);
}
inline
const_Vector<typename Promotion<T1, T2>::type,
expr::Binary<expr::op::Mult, Block1, Block2, true> const>
t_mul(const_Vector<T1, Block1> v1, const_Vector<T2, Block2> v2)
{
typedef typename Promotion<T1, T2>::type RT;
typedef expr::Binary<expr::op::Mult, Block1, Block2, true> block_t;
return const_Vector<RT, const block_t>(block_t(v1.block(), v2.block()));
}
lazy_Vector<T, Unary<Scale, BlockType> const>
scale(const_Vector<T, BlockType> input, T value)
{
Scale<BlockType> s(input.block(), value);
Unary<Scale, BlockType> block(s);
return lazy_Vector<T, Unary<Scale, BlockType> const>(block);
}
const_Vector<T, Unary<Scale, BlockType> const>
scale(const_Vector<T, BlockType> input, T value)
{
typedef Unary<Scale, BlockType> block_type;
Scale<BlockType> s(input.block(), value);
block_type block(s);
return const_Vector<T, Unary<Scale, BlockType> const>(block);
}
inline
const_Vector<T, expr::Unary<expr::op::Minus, Block, true> const>
t_neg(const_Vector<T, Block> v1)
{
typedef expr::Unary<expr::op::Minus, Block, true> block_t;
return const_Vector<T, const block_t>(block_t(v1.block()));
}
void
generic_prod(
const_Vector<T0, Block0> a,
const_Matrix<T1, Block1> b,
Vector<T2, Block2> r)
{
using namespace vsip_csl::dispatcher;
assert(r.size() == b.size(1));
assert(a.size() == b.size(0));
#ifdef VSIP_IMPL_REF_IMPL
Evaluator<op::prod_vm, dispatcher::be::cvsip,
void(Block2&, Block0 const&, Block1 const&)>::exec
(r.block(), a.block(), b.block());
#else
vsip_csl::dispatch<op::prod_vm, void,
Block2&, Block0 const&, Block1 const&>
(r.block(), a.block(), b.block());
#endif
}
void
operator()(const_Vector<T, Block1> in, Vector<T, Block2> out)
{
typedef impl::Layout<1, row1_type, impl::Stride_unit, impl::Cmplx_inter_fmt> layout_t;
impl::Ext_data<Block1, layout_t> in_ext (in.block(), impl::SYNC_IN);
impl::Ext_data<Block2, layout_t> out_ext(out.block(), impl::SYNC_OUT);
assert(in_ext.stride(0) == 1);
assert(out_ext.stride(0) == 1);
if (Dir == -1)
IppFFT<T>::forward(fft_, in_ext.data(), out_ext.data(), buffer_);
else
IppFFT<T>::inverse(fft_, in_ext.data(), out_ext.data(), buffer_);
}
T
sum_ext(const_Vector<T, Block> view)
{
vsip::dda::Data<Block, dda::in> raw(view.block());
float const *data = raw.ptr();
stride_type stride = raw.stride(0);
T sum = T();
for (index_type i=0; i<view.size(); ++i)
{
sum += *data;
data += stride;
}
return sum;
}
void out_of_place(BE *backend,
const_Vector<InT, Block0>& in,
Vector<OutT, Block1>& out)
{
{
dda::Data<Block0, dda::in> in_data(in.block());
dda::Data<Block1, dda::out> out_data(out.block());
backend->out_of_place(in_data.ptr(), in_data.stride(0),
out_data.ptr(), out_data.stride(0),
select_fft_size<InT, OutT>(in_data.size(0), out_data.size(0)));
}
// Scale the data if not already done by the backend.
if (!backend->supports_scale() && !almost_equal(scale_, scalar_type(1.)))
out *= scale_;
}
void by_reference(BE *backend,
const_Vector<InT, Block0>& in,
Vector<OutT, Block1>& out)
{
{
Ext_data<Block0> in_ext (in.block(), SYNC_IN);
Ext_data<Block1> out_ext(out.block(), SYNC_OUT);
backend->by_reference(
in_ext.data(), in_ext.stride(0),
out_ext.data(), out_ext.stride(0),
select_fft_size<InT, OutT>(in_ext.size(0), out_ext.size(0)));
}
// Scale the data if not already done by the backend.
if (!backend->supports_scale() && !almost_equal(scale_, scalar_type(1.)))
out *= scale_;
}