static int
input_zvmmul_row_f(lwp_functions* pf,
void* p_params,
void* entries,
unsigned int current_count,
unsigned int total_count)
{
unsigned int const FP = 1; // Split-complex data: 1 floats per point.
Vmmul_split_params* params = (Vmmul_split_params*)p_params;
unsigned long length = params->length;
unsigned long mult = params->mult;
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 0);
add_vector_f(pf, entries,
params->ea_input_matrix_re +
current_count * FP * mult * params->input_stride * sizeof(float),
length * mult);
add_vector_f(pf, entries,
params->ea_input_matrix_im +
current_count * FP * mult * params->input_stride * sizeof(float),
length * mult);
if (current_count == 0)
{
unsigned long dma_size = VSIP_IMPL_INCREASE_TO_DMA_SIZE(length, float);
add_vector_f(pf, entries, params->ea_input_vector_re, dma_size);
add_vector_f(pf, entries, params->ea_input_vector_im, dma_size);
}
int
output(
Plugin_functions* pf,
void* p_context,
void* p_params,
void* entries,
unsigned int current_count,
unsigned int total_count)
{
unsigned int const FP = 1; // Split-complex data: 1 floats per point.
Vmmul_split_params* params = (Vmmul_split_params*)p_params;
unsigned long length = params->length;
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_OUT, 0);
add_vector_f(pf, entries,
params->ea_output_matrix_re +
current_count * FP * params->output_stride * sizeof(float),
length);
add_vector_f(pf, entries,
params->ea_output_matrix_im +
current_count * FP * params->output_stride * sizeof(float),
length);
(pf->f_dtl_end)(entries);
return 0;
}
int
input(
Plugin_functions* pf,
void* p_context,
void* p_params,
void* entries,
unsigned int current_count,
unsigned int total_count)
{
unsigned int const FP = 1; // Split-complex data: 1 floats per point.
Vmmul_split_params* params = (Vmmul_split_params*)p_params;
unsigned long length = params->length;
#if PPU_IS_32BIT
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 0);
add_vector_f(pf, entries,
params->ea_input_matrix_re +
current_count * FP * params->input_stride * sizeof(float),
length);
add_vector_f(pf, entries,
params->ea_input_matrix_im +
current_count * FP * params->input_stride * sizeof(float),
length);
if (current_count == 0)
{
add_vector_f(pf, entries, params->ea_input_vector_re, length);
add_vector_f(pf, entries, params->ea_input_vector_im, length);
}
(pf->f_dtl_end)(entries);
#else
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 0);
add_vector_f(pf, entries,
params->ea_input_matrix_re +
current_count * FP * params->input_stride * sizeof(float),
length);
(pf->f_dtl_end)(entries);
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, length*sizeof(float));
add_vector_f(pf, entries,
params->ea_input_matrix_im +
current_count * FP * params->input_stride * sizeof(float),
length);
(pf->f_dtl_end)(entries);
if (current_count == 0)
{
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 2*length*sizeof(float));
add_vector_f(pf, entries, params->ea_input_vector_re, length);
(pf->f_dtl_end)(entries);
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 3*length*sizeof(float));
add_vector_f(pf, entries, params->ea_input_vector_im, length);
(pf->f_dtl_end)(entries);
}
#endif
return 0;
}
static int
input_zfft_f(lwp_functions* pf,
void* params,
void* entries,
unsigned int iter,
unsigned int iter_max)
{
Fft_split_params* fftp = (Fft_split_params *)params;
unsigned int size = fftp->size;
unsigned int chunks = fftp->chunks_per_wb;
unsigned int cur_chunks;
unsigned int i;
alf_data_addr64_t ea;
if (iter == iter_max-1 && iter_max * chunks > fftp->chunks_per_spe)
cur_chunks = fftp->chunks_per_spe % chunks;
else
cur_chunks = chunks;
if (size == fftp->in_blk_stride)
{
size *= cur_chunks;
cur_chunks = 1;
}
// Transfer input.
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 0);
for (i=0; i<cur_chunks; ++i)
{
ea = fftp->ea_input_re +
(iter * chunks + i) * fftp->in_blk_stride * sizeof(float);
add_vector_f(pf, entries, ea, size);
}
for (i=0; i<cur_chunks; ++i)
{
ea = fftp->ea_input_im +
(iter * chunks + i) * fftp->in_blk_stride * sizeof(float);
add_vector_f(pf, entries, ea, size);
}
(pf->f_dtl_end)(entries);
return 0;
}
int
input(
Plugin_functions* pf,
void* p_context,
void* p_params,
void* entries,
unsigned int current_count,
unsigned int total_count)
{
unsigned int const FP = 1; // Split-complex data: 1 floats per point.
Vmmul_split_params* params = (Vmmul_split_params*)p_params;
unsigned long length = params->length;
// Set v_len to minimum of 4 since transfers must be a multiple of 16B.
unsigned long v_len = total_count;
v_len += params->shift;
if (v_len == 0) v_len = 4;
else if (v_len % 4 != 0) v_len += 4 - (v_len % 4);
#if PPU_IS_32BIT
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 0);
add_vector_f(pf, entries,
params->ea_input_matrix_re +
current_count * FP * params->input_stride * sizeof(float),
length);
add_vector_f(pf, entries,
params->ea_input_matrix_im +
current_count * FP * params->input_stride * sizeof(float),
length);
if (current_count == 0)
{
add_vector_f(pf, entries, params->ea_input_vector_re, v_len);
add_vector_f(pf, entries, params->ea_input_vector_im, v_len);
}
(pf->f_dtl_end)(entries);
#else
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 0);
add_vector_f(pf, entries,
params->ea_input_matrix_re +
current_count * FP * params->input_stride * sizeof(float),
length);
(pf->f_dtl_end)(entries);
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, length*sizeof(float));
add_vector_f(pf, entries,
params->ea_input_matrix_im +
current_count * FP * params->input_stride * sizeof(float),
length);
(pf->f_dtl_end)(entries);
if (current_count == 0)
{
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, 2*length*sizeof(float));
add_vector_f(pf, entries, params->ea_input_vector_re, v_len);
(pf->f_dtl_end)(entries);
(pf->f_dtl_begin)(entries, ALF_BUF_OVL_IN, (2*length+v_len)*sizeof(float));
add_vector_f(pf, entries, params->ea_input_vector_im, v_len);
(pf->f_dtl_end)(entries);
}
#endif
return 0;
}