constexpr const int pool_size = 3;
const float32x4_t top_data = vld1q_f32(reinterpret_cast<const float *>(input_top_ptr + input.offset()));
const float32x4_t middle_data = vld1q_f32(reinterpret_cast<const float *>(input_middle_ptr + input.offset()));
const float32x4_t bottom_data = vld1q_f32(reinterpret_cast<const float *>(input_bottom_ptr + input.offset()));
float32x2_t res = {};
if(pooling_type == PoolingType::AVG)
{
// Calculate scale
float scale = calculate_avg_scale(id, pool_size, upper_bound_w, upper_bound_h, pool_pad_x, pool_pad_y, pool_stride_x, pool_stride_y);
const float32x2_t scale_v = vdup_n_f32(scale);
// Perform pooling
const float32x4_t sum_data = vaddq_f32(vaddq_f32(top_data, bottom_data), middle_data);
res
NEON 3x3 max pool示例代码
最新推荐文章于 2022-11-09 14:52:43 发布