Line data    Source code

       1              : /**
       2              :  * @file AVXFrugalPedestalSubtractProcessor.cpp
       3              :  *
       4              :  * @copyright This is part of the DUNE DAQ Software Suite, copyright 2020.
       5              :  * Licensing/copyright details are in the COPYING file that you should have
       6              :  * received with this code.
       7              :  */
       8              : 
       9              : #include "tpglibs/AVXFrugalPedestalSubtractProcessor.hpp"
      10              : 
      11              : namespace tpglibs {
      12              : 
      13           15 : REGISTER_AVXPROCESSOR_CREATOR("AVXFrugalPedestalSubtractProcessor", AVXFrugalPedestalSubtractProcessor)
      14              : 
      15           50 : void AVXFrugalPedestalSubtractProcessor::configure(const nlohmann::json& config, const int16_t* /* plane_numbers */) {
      16              :   // Configure common metric collection parameters
      17              :   // Register pointers to the ACTUAL member variables, not copies
      18              :   // Use shared_ptr with no-op deleter to avoid double-free
      19           50 :   m_internal_state_name_registry.register_internal_state("pedestal", 
      20          100 :     std::shared_ptr<__m256i>(&m_pedestal, [](auto*){}));
      21           50 :   m_internal_state_name_registry.register_internal_state("accum", 
      22          100 :     std::shared_ptr<__m256i>(&m_accum, [](auto*){}));
      23              :     
      24           50 :   configure_internal_state_collection(config);
      25              :   
      26           50 :   m_accum_limit = config["accum_limit"];
      27           50 : }
      28              : 
      29          650 : __m256i AVXFrugalPedestalSubtractProcessor::process(const __m256i& signal) {
      30              :   // Update sample counter and write internal states to buffer for harvesting
      31          650 :   m_samples++;
      32          650 :   if (m_collect_internal_state_flag && (m_samples % m_sample_period == 0)) {
      33          650 :     m_internal_state_buffer_manager.write_to_active_buffer();
      34              :   }
      35              : 
      36              :   // Find the channels that are above or below the pedestal.
      37          650 :   __m256i is_gt = _mm256_cmpgt_epi16(signal, m_pedestal);
      38          650 :   __m256i is_lt = _mm256_cmpgt_epi16(m_pedestal, signal);
      39              : 
      40              :   // Update m_accum.
      41          650 :   __m256i to_add = _mm256_setzero_si256();                                    // Assumes equal to pedestal.
      42          650 :   to_add = _mm256_blendv_epi8(to_add, _mm256_set1_epi16(1), is_gt);           // Set the above pedestal case.
      43          650 :   to_add = _mm256_blendv_epi8(to_add, _mm256_set1_epi16(-1), is_lt);          // Set the below pedestal case.
      44              : 
      45          650 :   m_accum = _mm256_add_epi16(m_accum, to_add);
      46              : 
      47              :   // Check the accum limit condition.
      48          650 :   is_gt = _mm256_cmpgt_epi16(m_accum, _mm256_set1_epi16(m_accum_limit));      // m_accum > +limit.
      49          650 :   is_lt = _mm256_cmpgt_epi16(_mm256_set1_epi16(-1*m_accum_limit), m_accum);   // m_accum < -limit = -limit > m_accum.
      50              : 
      51          650 :   to_add = _mm256_setzero_si256();
      52          650 :   to_add = _mm256_blendv_epi8(to_add, _mm256_set1_epi16(1), is_gt);
      53          650 :   to_add = _mm256_blendv_epi8(to_add, _mm256_set1_epi16(-1), is_lt);
      54              : 
      55              :   // Update pedestal.
      56          650 :   m_pedestal = _mm256_adds_epi16(m_pedestal, to_add);
      57              : 
      58              :   // Reset too high/low m_accum channels.
      59          650 :   __m256i need_reset = _mm256_or_si256(is_lt, is_gt);
      60          650 :   m_accum = _mm256_blendv_epi8(m_accum, _mm256_setzero_si256(), need_reset);
      61              : 
      62          650 :   return AVXProcessor::process(_mm256_sub_epi16(signal, m_pedestal));
      63              : }
      64              : 
      65              : } // namespace tpglibs