Description

Addresses issue #43

Changes Made

1. AVX Threshold Processor
   File: AVXThresholdProcessor.cpp
   Optimization: Replaced _mm256_blendv_epi8(_mm256_setzero_si256(), signal, mask) with _mm256_and_si256(signal, mask).
   Latency/Throughput Improvement: Reduced a variable-blend instruction down to a single bitwise AND instruction, increasing throughput and port availability (releasing Port 5 constraints).

2. AVX Frugal Pedestal Subtract Processor
   File: AVXFrugalPedestalSubtractProcessor.cpp
   Optimizations: Replaced the two 3-instruction blocks generating to_add = (is_gt ? 1 : 0) + (is_lt ? -1 : 0) (which materialized zero, 1, -1 and ran blendv) with _mm256_sub_epi16(is_lt, is_gt). Since comparison masks contain elements that are either 0 or 0xFFFF (which is signed -1), is_lt - is_gt mathematically computes the exact desired addition/subtraction.
   Replaced _mm256_blendv_epi8(m_accum, _mm256_setzero_si256(), need_reset) with _mm256_andnot_si256(need_reset, m_accum).
   Latency/Throughput Improvement: Avoided 4 slow blendv instructions, 2 zero-vector initializations, and 4 constant-vector initializations. The operations now use single-cycle arithmetic subtraction and bitwise ANDNOT.

3. AVX Pipeline
   File: AVXPipeline.cpp
   Optimizations: Replaced _mm256_blendv_epi8(_mm256_setzero_si256(), m_ones_register, active) with _mm256_and_si256(m_ones_register, active).
   Replaced 5 blendv calls masking fields in generate_tps with fast _mm256_and_si256.
   Replaced 5 blendv calls resetting fields in generate_tps with fast _mm256_andnot_si256.
   Latency/Throughput Improvement: Eliminated 11 slow variable-blend (blendv) intrinsics in the critical path of trigger primitive generation.

Standalone Benchmark and Verification App
A self-contained validation and performance benchmark app is located in /test/app/test_benchmark_tpg.cxx

You can build this benchmark application on any Linux system with GCC and AVX2 support without needing the DUNE-DAQ environment:

Compile

g++ -O3 -mavx2 test/apps/test_benchmark_tpg.cxx -o tpglibs_benchmark

Run

./tpglibs_benchmark

Type of change

• Documentation (non-breaking change that adds or improves the documentation)
• New feature or enhancement (non-breaking change which adds functionality)
• Optimization (non-breaking change that improves code/performance)
• Bug fix (non-breaking change which fixes an issue)
• Breaking change (whatever its nature)

Testing checklist

• Unit tests pass (e.g. dbt-build --unittest)
• Minimal system quicktest passes (pytest -s minimal_system_quick_test.py)
• Full set of integration tests pass (dunedaq_integtest_bundle.sh)
• Python tests pass if applicable (e.g. python -m pytest)
• Pre-commit hooks run successfully if applicable (e.g. pre-commit run --all-files)

Comments here on the testing

Further checks

• Code is commented where needed, particularly in hard-to-understand areas
• Code style is correct (dbt-build --lint, and/or see https://dune-daq-sw.readthedocs.io/en/latest/packages/styleguide/)
• If applicable, new tests have been added or an issue has been opened to tackle that in the future.
  (Indicate issue here: # (issue))