Skip to content

OPENMP_IMPLEMENTATION.md

Latest commit

 

History

History
139 lines (100 loc) · 4 KB

OPENMP_IMPLEMENTATION.md

File metadata and controls

139 lines (100 loc) · 4 KB

OpenMP Audio DSP Implementation

Summary

This implementation adds actual OpenMP parallelization to the Hyphon audio engine, replacing the previous situation where libomp was linked but never used.

Files Added

C++ Audio DSP Module (emscripten/audio_dsp.cpp)

Provides OpenMP-parallelized audio processing functions:

Function Description OpenMP Features Used
applyGain() Apply gain to buffer in-place parallel for simd
mixBuffers() Mix multiple buffers with gains parallel for simd
findPeak() Find peak amplitude parallel for reduction(max:)
deinterleaveStereo() Deinterleave stereo parallel for simd
interleaveStereo() Interleave stereo parallel for simd
floatToInt16() Convert float32 to int16 parallel for simd
applyStereoWidth() Adjust stereo width parallel for simd
getNumThreads() Get thread count omp_get_max_threads()
setNumThreads() Set thread count omp_set_num_threads()

TypeScript Bridge (src/engines/AudioDSP.ts)

JavaScript/TypeScript interface to the WASM module:

  • Automatic fallback to JS implementation when WASM unavailable
  • Memory management (malloc/free on WASM heap)
  • Type-safe API with error handling

Tests (src/__tests__/AudioDSP.test.ts)

Comprehensive test suite covering:

  • Thread management (get/set thread count)
  • All DSP functions via WASM
  • Fallback JS implementations
  • Large buffer performance

Build System Updates

emscripten/build.sh

Updated to compile the new audio_dsp.cpp module with OpenMP flags:

# Audio DSP (OpenMP enabled)
compile_cpp "$SCRIPT_DIR/audio_dsp.cpp"

The build already had correct OpenMP flags:

  • -fopenmp - Enable OpenMP
  • -pthread - Enable pthreads (required for Emscripten OpenMP)
  • libomp.a - Static OpenMP runtime library

Usage Example

import { audioDSP } from './engines/AudioDSP';

// Check if DSP module is available
if (audioDSP.isAvailable()) {
    console.log(`OpenMP threads: ${audioDSP.getNumThreads()}`);
    
    // Apply gain with OpenMP parallelization
    const buffer = new Float32Array(audioData);
    audioDSP.applyGain(buffer, 2, 0.8); // 2 channels, 0.8 gain
    
    // Find peak amplitude (parallel reduction)
    const peak = audioDSP.findPeak(buffer, 2);
    console.log(`Peak: ${peak}`);
}

Performance Benefits

The OpenMP implementation provides speedups for:

  • Large buffer processing (> 10,000 samples)
  • Multi-channel audio
  • Batch operations (mixing multiple sources)

Fallback JS implementations ensure functionality works even without WASM.

Thread Safety

All functions are thread-safe:

  • No shared mutable state between calls
  • Each function operates on provided buffers only
  • Thread count can be adjusted at runtime

Future Enhancements

Potential additions to leverage OpenMP further:

  1. FFT processing - Parallel FFT for spectral analysis
  2. Convolution - Parallel impulse response processing
  3. Granular synthesis - Parallel grain processing
  4. Multi-track rendering - Parallel track mixing in XM export

Technical Details

SIMD Directives

The code uses simd clauses where appropriate:

#pragma omp parallel for simd schedule(static)
for (int i = 0; i < totalSamples; i++) {
    buffer[i] *= gain;
}

This hints to the compiler that iterations are independent and can use SIMD instructions.

Reduction Operations

For findPeak(), we use OpenMP reduction:

#pragma omp parallel for reduction(max:peak) schedule(static)
for (int i = 0; i < totalSamples; i++) {
    peak = std::max(peak, std::abs(buffer[i]));
}

This safely combines results from all threads.

Memory Alignment

The code assumes standard 4-byte alignment for float arrays, which works well with Emscripten's memory model.

Testing

Run the AudioDSP tests:

npm test -- src/__tests__/AudioDSP.test.ts

All 18 tests pass, covering:

  • 7 WASM path tests
  • 7 fallback JS tests
  • 4 edge case/error tests