Buffer Operations¶

Optimized SIMD-accelerated buffer operations for real-time audio processing.

Features¶

SIMD Optimized: SSE/AVX vectorization with scalar fallbacks
Header-Only: No linking required
Real-Time Safe: No allocations in processing functions
Tested: Comprehensive correctness and performance tests

Modules¶

buffer_math.hpp¶

Mathematical operations on audio buffers: - buffer_add() - Add buffers element-wise - buffer_multiply() - Multiply buffers element-wise - buffer_scale() - Scale by constant - buffer_mix() - Mix with gain - buffer_peak() - Find maximum absolute value - buffer_rms() - Calculate RMS level

buffer_copy.hpp¶

Copy and format conversion: - buffer_copy() - Optimized memory copy - buffer_deinterleave() - LRLR... → [L...][R...] - buffer_interleave() - [L...][R...] → LRLR... - buffer_copy_with_gain() - Copy with gain applied - buffer_reverse() - Reverse buffer in-place

buffer_fill.hpp¶

Fill patterns: - buffer_clear() - Zero buffer - buffer_fill() - Fill with constant - buffer_ramp() - Linear ramp - buffer_ramp_exponential() - Exponential curve - buffer_ramp_scurve() - Smooth S-curve

Usage¶

#include "buffer_math.hpp"
#include "buffer_copy.hpp"
#include "buffer_fill.hpp"

using namespace audiolab::core::buffer;

// Process audio
float left[512], right[512];
float output[1024];

// Mix channels
buffer_mix(left, right, 0.5f, 512);  // left += right * 0.5

// Interleave for output
buffer_interleave_stereo(output, left, right, 512);

// Check levels
float peak = buffer_peak(output, 1024);
float rms = buffer_rms(output, 1024);

Performance¶

All operations have SIMD-optimized paths:

Operation	SIMD	Speedup
buffer_add	AVX	~8x
buffer_multiply	AVX	~8x
buffer_scale	AVX	~8x
buffer_mix	AVX	~6x
buffer_peak	AVX	~4x
buffer_rms	AVX	~6x
deinterleave	SSE	~3x
interleave	SSE	~3x

Benchmarks on Intel i7 @3.6GHz, 4096 samples

Building¶

mkdir build && cd build
cmake ..
cmake --build .

# Run tests
ctest

# Run benchmarks (optimized)
./benchmark_buffer_operations

SIMD Support¶

The library automatically detects and uses available SIMD instructions:

AVX: 8 floats per operation (most modern CPUs)
SSE: 4 floats per operation (fallback)
Scalar: No SIMD (universal fallback)

To force specific SIMD level:

# AVX
cmake -DCMAKE_CXX_FLAGS="-mavx" ..

# SSE only
cmake -DCMAKE_CXX_FLAGS="-msse -mno-avx" ..

# No SIMD
cmake -DCMAKE_CXX_FLAGS="-mno-sse" ..

Integration¶

Add to your CMakeLists.txt:

add_subdirectory(buffer_operations)
target_link_libraries(your_target PRIVATE buffer_operations)

Design Principles¶

Zero-cost abstraction - No performance penalty vs hand-written code
Explicit SIMD paths - Clear code with #ifdef __AVX__ guards
Safe scalar fallbacks - Works on any platform
Cache-friendly - Sequential memory access patterns
Alignment-agnostic - Works with unaligned data

License¶

Part of AudioLab project.