Performance Optimization¶

Guidelines for getting the best performance from PFC.

Choose the Right Codec¶

Different codecs have different performance characteristics:

Unary: Fastest encoding/decoding, worst compression
Fixed-width: Predictable, no branching
Elias codes: Good balance of speed and compression
Huffman/LZ77: Best compression, slower

Batch Operations¶

Process multiple values together:

// Slower: Individual operations
for (auto value : values) {
    packed_vec.push_back(value);
}

// Faster: Batch insert
packed_vec.reserve(values.size());
std::copy(values.begin(), values.end(),
          std::back_inserter(packed_vec));

Use Parallel Algorithms¶

With TBB:

#include <execution>

packed_transform(std::execution::par_unseq,
                data.begin(), data.end(),
                output.begin(),
                expensive_function);

Profile First¶

Use profiling tools to find bottlenecks:

perf record ./your_app
perf report

Common hotspots: - Bit I/O operations - Decode-on-access in tight loops - Unnecessary copies

Memory Layout¶

Consider alignment and cache behavior:

// Good: Sequential access
for (size_t i = 0; i < vec.size(); ++i) {
    process(vec[i]);
}

// Bad: Random access in packed data
for (size_t i : random_indices) {
    process(vec[i]);  // O(n) worst case
}

See Design Principles for performance characteristics.