Boost C++ Performance: 20% Gains with gprof Profiling and Smart Optimizations

Optimization Guidelines

1. Pareto principle: focus on the ~20% of code that consumes most time. Example: function A is large but called once; function B is small but called 1000 times, so prioritize B.

2. Write code first, optimize later; premature optimization can hurt readability and development efficiency.

Tools

gprof – GNU profiling tool for C/C++ on Linux, AIX, Sun etc. It records function call counts, CPU time, and call graphs.

Using gprof

1) Compile with -pg flag, e.g., g++ -pg -o test.exe test.cpp.

2) Run the executable; it generates gmon.out with profiling data.

3) Analyze with gprof test.exe gmon.out (or redirect output to a file).

Practice

Faced with performance bottlenecks, we used gprof and identified two major issues.

1. Expensive Large Object Initialization

Object VOBJ1 was constructed 307 times, consuming 6.5% of total time. Moving initialization to the constructor initializer list reduced its cost to 1.8%.

2. Inefficient Map Usage

Recordset::GetField called 89 times, costing 6.8%. The original code used map["a"] which inserts missing keys. Replacing with find and switching from map (red‑black tree) to unordered_map cut the cost to 1.4% and halved execution time.

string Recordset::GetField(const string &strName)
{
    unordered_map::iterator iter = m_fields.find(strName);
    if (iter == m_fields.end())
    {
        LOG_ERR("[Recordset::GetField] " << strName << " Not Find!!");
    }
    return m_records[nIndex].GetValue(iter->second - 1);
}

Summary

By modifying fewer than 30 lines of code, overall performance improved by about 10%. Targeted profiling and focusing on the critical hotspots—like heavy object construction and suboptimal container choices—delivers significant gains.