StringInt-Benchmarks

Here's a few benchmarks on a few common operating system/compiler platforms. Although feel free to try it out on more and let me know if you do/what scores you get! Cheers! (Click each of the badges below to be taken to the respective benchmarks). Also I will add some instructions for building below.

The Libraries

• void-numerics
• jeaiii_itoa
• fmt
• std Library

Building

Full build instructions — including the PGO/LTO pipeline used to produce every result on this page — live in Compilation Method.

By OS/Compiler:

---

Linux/Clang:

---

• int-to-str
• str-to-int
• str-to-int (leading zeros)

Linux/GCC:

---

• int-to-str
• str-to-int
• str-to-int (leading zeros)

Windows/MSVC:

---

• int-to-str
• str-to-int
• str-to-int (leading zeros)

macOS/Clang:

---

• int-to-str
• str-to-int
• str-to-int (leading zeros)

macOS/GCC:

---

• int-to-str
• str-to-int
• str-to-int (leading zeros)

Pre-requisites:

---

1. CMake 3.28 or later.
2. A C++20 compliant compiler.
3. Python3 and Seaborn for generating the graphs.

Compilation_Instructions.md

Note:

Two separate things are doing work here. Timing methodology — adaptive epoch sampling, Welch's t-test for statistical tie detection, Bessel-corrected variance — comes from the CPU benchmark library benchmarksuite. Making sure the binaries being timed are a fair fight is this repo's own job: every library under test, including the standard library, is built through the same three-stage PGO/LTO pipeline (instrumented GENERATE build → training run → profile-consuming USE build) before it's ever measured, so no library gets a hand-optimized build going up against everyone else's defaults.

"Anytime someone comes to you with one benchmark? They have gamed that benchmark." — Matt Kulukundis, CppCon 2017: "Designing a Fast, Efficient, Cache-friendly Hash Table, Step by Step"