Skip to content

COMPILE.md

Latest commit

 

History

History
172 lines (123 loc) · 3.79 KB

COMPILE.md

File metadata and controls

172 lines (123 loc) · 3.79 KB
Raw

Installation Options

Option 1: Download Pre-built JAR (Recommended)

The easiest way to use FastFloat is downloading the pre-built JAR with integrated native library:

GitHub Releases

Download the latest release: Releases Page

# Download JAR with integrated DLL/SO/DYLIB
wget https://github.com/andrestubbe/FastFloat/releases/download/v1.2.0/fastfloat-1.2.0.jar

# Or download JAR + separate native library
wget https://github.com/andrestubbe/FastFloat/releases/download/v1.2.0/fastfloat-1.2.0.jar
wget https://github.com/andrestubbe/FastFloat/releases/download/v1.2.0/fastfloat.dll     # Windows
wget https://github.com/andrestubbe/FastFloat/releases/download/v1.2.0/libfastfloat.so   # Linux
wget https://github.com/andrestubbe/FastFloat/releases/download/v1.2.0/libfastfloat.dylib # macOS

Maven (JitPack)

<repositories>
    <repository>
        <id>jitpack.io</id>
        <url>https://jitpack.io</url>
    </repository>
</repositories>

<dependencies>
    <dependency>
        <groupId>com.github.andrestubbe</groupId>
        <artifactId>fastfloat</artifactId>
        <version>v1.2.0</version>
    </dependency>
</dependencies>

Gradle (JitPack)

repositories {
    maven { url 'https://jitpack.io' }
}

dependencies {
    implementation 'com.github.andrestubbe:fastfloat:v1.2.0'
}

Building from Source

Prerequisites

  • JDK 17+
  • Maven 3.9+
  • Windows: Visual Studio 2019+ or Build Tools
  • Linux: GCC 9+ or Clang 10+
  • macOS: Xcode Command Line Tools

Build

Windows

compile.bat
mvn clean package

Linux/macOS

chmod +x compile.sh
./compile.sh
mvn clean package

The build script auto-detects CPU features (AVX2, AVX-512, FMA3) and compiles with optimal flags (-O3 -march=native -ffast-math).

Cross-Platform Build (All Platforms)

For building native libraries for multiple platforms:

# Windows (MSVC)
compile.bat

# Linux (GCC/Clang)
./compile.sh

# macOS (Clang)
./compile.sh

The native library will be placed in build/ (Windows: fastfloat.dll, Linux: libfastfloat.so, macOS: libfastfloat.dylib).

Compiler Optimization Flags

Windows (MSVC)

cl /O2 /arch:AVX2 /fp:fast /GL /LTCG fastfloat.cpp

Flags explained:

  • /O2 - Maximum optimization
  • /arch:AVX2 - Enable AVX2 instructions
  • /fp:fast - Fast floating-point model (20-30% speedup)
  • /GL - Whole program optimization
  • /LTCG - Link-time code generation

Linux/macOS (GCC/Clang)

g++ -O3 -march=native -ffast-math -fPIC -shared -o libfastfloat.so fastfloat.cpp

Flags explained:

  • -O3 - Maximum optimization
  • -march=native - Optimize for host CPU
  • -ffast-math - Aggressive floating-point optimizations
  • -fPIC - Position-independent code

Run Examples

cd examples/00-basic-usage
mvn compile exec:java

Troubleshooting

JNI UnsatisfiedLinkError

If you get UnsatisfiedLinkError, the native library was not found:

  1. Check that the DLL/so/dylib exists in build/
  2. On Windows, ensure the DLL is in PATH or copy to C:\Windows\System32
  3. On Linux/macOS, set LD_LIBRARY_PATH or DYLD_LIBRARY_PATH

CPU Feature Detection Failed

If AVX2/AVX-512 detection fails, you can force specific optimizations:

# Force SSE4.2 only (older CPUs)
./compile.sh --sse42

# Force AVX2
./compile.sh --avx2

# Force AVX-512
./compile.sh --avx512

Multi-Binary Runtime Dispatch

For optimal performance across different CPUs, build multiple binaries:

# Build all variants
./compile.sh --multi

# This creates:
# - build/libfastfloat_sse42.so (baseline)
# - build/libfastfloat_avx2.so (AVX2 optimized)
# - build/libfastfloat_avx512.so (AVX-512 optimized)

# The library auto-selects the best version at runtime based on CPUID