Skip to content

FastSIMD 0.1.1

Choose a tag to compare

View all tags
@andrestubbe andrestubbe released this 14 Aug 06:54
· 21 commits to main since this release
0.1.1
7eb6897

TITLE:
FastSIMD: Unified Hardware Vector Acceleration Engine for Java 🚀

DESCRIPTION:
Demonstrating FastSIMD, the core hardware vector engine of the FastJava ecosystem. Powered by AVX2, AVX-512, and ARM NEON intrinsics, it provides 32-byte parallel byte scanning, 256-bit memory copying, and SIMD vector math for Java applications.

TECHNICAL HIGHLIGHTS:

  • 32-Byte Parallel Matching: Uses _mm256_cmpeq_epi8 and _mm256_movemask_epi8 for ultra-fast delimiter scanning.
  • 256-Bit Memory Engine: Accelerates off-heap buffer operations using AVX2 SIMD load/store instructions.
  • Cross-Platform Hardware Abstraction: Unified C++ header interface covering x64 (AVX2/AVX-512) and ARM64 (NEON).
  • JNI Interoperability: Zero-overhead JNI bridge via FastCore loader.

RESULTS:

  • SIMD Byte Search: Up to 15.3x speedup over standard Java loops
  • Throughput: Exceeds 100,000 ops/s on 100MB buffer searches

FASTJAVA SYSTEM REPOSITORIES:

  1. FastPointer: https://github.com/andrestubbe/FastPointer
  2. FastMemory: https://github.com/andrestubbe/FastMemory
  3. FastSIMD: https://github.com/andrestubbe/FastSIMD
  4. FastSharedMemory: https://github.com/andrestubbe/FastSharedMemory

Ecosystem Overview: https://github.com/andrestubbe/FastJava

Part of the FastJava Ecosystem — Making the JVM faster.
Made with ⚡ by Andre Stubbe

TAGS:
Java, FastJava, FastSIMD, SIMD, AVX2, Hardware Acceleration, Vector Math, C++, JNI, High Performance

Assets 3
Loading