TeraXLang (formerly Triton Xtra Language)

TeraXLang is a Triton-based language extension that provides additional features for GPU programming, including:

• Shared Memory Operations: Efficient smem/tmem operations with fragment-based loading/storing
• TMA (Tensor Memory Accelerator): Hardware-accelerated memory operations
• Warp-level Primitives: Warp-level reductions and synchronization
• Mbars (Mailboxes): Fine-grained thread synchronization

API Reference

• Core API - Low-level TeraXLang primitives
• JIT API - JIT compilation utilities

Tools

• IR Viewer - Interactive binding viewer for TTIR/TTGIR/PTX and Python source