For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo module

tma

NVIDIA Tensor Memory Accelerator (TMA) module.

Provides types and functions for working with NVIDIA's Tensor Memory Accelerator, which enables efficient asynchronous data movement between global and shared memory on GPUs with Hopper architecture and newer.

The TMA hardware provides hardware-accelerated multi-dimensional memory copies with features like swizzling for bank conflict avoidance, L2 cache promotion hints, and support for various data types and memory layouts.

Structs

• ​TensorMapDataType: Data type enumeration for TMA tensor map descriptors.
• ​TensorMapFloatOOBFill: Out-of-bounds fill mode for floating-point TMA operations.
• ​TensorMapInterleave: Interleave mode for TMA tensor map descriptors.
• ​TensorMapL2Promotion: L2 cache promotion hint for TMA tensor map descriptors.
• ​TensorMapSwizzle: Swizzle mode for TMA tensor map descriptors.
• ​TMADescriptor: TMA tensor map descriptor.

Functions

• ​create_tma_descriptor: Creates a TMA descriptor for tiled memory operations.
• ​prefetch_tma_descriptor: Prefetches a TMA descriptor into the constant cache.