LibraryActive

CUDA Tile C++

Tile programming extension to CUDA C++ for intra-block parallelism via elementwise tile operations.

Open source page ↗

Field note

What it does

CUDA Tile C++ provides high-level tile abstractions that the compiler auto-parallelizes across threads, leveraging TMA and Tensor Cores for performance portability across NVIDIA GPU architectures.

Capabilities

Available capabilities

Performance portability across GPU architecturesAuto-parallelizationTile programmingTensor Core utilizationTensor Memory Accelerator utilization

Tags

Tags

CUDAC++GPU

Ways to use it

Ways to use it

sdk

C++ / https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html

Product features

Product features

1. GDS cuFile API Reference #

2. Introduction #

3.1. Dynamic Interactions #

3.2. Driver, File, and Buffer Management #

3.3. cuFile Compatibility Mode #

3. Usage #

4.1.1. Declarations and Definitions #

4.1.2. Typedefs #

4.1.3. Enumerations #

4.1. Data Types #

4.2. cuFile Driver APIs #

4. cuFile API Specification #