NVPL TENSOR: Developer Guide and Reference

Welcome to the NVPL TENSOR library documentation.

NVPL TENSOR (NVIDIA Performance Libraries TENSOR) is part of NVIDIA Performance Libraries that provides tensor primitives.

NVPL TENSOR works on any 64-bit Arm based processors with Armv8.1-A or later architecture extension and is specifically optimized for:

• Arm Neoverse V2 based CPUs, such as NVIDIA Grace

• Arm Neoverse V1 based CPUs, such as Amazon (AWS) Graviton3

Key Features

• Support for up to 64-dimensional tensors.

• Arbitrary data layouts.

• Main computational routines:

  • Tensor contractions.

  • Tensor reductions (including partial reductions).

  • Element-wise tensor operations:

    • Arbitrary tensor permutations.

The documentation consists of three main components:

• A User Guide that introduces important basics of cuTENSOR including details on notation and accuracy.

• A Getting Started guide that steps through a simple tensor contraction example.

• An API Reference that provides a comprehensive overview of all library routines, constants, and data types.

Contents

• Release Notes
  • NVPL TENSOR 0.3.0
  • NVPL TENSOR 0.2.0
  • NVPL TENSOR 0.1.0
• User Guide
  • Nomenclature
  • Einstein Notation
  • Performance Guidelines
  • Accuracy Guarantees
  • Scalar Types
  • Supported Operators
  • Logging
• Getting Started
  • Headers and Data Types
  • Define Tensor Sizes
  • Initialize Tensor Data
  • Create Tensor Descriptors
  • Create Contraction Descriptor
  • Determine Algorithm and Workspace
  • Plan and reduce workspace
  • Execute
• API Reference
  • NVPL TENSOR Data Types
  • NVPL TENSOR Functions
• Software License Agreement
• Third Party License Agreements
  • HPTT

Indices And Tables

• Index

• Search Page