.rst .pdf

oneDTL Specification

The oneAPI Data Transformation Library (oneDTL) is an extensive library of ready-to-use, data compression functions that are highly optimized for diverse architectures. As a part of oneAPI, oneDTL is designed to allow execution on a wide variety of computational devices: CPUs, GPUs, and other accelerators.

1. Take advantage of vectorization and SIMD (single instruction, multiple data) instructions on CPU and GPU.

2. Take advantage of CPU threading and GPU SIMT (single instruction, multiple threads).

3. Improve the performance of computation-intensive applications with pipelines involving data compression and other operations provided as DPC++ kernel or oneAPI libraries.

• oneDTL Architecture
  • API Design
    • oneDTL Namespaces
    • Standard C++ Data Types
    • DPC++ Data Type Usage
  • Memory Model
  • Execution Model
    • Use of Queues
    • Device Usage
    • Asynchronous Execution
    • Host Thread Safety
  • Error Handling
    • Exception Classification
      • Common Exceptions
  • Other Features
    • Current Version of this oneDTL Specification
    • Pre/Post Condition Checking
• Data Compression
  • Lossless Compression
  • Lossy Compression
    • ZFP
      • oneDTL Specific
      • Chunks
      • Index Array
      • Contents

next

oneDTL Architecture