An OpenMP* Transition for CPUs and GPUs

In the high-performance computing (HPC) realm, OpenMP* is among the most popular parallel programming models for AI and HPC applications and hardware accelerators (including GPUs). It also provides a portable solution to unpack the possibilities of our CPU to GPU offloading model.

In this session, senior principal engineer and compiler architect Xinmin Tian discusses:

• A high-level comparison on the Intel® C++ Compiler Classic vs. Intel® C++ Compiler drivers on OpenMP versions 4.5, 5.0, and 5.1 on CPUs
• New features in Intel C++ Compiler, including:
  • OpenMP Single Instruction Multiple Data (SIMD) for GPUs
  • Unified shared memory (USM)
  • A declare mapper, class member functor, and function pointer in the offload region
  • Asynchronous offloading
• Migrate the Intel C++ Compiler optimization report from Intel C++ Compiler Classic

To access the Intel® oneAPI DPC++/C++ Compiler and more:

• Download the Intel oneAPI DPC++/C++ Compiler as part of the Intel® oneAPI Base Toolkit. This is a foundational set of 15 tools and libraries for developing high-performance, data-centric applications across CPUs, GPUs, FPGAs, and other accelerators. More Information
  
• Intel® Parallel Studio XE or Intel® System Studio users can migrate to the next generation of these powerful suites. Each one includes the Intel oneAPI DPC++/C++ Compiler. More Information