Contents

• 1. Overview
  • 1.1. System Prerequisites
  • 1.2. Prepare Your System
  • 1.3. Supporting Documentation and Examples
• 2. Using OpenACC with the NVIDIA HPC Compilers
  • 2.1. CUDA Versions
  • 2.2. Compute Capability
  • 2.3. PTX JIT Compilation
  • 2.4. C Structs in OpenACC
  • 2.5. C++ Classes in OpenACC
  • 2.6. Fortran Derived Types in OpenACC
  • 2.7. Fortran I/O
    • 2.7.1. OpenACC PRINT Example
  • 2.8. OpenACC Atomic Support
  • 2.9. OpenACC Declare Data Directive for Global and Fortran Module Variables
  • 2.10. OpenACC Error Handling
  • 2.11. C Examples
  • 2.12. Fortran Examples
    • 2.12.1. Vector Addition on the GPU
    • 2.12.2. Multi-Threaded Program Utilizing Multiple Devices
  • 2.13. Troubleshooting Tips and Known Limitations
• 3. Implemented Features
  • 3.1. OpenACC specification compliance
  • 3.2. Defaults
  • 3.3. Environment Variables
  • 3.4. OpenACC Fortran API Extensions
    • 3.4.1. acc_malloc
    • 3.4.2. acc_free
    • 3.4.3. acc_map_data
    • 3.4.4. acc_unmap_data
    • 3.4.5. acc_deviceptr
    • 3.4.6. acc_hostptr
    • 3.4.7. acc_is_present
    • 3.4.8. acc_memcpy_to_device
    • 3.4.9. acc_memcpy_from_device
    • 3.4.10. acc_get_cuda_stream
    • 3.4.11. acc_set_cuda_stream
  • 3.5. Known Limitations
    • 3.5.1. ACC routine directive Limitations
    • 3.5.2. C++ and OpenACC Limitations
    • 3.5.3. Other Limitations
  • 3.6. Interactions with Optimizations
    • 3.6.1. Interactions with Inlining