關於支援OPenACC的編譯器說明

cuda_study發表於2018-03-06
編譯
很多開發者都在關注支援OPenACC的編譯器都有哪些，下面給大家彙總一下，share給大家。

關於支援OPenACC的編譯器說明
支援OpenACC的編譯器：PGI, Cray, CAPS 和 GCC。下面是 gcc 對 OpenACC 支援的介紹。
All – GCC 6.1 was released withimproved OpenACC support – see below for key OpenACC highlights.  
PGI is still NVIDIA’s go-tocompiler for OpenACC.
GCC 6.1’s improved supportfor OpenACC is an important step in making the standard pervasive.
Noteworthy as well, GCC 6.1C++ default is now C++14 (-std=gnu++14 instead of -std=gnu++98)
GCC 6.1 OpenACC highlights
·           GCC 6.1 wasreleased 04/27 with improved support for OpenACC:
o    Standard supported: OpenACC 2.0a
o    Compilers: C, C++ and Fortran
o    Accelerators: NVIDIA GPUs
o    Hosts:
§  x86
§  Power8 support is broken in GCC 6.1 – though a functional March snapshot isavailable from Mentor Graphics
·           OpenACC 2.0asupport status in GCC 6.1:
o   Read the cautions and capabilities statedhere: https://gcc.gnu.org/gcc-6/changes.html
o   Recommend using parallel constructs ifpossible:
§  Additionally,specify explicitly number of gangs, workers, and vector length clauses forbetter performance
o   Avoid kernels constructs as there are performance andfunctionality limitations
·           How to get GCC 6.1?
o   As of end of April 2016, GCC 6.1 is not yetadopted in new Linux distributions
o   Fedora 24 (ETA June 2016) will adopt GCC6.1 as a default compiler, it is anticipated (not POR) that Ubuntu 16.10 willadopt GCC 6.1 or later version
o   GCC 6.1 may be built from source (see https://gcc.gnu.org/wiki/Offloading)         
·         GCC OpenMP 4.5status:
o    The OpenMP 4.5 specification is fullysupported in GCC 6.1 for the C and C++ compilers
§  thecompiler can be configured for OpenMP offloading to Intel XeonPhi KnightsLanding and AMD HSAIL
§  Supportfor NVPTX is under development in the gomp-nvptx branch, not supported in GCC6.1.
o    GCC can now generate HSAIL (Heterogeneous System Architecture IntermediateLanguage) for simple OpenMP device constructs
·          OpenACCsupport beyond GCC 6.1 – future work for GCC 7:
o   Performance improvements
·       ImproveGCC`s parallelization of inner loops
·       Defaultbehavior of number of gangs, workers and vector length for parallel constructs
o   OpenACC 2.5 support
原文釋出時間為：2016-6-24 12:51:06
原文由：nvadmin 釋出，版權歸屬於原作者 
本文來自雲棲社群合作伙伴NVIDIA，瞭解相關資訊可以關注NVIDIA官方網站


相關文章