A summary of new features in each version of the PGI 2017 release.
PGI releases version update, as often as monthly, which may include both new features and bug fixes. The new features added with each version are listed below. Bug fixes are listed on the PGI Release 2017 Technical Problem Reports summary page.
Complete details on the new features listed here are included in the PGI Release Notes for x86-64, PGI Installation Guide and Release Notes for OpenPOWER, and the PGI Visual Fortran® Release Notes.
PGI 2017 version 17.10 (released November 2, 2017)
- PGI Community Edition release
- Integrated support for the production release of CUDA Toolkit 9.0
- CUDA 9.0 cooperative groups feature added to CUDA Fortran
- Added OpenMP 4.5 taskloop construct and all its clauses except firstprivate and lastprivate. OpenMP 4.5 supported on OpenPOWER and the x86/LLVM beta releases only.
- Updated PGI-built Open MPI libraries to version 2.1.2 for Linux
- Support for macOS High Sierra 10.13
- Plus 18 user-requested enhancements and fixes.
PGI 2017 version 17.9 (released September 27, 2017)
- PGI Accelerator Features and Enhancements
- Integrated support for the CUDA Toolkit 9.0 Release Candidate (RC). CUDA 7.5 remains the default.
- Plus 27 user-requested enhancements and fixes
PGI 2017 version 17.7 (released August 8, 2017)
- PGI Accelerator Features and Enhancements
- OpenACC and CUDA Fortran now support Tesla V100 (Volta) GPUs. Added initial support for the CUDA 9 toolkit.
- Added support for CUDA Unified Memory; new pool allocator.
- Beta support for automatic deep copy of Fortran derived types in OpenACC
- Now call optimized cuSolverDN routines from CUDA Fortran and OpenACC Fortran using the PGI-supplied interface module and cuSOLVER libraries
- All PGI Compilers
- Beta LLVM code generator and OpenMP runtime for the PGI Linux/x86-64 compilers include initial support for OpenMP 4.5 syntax and features on multicore CPUs
- Improved inlining with the -Minline option.
- Program Analysis Summary Output (PASO) feature enables exporting of information gathered from various compilation stages in an organized, machine readable format.
- C++ Compiler
- Added incremental support for C++17 features.
- Added support for lambdas with capture in OpenACC GPU-accelerated compute regions.
- Improved LCALS performance.
- GNU 6.3 interoperability.
- PGI C++ compilers supported as a CUDA 9.0 NVCC host compiler.
- New CUDA 9 features include:
- Enhanced unified memory profiling
- OpenACC details table
- Support for cooperative kernel launches
- NVLink events
- OpenACC profiling on all multicore systems
PGI 2017 version 17.5 (released May 30, 2017)
- CUDA Toolkit — updated nvlink linker utility to address an issue with the link ordering of object files on Linux x86-64.
- Expanded Operating System support to include Ubuntu 17.04 and Redhat Enterprise Linux 6.9 distributions.
- Plus 12 user-requested enhancements and fixes
PGI 2017 version 17.4 (released April 30, 2017)
- PGI Community Edition release including first release for Microsoft Windows.
- PGI Accelerator Compilers
- Atomic add and subtract support of Fortran single precision complex data types in device code.
- Device support for floating point classification macros isfinite, isnan, isinf, and round function.
- Added support for CUDA Fortran warp-vote operations.
- Changed the impact of -g on the optimization level used when compiling device code.
- Updated the version of nvlink to 8.0.73.
- PGI Tools
- Improved debugging support of C++ programs on recent macOS versions.
- Plus 20 user-requested enhancements and fixes
PGI 2017 version 17.3 (released March 24, 2017)
- First PVF release for 2017
- Added debugger support for macOS version 10.12 (Sierra).
- Updated the PGI C/C++ compiler plug-in for Eclipse versions Neon and Mars; the Eclipse plug-in is available for Linux x86-64 and OpenPOWER platforms.
- Plus 34 user-requested enhancements and fixes
PGI 2017 version 17.1 (released February 10, 2017)
- Comprehensive OpenACC 2.5 support; improved cache directive performance
- PGC++ C++14 support; GCC 5.1 through 6.2 interoperability; improved exception handling
- PGC++ as an CUDA host compiler now supports C++11
- Support for CUDA 8 and Tesla P100
- Improved inlining
- Additional tuned numerical math intrinsic functions for Haswell and Broadwell CPUs
- Updated pre-compiled MPICH, MVAPICH and Open MPI libraries
- Updated pre-compiled netCDF, parallel netCDF and ESMF libraries
- Added variable rollover support to debugger GUI; updated Eclipse plug-ins
- Expanded operating system support
- Dropped support for 32-bit, GPUs on macOS, CUDA 7.0, macOS Lion/Mountain Lion
- Plus 24 user-requested enhancements and fixes
See also the PGI 2017 New Release Information.