updated 11:00 pm EST, Thu November 12, 2009
New version supports up to 256 cores
The Portland Group has announced the 2010 version of its PGI compilers and development tools. The new release is the first to fully support the PGI Accelerator Programming model 1.0 standard for x64 processor systems that use NVIDIA CUDA-enabled GPUs. CUDA Fortran, which is the primary GPU interface, is also now included to give developers direct control of NVIDIA GPUs.
The PGI 2010 release also adds GPU device-resident data, which allows developers to define and leave data on the GPU across accelerator regions. The new version now supports Complex and Double Complex Fortran data types, C structs, Fortran derived types and nested accelerator regions within OpenMP parallel regions.
Other new features in the 2010 update include additional support for Fortran 2003 incremental functionality, the newest EDG 4.1 C++ front-end, OpenMP parallel programming for up to 256 cores, and AVX code generation. A major update to the PGPROF performance profiler enables profiling of binary executables.
PGI 2010 will support Mac OS X 10.6 Snow Leopard and Windows 7, with availability beginning Tuesday, Nov 17.