Исправления:
https://www.pgroup.com/support/release_tprs_2011.htmPGI 2011 version 11.9•
CUDA-x86 code generator now produces code that is optimized for the latest x64 platforms.
• Support is available for
AMD’s Bulldozer based processor. The option
-tp now supports the suboption
bulldozer.
• Hardware count-based
profiling support for both Sandybridge and Bulldozer based processors is available.
Use
pgcollect -hwtimePGI 2011 version 11.8• PGI 11.8 includes an update of Cygwin to version 1.7.9-1.
• PGI 11.8 includes support for the Mac OS X Lion operating system.
•
PGI Accelerator compilers and
CUDA Fortran include these modifications and enhancements:
• CUDA Fortran supports the option
-mcmodel=medium.
• PGI provides a module which defines interfaces to the CUBLAS Library from PGI CUDA Fortran. These
interfaces are made accessible by placing the following statement in the CUDA Fortran host-code
program unit:
use cublas__________________________________________________________________________________________________
PGI 2011 version 11.7•
PGI compilers — 256 bit wide vector intrinsics called for AVX-enabled targets; other AVX optimizations.
•
CUDA Fortran — CUDA 4.0 API updates; updated CUDA Fortran Programming Guide and Reference.
•
Fortran 2003 — added support for Final Procedures.
•
PGDBG® Debugger — new Call Stack display and capabilites; added support for up-arrow and command line history.
• Plus 17 user-requested enhancements and fixes
PGI 2011 version 11.6•
PGI compilers — support for Intel AVX-enabled Sandy Bridge processors.
•
PGI Accelerator™ compilers — support for calling subroutines or CUDA kernels using device-resident data.
•
CUDA Fortran — support for shared memory automatics in device code, limited support for printing from device code, and added a module for using the NVIDIA CUBLAS library.
•
CUDA-x86 — emulated version of the CUBLAS library.
•
Fortran 2003 — support for sourced allocation.
•
PGI C++ compiler — significant improvements in the treatment of throwing and non-throwing functions.
• Plus 19 user-requested enhancements and fixes.
PGI 2011 version 11.5•
PGI Accelerator C/C++ — added support for compiling CUDA C/C++ to run on x86 targets with or without an NVIDIA GPU accelerator. This is a functional release; x86 performance has not been optimized (due later in 2011). CUDA features not supported in this release include:
• no support for the CUDA driver API (no plans)
• no support for texture memory (Q4, 2011)
• no support for OpenGL interoperability (to be determined)
• no support for warp-synchronous programming (no plans)
• no support for CUBLAS and CUFFT libraries (due in 11.6)
• no support for the Thrust C++ library (to be determined)
• Plus 24 user-requested enhancements and fixes.
PGI 2011 version 11.4•
PGI Accelerator — expanded support for libm math routines and added support for most of the NVIDIA device built-in routines, added the ability to access the GPU timers during execution and added new routines to simplify allocating memory in C. Using the new deviceptr( ) clause, this memory can be used within PGI Accelerator regions with no implicit data movement.
•
PGI Visual Fortran — support for Visual Studio 2010 SP1.
•
PGPROF® Profiler and PGDGB Debugger — support user-specific preferences, standardized menus, updated names and better implementation for choosing and setting fonts. PGDBG debugger also includes a new docking framework that allows tabs to be moved and "torn off" into separate windows.
• Plus 19 user-requested enhancements and fixes.