Next: download, Previous: about, Up: Introduction [Contents][Index]
For up-to-date information on supported platforms, please consult the platforms area on the Blitz++ home page:
http://oonumerics.org/blitz/platforms/
The information in this document may be out of date.
Blitz++ was developed and tested using KAI C++ under AIX. It should (in theory) port to other KAI C++ platforms (Cray, SGI, HP, Sun, Linux, Compaq) without difficulty. Since KAI C++ uses an EDG front end, other EDG front-ended compilers (e.g. Comeau) should be able to compile Blitz++.
Recommended compile flags are:
+K3 -O2 --restrict --abstract_pointer --abstract_float -tused
Note that you cannot compile with -tall (this will generate lots of
errors).
Under Linux, you may need the flag -D__signed__=. You should omit
-tused since this template instantiation model is not supported by
gcc, which is used as the back-end compiler.
Please note that since the purchase of KAI by Intel several years ago, the
use of this compiler has been gradually phased out, with most of the KAI
compiler technology being transferred into the Intel C/C++ compiler. Thus,
Blitz++ is no longer being routinely tested under KCC.
Blitz++ compiles under fairly recent versions of the Intel C++ compiler
(version 7.1, 8.x or 9.0) for Linux platforms, as well as the comparable
plug-in compiler for Windows that can be used within the Microsoft
Visual Studio IDE. We recommend use of the -ansi flag for compiler
adherence to the ANSI C++ standard. The -strict_ansi flag is not
used, as this can lead to many compiler error messages originating from
code in the GNU C++ standard library header files.
More information:
http://www.intel.com/software/products/compilers/clin
http://www.intel.com/software/products/compilers/cwin
Blitz++ has been ported to the C++ compiler within the Microsoft VS.NET 2003 compiler and IDE package. We provide a zip archive containing an appropriate configuration header file and project files for building the Blitz library and all of the testsuite codes. Previous versions of the Microsoft C++ compiler within Visual Studio do not have the required C++ features needed by Blitz++ and are not supported. Blitz can be compiled under Visual Studio by using the Intel plug-in C++ compiler for Windows.
GCC (g++) is a free GNU C++ compiler. It compiles Blitz++
reliably on almost any platform running Linux or another Unix variant
(including the Darwin OS). In fact, most Blitz++ development work is done
with g++.
Work has been done in the latest Blitz++ release to ensure compatibility with the gcc-4.x release, which contains some key changes in enforcement of the C++ ANSI standard related to the use of anonymous enums.
gcc may be downloaded from http://www.gnu.org/software/gcc/gcc.html.
If you are using gcc under Solaris, SunOS, or OSF/1, please see the README.binutils file included in the distribution.
We have just added support for the PathScale pathCC compiler in blitz-0.9. This compiler is being provided on many of the newer Opteron x86_64 systems. We welcome feedback on support for this compiler.
The Portland Group PGI compiler is supported on Linux x86 platforms, and more recent versions of pgCC such as 5.x are capable of compiling blitz. However, performance is not particularly good and PGI support is not being actively tested or maintained.
Absoft markets a variant of the IBM xlC compiler called xlc++ that
has been ported to the Darwin operating system (Mac OS X). We are providing
support for compiling blitz using this compiler. Although we have noted
and reported some bugs with this compiler, it is capable of producing fairly
high quality optimized code for the Darwin platform.
Metrowerks is sort-of supported; see the platforms web page and the mailing lists for more information. Support for Metrowerks is no longer being actively maintained.
The Compaq C++ compiler version 6.x is supported, and we recommend use of
the -model ansi flag in order to obtain standard C++ compiler behavior
and proper name mangling. We have inserted a workaround for the lack of
standard ostream support for the long double type. At this point,
performance of blitz code under the cxx compiler is somewhat disappointing.
The IBM Visual Age C++ compiler is supported as of version 6.x. However, you may run into a variety of annoying compiler bugs that prevent certain portions of the Blitz++ library code from compiling. Therefore, it is strongly recommended that you obtain xlC version 7.0.0.3 or later, in order to get all the necessary patches that IBM has produced in response to our bug reports. Also note that compile times can be fairly long with full optimization enabled.
The SGI MIPSpro-7.x version CC compiler is supported, although active testing is no longer being done.
As of version Sun Studio 10.x, the Sun CC compiler should be capable of compiling blitz. However, no active support is being provided.
As of Version 0.2-alpha-02 of Blitz++, Version 3.0.0.0 of the Cray C++
compiler is supported (well, tolerated anyway). It seems to be based on an
older version of the EDG front end, so some kludges are required. It
doesn’t support partial ordering of member templates, so slicing arrays
requires the workaround described in Section Array slicing. Portions
of the standard library are missing, such as <limits>,
<complex>, and <set>. This means you won’t be able to use
complex numbers (well, not the ISO/ANSI C++ versions anyway), numeric
inquiry functions, or fast traversal orders.
These compilation flags are recommended:
-h instantiate=used
For optimization, you’ll want:
-O3 -h aggress
The ability of the Cray C++ compiler to optimize away temporary objects is disappointing. It’s not able to optimize away expression templates overhead or comma-delimited array initializers. Please note that support for compiling Blitz++ under the Cray C++ compiler is no longer being actively maintained.
This platform and compiler are no longer being actively supported.
Next: download, Previous: about, Up: Introduction [Contents][Index]