User forums > General (but related to Code::Blocks)
TDM's unofficial GCC 4.2.0 for MinGW (now with OpenMP, Fortran and Obj-C)
snija:
That`s not true, for people stick to science computation(like gfortran) and target to dual core and quad core cpu , it`s the most easiest way to utilize the resource you have.
Another important feature in the coming version is auto-vectorize .
Soultaker:
I tried to use this build, but unfortunately it's missing OpenMP support, which was the specific reason why I wanted to use GCC 4.2 under Windows. OpenMP is an important feature, in my opinion, especially since GCC I believe is the first free compiler to offer it on Windows. (Microsoft Visual C/C++ compiler has OpenMP support, but the free version does not include the necessary libraries.)
20-40 has posted a more complete build of GCC 4.2 here: GCC4.2; it has working OpenMP support (although I didn't test it extensively, the basics seem to work just fine). It's a lot older, but are there any other issues? Is there a reason to prefer your packages over this one, or vice versa?
Something else: I saw you needed a mirror, so I put the files on my S3 site:
* gcc-4.1.2-i386-pc-mingw32.7z
* gcc-4.1.2-i386-pc-mingw-src.7z
* gcc-4.2.0-tdm-1.7z
* gcc-4.2.0-tdm-1-src.7zFeel free to link to these on your website.
Soultaker:
Ah, in retrospect, that seems to be the one that I actually ended up using, not the older release.
So I guess it's just as recent as TDragon's, but with OpenMP support. I'll stick with that one then. Thanks!
Grom:
Recently I tested OpenMP in "MinGW" gcc 4.2.0. On dual core Xeone under WinXP I've got slower code, compare to non parallel version :shock:. With 2005 studio I've got 30% faster code. Probably threading was not don in efficient way...
Grom:
I've combined gcc 4.2.0 from two distributions. One of them had one piece of omp another one had second piece. After that I parallelized the main loop in my program. VC++ gave 30% on two CPUs, gcc nothing with full load of both CPUs. Seems to be the problem is the speed of switching between one CPU usage and 2 CPUs usage. Looks like behind omp was implemented a thread model with new delete for every one parallel subblock.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version