Linked by Thom Holwerda on Wed 15th Jun 2011 14:23 UTC, submitted by Valhalla
General Development "PathScale announced today that the EKOPath 4 Compiler Suite is now available as an open source project and free download for Linux, FreeBSD and Solaris. This release includes documentation and the complete development stack, including compiler, debugger, assembler, runtimes and standard libraries. EKOPath is the product of years of ongoing development, representing one of the industries highest performance Intel 64 and AMD C, C++ and Fortran compilers." More here.
Thread beginning with comment 477394
To view parent comment, click here.
To read all comments associated with this story, please click here.
Member since:

Is this wall-clock or cpu-time?

Wall clock time

Your timings are meaningless without details!
What compiler versions? What compiler options?

These weren't with super aggressive optimizations but with "safe" optimizations. Trying to do things like -Ofast on most compilers will often break the numerical accuracy of the LAPACK code so anything over O2 is recommended to be avoided.
gfortran 4.4.0 : -O2 -march=native
ifort 12.0 : -O2 -xHost
pgif90 13.1 : -O2 (pgi does the equiv. of march=native by default)
pathf90 4.0.10 : -O2 -march=native

With pathf95, built from today sources, I
cannot compile lapack from netlib without
hitting an internal compiler error.

Perhaps an issue to take up with PathScale and your particular setup but two of the nightly LAPACK svn builds are with the previously linked to 4.0.10 nightly build from 2 days ago.

This, of course, means that the execution time of any lapack benchmark cannot be obtained.

Please note that this is not designed to be a benchmark but simply the test suite for the code base. It was just one example of code that I've seen and worked with that shows great gains from the PathScale compiler.

And for those who might be wondering, this is not using an optimized BLAS imlementation but instead the netlib fortran BLAS was built as part of the LAPACK build. On larger datasets and matricies building against an optimized BLAS (like goto, atlas, mkl, or acml) can make a big difference but for the small data sizes in the software test suite the difference is not significant.

Reply Parent Score: 2