GCC & LLVM Clang Performance On The Intel Atom

On all three x86_64 Linux systems in our earlier GCC/Clang performance article the GraphicsMagick resizing performance improved remarkably between GCC 4.3 and 4.4 and it remained that way up through the most recent GCC 4.6 snapshot at the time. However, with the Intel Atom i686 netbook, the performance was virtually unchanged between GCC 4.2.4 and 4.6.0-20101120 and it was not any different (for good or bad) with Clang using LLVM 2.8.

The image sharpening performance with GraphicsMagick on the Atom netbook was hard to gauge due to the slow pace at which it handled the sample photographs.

Clang on the Atom netbook had not worked for the timed compilation test profiles, but with the timed Apache compilation test the time it took to build did not differ much between GCC 4.2 and 4.5.1. It did, however, drop with GCC 4.6. This is even after building GCC 4.6 with the --enable-checking=release argument at the request of Jan Hubicka, one of the SUSE/Novell compiler developers.

GCC 4.6 was still slower at building ImageMagick.

Next up in our compiler tests will be a look at GCC and LLVM/Clang on the ARM architecture as we attempt to try out these open-source compilers on a Nokia N900. Ideally, this will be done without any cross compiling, but so far, there have been some issues on the Nokia N900 smart-phone.