LLVM Clang vs. GCC Compilers For AMD's Steamroller

Besides the interesting but disappointing AMD Kaveri Gallium3D vs. Catalyst Linux driver benchmarks published this morning, here's some more AMD A10-7850K "Kaveri" benchmarks for your Sunday viewing pleasure.

A more proper compiler comparison from the AMD APU with "Steamroller" CPU cores will come in the weeks ahead when more time allows, but it seems a lot of Phoronix readers are curious about the Steamroller performance and separately about the latest Clang C/C++ compiler improvements. I've spent some of today running some GCC and Clang compiler tests.

From an Ubuntu 13.10 x86_64 installation I tested GCC 4.8.2 as packaged in the "Saucy" archive followed by using Saucy's LLVM Clang 3.2 compiler followed by Clang 3.3. I also ran some LLVM Clang 3.4 and Clang 3.5 benchmarks using the LLVM.org Debian repository binaries. With my more comprehensive testing all compilers will be built locally from source and there will also be an even wider selection of code compilers benchmarked.

For those curious about these A10-7850K compiler benchmarks I uploaded all of the raw data to 1401197-PL-AMDKAVERI37 after running all of these tests under the Phoronix Test Suite.
There are some Clang performance improvements to note in some of the results to make it more competitive with the GNU Compiler Collection.
For some tests, Clang is still trailing GCC.
Clang easily wins some tests.
OpenMP support is still lacking in mainline LLVM/Clang.
The compile times for LLVM/Clang are still excellent.
There are some regressions.
Now check out the rest of the compiler results.