GCC 4.8.2 pulled ahead of LLVM Clang 3.4 when it came to the C-Ray multi-threaded ray-tracer. Over the past few GCC releases the developers have been doing some tuning optimizations that particularly benefit C-Ray.
GCC also had a slight advantage over Clang when it came to the FLAC and MP3 audio encoding performance.
Our Tegra K1 compiler benchmarking comparison ended today with the basic Apache web-server test where the performance was close to the same between Clang 3.4 and GCC 4.8.2.
Overall, Clang performed quite admirably on the Tegra K1 quad-core Cortex-A15 SoC. Generally, Clang was either generating binaries that ran at the same speed as the GCC binaries or better. There were only a few cases of Clang being beaten out by GCC by a wide margin and then of course Clang's current lack of mainline OpenMP support. The 32-bit ARM performance out of LLVM Clang 3.4 is very good, which should be expected given that many ARM vendors in particular have been investing in the LLVM stack. Coming up next when it comes to ARM compiler benchmarks will be tests of built-from-source LLVM Clang 3.5 SVN and GCC 4.9.0.
If you appreciate the extensive benchmarking done at Phoronix for covering important open-source and Linux areas of performance, please consider subscribing to Phoronix Premium.