GCC vs. Clang Compiler Benchmarking On Intel's Skylake CPU

[phoronix]

Phoronix: GCC vs. Clang Compiler Benchmarking On Intel's Skylake CPU

Continuing in our compiler benchmarks this week are some GCC vs. Clang C/C++ compiler performance benchmarks on Intel's new Skylake processor while testing from Ubuntu Linux 64-bit.

http://www.phoronix.com/vr.php?view=22129

[wargames]

GCC FTW!

[mlau]

GCC FTW!

yep, it's a very good compiler, and I don't care that it doesn't generate better code than clang on synthetic benchmarks.

[carewolf]

Love the back and forth, but it is really surprising to see so big differences both ways. The tests must be really one note. One single innerloop can easily be twice as slow, but you should be testing more than one innerloop in each test.

[Sahastra]

Michael should you not be adding -mcpu=skylake for GCC otherwise GCC doesnot optimise and builds binary for generic processor which naturally doesnot use full potentials of latest processors Whereas for Clang flag is there

Host CPU: skylake

[milkylainen]

GCC FTW!

Indeed. GCC is part of an amazing compiler suite. One thing that seldom gets mentioned is GCCs target capabilities.
It is by far GCC's best asset.
One compiler suite to rule them all.

[jake_lesser]

GCC's pounding CLANG all the way. GCC is still relevant and CLANG's the irrelevant one.

[geearf]

Indeed. GCC is part of an amazing compiler suite. One thing that seldom gets mentioned is GCCs target capabilities.
It is by far GCC's best asset.
One compiler suite to rule them all.

GCC is the collection
GCC stands for GNU Compiler Collection.
As for the benchmark, Michael, I think it would be nice to get compile time for each binaries per compiler as well. Forgetting debug messages, it's also one nice part of a compiler.

[jadrian]

Indeed. GCC is part of an amazing compiler suite. One thing that seldom gets mentioned is GCCs target capabilities.
It is by far GCC's best asset.
One compiler suite to rule them all.

There's more to a compiler suite (in the broader sense of the term) than just performance though. And clang appears to be further ahead in terms of static analysis capabilities.

[alaviss]

Michael configure llvm/clang with "--with-clang-default-openmp-runtime=libomp" to make clang generate code for libomp by default instead of libgomp (allowing to use -fopenmp normally without =libomp prefix)