A Linux Compiler Deathmatch: GCC, LLVM, DragonEgg, Open64, Etc... | Hacker News Reader

A Linux Compiler Deathmatch: GCC, LLVM, DragonEgg, Open64, Etc...

phoronix.com

35 pointsbconway10 comments

Link to the results:

http://global.phoronix-test-suite.com/index.php?k=profile...

Scroll down for bar graphs of random polarity with occasional spurious entries using a pittance of available visual coding.

I think giving each compiler a hue and encoding the optimization level as saturation would have been a useful addition to the spatial organization. Rephrasing the data so that "big = good" would also save much of the mental "what am I looking at?" diversion at each chart view.

If only there was some sort of Standard Performance Evaluation Corporation out there with a wide variety of CPU intensive benchmarks to evaluate architectures and compilers.

Not getting a SPEC license makes this strictly amateur hour. And yes, I know compilers are tuned for SPEC, but you'd at least get repeatable high-quality results on a wide range of different programs.

SPEC isn't necessarily representative of every program out there, but his weird (and small) collection of microbenchmarks and random fragments of code is even less interesting than SPEC.

Is there some actual content on the middle of all these ads? I gave up on page 2.

Some points that appear to come out of the data:

tcc (tiny c compiler) appears to be really good - it scores very well in both compilation time and execution speed

open64 (added in second batch of tests) also appears very good, for the same reasons

clang is quite slow.

Programs compiled with gcc -O2 often run faster than with -O3.

I think you're referring to another benchmark. AFAICS, this one doesn't test clang or tcc.

From the main article link I clicked through to the forum thread[1] where the tester posted his benchmarks. There's more than one there as he progressively tested different compilers and different setups; I looked at the first 2-3.

[1] http://phoronix.com/forums/showthread.php?29090-CompilerDeat...

The forum thread on page 3[1] however reads:

  side-by-side comparisons do however indicate that Clang performed as good or better than ICC or GCC when -O2 is used!

And through the link by yread you can see that clang really performs very well overall.

[1]: http://phoronix.com/forums/showthread.php?29090-CompilerDeat...

tcc is seriously awesome. Read the source - this is how C compilers should be written. It can add bounds checking to your programs. It's small enough to fit into a library you can use for dynamic code generation.

IMO tcc with support for more architectures is just as viable an alternative to gcc as clang is.

They're testing on an Atom platform, which is pretty horrible if you're comparing performance.

Many of the optimizations that compilers can do nowadays rely on better than an in-order "barely better than a Pentium" CPU.

Reply on news.ycombinator.com