GCC 2010 Summit Presentations Now Online

[phoronix]

Phoronix: GCC 2010 Summit Presentations Now Online

For those interested in compilers, particularly GCC, or are interested in some technical slides to look over this weekend, the presentations from the 2010 GNU Compiler Collection Summit are now available online...

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

[curaga]

POV-Ray got 5% faster and 2.3% smaller by using LTO. Fairly impressive.

[markg85]

I wonder if GCC will have any future with the seemingly superior LLVM lurking around the corner?

@Michael when are you gonna make a new GCC vs LLVM benchmark series? Since i really would like to know the current state because LLVM 2.8 is supposedly (a lot?) faster then 2.7 thus probably catching up or even beating GCC..

Interesting note is that chromium is already faster on llvm-clang then on gcc : http://lists.cs.uiuc.edu/pipermail/c...er/011525.html (though that might be misleading since it's ran against gcc 4.2.1 (latest gplv2 gcc release))

[Michael]

@Michael when are you gonna make a new GCC vs LLVM benchmark series? Since i really would like to know the current state because LLVM 2.8 is supposedly (a lot?) faster then 2.7 thus probably catching up or even beating GCC.

I'm actually in the process of doing so right now.

[markg85]

I'm actually in the process of doing so right now.

Awesome!
Any idea when the article is coming online?

Thanx a lot

[Michael]

Testing should be done next week but it won't hit the publishing queue until probably the second week of November.

I also had GCC vs. LLVM tests on ARM I was going to do until the B3 overheated and seems to be dead

[markg85]

Testing should be done next week but it won't hit the publishing queue until probably the second week of November.

I also had GCC vs. LLVM tests on ARM I was going to do until the B3 overheated and seems to be dead

One testcase less for you to test

[XorEaxEax]

@Michael when are you gonna make a new GCC vs LLVM benchmark series? Since i really would like to know the current state because LLVM 2.8 is supposedly (a lot?) faster then 2.7 thus probably catching up or even beating GCC..

Ehh, where did you hear this? From my benchmarks and what I've read in the release notes there was precious little work on optimization in llvm 2.8

Please point to the source of this? From my tests llvm 2.8 is still slower than gcc 4.5 (code performance, not compiling) so have I missed some new optimization flag or what?

[XorEaxEax]

As for the GCC summit papers, interesting read on LTO and Firefox, especially the reason for the slow cold startup.

[smitty3268]

Ehh, where did you hear this? From my benchmarks and what I've read in the release notes there was precious little work on optimization in llvm 2.8

Please point to the source of this? From my tests llvm 2.8 is still slower than gcc 4.5 (code performance, not compiling) so have I missed some new optimization flag or what?

No, it's definitely still slower. The reasons to go with LLVM right now are better compile times, error messages, and general hackability. Performance still belongs to GCC.