LLVM performance test

Hi all,

I did a performance test of two real applications (FFMPEG and GTK) on
ARM. For more details see:

Lauro

This is pretty awesome. Thanks Lauro!

Evan

Hi Lauro,

Hi all,

I did a performance test of two real applications (FFMPEG and GTK) on
ARM. For more details see:
LLVM performance tests | Lauro

Could you give me some more information about the applications you compiled and ran? Where can I obtain source codes, which input did you provide, ... These two seem interesting for some future work of mine (comparing real applications that are used by normal users with industry standard benchmarks like SPEC CPU).

Tips for additional applications are also welcome :slight_smile:

greetings,

Kenneth

This was meant for Lauro only, my apologies.

Hi,

I have two related questions: Are there LLVM developers on this list that are
interested in improving LLVM performance? I have examples in which LLVM is
slower than gcc (e.g., STM linked list microbenchmark, 1.8 (gcc) vs. 1.3
(llvm), more is better). I've never done compiler performance tuning, so I
don't know how much effort is required.

Second question: does anyone keep track of performance results?

Thanks,
Torvald

Benchmark info is submitted by the nightly testers.

http://llvm.org/nightlytest/

— Gordon

Hello, Torvald.

I have two related questions: Are there LLVM developers on this list that are
interested in improving LLVM performance?

Yes, surely.

I have examples in which LLVM is slower than gcc (e.g., STM linked list microbenchmark,
1.8 (gcc) vs. 1.3 (llvm), more is better).

Please provide link to benchmark.

Second question: does anyone keep track of performance results?

This is usually done via nightly tests.

Yes absolutely. If you can narrow down the example to a simple testcase and attach that to a bugzilla, that would be wonderful.

-Chris

Benchmark info is submitted by the nightly testers.

http://llvm.org/nightlytest/

Hmm, I could not really see benchmark info there. For example,
when I look at
http://llvm.org/nightlytest/machine.php?machine=153, where is
this info?

Maybe it's implicitly in the column "dejagnu time wall", because
this amount has decreased from 2007-08-09 to 2007-08-19 from 200
seconds to 180 seconds. But that would be a lousy benchmark ...

Raw data is here:
http://llvm.org/nightlytest/fulltest.php?machine=153&night=3458

“Significant” changes are hilighted on this page:
http://llvm.org/nightlytest/test.php?machine=153&night=3458

Be patient; they take a while to load.

— Gordon