A Linux Compiler Deathmatch

[Jimbo]

so do you have a suggested set of flags I should use for ICC in my final round of tests? I am about to have the -Os round finished (only open64 left to run) and after that I was going for a round of compiler-specific and supposedly optimized flags.

I would suggest to create a set of profiles for each compiler and test them. Something like this:


ICC version 12 (parallel studio xe 2011)
most agresive flags: -fast -no-prec-div -fp-model fast
more agresive flags: -O3 -xHOST -ip -no-prec-div -fp-model fast
agresive flags: -O3 -xHOST -ip
normal O3: -O3
normal O2: -O2
normal Os: -Os

open64 4.2.3
most agresive flags: -ofast
more agresive flags: -O3 -ffast-math -lno -apo -mso -march=wolfdale
agresive flags: -O3 -lno -apo -mso -march=wolfdale
normal O3: -O3
normal O2: -O2
normal Os: -Os

GCC 4.5.2
most agresive flags: -O3 -ffast-math -march=native -mtune=native -fexcess-precision=fast -ftree-parallelize-loops=5 -floop-parallelize-all -floop-interchange -floop-strip-mine -floop-block -flto
more agresive flags: -O3 -ffast-math -march=native -mtune=native -fexcess-precision=fast -ftree-parallelize-loops=5 -floop-parallelize-all -floop-interchange -floop-strip-mine -floop-block -flto
agresive flags: -O3 -march=native -mtune=native -ftree-parallelize-loops=5 -floop-interchange -floop-strip-mine -floop-block -flto
normal O3: -O3
normal O2: -O2
normal Os: -Os


If u are not in the mood to do alll this , I would suggest to try more agresive flags or agresive flags profile.

note1: Most agresive flags and more agresive flags perform not acurate IEEE maths operations, but this should be fine for most programs.
note2: This is my personal choice, I only toyed a little with GCC vs ICC in the past, so those flags could be wrong or not optimal.

[staalmannen]

If u are not in the mood to do alll this , I would suggest to try more agresive flags or agresive flags profile.

note1: Most agresive flags and more agresive flags perform not acurate IEEE maths operations, but this should be fine for most programs.
note2: This is my personal choice, I only toyed a little with GCC vs ICC in the past, so those flags could be wrong or not optimal.

Heh.. I was basically planning on doing just a single round more after the -Os one... so I will go for the "most aggressive" choice :P

Got some tips about how to format the data nicer from mtippet, which I will try for the -Os merge.

[Jimbo]

... so I will go for the "most aggressive" choice :P

Nice! this should be the interesting round!!

Enable -flto on GCC is a little tricky, it seems that your LDFLAGS should be the same that CFLAGS, so you should edit the makefile and copy paste the CFLAGS flags into LDFLAGS, but I am not really sure at all about -flto flag, sorry.

[Jimbo]

open64 4.2.3
most agresive flags: -Ofast -ffast-math -lno -apo -mso -march=wolfdale

If most aggressive choice doesn't compile , change to more aggressive choice.

[staalmannen]

Community CompilerDeathMatch updated with -Os

A new merge with the -Os results are hereby presented:

http://global.phoronix-test-suite.com/index.php?k=profile&u=staalmannen-231-8340-1312

Enjoy!

Next up: Aggressive and compiler-specific settings (are there any tweaks that can be made to TCC?).

[staalmannen]

Intermediate update, specific optimizations.

The final round of the CompilerDeathMatch has started.

An intermediate update can be found here:

http://global.phoronix-test-suite.com/index.php?k=profile&u=staalmannen-26463-30048-18506

Changes:
- Optimization for PCC (suggested by Anders Magnusson, PCC mantainer)
- the suggested most aggressive optimizations for ICC and GCC
- SolarisStudio on default settings

The SolarisStudio is a late commer to the party, but will hopefully catch up soon.

[staalmannen]

Final results in new thread

I have put the final results for discussion in a new thread