Arch Linux

el mariachi

Member

Registered: 2007-11-30

Posts: 595

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

@codestr0m: Thanks for all the info Sorry if I sounded ungrateful or something, I'm not. I would love to help, but since I'm no developer and I don't know any language (besides the hackish stuff I do to DWM), I feel I won't contribute much by saying that app X didn't compile...

Anyway, I hope the project matures and that it makes my already reasonably fast laptop into something even better

Grinch

Member

Registered: 2010-11-07

Posts: 265

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

If you have an i3/i5/i7 CPU you may have been hit by this bug:
https://github.com/path64/compiler/issues/34
It causes this CPU not to be recognized and many, many optimizations not working. So you have to set -march=... for your CPU.

Indeed I do have an i5, and given that I went with -march=native this could have been the reason I didn't see any good performance results. Back to testing I think, thanks for the heads up.

That so many software doesn't compile seems to often be the fault of the makefiles and configure scripts...

Not sure what you mean here, ekopath is supposed to be compatible with gcc, so basically I should just need to change the CC,CXX,LD variables to pathcc/pathCC and it should work, or am I missing something?

codestr0m

Member

Registered: 2010-03-18

Posts: 4

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

@codestr0m: Thanks for all the info Sorry if I sounded ungrateful or something, I'm not. I would love to help, but since I'm no developer and I don't know any language (besides the hackish stuff I do to DWM), I feel I won't contribute much by saying that app X didn't compile...

Anyway, I hope the project matures and that it makes my already reasonably fast laptop into something even better

Let us know which program it is.  We'll possibly give some hint comments for how a non-developer can easily dig and reduce the problem.

@6xx - that's not a stable version since it's in the nightly directory.  It's likely only there by accident

@Grinch - We do best effort for GCC compat, but there's some flags we don't support and some of the micro flags like -O2 and -O3 behave a bit differently.  We also have -ipa and things like -Ofast which can give performance gains, but gcc don't support.  Also a small note on C++ - We used to be STL link compatible, but since v4 not anymore.

el mariachi

Member

Registered: 2007-11-30

Posts: 595

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

How can I make "make" use pathcc instead of gcc? It's boring doing it by hand

Last edited by el mariachi (2011-08-19 17:03:27)

Grinch

Member

Registered: 2010-11-07

Posts: 265

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

Actually, from gcc 4.6 -Ofast is supported (-O3 + -ffast-math right now afaik). Yes, I realize that -O2/-O3 doesn't match 1:1 between compilers, however the options which are supposed to create the fastest code (in this case -Ofast) would be a fair comparison (although in practice it's not always that a higher optimization level end up faster due to the difficulties in optimization heuristics). However the tests I did previously were done with -O3, so I really should do the comparisons again using -Ofast. Something to play around with during the weekend I guess (yes I am that boring! )

An interesting comparison would be gcc vs pathcc with -Ofast, interprocedural optimization (ipa/flto), and feedback optimization (fb/fprofile) as that should provide the best basis for optimization I think.

Anyway, thanks for the info, and thanks for open source'ing ekopath, more compilers in the FOSS arsenal is always a welcome thing!

Cdh

Member

Registered: 2009-02-03

Posts: 1,098

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

How can I make "make" use pathcc instead of gcc? It's boring doing it by hand

Unortunately it depends.
When the makefiles are well written you can simply specify the variables CC and CFLAGS for the c compiler and CXX and CXXFLAGS (I think) for the c++ compiler.
But there are many that have gcc hardcoded. For configure scripts there sometimes is a --cc= option, but often not..

Maybe it would be a good idea creating symlinks called gcc and g++ in the ekopath bin directory (assuming it is in /opt/) and then just compile with a different PATH. Or maybe not.

Is -march=native now supported? When it first was released it failed with native because it was called -march=auto if I remember correctly. I remember many configure scripts failing because of compiler flags not supported by pathcc...

---

฿ 18PRsqbZCrwPUrVnJe1BZvza7bwSDbpxZz

el mariachi

Member

Registered: 2007-11-30

Posts: 595

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

Thanks! I'll try that later, now it's working with export CC=pathcc. -march=native isn't working, but -Ofast compiles most stuff.
The problem really seems to be the Makefiles and, well, lack of support for some flags. But I already compiled some stuff with it, but they're so light already, no big difference can be felt.

kfgz

Member

From: Supraśl, Poland

Registered: 2009-03-02

Posts: 114

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

I need some help with PKGBUILD. With new version of ekopath (that is 2011-11-04) makepkg fails with the following error:

Problem running post-install step. Installation may not complete correctly
 Error writing file /usr/sbin/subserver

but it installs when I manually run the installer

./ekopath-2011-11-04-installer.run -mode unattended --prefix /home/xyz/ekopath

I think that problem is within fakeroot. How can I modify PKGBUILD to make build environment 100% vanilla (no LDFLAGS, no nothing)?

EDIT:

PKGBUILD - http://kfgz.mydevil.net/ekopath-bin-201 … src.tar.gz

Last edited by kfgz (2011-11-05 19:09:04)

falconindy

Developer

From: New York, USA

Registered: 2009-10-22

Posts: 4,111

Website

Re: PathScale Open-Sources The EKOPath 4 Compiler Suite

Changing compiler flags for a precompiled binary isn't going to accomplish anything. I suggest asking upstream about it, because it seems like a bug in that this particular file isn't obeying the install prefix.

---

blog | ±github