Open64 on Arch64 anyone?

ashyanbhog · 2010-03-07 05:55:51

AMD released opensource x32 / 64 compiler recently.

http://developer.amd.com/cpu/open64/Pages/default.aspx

AMD says this compiler suite optimizes performance for AMD64 WITHOUT crippling Intel CPU's, unlike ICC and GCC compilers which cripple AMD64.

Some programs are seeing a dramatic 25% jump in run-time performance using this compiler vs gcc

http://www.enigmaathome.net/forum_thread.php?id=106

The only limitation so far seems to be the front-end which is gcc 4.2 based.

No Archer seems to have tried this compiler so far, anybody intersted in changing that state?

AMD support forum for the compiler

http://forums.amd.com/devforum/messagev … erthread=y

nbyers · 2010-03-09 06:21:20

I would love to see this package in the repos. Very interesting project.

jumzi · 2010-03-09 17:39:03

Wonder why they don't just work closer with gcc instead of reinventing the wheel if the compiler is opensource anyway?

skottish · 2010-03-12 02:49:57

The last time that I checked, that compiler needs registration to download, so it's not going to work well with standard Arch PKGBUILDs.

ashyahbhog,

It's true that Intel intentionally cripples ICC for AMD, but GCC? I've never heard of that and can hardly believe that it would ever happen.

ashyanbhog · 2010-03-12 11:31:23

skottish wrote:

ashyahbhog,
It's true that Intel intentionally cripples ICC for AMD, but GCC? I've never heard of that and can hardly believe that it would ever happen.

I didn't mean intentionally cripple. Could have something to do with gcc's architecture.

Open64 is written with 64bit in mind, compared to truckloads of system architectures supported by GCC

kevku · 2010-03-12 16:03:22

skottish wrote:

The last time that I checked, that compiler needs registration to download, so it's not going to work well with standard Arch PKGBUILDs.
..../

they have svn too
svn co https://svn.open64.net/svnroot/open64/b … en64-4.2.3 for version 4.2.3
https://svn.open64.net/svnroot/open64/ for main trunk

but gonna be a pain downloading 66MB via svn.

skottish · 2010-03-13 02:10:49

kevku wrote:

skottish wrote:
The last time that I checked, that compiler needs registration to download, so it's not going to work well with standard Arch PKGBUILDs.
..../
they have svn too
svn co https://svn.open64.net/svnroot/open64/b … en64-4.2.3 for version 4.2.3
https://svn.open64.net/svnroot/open64/ for main trunk
but gonna be a pain downloading 66MB via svn.

Thanks. It also turns out that it can be downloaded without registration now. I may just have to try this thing out.

codestr0m · 2010-03-18 06:37:47

Open64 should be freely downloadable from Open64.net, but that's not the same version which was originally referenced. (I do believe the AMD version requires registration, but I could be mistaken)

Disclaimer - I work at PathScale (which was indirectly the largest contributor to Open64 for x86 the past number of years)

Generally, Open64 is not recommended for production and is mostly research and benchmark oriented. PathScale will be releasing a high quality beta of our compiler in order to receive a broader amount of testing. We would be happy to make this available to the whole Arch community if anyone is interested to help us get this going. Just ping me on irc or via email.

Thanks

./Christopher

irc.freenode.net #pathscale

codestr0m · 2010-03-18 06:51:20

ashyanbhog wrote:

skottish wrote:
ashyahbhog,
It's true that Intel intentionally cripples ICC for AMD, but GCC? I've never heard of that and can hardly believe that it would ever happen.
I didn't mean intentionally cripple. Could have something to do with gcc's architecture.
Open64 is written with 64bit in mind, compared to truckloads of system architectures supported by GCC

I can't help, but be curt in response to this.. Open64/Pro64/PathScale has a *long* history which predates 64bit processors even being dreamed about.. (ref: ucode compiler) The framework for the compiler is very flexible and could provide a highly optimized compiler for probably any target. [1] It was only a dedicated effort to specifically optimize for AMD64 processors that would lead to this misconception. With that in mind I should note that I totally defend Intel with the decisions about ICC. It takes a huge amount of effort to test, tune and verify any new processor, isa and target. For them to disable optimizations was intelligent, sane and at worst overly conservative, but probably not malicious.

Hopefully this clears up a bit of FUD

./C

[1] PathScale only currently (publicly) supports AMD64/Intel64, and the (SiCortex) MIPS 5kf processors, but Open64 and other companies working with it have retargets of varying quality for for Simplight, Loongson, Power G4, DSP's, DPU's, GPU's, xtensa, IA64, ARM and truckloads of others.

Arch Linux

#1 2010-03-07 05:55:51

Open64 on Arch64 anyone?

#2 2010-03-09 06:21:20

Re: Open64 on Arch64 anyone?

#3 2010-03-09 17:39:03

Re: Open64 on Arch64 anyone?

#4 2010-03-12 02:49:57

Re: Open64 on Arch64 anyone?

#5 2010-03-12 11:31:23

Re: Open64 on Arch64 anyone?

#6 2010-03-12 16:03:22

Re: Open64 on Arch64 anyone?

#7 2010-03-13 02:10:49

Re: Open64 on Arch64 anyone?

#8 2010-03-18 06:37:47

Re: Open64 on Arch64 anyone?

#9 2010-03-18 06:51:20

Re: Open64 on Arch64 anyone?

Board footer