Slashdot Mirror
Gcc for the IA-64.
Anonymous Coward writes "Cygnus will port the gcc compiler (egcs, soon to be renamed gcc3) to the IA-64 architecture, notably for the Linux OS and applications. Check the press release here.
" Essentially, Cygns is working with Intel to port GNUPro Software to IA-64 is the gist of deal. GOod to see people planning ahead, I s'pose.
Compiler technology is usually seen as the hard part in an architecture such as IA-64. Supporting it in gcc will need some hard work. It's good that Intel contributes to our free compiler, so that we don't have to wait for the IA-64 specs, and gcc will be ready when systems ship.
It's about time. A couple of weeks ago I said that the proposed Linux port to IA-64 would die w/o a decent compiler (or any compiler at all for that matter). It's good to see that somebody apparently listened. I'm not convinced we'll actually see something that ships before 2001 though. At that point in time, the whole architecture could be a piece of garbage compared to competing platforms. It'll be a bumpy ride, but at least this project is no longer completely doomed to fail.
I guess Intel is also working with MS on the compilers. But if they can't do it right, they could still ship a gcc-compiled Windows!
"Cygnus will offer GNUPro tools for IA-64 Architecture through the GNUPro subscription model that include regular software upgrades,
From what I can tell they are going to SELL GPL'd code BEFORE making the source public. This sounds like a clear violation of the terms of GPL.
Can anyone clarify?
think of all the source files 'hard coded' for
CC = gcc, CC = egcc, that is likely many Makefiles.
The FAQ about GNUPro includes a question about why shouldn't I go with
Red Hat's bundled tools, and answers that their compilers
have superior optimizations, enhancements, etc. If this
compiler is egcs or based on gcc, surely it has to be
distributed under the GNU licence? Maybe the FAQ is
misleading and the compiler is the same, but the other members
of GNUPro are not GPL'd (you have to pay for the package,
no redistribution). Even so, how ethical is it to call
something GNU and not GPL it? I don't like it.
They also invested in Be and sent 2 of their developers there to make the port on x86 processors. Intel is usually compared to Microsoft, but they DO have competition, and they do make good stuff too (ok, the x86 architecture is a bit old, but their processor manage to be as fast as some more modern RISC chips, which proves they are pretty good at what they do).
Sorry that appeared three times. One I can explain:
I cancelled the submit to insert the URL, but the proxy
server probably sent it anyway. I don't know why the other.
gcc doesn't really work that way. It has a (mostly) platform-independent "front-end" and a platform-specific "back-end". Getting the front-end to run on just about anything is fairly trivial, and once a back-end is written for an architecture, it can (again, usually, since there are a few weird dependences) be used to emit code for that architecture when run on any computer which can run the front-end. There's nothing significantly different between a gcc cross-compiler and a gcc native compiler. There are several platforms for which a back-end is written but the front-end is not ported, but to my knowledge these were all embedded systems (pagers, cellphones, etc) which couldn't run a compiler anyway.
-
I don't think Intel would be paying Cygnus good money unless they were writing a gcc back-end for IA64, and given an IA64 back-end, a native IA64 compiler and myriad [X]-to-IA64 cross-compilers follow pretty easily.
Yes, I was also there then, and recall a small group of
engineers being funded to target GCC for the 80960 as well.
someone said that the RISC IA-64 port will be much more difficult than the IA-32 port was, and that the project is mamoth. As a compiler writing student, I know:
That wouldn't work out. Cygnus makes most of its business dealing with non-Intel architectures.
I keep hearing that intel ports linux to IA-64,but the fact is that HP already has linux running on an IA64 emulator and has an egcs based cross-compiler. HP has full time developers working on Linux.
I believe that it's not intel who is working with cygnus too.
People should notice that the IA-64 is a joint intel-hp processor not an intel-only.
HP is really pushing linux and will even more in the future.
Gyula
M$ VC++ (5.x +) tends to producer better executables than any other compiler under Windows on a Pentium, the question is about the retargetabiliy of the compiler -- GCC was designed to be easily ported and retargetted, so this will be less of a problem -- VC++, I imagine, is heavily x86 orientated.
What's interesting is that Digital/Compaq hasn't jumped on the Merced bandwagon, nor have they announced a VLIW architecture of their own. Hmm... AFAIK, no other vendor is trying to commercially implement VLIW other than HP/Intel. Compaq already has a 64-bit native architecture, and it still has head room for some of the EPIC features, notably predication. The Alpha has 15 years left as an architecture design. Get with it.
The pgcc project predated egcs. RMS had estranged a lot of groups who splintered from the "main" FSF and did their own thing -- the Linux people, the pgcc people, the GNATS people, etc. The pgcc compiler was based on gcc and had a bunch of x86-specific optimizations hacked in without much consideration for portability to other architectures. Enter egcs: Cygnus used the egcs to re-unify all the splinters under one source base again. The pgcc people were invited to apply their x86-specific optimizations to egcs (but this time keep the code portable), and by and large they did. There are still some things that pgcc does better than egcs, but IMO as a rule of thumb you're better off with egcs.
the EGCS project is the new maintainer of GNU GCC now. The old name egcs is dropped, so forget about it, from now its only GCC.
I believe the backend (code generation part) of their Alpha compiler is supplied by DEC/Compaq. No way DEC would let them use a MS version which would be nowhere near as tuned as the DEC version.
MS code is not ANSI C or standard C++ clean. I also believe that GCC/EGCS' support for C++ templets is still less mature than the MS C++ compiler. There is also some other propritary changes that EGCS does not impliment/support. Isn't there some MS changes to C++ syntax to make dealing with Unicode structures easier?
This also seems to be a problem for WINElib. While a library could be made to make Linux/X11 API compatible with Win16/Win32 code, MS C++ source code compatiblity would still require a compiler which impliments MS changes to the C++ syntax.
A you know, Intel is investing in RedHat, so they already acknowledged the existence/importance of OSS. And as GCC/EGCS will be used more and more in the server-market, Intel needs to cooperate with the OSS-community. The server-market will be important to Intel's IA64 sells.
Draw your own conclusions.
How complex is "extremely complex?" Nowhere near the level of IA-64, I would guess. I found it odd that Taschek would bash HP for their compiler support. I think they have the best chance of producing a decent VLIW-style compiler, given all the research that has come out of HP labs.
What's interesting is that Digital/Compaq hasn't jumped on the Merced bandwagon, nor have they announced a VLIW architecture of their own. Hmm...
Having GCC working well on IA-64 will go a long way to further the impact of Linux, especially if GCC works much better than any compiler out of Redmond. Even if Windows 2000 runs on IA-64, if the applications crawl, not too many people are going to choose that OS over another.
I disagree with this and the meat of Taschek's article. If businesses choose Windows, they will choose Windows. I don't think a compiler will make or break an operating system. Established use and ingrained practices are much larger factors. If a good compiler targeting Windows on IA-64 is not available, Merced will be relegated to a niche market. If the free software community is counting on Merced being the end of proprietary operating systems, we've lost.
It's also interesting that people assume Microsoft will be providing the IA-64 compiler for Windows. Intel is producing their own compiler. They're having problems, but chances are there will be more than one IA-64 compiler for Windows.
Posted by Mike@ABC:
How much do you want to bet that Intel will buy up Cygnus inside of a year? They already worked together to tune Linux for Intel Pentium workstations, and now they're working on the IA-64 architecture. Intel is increasingly Linux-friendly (one of their execs is a keynote speaker at the next LinuxWorld) and Cygnus is a key player in making sure that Linux runs well on Intel's architecture. I think Intel has seen the writing on the wall, and they have a tendency to buy up useful companies.
Just a thought. We'll see.
Seems kind of stupid to rename it from egcs to gcc3.
They should go with either straight old gcc or leave it as egcs. What's the point of introducing yet another name for a GNU compiler? Either it's gcc or it isn't. If it isn't, then keep it egcs. If it's the new GNU compiler, then call it gcc. What could the possible advantage be to introducing yet another compiler name?
Those are two different questions.
The answer to the first is, very good.
The answer to the second is, pretty good.
IA64 is rumored to be quite different from current 64-bit architechtures.
Why? EGCS (the next gcc) already supports several 64-bit architectures like Alpha and UltraSparc, and incorporates the Haifa scheduler which can be a big performance boost.
I must've missed that. Could you give a reference?
There are other ways of keeping people from implementing your instruction set architecture, e.g. patents; there are a couple of patents that might apply to IA-64 implementations, namely US patent 5638525 and US patent 5859999.
(I also wouldn't assume there's no cleverness in the silicon; punting a lot of the work to the compiler doesn't necessarily mean getting a CPU "so simple even a child could do it".)
How good is gcc/egcs at compiling? I mean how optimum is it on x86 stuff.
"Think of it as evolution in action."
Of course, porting will need to be done to get gcc to RUN on a new architecture even as a crosscompiler. However, adding support for compiling stuff for that new architecture is NOT porting - you could be writing a cross compiler.
Don't let the MS weenies confuse people with their cunning terminological inexactitude! :)
The compiler we call "GCC" right now is GCC version 2.x, or gcc2 for short. The compiler currently called "EGCS" will become version 3.x of GCC, or gcc3 for short.
It is also worth noting that EGCS' current version numbers are 2.9x.xx, which implies that they are pre-release versions of 3.0.
So, in short, EGCS will be the new GCC, and will be called, simply, GCC. In saying "egcs will become gcc3", they mean that egcs will become the new version of GCC.
Hope that helps!
John Taschek over at PCWeek has an interesting editorial about just how important getting a good compiler will be for all OS's, not just Linux. He suggests that MS is going to have serious problems with the port, since the EPIC architecture is vastly different from the traditional Intel CISC architecture that MS is used to. For RISC vendors, they are used to dealing with extremely complex compilers.
Having GCC working well on IA-64 will go a long way to further the impact of Linux, especially if GCC works much better than any compiler out of Redmond. Even if Windows 2000 runs on IA-64, if the applications crawl, not too many people are going to choose that OS over another.
well they said they had GNU/Linux working already on IA64 a while back the thing is that with IA64 all of the Brains is in the compiler
so ?
well intel will want to keep all of its dirty little secrets of the compiler to itself or anybody can reproduce the silicon easy peasey because the machine is dumb and the compiler knows all !!
Hmm so whats going on
??
anybody
cheers john
when IA-64 specs become publically available, suggesting that
they have promised Intel not to release the source before that time.
This is consonant with the GPL as long as they do not release binaries
either. But then the work Cygnus will do till the source-code release
is not after the bazaar development model since only employees of
Cygnus will be doing it.
Is that about right?
The article also mentions that HP has run "successful tests of the Oracle 8i database on an HP-UX based IA-64 simulator."
At least regarding the compiler (and 64-bit kernel / memory model, I would think), Linux is not far behind. I believe that these vendors are hoping that IA-64 will allow them to place some more distance between their products and Linux. I also think that they have another thing coming.
Let the code run free.
It's great to see Intel cooperating on such an important project. Having the gcc compiler for that architecture is vital for the development of Linux on the platform. All I can say is it's great to seeIntel taking OSS seriously.
Cool Linux PC Badges!
...but it's all "cc" to me :)
There's a (bad) country/western song in there somewhere.
Save the whales. Feed the hungry. Free the mallocs.
IIRC, there was the pgcc project, an egcs derivative where they experimented with specific optimizations for the Pentia. I haven't checked up on them recently, though, and the URL I have is no longer valid.
-jgj
--- Premature complacency is the evil of all roots
Intel has a history of supporting GCC for new chips. I was involved with a 1989 project where Intel gave money and machines to RMS to get GCC running well on the 386 as part of a plan to get Mach ported (when SystemV and Xenix were the only games in town). Intel's position was simple: all the 386 compilers suck - maybe with a small dose of cash (especially compared with what they paid Green Hills) these longhairs will come up with something.
The concept of "porting" GCC to something so different from current 32-bit machines as IA64 is going to be a major strain on the backend. Getting performance to even the current GCC standard (ie good but not best-in-class) will take at least a couple o years once hardware is available.
You can't fight in here - this is the war room!