
Intel Makes Linux Move - Enhancing Compilers 87
Several folks sent this in, but an AC got it first:
Intel is throwing more of its substantial bulk
behind Linux.
They will be assisting Cygnus in creating compilers optimized
for the P2 and P3 processors for Linux. With all the
recent intel blunders (cpu id's, the p3's pathetic performance
etc) its cool to see something getting done right.
Very Bad, If not gcc/egcs (Score:1)
Intel is so much smarter than AMD (Score:1)
Remember the PIII/K6-3 comparison on Tom's Hardware? Much of the software used to do the benchmarking used Intel's streaming SIMD extensions but did not support AMD's 3DNow! extensions (Tom made that very clear).
Now we're going to have a compiler for Linux that will no doubt enable (well, simplify) the creation of similarly optimized programs.
AMD is doing very well in chip design, less well in fabrication, and very badly in all the other things Intel does so well.
--
Shocking (Score:1)
Since when ?
Otherv architectures (Score:1)
Now I would like To see Motorola doe the Same so that LinuxPPC would get faster (and so would all the BeOS gcc stuffs).
Arm could help too. As should all Risc vendors. Remember one off the concept of the Risc architecture is : let the compiler do the work, not the processor -- (Since P's are suppose to be CISCs I think other architectures would get Better boost on that kind of work then the x86 arch).
Ludo
It's about time. (Score:1)
--
Yeah! It's the compilers, stupid (Score:1)
Motorola, IBM, Compaq, etc. have yet to figure this out.
MMX competes with Cyrix's 3DNow!(?) (Score:1)
I wonder when Cyrix will create this 3DNow! Technology to compete with MMX?
Silly journalists.
sweet (Score:1)
This compiler is not a big threat. (Score:1)
Of course, the Linux community is not the only market for processors, so companies might be willing to sacrifice some processor revenue from Linux users in exchange for increased revenue from compilers. However, this environment would not allow Intel to raise prices significantly, because even if they could force it out of Linux users--although I don't think they could--they would lose enough sales in other markets to make the price increase unprofitable.
Sorry, you didn't do your homework (Score:1)
It does, however, say - at least as I read clause 6 of the GPL - that if you provide a GPLed program to anybody, the recipient can redistribute it:
"These terms and conditions" don't include "they have to give you money when they redistribute it", and that, combined with "You may not impose further restrictions...", makes it look as if, when you sell GPLed software to somebody, they can put it up for FTP for free and announce it to the world, and you have to live with that.
I.e., it doesn't say you have to release it publicly, but it lets anybody else do so, even if they paid you a large amount of money for it.
Beowulf... (Score:1)
built with this...
:P
--Corey
This will work (Score:1)
AMD and Intel (Score:1)
--
Aaron Gaudio
"The fool finds ignorance all around him.
Intel is the coolest (Score:1)
Celery 300A O/C to 450 MHZ. Fast and cheap.
Not if you download a precompiled optimized disto (Score:1)
Yeah! It's the libraries, stupid (Score:1)
Honest questions here... (Score:1)
For the 'x86 chips, I like the Cygnus tools well enough, but what about the *rest* of the world, also known as Linux PPC, G3/G4, SA-110, etc.?
Secondarily, can any
I know it's not freeware, but their compiler is the only one I know of that has benchmarked C++ code running competitively with Fortran.
Intel is the coolest (Score:1)
Sweet (Score:1)
Of course, most people who use Redhat and their handy dandy RPMs will still be stuck with 386 binaries, as RH doesn't really make a point of using RPM_OPT_FLAGS in all its SRPMs.
You retard (Score:1)
I can just see it now. Some overclocker is going to be messing around with the CPU and is going to get his fingers frozen off.
The PIII isn't going to bread 700MHz.
Simply because they have to... (Score:1)
It is the only way the PIII will achieve anything close to
the performance gain you would expect a new CPU to have.
AMD and Intel (Score:1)
egcs is open source (Score:1)
egcs is open source (Score:1)
gcc 2.7.xxxx. Anyhow, no offense was intended to Cygnus or egcs.
It is gcc, only optimized (Score:1)
http://egcs.cygnus.com/
AMD,Cyrix,IDT, Intel... (Score:1)
-----------------------------------------------
"Expect more companies to hop on that bandwagon. Cyrix Corp., which makes low-cost computer processors, says it also is interested in working on Linux, once it finalizes its 3D-Now! technology, a rival version of MMX."
-----------------------------------------------
My response to the journalist:
-----------------------------------------------
Hi,
I'd like to correct a misconception that you printed regarding 3D-Now! technology and Cyrix.
3D-Now! is a joint venture, spearheaded by AMD corp - and currently supported by their K6-2 and K6-III processors. Cyrix and IDT are also supporters of the technology and will be or already have integrated the technology into their CPUs. The WinChip2 available from IDT already has 3D-Now! support.
3D-Now! was released with the K6-2 in June of 998, and is actually more similar to the recently release SSE (formerly KNI) instructions found in
the Pentium-III chip from Intel. Both instruction sets are intended to improve floating point performance - important for 3D-Games, Voice
Recognition, Video Compression and other similar applications.
They use a technique known as Single Instruction, Multiple Data (SIMD) to achieve higher performance by allowing instructions issued by the processor to actually affect more than one bit of data simultaneously. In the case of both 3D-Now! and SSE, the improvement is the inclusion of 2 execution units each capable of calculating 2 32-bit floating point values simultaneously - thus, each processor can process 4 floating
point instructions at the same time.
Intel's competitors have a rough enough time dealing with Intel's marketing campaign without having to deal with bad press.
-----------------------------------------------
While I will certainly admit - and this is going over it all one more time for those of you unfamiliar with the architectures of the x86 CPUs (I'm really only familiar with the k6, P2 and M2 cores).
FPU performance:
P2: 2 Execution units, variable latency of 3 to 5 cycles. Per clock cycle peak efficiency: 2 instructions per 3 cycles - 66%.
K6: 1 Execution Unit, low latency of 2 cycles. Per clock cycle peak efficiency: 1 instruction every 2 cycles 50%.
M2: 1 Execution Unit, fixed latency of 4 cycles. Per clock cycle peak efficiency: 1 instruction every 4 cycles 25%.
So, take all three CPUs, and assume everything is equal, and no really tricky tricks are used, and run them at the same speed of 300MHz. That means that you'll get roughly:
198 MFLOPs for a PII core.
150 MFLOPs for a K6 core.
75 MFLOPs for an M2 core.
In actuality the numbers usually aren't far off.
The flip side of the equation is that the Cyrix and AMD chips generally have better integer performance - a holdover in their design plans from 2 years ago when the office application benchmark was king - and office apps use mostly integer instructions. So usually per clock the Cyrix and AMD chips will produce more MIPS than the P-2, but less MFLOPS.
However, coupled with the fact that the P2 core also has some additional features in terms of memory access, and whatnot and the real world bench marks change up somewhat. But not too much really. Especially since the K6-2 with CXT core implements some of those features.
Check out Tom's hardware for some benchmarks of the K6-III and P-III. It's got some great benchmarks of how much SSE instructions will improve the P-II core. However, it's a little light on the 3D-Now! optimized titles for comparison.
http://www.tomshardware.com/releases/99q1/99022
Check out Ace's hardware for some K6 core info. It gives you some very real insight into how the cores of the P2 and K6 work, and what effects cache memory has on them.
http://www.aceshardware.com/articles/x86/k6-3_i
- Porter
I think I will stick with GCC (Score:1)
CPUID (Score:1)
Sparcs have had cpuids forever, and I think mips and PPC's do too... It's useful for copy protection of expensive programs. And although it's nice to have GPL'd stuff, if you're designing a specialized CAD program for a certain type of application, and you're only going to sell it to a few companies, you may very well need to sell it for $5,000 or more a license, and you'd want to be sure that you were getting all your money...
Anyway, it's not such a big problem, although if it was placed on mainstream computer parts I could see how it might be used for more sinister motives.
Sorry, you didn't do your homework (Score:1)
--
Aaron Gaudio
"The fool finds ignorance all around him.
Sweet (Score:1)
Sorry, you didn't do your homework (Score:1)
I'm very concerned (Score:1)
Even if they don't, then any optimizations that are part of GPLed code must be made available to their customers in source form, and their customers can give the source (or binaries) away for free, as I read clauses 3 (must make source available) and 6 (can't stop people you provided it to from giving it away) of the GPL.
The press release [cygnus.com] says:
The libraries may be Cygnus-proprietary (if they're not GLIBC-based, say), but the changes to the compiler and assembler won't be, unless they throw out GCC/EGCS, GAS, GLD, etc.
They don't say anything about KNI support in the compiler, however.
It's nothing new. (Score:1)
I applied the patches but they proved to be defective.
I tried the patched compiler it on PovRay and it crashed. I remember having to use the ordinary compiler for one of the translation units because the patched compiler died.
no, egcs is not "Cygnus' egcs" (Score:1)
Cygnus is donating resources to egcs, but it is not "Cygnus's egcs". egcs legally belongs to the Free Software Foundation, and it is a net-wide project, not a Cygnus project.
egcs has a steering committee that controls it, and Cygnus does not have a majority on that committee. So if Cygnus tried to do something with egcs that outsiders don't like, they would be outvoted. (I doubt if this would happen as they are good guys).
Cygnus's role in egcs is much like Red Hat's role in Gnome: they provide resources and pay people to work on it, but there are many outside developers.
Also, egcs is not Cygnus's only compiler effort. Typically they develop gcc enhancements under contract and provide those enhancements to their paying customers first. Later on these changes are integrated into egcs. (The paying customers are allowed to pass the changes on to others under the GPL terms, but they are not obliged to do so and typically they don't, meaning that they get the new code sooner than anyone else in exchange for their money).
Of course this is GCC (Score:1)
One compiler to build them all, one compiler to...oops, nevermind.
Linux Headlines on ABC News site (Score:1)
right in their headlines!
Intel is the coolest (Score:1)
Haven't you _noticed_ that the performance of x86 is no longer scaling linearly with mhz? It's hitting a wall.
No additional charges (Score:1)
Far as I know, Cygnus has no plans to change its business model because of this deal -- same availability as before. And Intel says they're investing in this because it's a smart business move. The vast majority of Linux machines are running on the x86 architecture, so if they optimize, they figure they can sell more machines to Linux users.
I hope this is GCC? (Score:1)
Does anyone have the info to soothe and comfort me?
Sorry, you didn't do your homework (Score:1)
Honest questions here... (Score:1)