Message boards : Number crunching : Slashdot article states that Intel compiler deliberately creates unoptimized code.
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Alex

Send message
Joined: 2 Sep 04
Posts: 378
Credit: 10,765
RAC: 0
Message 8413 - Posted: 12 Jul 2005, 16:06:40 UTC
Last modified: 12 Jul 2005, 16:25:12 UTC

This might be something to keep in mind if people are compiling CPU specific 'optimized' clients like they do at the seti project.. it depends on which compiler you use.

Slashdot is an interesting place to see people rant...
http://yro.slashdot.org/article.pl?sid=05/07/12/1320202&threshold=-1&tid=142&tid=118&tid=123

Copy of the AMD legal complaint:
http://www.amd.com/us-en/assets/content_type/DownloadableAssets/AMD-Intel_Full_Complaint.pdf

Update:skimmed the legal complaint: compiler notes start in paragraph 123, page 40.
ID: 8413 · Report as offensive     Reply Quote
Profile Jim Baize
Avatar

Send message
Joined: 17 Sep 04
Posts: 103
Credit: 38,543
RAC: 0
Message 8414 - Posted: 12 Jul 2005, 16:50:56 UTC - in response to Message 8413.  
Last modified: 12 Jul 2005, 16:51:35 UTC

ID: 8414 · Report as offensive     Reply Quote
Profile The Gas Giant

Send message
Joined: 2 Sep 04
Posts: 309
Credit: 715,258
RAC: 0
Message 8415 - Posted: 12 Jul 2005, 20:42:23 UTC

I thought I read on the seti boards that the optimised seti application compiled using an Intel compiler for an AMD and it's instruction set actually ran a little faster than the equivalent for an Intel chip and it's instruction set. If wonder if the Intel lawyers will come across the seti optimised thread.

Live long and crunch.

Paul
(S@H1 8888)
BOINC/SAH BETA
ID: 8415 · Report as offensive     Reply Quote
Profile Jim Baize
Avatar

Send message
Joined: 17 Sep 04
Posts: 103
Credit: 38,543
RAC: 0
Message 8416 - Posted: 12 Jul 2005, 21:47:10 UTC - in response to Message 8415.  

As I understand it, if the instructions are compiled on an Intel, regardless of which chip it runs on, it will run fine. HOWEVER, if the code is compiled on an AMD, the compiler then inserts the bad code.

Jim

> I thought I read on the seti boards that the optimised seti application
> compiled using an Intel compiler for an AMD and it's instruction set actually
> ran a little faster than the equivalent for an Intel chip and it's instruction
> set. If wonder if the Intel lawyers will come across the seti optimised
> thread.
>
> Live long and crunch.
>
>
ID: 8416 · Report as offensive     Reply Quote
Travis DJ

Send message
Joined: 29 Sep 04
Posts: 196
Credit: 207,040
RAC: 0
Message 8425 - Posted: 13 Jul 2005, 16:08:12 UTC - in response to Message 8416.  
Last modified: 13 Jul 2005, 16:09:00 UTC

> HOWEVER, if the code is compiled on an AMD, the compiler then inserts the bad
> code.

...unless the programmer overrides the CPUID to always report "GenuineIntel" and then AMD architectures receive the same optimizations any Intel chip would. The article told some of the story and the discussion was even more revealing.

The part that interests me is something I had forgotten. MMX(2), SSE, SSE2, SSE3 are all Intel technologies which AMD had to license in order to remain "compatible" with the Pentium 4's newer features. This basically means the technology has to be implemented in AMD's processor line in a very specific way, Intel's way. So the argument really is "If AMD's Athlon line of processors contain Intel-licensed technology, why not could the Intel compiler support the very technologies in non-Intel CPUs which it uses in its own products as well as other companies' products which contain those technologies?" I could certainly understand Intel's position about not supporting other CPUs if the target CPU doesn't have the target technology which is being programmed for (i.e. if the Athlon64 doesn't have SSE3 and SS3 instructions are used then it *should* use an alternate codepath) - but we're talking Intel not supporting Intel technology in non-Intel products. It will be interesting watching this unfold in court.

ID: 8425 · Report as offensive     Reply Quote

Message boards : Number crunching : Slashdot article states that Intel compiler deliberately creates unoptimized code.


©2025 CERN