new a64 optimized compiler

[saaya]

http://www.theinquirer.net/?article=14077

its a hacked intel compiler, he claims it gives an up to 22% performence boost over previous compilers!

please post results

[saaya]

c'mon anybody give it a try. should give a nice performence boost in superpi pifast and f@h etc...

[Overlag]

i dont get it... how does this work?

[doctorcod]

hmpf! if i am going to have to recompile every program that i want the performance boost in, which you would, then no thanks lol. but yeah for benches, lets see if it really works.

[saaya]

the compiler translates requests from any app into the cpus native language. its like the cpu is an illegal mexican worker and you need some dude to translate him what he is supposed to do. now you just found somebody who can translate to spanish faster and doesnt make that many mistakes as the translator before so our mexican friend can work even harder and faster :D

[dropadrop]

Originally posted by saaya
the compiler translates requests from any app into the cpus native language. its like the cpu is an illegal mexican worker and you need some dude to translate him what he is supposed to do. now you just found somebody who can translate to spanish faster and doesnt make that many mistakes as the translator before so our mexican friend can work even harder and faster :D

:D

Great explanation... I heard of some even more efficient complilers being released though, problem is they are commercial and fairly expensive. These compilers will start making more of a diferance when companies port their apps to the x86-64 version of windows. Even if the windows itself won't give a performance advantage, using optimised compilers will.

[Maxvla]

Originally posted by saaya
the compiler translates requests from any app into the cpus native language. its like the cpu is an illegal mexican worker and you need some dude to translate him what he is supposed to do. now you just found somebody who can translate to spanish faster and doesnt make that many mistakes as the translator before so our mexican friend can work even harder and faster :D
and for even less money :x

[enzoR]

lol great explanation saaya!

[saaya]

hehe :D thx

this compiler is still 32bit though, its mainly about sse2 i think...
funny thing is amd doesnt even work on a compiler afaik, they just use intels compilers and still beat them at their own game :rotf:

[OwcA]

Intel Compiler can produce 64 bit code but only for Itanium. This can all well change if Intel is planning on releasing a x86 64 bit processor of their own.

[saaya]

wee i was talking crap, of course amd is working on a compiler for x86-64 :P

[OwcA]

The increasing differences (and therefore gains from platform specific optimizations) between processors is wind in the sails of opensource comunity.

[DMOS]

The problem with this is that you need the source code binaries for the applications you plan to recompile. You aren't going to have that for most programs, with the exception of the open source ones as mentioned by OwcA.

[MentholMoose]

This is not a hacked compiler. The guy wrote a utility that will patch Intel-optimized binaries to work on non-Intel CPUs. So all you need to do is use his utility on a binary that meets this requirement; you don't necessarily need the source code (you only need it if there was no Intel-optimized version of the binary and you had to compile one yourself).

/*
* iccOut 1.0
*
* This program enables programs compiled with the intel compiler using the
* -xN flag to run on non-intel processors. This can sometimes result in
* large performance increases, depending on the application. Note that even
* though the check will be removed, the CPU running the application *MUST*
* support both SSE and SSE2 or the program will crash.
*
*/

[IvanAndreevich]

So, is anyone going to compile a fresh version of some pi-calculating software to see any benefits of this?

[Overlag]

or seti 64 :D

[saaya]

Originally posted by MentholMoose
This is not a hacked compiler.

The guy wrote a utility that will patch Intel-optimized binaries to work on non-Intel CPUs.

so hows that not hacked?

[MentholMoose]

Originally posted by saaya
so hows that not hacked? The utility doesn't do anything to the compiler at all.

[QuadDamage]

Originally posted by Overlag
or seti 64 :D

i'll do it, just tell me how:D

[saaya]

it DOES change the code to make it do something it was not supposed to do, run on amd cpus... thats what i call hacked...

[Vlad Draculea]

it is hacked but it is not a compiler, that was his point i think, and that could mean a lot for amd if it is as easy as it seems, no sane developer would doubt for a minute to put his binaries there and push the button (they should first insert a sse2/see check so it doesnt crash in old cpus).
no need to delete/edit, this doesnt seem like a good new to intel, so it must make current owners unhappy.

[bigjohns97]

Originally posted by saaya
the compiler translates requests from any app into the cpus native language. its like the cpu is an illegal mexican worker and you need some dude to translate him what he is supposed to do. now you just found somebody who can translate to spanish faster and doesnt make that many mistakes as the translator before so our mexican friend can work even harder and faster :D

Are you sure you are in Germany and not in Texas?

[saaya]

hehe lol :D

[bigjohns97]

what funny about this whole situation is that the compiler is not optomised for amd64 it's just not letting intel use ss3 while amd dosen't. In other words it lets amd 64 chips take advantage of sse optomisations. I bet you money if amd wrote a compiler it would smear intel benchmarks and vise versa.

[zabomb4163]

intel wrote the complier

[OwcA]

Let's nt forget, that Intel Compiler is above all a vector compiler. Comapred to say gcc with full P4 optimizations it will still offer substantialy better results (ofcurse only in situations witch benefit from vectorization).

[agenda2005]

I used both intel EMT64 compiler v8.1 and gcc on A64 3500+ with SuSe Linux 9.1, surprisingly, the intel binary was twice the size of gcc own. Intel compiled binary also run twice slower even with a highly vectorized code.
Anyway, I just found this(Intel compiler patch for AMD A64) a while ago, I hope Intel is not shooting herself in the leg. http://www.theinquirer.net/?article=22750

[HARDCORECLOCKER]

:confused: Anyone tried it so far? Results?

:toast:

[[XC] leviathan18]

anyone tried???? it would be great if it gives me 22% speed up in super pi

[Walrusbonzo]

Hi All

I've downloaded the source code to this program and compiled it. I'm currently running through loads of EXEs to see which ones can be patched and which can't.

So far, the ones you may or may not be interested.

Super Pi - No Patch
PiFast 4.3 - No Patch
3DMark2001SE - PATCH! :D I wonder what difference it makes. Rendered Useless
3DMark2003 - No Patch
3DMark2005 - Patch, Rendered Useless
PCMark2004 - Patch, Rendered Useless
FarCry.exe - No Patch
Editor for Farcry - PATCH
HL2 - No patch
United Devices, ALL the EXEs - No Patch
WINRAR, WinRAR.exe, Unrar.exe and RAR.exe - ALL can be patched! Hoping for faster file archiving now :D Works good so far, dunno if it's any faster though.
Prime 95 - No Patch
Speedfan - Patch - Rendered Useless
WMPLAYER.EXE - No Patch
Explorer.exe(Win XP x64) - No Patch
MPLAYER2.exe - No Patch
Lame.exe - Patch, Rendered Useless
SANDRA 2005 - No Patch

I'll post more as I know more.

I've attached the program to this post as well so you can try it for yourself! As usual, I accept no responsibility for any damage it may cause when you use it!

[airwolves]

Anyone else??

[WeStSiDePLaYa]

so can someone tell me what to do with this??? does it work with normal winxp pro? how much benifits i looking at?

[Ubermann]

If this works it ruins everything and make the whole scene a mess ?

[matt9669]

It's not a hack saaya (at least not in the illegal or reverse engineering sense), it merely disables a check for an Intel processor in cases where SSE and SSE2 compatibility are issues. Programs that make heavy use of SSE/SSE2 and do NOT check for these feature flags but merely check for the correct Intel processor, stand to see large improvements in performance.

Had the processor check been written correctly, these programs would already be running at full steam. ;)

And yes, unfortunately the A64 does not handle packed/vectorized SSE2 with greater efficiency than scalar :doh:

[metro.cl]

great, i hope this works amd 22% faster for free :) well on the apps that use the optimization

[craig588]

Programs that make heavy use of SSE/SSE2 and do NOT check for these feature flags but merely check for the correct Intel processor, stand to see large improvements in performance.

I think things based on the Quake 3 engine do this. Probably the UT engine too. I know that I usually need to manaully force SSE to on when I play games using those engines. Maybe if you find a game where it doesn't allow you to force flags it would help alot.

[STEvil]

quake 3 performs flag checks, and I assume Unreal does too.. its quite bad form not to, really.

[WeStSiDePLaYa]

ummm....... so anybody help me out, what to do with this proggy?

[Walrusbonzo]

You need to run in it in command prompt.

"iccout filename.exe"

It will say "Subtituting Code" if it actually finds the Intel only instruction, it replaces it with a NOP(No operation).

Just beware, from my testing so far, it renders a lot of programs useless, i.e. the Intel only instruction is in there for a reason!

EDIT: Just tested WinRAR file compression. It's actually slower(About 5%) with the patched version. This suggests to me that some software may be using this Intel only check to determine which codepath to use. This also explains why some software becomes useless after the patch.

[***Deimos***]

You need to run in it in command prompt.

"iccout filename.exe"

It will say "Subtituting Code" if it actually finds the Intel only instruction, it replaces it with a NOP(No operation).

Just beware, from my testing so far, it renders a lot of programs useless, i.e. the Intel only instruction is in there for a reason!

EDIT: Just tested WinRAR file compression. It's actually slower(About 5%) with the patched version. This suggests to me that some software may be using this Intel only check to determine which codepath to use. This also explains why some software becomes useless after the patch.

I'm no expert, but dont AMD and Intel BOTH support SSE/SSE2 (in P4 and Athlon64), and SSE3 in Venice/Prescott. What "Intel-only" instructions could there be, other than the 4-5 that are reserved for Intel Hyperthreading... I ask you since you seem to be on the ball with this stuff..

[Walrusbonzo]

I'm no expert, but dont AMD and Intel BOTH support SSE/SSE2 (in P4 and Athlon64), and SSE3 in Venice/Prescott. What "Intel-only" instructions could there be, other than the 4-5 that are reserved for Intel Hyperthreading... I ask you since you seem to be on the ball with this stuff..

I'm not really that much on the ball, I just read the article and understood it.

Took the source code, compiled it and started to play with it :)

[DudeMiester]

This is only for programs that have been compiled with the Intel optimising compiler, and only with Intel specific optimisations turned on. Plus this is from over a year ago, there surely have been revisions to their compiler, probably rendering this program useless. There is little point of trying this thing out, it'll probably just corrupt your programs. Also, if the programs do checksums on themselves, they'll be rendered useless, as this modifies the exe.

[saaya]

HOLY THREAD RESURRECTION BATMAN! :eek: :lol:


It's not a hack saaya (at least not in the illegal or reverse engineering sense), it merely disables a check for an Intel processor in cases where SSE and SSE2 compatibility are issues. Programs that make heavy use of SSE/SSE2 and do NOT check for these feature flags but merely check for the correct Intel processor, stand to see large improvements in performance.

Had the processor check been written correctly, these programs would already be running at full steam. ;)

And yes, unfortunately the A64 does not handle packed/vectorized SSE2 with greater efficiency than scalar :doh:

well they (intel) called it hack :D i thougt its funny haha i agree its not really hacked, at least i wouldnt call it that... but i think officially it is hacked, no? they mean hacked as to manipulated without approval of the owner.

you can also call it patched :D

and yeah, the a64 is bad with sse and sse2, BUT sse and sse2 are more universal... so if the programs have unefficient code sse and sse2 might still be faster on the a64, no?

its mostly about compability though so apps that use sse2 can work alright on a64s i guess. the latest compiler story is hilarious! :lol:
intel will have to update their won compiler to work with a64s :rofl:

[saaya]

and saaya, please post in the right section next time! this surely doent belong in the xoc section! :slapass:

:D

[STEvil]

actually they are pretty good with sse2, thats why sse3 (re-organized sse2) doesnt give much of a boost.