64-Bit / 32 bit, SSE3 and other extensions?
Message boards : Application Info : 64-Bit / 32 bit, SSE3 and other extensions?
| Author | Message | |
|---|---|---|
|
What optimizations are currently being used for the 32-bit windows application? Any?
|
||
ID:
2933 | Rating: 0
| rate:
/
|
||
Andre Kerstens
Forum moderator Project tester Volunteer tester 
Joined: Sep 11 06 Posts: 749 ID: 1 Credit: 15,199 RAC: 0 |
||
|
No cpu specific optimizations are utilized because the application has to run on so many different processors without crashing. And yes, this is a restriction, but hardly one that can be avoided in a boinc project.
What optimizations are currently being used for the 32-bit windows application? Any? ____________ D@H the greatest project in the world... a while from now! |
||
|
ID:
2934 | Rating: 0
| rate:
/
|
||
|
All 64 bit processors support SSE2 so that optimization can safley go into a 64 bit app. I wish you luck on 64 bit as its what i use these days and the wrapper while it works is not the best solution.
|
||
|
ID:
2938 | Rating: 0
| rate:
/
|
||
|
All 64 bit processors support SSE2 so that optimization can safley go into a 64 bit app. I wish you luck on 64 bit as its what i use these days and the wrapper while it works is not the best solution. Er.. well, couldn't you just run the same compile again but optimise for SSE2 and whatever other optimizations? Then it would be like an optimised 32-bit application, but only sent to the 64-bit clients. I know.. sounds simple just to say it like that.. hahaha but.. ? This would kindof be a middle ground without having a 64-bit application, as I don't think it would take many modifications just to switch the compilier to optomize for processor extensions. It still wouldn't be a 64-bit application, but it would be an improvement over the 32-bit 'wrapper' app, correct? I know I asked this once in the BOINC forums, but when you run the compiler, there are options so that it will make the SSE2 /MMX/ ETC.. optomizations for you right? It's all automated? I mean.. to me this seems like a totally effortless way to get 10-20% faster work, all you have to do is just click a few things and let the compiler run for a few hours (or however long it takes), then replace the application that gets sent to 64-bit clients with the new one right? If there is no real performance increase, I could let you know about it, fairly quickly, and then we'd know if you should run the compile for future iterations of the Charmm app or if it was a waste of time...? |
||
|
ID:
2941 | Rating: 0
| rate:
/
|
||
|
David Ball
Forum moderator Volunteer tester 
Joined: Sep 18 06 Posts: 274 ID: 115 Credit: 1,634,401 RAC: 0 |
||
|
No cpu specific optimizations are utilized because the application has to run on so many different processors without crashing. And yes, this is a restriction, but hardly one that can be avoided in a boinc project. Just wanted to make a few comments: 1. It would probably mean more HR classes. 2. Don't the Intel compilers support compilation for those features and falling back to not using them if they detect an AMD cpu or lack of support for the feature at runtime? If it means a big speed increase, it might be worth doing. 3. Alternately, the application could include 2 or 3 versions of Charmm and the wrapper could decide which to execute. It would mean that everyone downloaded the 2 or 3 versions of Charmm and only used one of them, but the big download would only occur when versions of Charmm change. 4. There are now going to be SSE4 instructions. Happy Crunching! -- David ____________ The views expressed are my own. Facts are subject to memory error :-) Have you read a good science fiction novel lately? |
||
|
ID:
2942 | Rating: 0
| rate:
/
|
||
|
Andre Kerstens
Forum moderator Project tester Volunteer tester
Joined: Sep 11 06 Posts: 749 ID: 1 Credit: 15,199 RAC: 0 |
||
|
I'll ask Memo to look into this. Enabling some optimization for 64-bit clients makes sense to me. We'll have to see if we can enable extensions so specifically with compiler options and whether the app will still work if that extension wouldn't be available. But I'm sure we can figure that out by doing a bit of research :-)
|
||
|
ID:
2946 | Rating: 0
| rate:
/
|
||
|
I wait with baited breath :) Another project to turn my C2D on 64 bit Linux loose on. Better still if i ever finish building my c2Q something to really let it loose on.
|
||
|
ID:
2971 | Rating: 0
| rate:
/
|
||
|
What about the intel-mac application? No need to worry about old processor support. The *oldest* chips is the Core Solo/Duo. Which means that all intel Macs support the following:
|
||
|
ID:
2973 | Rating: 0
| rate:
/
|
||
|
What about the intel-mac application? No need to worry about old processor support. The *oldest* chips is the Core Solo/Duo. Which means that all intel Macs support the following: But Core Duo don't support 64bit so for maximum support, you still need 32bit apps. |
||
|
ID:
2986 | Rating: 0
| rate:
/
|
||
|
But Core Duo don't support 64bit so for maximum support, you still need 32bit apps. Not sure I follow. I didn't say anything about 64 bit. I asked about optimization used for the Mac applications. ____________ Dublin, CA Team SETI.USA |
||
|
ID:
2989 | Rating: 0
| rate:
/
|
||
|
But Core Duo don't support 64bit so for maximum support, you still need 32bit apps. I agree. There certainly aren't any Intel / Mac computers around that could not take advantage of many of the same extensions. I hear that in some cases the processing time can be halved or more. The compiles have to be made anyway, they might as well contain these optimizations. |
||
|
ID:
2999 | Rating: 0
| rate:
/
|
||
|
I hear that in some cases the processing time can be halved or more. The speed up from 64 bits and the compiler flags differ heavly on the application and on the architecture. Some applications could also run slower with 64 bit compared with 32 bits. My expiriance is about -5 to 10 % faster with heavy optimized flags. Also about -15 to +15 % with 64 bits and good optimisation. If you get more, you have luck... |
||
|
ID:
3002 | Rating: 0
| rate:
/
|
||
|
What about the intel-mac application? No need to worry about old processor support. The *oldest* chips is the Core Solo/Duo. Which means that all intel Macs support the following: Bump? Do the Intel Mac and/or the PPC Mac applications take advantage of things like SSE2 and/or AltiVec? If not, any plans to do so? Seems a waste not to, since no legacy support is required for either. ____________ Dublin, CA Team SETI.USA |
||
|
ID:
3018 | Rating: 0
| rate:
/
|
||
|
But Core Duo don't support 64bit so for maximum support, you still need 32bit apps. Oops sorry. I thought we were talking about 64 bit apps as the title suggest.My bad. |
||
|
ID:
3031 | Rating: 0
| rate:
/
|
||
|
Andre Kerstens
Forum moderator Project tester Volunteer tester
Joined: Sep 11 06 Posts: 749 ID: 1 Credit: 15,199 RAC: 0 |
||
|
We compile the Mac Intel and PPC app with the -O2 compiler flag set, which does a lot of optimization of code. We'll check into the sse2/altivec optimizations specifically and enable these if not done by O2 yet. Thanks Andre ____________ D@H the greatest project in the world... a while from now! |
||
|
ID:
3081 | Rating: 0
| rate:
/
|
||
|
We compile the Mac Intel and PPC app with the -O2 compiler flag set, which does a lot of optimization of code. We'll check into the sse2/altivec optimizations specifically and enable these if not done by O2 yet. Great! Please keep us updated! ____________ Dublin, CA Team SETI.USA |
||
|
ID:
3087 | Rating: 0
| rate:
/
|
||
|
I'll ask Memo to look into this. Enabling some optimization for 64-bit clients makes sense to me. We'll have to see if we can enable extensions so specifically with compiler options and whether the app will still work if that extension wouldn't be available. The AMD64 ABIs for both Linux and Windows require SSE2 support. Therefore, when a 64-bit application is built by default it uses SSE and SSE2 for all single and double-precision floating-point calculations. And given that SSE and SSE2 can be taken for granted on AMD64 systems, it follows that vectorization can also be used. So, Aaron is right on suggesting that the 32-bit application sent to AMD64 clients could have SSE and SSE2 enabled as well as vectorization. IIUC, Docking uses the GCC compiler. Therefore, make sure to add "-msse2 -mfpmath=sse -ftree-vectorize" (-msse2 implies -msse) when building a 32-bit application for AMD64 clients. The Intel compiler generates code for several processors and at run-time runs the code better suited depending on the features supported. However, it's known to run the slowest code the possible when run on AMD processors. The MS compiler however doesn't support vectorization, but it does support SSE and SSE2 through the option "/arch:sse2". HTH ____________ |
||
|
ID:
3088 | Rating: -1
| rate:
/
|
||
|
Andre Kerstens
Forum moderator Project tester Volunteer tester
Joined: Sep 11 06 Posts: 749 ID: 1 Credit: 15,199 RAC: 0 |
||
|
We only use the gcc compiler (g95 for the fortran code) on Mac PPC. For all the other platforms we use the Intel compiler.
____________ D@H the greatest project in the world... a while from now! |
||
|
ID:
3089 | Rating: 0
| rate:
/
|
||
|
any update on a x64 sse windows app?
|
||
|
ID:
5103 | Rating: 0
| rate:
/
|
||
Message boards : Application Info : 64-Bit / 32 bit, SSE3 and other extensions?
Database Error: The MySQL server is running with the --read-only option so it cannot execute this statement
array(3) {
[0]=>
array(7) {
["file"]=>
string(47) "/boinc/projects/docking/html_v2/inc/db_conn.inc"
["line"]=>
int(97)
["function"]=>
string(8) "do_query"
["class"]=>
string(6) "DbConn"
["object"]=>
object(DbConn)#24 (2) {
["db_conn"]=>
resource(102) of type (mysql link persistent)
["db_name"]=>
string(7) "docking"
}
["type"]=>
string(2) "->"
["args"]=>
array(1) {
[0]=>
&string(51) "update DBNAME.thread set views=views+1 where id=210"
}
}
[1]=>
array(7) {
["file"]=>
string(48) "/boinc/projects/docking/html_v2/inc/forum_db.inc"
["line"]=>
int(60)
["function"]=>
string(6) "update"
["class"]=>
string(6) "DbConn"
["object"]=>
object(DbConn)#24 (2) {
["db_conn"]=>
resource(102) of type (mysql link persistent)
["db_name"]=>
string(7) "docking"
}
["type"]=>
string(2) "->"
["args"]=>
array(3) {
[0]=>
object(BoincThread)#3 (16) {
["id"]=>
string(3) "210"
["forum"]=>
string(2) "11"
["owner"]=>
string(3) "367"
["status"]=>
string(1) "0"
["title"]=>
string(43) "64-Bit / 32 bit, SSE3 and other extensions?"
["timestamp"]=>
string(10) "1246562711"
["views"]=>
string(4) "2409"
["replies"]=>
string(2) "18"
["activity"]=>
string(22) "1.2315024685712999e-87"
["sufferers"]=>
string(1) "0"
["score"]=>
string(1) "0"
["votes"]=>
string(1) "0"
["create_time"]=>
string(10) "1175656419"
["hidden"]=>
string(1) "0"
["sticky"]=>
string(1) "0"
["locked"]=>
string(1) "0"
}
[1]=>
&string(6) "thread"
[2]=>
&string(13) "views=views+1"
}
}
[2]=>
array(7) {
["file"]=>
string(63) "/boinc/projects/docking/html_v2/user/community/forum/thread.php"
["line"]=>
int(184)
["function"]=>
string(6) "update"
["class"]=>
string(11) "BoincThread"
["object"]=>
object(BoincThread)#3 (16) {
["id"]=>
string(3) "210"
["forum"]=>
string(2) "11"
["owner"]=>
string(3) "367"
["status"]=>
string(1) "0"
["title"]=>
string(43) "64-Bit / 32 bit, SSE3 and other extensions?"
["timestamp"]=>
string(10) "1246562711"
["views"]=>
string(4) "2409"
["replies"]=>
string(2) "18"
["activity"]=>
string(22) "1.2315024685712999e-87"
["sufferers"]=>
string(1) "0"
["score"]=>
string(1) "0"
["votes"]=>
string(1) "0"
["create_time"]=>
string(10) "1175656419"
["hidden"]=>
string(1) "0"
["sticky"]=>
string(1) "0"
["locked"]=>
string(1) "0"
}
["type"]=>
string(2) "->"
["args"]=>
array(1) {
[0]=>
&string(13) "views=views+1"
}
}
}
query: update docking.thread set views=views+1 where id=210