Faster memory copy (optimized for AMD)

Max.Â² · Post by **Max.Â²** » Thu Nov 25, 2004 9:34 pm

Bonne_den_kule wrote:Looks like it is big difference between a AthlonXp 2400 and 2500. I have only 512MB, cheap memory. Here is my results.

Guess it mainly depends on the frontside bus and the chipset. XP 2500 are generally FSB 333, while XP 2400 I think were available at 266 and 333.

I got an XP 2400/266 and therefor was a bit surprised by traumatic's results, as XP 1600 always are 266 too. Memory is pretty fast for my machine (CL2) so probably the OS and the chipset jump in.

Bonne_den_kule · Post by **Bonne_den_kule** » Fri Nov 26, 2004 3:17 pm

Max.Â² wrote:
Bonne_den_kule wrote:Looks like it is big difference between a AthlonXp 2400 and 2500. I have only 512MB, cheap memory. Here is my results.
Guess it mainly depends on the frontside bus and the chipset. XP 2500 are generally FSB 333, while XP 2400 I think were available at 266 and 333.

I got an XP 2400/266 and therefor was a bit surprised by traumatic's results, as XP 1600 always are 266 too. Memory is pretty fast for my machine (CL2) so probably the OS and the chipset jump in.

I am using 333 fsb (166hz). but 2700 512 mb ram (one block).

fsw · Post by **fsw** » Fri Nov 26, 2004 11:11 pm

Excellent work

BTW: I don't want to offend anybody, specially not Fred - HI FRED - but shouldn't stuff like that be covered by the PureBasic compiler

Under 'compiler options', in the editor, it suggests that it's able to OPTIMIZE for ALL CPU's, MMX, SSE, SSE2 and 3DNOW, but actually it seems it doesn't 8O

fsw · Post by **fsw** » Sat Nov 27, 2004 2:27 am

Forgot to post my results:

---------------------------
Test Results 1st run:
---------------------------
--- 64 MB tranfer test ---
AMD Function : 281
Pure Function : 4606
AMD Function is 16 times faster.

--- 1 MB tranfer test ---
AMD Function : 2464
Pure Function : 61278
AMD Function is 24 times faster.

--- 100kb tranfer test ---
AMD Function : 30
Pure Function : 5538
AMD Function is 184 times faster.

Impressive 8O

---------------------------
Test Results 2nd run:
---------------------------
--- 64 MB tranfer test ---
AMD Function : 251
Pure Function : 3395
AMD Function is 13 times faster.

--- 1 MB tranfer test ---
AMD Function : 2473
Pure Function : 55310
AMD Function is 22 times faster.

--- 100kb tranfer test ---
AMD Function : 20
Pure Function : 5087
AMD Function is 254 times faster.

Even More Impressive 8O

SONY Notebook Duron950, 256mb ram, WinXP

Post by **Fred** » Mon Nov 29, 2004 12:42 pm

That's indeed impressive results. As FSW, I will add this optimization to the CopyMemory() command, so it can take advantage of new CPU trough the compiler switch.

GeoTrail · Post by **GeoTrail** » Mon Nov 29, 2004 12:53 pm

Hi Fred.
It's great to see someone like you taking note of tests we users do and do something about it. More developers should be alot more like you

fsw · Post by **fsw** » Mon Nov 29, 2004 8:05 pm

Fred wrote:That's indeed impressive results. As FSW, I will add this optimization to the CopyMemory() command, so it can take advantage of new CPU trough the compiler switch.

Thanks

coma · Post by **coma** » Mon Dec 06, 2004 11:15 pm

Excellent El_Choni, if you have any other tips like this..... :)

akee · Post by **akee** » Tue Dec 07, 2004 9:45 am

GeoTrail wrote:Hi Fred.
It's great to see someone like you taking note of tests we users do and do something about it. More developers should be alot more like you

That's why PB is a "kick butt" translator. I hardly click on my Visual Studio icon these days...

Dreglor · Post by **Dreglor** » Fri Dec 17, 2004 4:01 am

WOAH...
isn't there anything that that some one can't opimaize!

Psychophanta · Post by **Psychophanta** » Fri Dec 17, 2004 11:46 am

Fred wrote:That's indeed impressive results. As FSW, I will add this optimization to the CopyMemory() command, so it can take advantage of new CPU trough the compiler switch.

Hey, nice

But don't forget, please

Froggerprogger · Post by **Froggerprogger** » Fri Dec 17, 2004 11:59 pm

Here are results on an Pentium Mobile 2,4GHz 512 DDR:

--- 64 MB tranfer test ---
AMD Function : 200
Pure Function : 3175
AMD Function is 15 times faster.

--- 1 MB tranfer test ---
AMD Function : 2814
Pure Function : 53356
AMD Function is 18 times faster.

--- 100kb tranfer test ---
AMD Function : 40
Pure Function : 581
AMD Function is 14 times faster.

Cool!
2814ms instead of 53356ms !

newbie · Post by **newbie** » Sat Dec 18, 2004 5:46 pm

AMD Athlon XP 2000+ (+DDR 333Mhz)

hooah

EDIT : I have a question :
Does that optimization means that it is by now faster to write :

Code: Select all

var1.s = "toto"
var2.s = ""

CopyMemoryAMD(@var1, @var2, 4)

than to use the normal way :

Code: Select all

var2 = var1

??

El_Choni · Post by **El_Choni** » Sat Dec 18, 2004 8:31 pm

BTW, Fred, if you plan to add this routine to PB, remember to include the "align" directives I had to comment in the code due to the fact that "align" is redefined by PB to be used only in the data section.

Regards,

Andras · Post by **Andras** » Tue Dec 21, 2004 10:46 pm

AMD-Athlon XP 2400+

---------------------------
Test Results
---------------------------
--- 64 MB tranfer test ---

AMD Function : 94

Pure Function : 1734

AMD Function is 18 times faster.

--- 1 MB tranfer test ---

AMD Function : 578

Pure Function : 27156

AMD Function is 46 times faster.

--- 100kb tranfer test ---

AMD Function : 16

Pure Function : 406

AMD Function is 25 times faster.