XProfan

- Page 3 -

Julian
Schmidt

iF (29.03.13)

it'll The speed of FPU-Operationen
measured, The from the CPU on The FPU dependent go.

and for Dau.

iF (29.03.13)

what comes because with you out? Perhaps. can the Yes herewith vergleichen:

170GFlops with the oberen bill.

otherwise

170 * 8 * 4 = 5440
465 * 4 * 3,4 = 6324

mflops.exe	2.806 kB

Hochgeladen:	03/29/13

Downloadcounter:
		Download

03/29/13 ▲

Öhm I bezweifle first that with you 170GFlops measured go,
to that comparison access my I7 thereafter strain time 465MFlops.

03/29/13 ▲

Julian
Schmidt

is correct are 170MFlops. my Error.

03/29/13 ▲

Julian
Schmidt

yet with the thing or has the Topic for you Done?

03/30/13 ▲

E.T.

Perhaps. isses yes a Flop with the Flop's

XProfan X2
Grüße aus Sachsen... Mario
WinXP, Win7 (64 Bit),Win8(.1),Win10, Win 11, Profan 6 - X4, XPSE, und 'nen schwarzes, blinkendes Dingens, wo ich das alles reinschütte...

03/30/13 ▲

>> wieviele Flops(Floating Point Operations Per Second) one processor has

The gepostete function misst, wieviele FlOps the Process abrufen can. thereby
is naturally To deliberating, what very this worth at all wiederspiegelt, too
because the Messprogramm of a OS dependent becomes, that in addition others
processes run out can and with Mehrkern-Prozessoren The Last too distributed.

i'd the value therefore sooner not the Number of Kerne multiplizieren
separate it so let and weiterverarbeiten How he's, there it against well
wiederspiegelt, what a normal windows-Process maximum abrufen could.

so would the worth to that comparison of processor apiece windows-versions suitable.

03/31/13 ▲

- Page 4 -

Nu what draus become? oO

05/16/13 ▲

Nu what ( [...]

) draus become? oO

05/16/13 ▲

Julian
Schmidt

Jup, habs time on different computer run. yields itself no comparison To others Benchmarks.

05/16/13 ▲

I wealth that the on it lying, that other Tests Results by
Mischkalkulationen herbilden. The question wieviele Flops a normal windows-
User-APP put down can answers The function imho integrally correctly..

05/17/13 ▲

p.specht

small Bench bzgl. FLOPS
=====================
{$clq}
goto "AA"
CC:
t&=&gettickcount
e!=mult(n&,r!,p!,q!)
t&=&gettickcount-t &

print stature$("#0.0000",4*n&/(t &+10^-35)),"kFLOPS",stature$("%g",e!-r!)

waitinput 1
case %key=27:end
case %csrlin>26:locate 1,1
BB:
goto "CC"

AA:
cls
font 2
var p!=pi()
var q!=1/pi()
var r!=exp(1)
var n&=100000000 'anhängen for xpse
var e!=exp(1)
declare t &
goto "BB"

'proc mult ' Interpreter, ProfComp.exe
nproc mult
parameters n&,r!,p!,q!
whileloop n&
r!=p!*q!*r!*q!*p!
endwhile
return r!
endproc

/* Results:
legend to the nachfolgenden data:

any Testwerte relating itself on 1 core!
1.: worth for Interpreter (kFlops)
2.: Profan-Compiler (kFlops)
3.: nProc MFlops
4.: xpia/JWasm (MFlops)
5.: Sonstiges, Notes How 'SSE4.2 benutzt' etc.
TESTKANDIDATEN
============
A: Laptop Medion Akoya E6220 with P6000 2-Core CPU 1,87 GHz/4GB,
Win64-SP1/Profan11.2a(32bit): 83 / 342 / 107 / 111 (Basisregister+FPU)

B: Desktop Medion Akoya M5432 / i7-Quadcore 2.83 GHz /4 GB,
Win64-SP1/Profan11.2a(32bit): (Test runs)

C: ...
*/

Computer: Gerät, daß es in Mikrosekunden erlaubt, 50.000 Fehler zu machen, zB 'daß' statt 'das'...

08/29/13 ▲

with this Testformel beget You imho More Overhead as FPOps.

therefore I had herewith:

iF (29.03.13)

iF (29.03.13)

@Julian: so, GEHTS me now again rather -

have you not forget...

here one code of a 32-bit-Opcode created the 250 time 1.000.000 FPU command
anweist:

Download
external Download

Compile Mark Separation

 {$cleq}
cls
print flopsTest(1000000,250,57.29577951308232087)
waitinput
end

flopsTest(long opcodeOpsC,enumC,float theFloatValueToTestWith){

    long opcode=genFPUOpcode(opcodeOpsC,theFloatValueToTestWith)
    long tme=getTickCount

    whileLoop enumC { call(opcode-7) }

    tme=getTickCount-tme
    globalFree(opcode)
    return tme

}

nProc genFPUOpcode(long opcodeOpsC,float theFloatValueToTestWith){

    mul opcodeOpsC,2
    long floatVal=dim(8)
    setFloat(floatVal,0,theFloatValueToTestWith)
    long opcode=dim(8+opcodeOpsC)
    setByte(opcode,0,$B8)//mov eax
    setLong(opcode,1,floatVal)
    setByte(opcode,5,$DD)//fld ptr eax
    setByte(opcode,7+opcodeOpsC,$C3)//ret0
    add opcode,7

    whileLoop 0,opcodeOpsC-1,2 {

        setByte(opcode,loop,$DC)

        if rnd(2) {

            setByte(opcode+1,loop,$30)

        } else {

            setByte(opcode+1,loop,$08)

        }

    }

    return opcode

}

my computer rechnet thereafter 250.000.000 Gleitkommawerte in 538ms.

whom first Parameter with flopsTest not higher as 1mio settle since the function
otherwise To big would.

pure FLOps as Opcodes in a area written and a Call on it discontinued so one even not too yet pure CPOps mitmisst.

07/09/14 ▲