does my computer suck?

[sail3005]

i made a program, see thread "This code is pissing me off" that times how long it takes a computer to count to 1 billion. I ran it several times under different conditions, and the average was about 18 seconds. This seems very long for me, it seems that it should only take a few seconds. Any ideas?

my computer is a 450mhz PIII.

[kedaman]

I had to remove the optimations for msvc to get that i wanted to actually loop

; 234  : 	for(int c=0;c<1000000000;++c);

	mov	DWORD PTR _c$[ebp], 0
	jmp	SHORT $L8938
$L8939:
	mov	eax, DWORD PTR _c$[ebp]
	add	eax, 1
	mov	DWORD PTR _c$[ebp], eax
$L8938:
	cmp	DWORD PTR _c$[ebp], 1000000000		; 3b9aca00H
	jge	SHORT $L8940
	jmp	SHORT $L8939

Anyways it seems to take 6 instructions per loop i have a 1.33Ghz Thunderbird (running at 1.00Ghz currently) and it takes about 6 seconds. That makes sense: 1Billion*6~6G Cpu cycles.
This loop seems to be incredibly stupid anyway, I would use a register, and cut down the amount of instructions to 3. Probably something to do with the optimations.

Post your assembler listings and we'll see what's taking so long. I guess you have 8 cpu cycles (0.45*18=8)

[Wynd]

Originally posted by sail3005
does my computer suck?

Yes

[Zaei]

A billion is a pretty big number (ever counted to a thousand?). You might try making the loop counter "register" to force it into a register.

Z.

[kedaman]

Originally posted by Zaei
A billion is a pretty big number (ever counted to a thousand?). You might try making the loop counter "register" to force it into a register.

Z.

MSVC ignores that pretty much (ask parksie )

[jim mcnamara]

Now that sucks, to borrow parlance. This is a 'feature' that makes me mad.

Register absolutely does not work for MSVC++. The docset says it is completely ignored. Like we can't figure out how to optimize code. If something becomes 'register' it has nothing to do with whether you declare it that way or not. The compiler is smrter than us, just ask Kedaman

[Zaei]

Indeed that does suck. Well, that is why we have inline assembly.

From: MSDN
The 32-bit compiler does not honor user requests for register variables.

Samurai Zaei says: "What? You dare to not honor my request!? Prepare for death!"

Z.

[kedaman]

Originally posted by Zaei
Indeed that does suck. Well, that is why we have inline assembly.



Samurai Zaei says: "What? You dare to not honor my request!? Prepare for death!"

Z.

Problem is inline assembly and optimation does not work together, once you put something in asm everything the optimizer works so hard for to inline will be wasted. So inline asm is only good when you have something specific, so it's very uncomfortable to have around in the lower layers.

The compiler usually thinks it's smarter, but when you've finally finished your expression templates you'll notice what kinda dumbass it really is.

[CornedBee]

For a quick by-hand optimization:

	mov	eax, 0
$L8939:
	inc	eax
	cmp	eax, 1000000000		; 3b9aca00H
	jl	SHORT $L8939

You can make this faster with the LOOP instruction, but I don't know how to use it. The worst thing here is the jmp instruction, it flushes the prefetch queue (meaning the next instruction takes 5-7 cycles instead of 1).

[kedaman]

mov ecx, 1000000000
f:
loop f

That should do it

[sail3005]

Originally posted by kedaman

Post your assembler listings and we'll see what's taking so long. I guess you have 8 cpu cycles (0.45*18=8)

How do i get the ASM code? I wrote it in C++.

[kedaman]

You can select it to list asm code into a file in project>settings>C++>Category>listing files