Followers 0

# Fast sqrt

## 13 posts in this topic

I've been looking through the various threads on this forum about fast sqrt and inverse sqrt approximations, and have also been looking at the relevant code in the Doom 3 sdk, and have a couple of questions. 1. I understand enough about the methods used to know that it relies on the specifics of 32 bit IEEE floating point representation. So I assume that means the code breaks if the representation changes? What happens on 64 bit machines like the Mac G5? 2. And, is it worth it? Or would you generally be better off (and close to as fast) using the standard functions, which I assume are gauranteed to be portable and consistent from platform to platform? (Please excuse me if the question seems naive, but it involves some areas I don't know much about, i.e. floating point architecture, etc.)
0

##### Share on other sites
there should be no need for it nowadays. and it's a fast inverse sqrt. Use the normal stuff, and see if it bottleneck your application. Nothing prevents you to have a float finvsqrt(float x) { ... } function in your math lib, then the implementation is left to you and the platform it's running on (so SSE/sqrt+div/Carmack for PC, and whatever for Macs).

I don't think yo ushould worry about it too much :)
0

##### Share on other sites
Look
float InvSqrt (float x){    float xhalf = 0.5f*x;    int i = *(int*)&x;    i = 0x5f3759df - (i >> 1);    x = *(float*)&i;    x = x*(1.5f - xhalf*x*x);    return x;}

And i got it from
here

Yes, it breaks when the representation changes. use the #defines!

You usually never need it. But if you need a lot of sqrts, use the above code.
From,
Nice coder
0

##### Share on other sites
it typically takes LOTS of sqrt calls to mean anything in terms of slowdown.

I suggest to all people that have a problem with using sqrt that they put a loop in their code somewhere that keeps calling sqrt. You can typically put it in thousands of times before you even see a FPS drop.
0

##### Share on other sites
I messed around with the functions a little and did some completely unscientific tests (if I remember correctly, in earlier threads on the subject people did some very rigorous comparisons).

I compared three versions: 1 / sqrtf(), the code from the above post, and the code from Doom 3. Doom 3 uses the same principal, but calculates the seed on the fly using some constants and a lookup table.

Amazingly enough (and unless I messed up somewhere) DoomInvSqrt() returned exactly the same results as 1 / sqrtf(). So no accuracy problems there. And Q3InvSqrt() was plenty close.

I just did a brute-force test - 10,000,000 calls to each function. The Q3 and Doom versions were about 1.8 times as fast as 1 / sqrt().

I would use the Doom version, but I imagine the code is copyrighted, and I don't understand it well enough to recreate it for myself. But I suppose the other code is fair game.
0

Uh...AP = jyk.
0

##### Share on other sites
can you copyright just a few lines of code? its probablly just an implementation of something discovered 200 years ago, if not longer ago.
0

##### Share on other sites
Is it Newton's Method you want to copywrite?
0

##### Share on other sites
Yes, I believe both versions (Q3 and Doom) use Newton's method. However, the Doom version uses some intricate and specific bit manipulation using lookup tables to find the seed. That's what I'm suspecting may be copyrighted.
0

##### Share on other sites
I did a benchmark on various float-type sqrt routines, and the fastest one was the 'sqrtf' in MSVC's standard library (equal to inline assembly). The post is here somewhere on gamedev, but I have no idea where.

The only way you'll get speedup is using the inverse square root formula to actually calculate the inverse square root, and iirc the win was barely one.

Well, you could also get a win using one of the functions that returns much less precision than sqrtf, but in that case you're not really comparing like things.
0

##### Share on other sites
There's a really cool paper about this fast sqrt trick, where everything is derived. It's from the thread posted here by Nice Coder.

invsqrt
0

##### Share on other sites
Isn't there a SSE sqrt instruction? Anyone know how that compares?
0

##### Share on other sites
I'm going to close the thread, since the topic has been discussed in immense detail throughout the archives.

In addition to the other thread created within the past few days on the same subject, I found many, many, MANY discussions on fast sqt dating back to 2001. I don't see much new in this thread. If anyone has a compelling argument why the thread should remain open, please send me a private message and make your case strongly. The topic of fast sqrt has been covered to death and I will double check any argument to reopen the thread against the archives to see if the argument holds water.
0

##### Share on other sites
OK,

I reopened the thread just to post this example from superpig. It seems like a good contribution that may be useful. The thread is closing back immediately.

Quote:
 From superpig via private messageSSE has both SQRT and RSQRT instructions, but you don't really get much benefit unless you're doing four of them at once. Say you want to get the lengths of four vectors, stored as a structure of arrays:__declspec(align(16)) struct blockOfFourVectors{float xValues[4];float yValues[4];float zValues[4];float lengths[4];}blockOfFourVectors data;__asm{ movaps xmm0, [data + 0x00] ; load x components movaps xmm1, [data + 0x10] ; load y components movaps xmm2, [data + 0x20] ; load z components ; square each component mulaps xmm0, xmm0 mulaps xmm1, xmm1 mulaps xmm2, xmm2 ; sum them into xmm0 addps xmm0, xmm1 addps xmm0, xmm2 ; sqrt to get length sqrtps xmm0, xmm0 ; save out movaps [data + 0x30], xmm0}That would calc all four lengths into data.lengths. For just a single vector it's not really worthwhile (and unless you store that vector in a SoA, would require a load of shuffling).

0