Comment Feed for David LeBlanc: Inside SafeInt (Going Deep on Channel 9)

Re: David LeBlanc: Inside SafeInt

David LeBlanc — Wed, 23 Sep 2009 04:15:30 GMT

Oh - I forgot to edit a couple of things - the suggested approach adds up to 5 clocks, not 6, which makes it 2.5x slower, not 3x.

Re: David LeBlanc: Inside SafeInt

David LeBlanc — Wed, 23 Sep 2009 04:07:22 GMT

Interesting question. I am not as good with perf as I would like, but in a situation like this, we're essentially counting instructions. Here's how it breaks down:

Method 1 (suggested) (11 instructions) 32-bit, not optimized

   unsigned char tmp = (unsigned char)i;
009A143E mov         al,byte ptr [i]
009A1441 mov         byte ptr [tmp],al
   if(tmp != i)
009A1444 movzx       eax,byte ptr [tmp]
009A1448 cmp         eax,dword ptr [i]
009A144B je          IntToUchar2+31h (9A1451h)
      return false;
009A144D xor         al,al
009A144F jmp         IntToUchar2+3Bh (9A145Bh)

   uc = (unsigned char)i;
009A1451 mov         eax,dword ptr [uc]
009A1454 mov         cl,byte ptr [i]
009A1457 mov         byte ptr [eax],cl
   return true;
009A1459 mov         al,1

Method 2 (original) (10 instructions)
   if(i < 0 || i > 255)
009A13DE cmp         dword ptr [i],0
009A13E2 jl          IntToUchar+2Dh (9A13EDh)
009A13E4 cmp         dword ptr [i],0FFh
009A13EB jle         IntToUchar+31h (9A13F1h)
      return false;
009A13ED xor         al,al
009A13EF jmp         IntToUchar+3Bh (9A13FBh)

   uc = (unsigned char)i;
009A13F1 mov         eax,dword ptr [uc]
009A13F4 mov         cl,byte ptr [i]
009A13F7 mov         byte ptr [eax],cl
   return true;
009A13F9 mov         al,1

Suggested method in release build:

01041050 0F B6 C8         movzx       ecx,al (3 clock)
01041053 3B C8            cmp         ecx,eax (1 clock)
01041055 75 03            jne         main+5Ah (104105Ah) (1 clock)

Current method in release build:

01041018 3D FF 00 00 00 cmp eax,0FFh (1 clock)
0104101D 77 06 ja main+25h (1041025h) (1 clock)

I also checked this in 64-bit, and it is the same instructions. In the suggested approach, we have an extra movzx, and it consumes a register, which would tend to cause surrounding code to optimize less efficiently on x86 (probably negligble on x64, but same effect). Due to the movzx being expensive (relative to cmp), the suggested approach is 6 cycles to the current approach's 2 cycles, or 3x worse. If the surrounding code were register constrained, as x86 often is, then you might need to push something on the stack to free a register, which would be more overhead.

To be fair, I chalk this up to pure, blind luck. While I did try to write the code to be as efficient as possible, readability and correctness were of higher importance. It is entirely possible that other operations may not be as efficient as possible. I think I did try to avoid temporary variables where possible, for just this reason, though in some cases it couldn't be avoided, and the effect would often be diluted by how expensive large multiplication and division operations are.

A large chunk of work that has not yet been done is to take all 64 combinations for each and every operation type and see exactly how they optimize, and dink with it to see if it can be better. As an aside, we considered using intrinsics for x64 multiplication, which would have been a lot faster, but decided not to in this release due to schedule constraints.

In addition, unless you happened to be using SafeInt inside of a rendering engine, or something else where a couple of cycles matter, a couple of cycles here and there are not going to add up to anything, considering that an allocation could cost several 10's of thousands of cycles, loading a COM object is much, much bigger, and so on.

Thanks for the question - I got to learn a bit about assembly that I didn't know before today.

Re: David LeBlanc: Inside SafeInt

jan.devaan — Wed, 02 Sep 2009 20:50:03 GMT

Nice video! It is a bit of a pity so many people are moving away C++ just when we are finding out how powerful it really is.

One question about the code though:

When casting from a large signed type to a smaller type (int to unsigned char), SafeInt now does a two-sided compare. Why not cast the int to a uchar, and check if its the same when cast again to an int? That's one if statement less. Should be smaller and a tiny bit faster.

Re: Re: Re: David LeBlanc: Inside SafeInt

sokhaty — Tue, 23 Jun 2009 02:52:34 GMT

Ale, thanks a ton for the pointers! Now I have to read them :)

Re: Re: David LeBlanc: Inside SafeInt

alecont — Sat, 20 Jun 2009 01:00:11 GMT

Hi sokhaty,

A book I really like on C++ templates is: C++ Templates: The Complete Guide (Hardcover), by David Vandevoorde, Nicolai M. Josuttis

Also, the classic Effective STL, by Scott Meyers is another great book.

David also suggests: STL Tutorial and Reference Guide, by David R. Musser, Gillmer J. Derge, Atul Saini.

HTH, and glad you liked the video! :-)

Ale Contenti
VC++ Dev Lead.

Re: David LeBlanc: Inside SafeInt

sokhaty — Thu, 18 Jun 2009 18:53:52 GMT

Ah, wonders of computing. Ideas are simple, yet 90+% of all the implementation work is about discovering and handling all the border cases :)

Btw, what were those (two?) good books mentioned in passing on the subject of partial template specialization (a.k.a. template meta programming)?

Re: David LeBlanc: Inside SafeInt

Christian Liensberger — Thu, 18 Jun 2009 16:20:17 GMT

Nice interview! I love these kind of things :)

Re: David LeBlanc: Inside SafeInt

LordKain — Wed, 17 Jun 2009 12:33:10 GMT

Another really interesting C++ video, thank you Charles! This class is really useful and beautifully written. I'll definitely start using it from now on in my projects as it is very easy to update existing code to take advantage of the features this class provides.

The C++ world is decidedly full of surprises, I'm looking forward for more C/C++ videos!

By the way, I'd love to see more videos going deep into some win32 stuff. There have been some interesting changes in these APIs during the last few years, and some new APIs will be included in Windows Seven (Web Service, Direct2D, etc.).