Comment Feed for Are Virtual Function Calls Still Expensive? (Coffeehouse on Channel 9)

Re: Are Virtual Function Calls Still Expensive?

AndyC — Fri, 24 Aug 2007 09:17:09 GMT

prencher wrote:

Yes, he did win. But he expended a whole lot more time and expertise to get it faster than Rico's basic C# application. That was the point of my post; If you're not an expert in the area, you will likely be able to get to your performance goals a lot easier and faster with e.g. XNA.

It's also worth noting that Rico really only lost in the end because of the startup time of the CLR. For a small sample like that, it's pretty much inevitable that it will impact significantly on the overall run time. For a more meaty application the time taken to start the CLR would be considerably less important.

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Fri, 24 Aug 2007 09:08:20 GMT

thumbtacks2 wrote:

littleguru wrote: Have you ever seen an intelligent fish
Yes, I have...they're called "game fish". They evade my hook more often than not...

I don't know how intelligent muskies are, but they do tend to survive a long time in most lakes and grow rather large (4-5 foot range in some cases).

Yay. But they eat just everybody else :P The sharks or whales in our game tend to survive also very long ;)

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 20:57:14 GMT

littleguru wrote:
Have you ever seen an intelligent fish

Yes, I have...they're called "game fish". They evade my hook more often than not... :(

I don't know how intelligent muskies are, but they do tend to survive a long time in most lakes and grow rather large (4-5 foot range in some cases).

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Thu, 23 Aug 2007 18:36:34 GMT

thumbtacks2 wrote:

littleguru wrote: Tons of stupid fishes
What? No AI ability?

Have you ever seen an intelligent fish ;) They make sure not to collide, but sometimes they do and one gets eaten :P And well it was just 40 hours in total that we coded... The other dude really went into it... He enjoyed it so much - you know: night sessions until 8-9 in the morning.

Re: Are Virtual Function Calls Still Expensive?

prencher — Thu, 23 Aug 2007 18:34:41 GMT

DigitalDud wrote:

prencher wrote:
In many ways it's similar with C# / C++.. For a lot of stuff C# can be faster, simply because it takes a lot of effort to make the C++ faster (See RicoM and Raymonds blog series where raymond did a C app and Rico did a C# one).

Yeah and Raymond ended up winning, no? In my experience, the .NET JIT compiler does not optimize nearly as aggressively as does C compilers, like VC++. It's limited by its requirement that it perform compilation in a very short amount of time.

Also with startup (especially cold) and memory footprint, C# will really lose here.

Yes, he did win. But he expended a whole lot more time and expertise to get it faster than Rico's basic C# application. That was the point of my post; If you're not an expert in the area, you will likely be able to get to your performance goals a lot easier and faster with e.g. XNA.

Everything has it's place, obviously, and I'm a C++ buff myself, but you don't get significantly better perf "by default" by using C++, not by a long shot.

In the case of games, at least the non-"AAA" titles, XNA is in many cases very fitting because the bulk of the performance depends on the graphics, and there XNA is very close to native.

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 18:23:06 GMT

littleguru wrote:
Tons of stupid fishes

What? No AI ability?

;)

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 18:21:35 GMT

littleguru wrote:
If you have some time try to have a look at paralax mapping and when you have enough of that at relief mapping. Relief mapping is just awesome (go to page 44 and following). It produces so nice outputs.

Interesting...I'll have to check those out. Much of this also depends on the camera angle I pick...whether it is a FPS-type angle (from a ground perspective) or something elevated (like looking down at a 45 degree angle onto the terrain).

Re: Are Virtual Function Calls Still Expensive?

RoyalSchrubber — Thu, 23 Aug 2007 18:17:02 GMT

ZippyV wrote:

littleguru wrote: actually it were 2,000 to 10,000 triangles per fish...

That's a lot for a stupid fish.

Off-topic: Have you guys seen the demo videos from Nvidia?

Human head: http://us.download.nvidia.com/downloads/nZone/videos/nzm_humanhead.wmv

Cascades (totally awesome water and pixel shader effect:
http://us.download.nvidia.com/downloads/nZone/videos/nzm_Cascades_tech.wmv

HOLY SH!T

Pure awesomeness. We need screensavers like this!!

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Thu, 23 Aug 2007 18:12:08 GMT

ZippyV wrote:

littleguru wrote: actually it were 2,000 to 10,000 triangles per fish...

That's a lot for a stupid fish.

Off-topic: Have you guys seen the demo videos from Nvidia?

Human head: http://us.download.nvidia.com/downloads/nZone/videos/nzm_humanhead.wmv

Cascades (totally awesome water and pixel shader effect:
http://us.download.nvidia.com/downloads/nZone/videos/nzm_Cascades_tech.wmv

Tons of stupid fishes :D

Off-topic: Now if they would be able to make the drivers work everything would be fine. It's nice to see how much research is going on in the areas and what they output. Nice videos.

ATi has also some very nice demo videos. Like the Ruby: Whiteout one.

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Thu, 23 Aug 2007 18:02:41 GMT

thumbtacks2 wrote:

DigitalDud wrote:

thumbtacks2 wrote:
That's been a major consideration of mine as of late, because in this iteration, the engine (and the game) will take place mostly outdoors.
Just use an octtree to subdivide volume, nice and simple for outdoor environments.
I'm looking at the whole gambit of data structures/containers out there...from octtrees to BSP trees to a bunch of other ideas. If it doesn't hinder performance, and doesn't take too long to implement, I may also try using a custom container. It's not that I don't find use in BSP trees or things like that, but I want to experiment with a few things. But the proof will be in the implementation, I guess.

If you have some time try to have a look at paralax mapping and when you have enough of that at relief mapping. Relief mapping is just awesome (go to page 44 and following). It produces so nice outputs.

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 17:51:41 GMT

DigitalDud wrote:

thumbtacks2 wrote:
That's been a major consideration of mine as of late, because in this iteration, the engine (and the game) will take place mostly outdoors.
Just use an octtree to subdivide volume, nice and simple for outdoor environments.

I'm looking at the whole gambit of data structures/containers out there...from octtrees to BSP trees to a bunch of other ideas. If it doesn't hinder performance, and doesn't take too long to implement, I may also try using a custom container. It's not that I don't find use in BSP trees or things like that, but I want to experiment with a few things. But the proof will be in the implementation, I guess.

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Thu, 23 Aug 2007 17:50:55 GMT

thumbtacks2 wrote:

littleguru wrote: actually it were 2,000 to 10,000 triangles per fish... I forgot to mention that. Btw. try also to use some features of MMX or SSE (or even SSE2) if you can. That's also a booster in some areas.
Thanks. Were those triangles complete with textures also? Or just shading? I have yet to check out your game, btw, but will probably do so at some point...

All textured! Pixel shader and vertex shader runs also on all the fishes and the ocean ground.

Just download it and run it - no install required. And press F1 to get a little help (inlay) that shows you also the keys to switch into wireframe mode, switch the texture quality, mipmap quality, rendering mode (immediate mode, vertex arrays, VBOs), toggle frustum culling and shaders and switch through the available anisotropy levels.

It's very nice to compare performance of the different featurs, since we always output the fps. :)

Re: Are Virtual Function Calls Still Expensive?

ZippyV — Thu, 23 Aug 2007 17:49:56 GMT

littleguru wrote:
actually it were 2,000 to 10,000 triangles per fish...

That's a lot for a stupid fish.

Off-topic: Have you guys seen the demo videos from Nvidia?

Human head: http://us.download.nvidia.com/downloads/nZone/videos/nzm_humanhead.wmv

Cascades (totally awesome water and pixel shader effect:
http://us.download.nvidia.com/downloads/nZone/videos/nzm_Cascades_tech.wmv

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 17:41:26 GMT

littleguru wrote:
actually it were 2,000 to 10,000 triangles per fish... I forgot to mention that. Btw. try also to use some features of MMX or SSE (or even SSE2) if you can. That's also a booster in some areas.

Thanks. Were those triangles complete with textures also? Or just shading? I have yet to check out your game, btw, but will probably do so at some point...[H]

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Thu, 23 Aug 2007 17:25:01 GMT

thumbtacks2 wrote:

littleguru wrote: In our game (FishSalad) for university, we had like 60+ fishes (with 2,000-10,000 triangles and pixel + vertex shaders) on the screen. Each of these objects was an instance of a class that had virtual methods, because it inherited from a base class (our base model class). Well it worked very well.
That's good to know...I also found a nice 175 slide presentation from SIGGRAPH 2003 here. It's a little outdated, and maybe the info is a rehash of techniques found elsewhere.

littleguru wrote: What's the biggest problem is the GPU. You need to make sure that stuff that is not in visible range is not drawn etc. That will higher your fps a lot.
That's been a major consideration of mine as of late, because in this iteration, the engine (and the game) will take place mostly outdoors.

actually it were 2,000 to 10,000 triangles per fish... :) I forgot to mention that. Btw. try also to use some features of MMX or SSE (or even SSE2) if you can. That's also a booster in some areas. :)

Re: Are Virtual Function Calls Still Expensive?

DigitalDud — Thu, 23 Aug 2007 17:21:46 GMT

prencher wrote:

In many ways it's similar with C# / C++.. For a lot of stuff C# can be faster, simply because it takes a lot of effort to make the C++ faster (See RicoM and Raymonds blog series where raymond did a C app and Rico did a C# one).

Yeah and Raymond ended up winning, no? In my experience, the .NET JIT compiler does not optimize nearly as aggressively as does C compilers, like VC++. It's limited by its requirement that it perform compilation in a very short amount of time.

Also with startup (especially cold) and memory footprint, C# will really lose here.

Re: Are Virtual Function Calls Still Expensive?

DigitalDud — Thu, 23 Aug 2007 17:12:42 GMT

thumbtacks2 wrote:

That's been a major consideration of mine as of late, because in this iteration, the engine (and the game) will take place mostly outdoors.

Just use an octtree to subdivide volume, nice and simple for outdoor environments.

Re: Are Virtual Function Calls Still Expensive?

evildictaitor — Thu, 23 Aug 2007 16:33:49 GMT

thumbtacks2 wrote:

littleguru wrote:What's the biggest problem is the GPU. You need to make sure that stuff that is not in visible range is not drawn etc. That will higher your fps a lot.
That's been a major consideration of mine as of late, because in this iteration, the engine (and the game) will take place mostly outdoors.

Actually auto-culling is one of the most important areas of research in games because of the vast improvement they can produce to the FPS. Bear in mind that lots of techniques can be used here, from bounding-box to bounding-sphere to internal culling and detail-removal. Most release games have many of these features running alongside each other, because a very fast render cycle will hide a multitude of sins.

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 16:17:30 GMT

littleguru wrote:
In our game (FishSalad) for university, we had like 60+ fishes (with 2,000-10,000 triangles and pixel + vertex shaders) on the screen. Each of these objects was an instance of a class that had virtual methods, because it inherited from a base class (our base model class). Well it worked very well.

That's good to know...I also found a nice 175 slide presentation from SIGGRAPH 2003 here. It's a little outdated, and maybe the info is a rehash of techniques found elsewhere.

littleguru wrote:
What's the biggest problem is the GPU. You need to make sure that stuff that is not in visible range is not drawn etc. That will higher your fps a lot.

That's been a major consideration of mine as of late, because in this iteration, the engine (and the game) will take place mostly outdoors.

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 16:10:24 GMT

Stebet wrote:
Or use Vertex Buffer Objects. I think most video drivers translate display lists into VBO's internally anyways.

Thanks...I'll look into those also.

Re: Are Virtual Function Calls Still Expensive?

prencher — Thu, 23 Aug 2007 15:44:41 GMT

Minh wrote:
In my game, Star Ranger, I used low poly models, had a generic (virtual calls) scene graph, used C++ to talk directly to DirectX, and I can't maintain vsync.

That is, when I turned vsync on, sometimes it'd drop below 60fps & so the game stutters for a little bit, which I didn't think was possible w/ C++.

I think it was because I didn't bother to batch any of my models, using a separate vertex buffer for each. I was gonna re-write it (doing better GPU management) w/ XNA, but just got side tracked.

Wouldn't that be cool? An XNA game performing better than C++ ?

Re: Are Virtual Function Calls Still Expensive?

Christian Liensberger — Thu, 23 Aug 2007 15:07:05 GMT

In our game (FishSalad) for university, we had like 60+ fishes (with 2,000-10,000 triangles and pixel + vertex shaders) on the screen. Each of these objects was an instance of a class that had virtual methods, because it inherited from a base class (our base model class). Well it worked very well. In games there are actually a lot of other things you should consider first.

Like people (speaking of other groups) were loading 500 MB of textures and crap at the beginning of each level or some were loading the whole background sound in, when the level started. THAT was annoying!

What's the biggest problem is the GPU. You need to make sure that stuff that is not in visible range is not drawn etc. That will higher your fps a lot. Our CPU was always down at like 10-20 % when had vsync enabled. Now without vsync enabled we are at 150+ fps and 100% cpu.

Re: Are Virtual Function Calls Still Expensive?

Stefán Jökull Sigurðarson — Thu, 23 Aug 2007 15:06:05 GMT

thumbtacks2 wrote:

Minh wrote: With games, optimizing for the video card is going to be much more crucial than optimizing for the CPU.

For example, if you batch all the draws for those models (even if they have virtual calls), you would see much better result than if you draw those models separately (even if they don't have virtual calls).
By "batching" the draws, I'm assuming you mean using display lists (I'm using OpenGL)...? I'm getting there....

Or use Vertex Buffer Objects. I think most video drivers translate display lists into VBO's internally anyways.

Re: Are Virtual Function Calls Still Expensive?

thumbtacks2 — Thu, 23 Aug 2007 14:44:08 GMT

Minh wrote:
With games, optimizing for the video card is going to be much more crucial than optimizing for the CPU.

For example, if you batch all the draws for those models (even if they have virtual calls), you would see much better result than if you draw those models separately (even if they don't have virtual calls).

By "batching" the draws, I'm assuming you mean using display lists (I'm using OpenGL)...? I'm getting there....:)