Irrlicht Engine

hendu wrote:Too slow to be useful on a very-high-end card + requires compute ability = surefire win, must be

we are already half done with this system. Its running fine on the CPU, and currently being translated to the GPU.

here is a live test-screenshot, of the work in progress



prepare to tilt!

You claim to have something that runs too slow on a GTX 480 running faster on a cpu? Dear trolls, I call your bluff.

Should it be true, I'm confident your paper on the improvements will be published ASAP?

^because only idiots recompute everything every frame^

In case you were wondering, this is why I call you trolls - you claim the impossible, and when you're questioned, you insult people.

The original voxel cone paper *shock* did not recompute everything every frame.

Okay.. voxel cone tracing takes approx 40ms on 720p on 480 GTX, if I remember correctly

25-45% of that was voxelizing triangles, we already have voxels XD
We can't have compute capability so we traded in RAM for DX10

This actually saved us from traversing a hierarchy/octree for every voxel sample XD - which actually makes up 50% of the sampling cost

That kinda shaves my cost from 1.35 TFLOP for almost realtime per pixel to 300-350 GFLOP.... a decent CPU has that across all cores

Obviously S.H.I.T. doesnt run per pixel


First of all, I love spherical harmonics... so I compute spherical harmonic AO distribution at very high quality across many frames (using 12+ cones per SH)

In essence its like almost realtime lightmap generation

Then I use CPU/OpenCL for the ACTUAL sparse voxel octree of injected irradiance and gather indirect lighting 3 times (our indirect is limited in length to the length of the most powerful light)

Insult?

@devsh

Thanks.

BTW, I do think that you should publish that somewhere, since I don't think anyone's done that before. If not a paper, a long blog post or so.

@sodandk

Maybe re-read your posts in this thread. You're not as bad as devsh, but still.

Some quotes:

hehe.... dang !

we rule... and we are modest...

These piling on someone else's post come off as extremely mob-like and insulting.

thats not insults... we are having fun... its a fun project...

oh, and btw., devsh is very talented... good to work with... (did I say that ? ... dont read here matt

BTW, I do think that you should publish that somewhere, since I don't think anyone's done that before. If not a paper, a long blog post or so.

I started a long thread on our still-offline website about clearing up the confusion around Spherical Harmonics, I will post full Mathematica 9 notebooks and maths equations for SH manipulation

As well as my proof of the deriviation of convolution of a light distribution function with a specular reflectance function

new screenies over at: http://www.facebook.com/media/set/?set= ... 875&type=3

when the ingame online model-library is ready, we can generate worlds with users models / constructions.
think of a big city with 50 uniques houses... or New York... or Venice or...

I attempted to do occlusion queries with Voxel Cone Tracing

I failed.. do not do it, its a waste of time (its good for fuzzy visibility detection for AI though!)

I shall go back to occlusion queries and implement quad-tree hierarchy occlusion queries.

Future Technology ....

http://www.youtube.com/watch?v=SNImu3n2_Go

sodandk wrote:thats not insults... we are having fun... its a fun project...

devsh wrote:Yes.. I'm doing the whole thing by myself mofos

That IS an insult (http://www.urbandictionary.com/define.php?term=mofos). There's a line between having fun and being cocky. Your project is certainly something to be proud about, but you can still be polite about it. Remember, we're reading text. We can't tell your fun intentions in a post, especially if you don't add things like emoticons or "/sarcasm" or "jk".

------------------------- (Line, since I don't want to post twice)

The video looks good. I like the water color, although I notice the top isn't transparent when looking from below towards the surface. It makes the water look like a cave. Is that intentional or is that just because it's evening?
The water flow back looks jumpy, or at least different from Minecraft as I recall.
Anyways, it still looks nice from the top.

Another question: Is the force field just one large cube or is it 6 walls?

I had problem with this line, from you ... (which I took as an insult)
>> I take it poor devsh isn't doing all of this. How many guys are on your team?
was that fun or ... ? devsh isnt a poor thing... he is very talentet! so ...

fine, lets keep clean here ...

----------------

the top is transparent, when you are underwater, but currently not when not in water. might be fixed.
The water flow back... we actually have a 'waterfall-shader' for that, but its not enabled.
The forcefield is basically indestructable transparent cubes this way it uses the normal general algorithm for cubes.
The algo have many setting per block... for example, can it burn etc... I tried to set all blocks burnable... entire world burned away... even water! hahahaha
The most tuff thing cpu wise, is the function that finds detached blocks, and marks them for falling. This function we hope to move to OpenCL or OpenACC or so...
As you see here, it will work with ANY type of structure, mountain or what ever: http://www.youtube.com/watch?v=c_G64vrzqxE