Dynamic Cloud Shadows

[danielj]

OH,NO!Just ONE MORE THING that OBLIGATE you to BUY a Geforce 8!
Until the end of 2008,HOW MANY THINGS will not work in previous video cards?Soon,Celestia will be DEDICATED to Geforces 8...

FYI:

6600 / 163.71:


8800GTS / 163.69:


Adirondack

[ElChristou]

OH,NO!Just ONE MORE THING that OBLIGATE you to BUY a Geforce 8!

Danielj, there is no obligation in anything. You just skip this feature and all is fine!

[cartrite]

I found a way of getting rid of those ^M characters ( open with kwrite => Tools => End of Line => Unix => Save ) quite easily and was able to see the differences between the original files and the patch files. It seems like there is a lot of math done in the cpu with floating point and the result is sent to the gpu. Maybe the result is getting truncated.
cartrite

[chris]

OH,NO!Just ONE MORE THING that OBLIGATE you to BUY a Geforce 8!
Until the end of 2008,HOW MANY THINGS will not work in previous video cards?Soon,Celestia will be DEDICATED to Geforces 8...

The whole point of all this testing is to find out why the cloud shadows patch doesn't work on every graphics card and fix it so that it does work.

--Chris

[chris]

This probably doesn't affect the code (or could it?) but I was trying to use Kompare to see the differences between the new patch and cvs originals and I couldn't because Kompare shows the whole file shaddermanager.h as being different with respect to the original cvs version of the file because of the dreaded ^M character. I seen them in Emacs. Every line ends with ^M. The other 3 files have the ^M character too.
EDIT: Another thing I found is that the 8000 series started to support 128 bit floating point in case that matters.
cartrite

cartrite,

the ^M is nothing but the Windows line ending displayed in an UNIX editor. This happens since Chris L. works under Windows. Since some devs use Windows some use Linux, that's quite OK and taken care of.

The 128bit floating point might be an issue, of course. We really need to know whether there is a card just below G80 that does NOT show the blocking.

PS: When do you sleep actually?

I was wondering the same

Anyhow, the 128-bit float support on the G80 shouldn't make any difference. In fact, one of the features of my patch is that it's supposed to emulate high precision texture interpolation on graphics cards that don't support it natively (i.e. cards other than the GeForce 8 and Radeon X2xxx series.) But, it appears that the code is relying on some GeForce 8 feature anyhow, and I can't figure out what. The blocks in the shadows on non-G80 cards correspond with texels in the cubemap LUT. The sharp boundaries are baffling--even if the problem is inadequate precision, bilinear filtering should still be operating and preventing such clear lines from showing.

Here's the section of shader code that uses the cubemap LUT:

Code: Select all

vec4 cloudShadowRGBA = textureCube(rectToSphTex, cloudShadowTexCoord0);
vec2 cloudShadowUV;
cloudShadowUV.x = dot(cloudShadowRGBA.rg, vec2(0.996094, 0.003922)) + cloudShadowTexOffset;
cloudShadowUV.y = dot(cloudShadowRGBA.ba, vec2(0.996094, 0.003922));
shadow *= (1.0 - texture2D(cloudShadowTex, cloudShadowUV).a * 0.75);


Note that the constants are 0.996094 = 255/256, and 0.003922 = 1/255. To clarify, the shader is calculating:

cloudShadowUV.x = red * 255/256 + green * 1/255
cloudShadowUV.y = blue * 255/256 + alpha * 1/255

...where red, green, blue, and alpha are 8-bit values from the cubemap LUT. If I could rely on hardware support for filtering of 16-bit textures, I could just use the value retrieved from the cubemap directly, replacing the first four lines of the above code with:

Code: Select all

vec2 cloudShadowUV = textureCube(rectToSphTex, cloudShadowTexCoord0);


Anyone have ideas?

--Chris

[cartrite]

Anyone have ideas?

We still don't know if a 7000 series card works or not. Maybe you should make an executable available so someone who can't build it can test too.

Once we know which cards don't work, then go through the technical specs and see what could be supported on working cards that isn't supported on non working cards.
As far as technical specs, I can only find basic information on the NVIDIA site. And only for 7000 and 8000 cards. Do you know a better source for tech specs?
cartrite

[chris]

Anyone have ideas?

We still don't know if a 7000 series card works or not. Maybe you should make an executable available so someone who can't build it can test too.

Once we know which cards don't work, then go through the technical specs and see what could be supported on working cards that isn't supported on non working cards.
As far as technical specs, I can only find basic information on the NVIDIA site. And only for 7000 and 8000 cards. Do you know a better source for tech specs?
cartrite

I'm away from my Windows machine for a few days, so I can't upload a Windows build--in fact I'm posting from my iPhone right now. Fridger's Quadro is based on a GeForce 7 series chip, so I suspect no other GEfoece 7 cards will work, but I'd still like to see someone try a 7900. Or an ATI card.

--Chris

[cartrite]

I haven't built this on windows but I'll give it a try and upload the executable to my web site. That way someone could replace that file with the file they have in there installed copy. What would they need to have. 1,5pre4 or 1.5pre5?
cartrite

[t00fri]

Anyone have ideas?

We still don't know if a 7000 series card works or not. Maybe you should make an executable available so someone who can't build it can test too.

Once we know which cards don't work, then go through the technical specs and see what could be supported on working cards that isn't supported on non working cards.
As far as technical specs, I can only find basic information on the NVIDIA site. And only for 7000 and 8000 cards. Do you know a better source for tech specs?
cartrite

I'm away from my Windows machine for a few days, so I can't upload a Windows build--in fact I'm posting from my iPhone right now. Fridger's Quadro is based on a GeForce 7 series chip, so I suspect no other GEfoece 7 cards will work, but I'd still like to see someone try a 7900. Or an ATI card.

--Chris

True, and as I wrote, the G72 chip does show blocking as well. Still a FX 7900 might be worth a try. My card is roughly equivalent to a FX 7300. I might compile a Windows version for other people to try. Actually I have one already. I just have to post it....

Bye Fridger

[cartrite]

Anyone have ideas?

We still don't know if a 7000 series card works or not. Maybe you should make an executable available so someone who can't build it can test too.

Once we know which cards don't work, then go through the technical specs and see what could be supported on working cards that isn't supported on non working cards.
As far as technical specs, I can only find basic information on the NVIDIA site. And only for 7000 and 8000 cards. Do you know a better source for tech specs?
cartrite

I'm away from my Windows machine for a few days, so I can't upload a Windows build--in fact I'm posting from my iPhone right now. Fridger's Quadro is based on a GeForce 7 series chip, so I suspect no other GEfoece 7 cards will work, but I'd still like to see someone try a 7900. Or an ATI card.

--Chris

True, and as I wrote, the G72 chip does show blocking as well. Still a FX 7900 might be worth a try. My card is roughly equivalent to a FX 7300. I might compile a Windows version for other people to try. Actually I have one already. I just have to post it....

Bye Fridger

I'll hold off on building windows then. I'm still on Linux.
cartrite

[t00fri]

So this is a Win XP executable (Celestia.exe) built with Chris' latest cloud shadow patch with Vc++.Net 2003 from the latest CVS code. People who want to try, have to exchange it with their installed Celestia.exe of 1.5.0pre5, say or perhaps also 1.5.0pre4. Note this is JUST the Celestia executable!!!!

http://www.celestiaproject.net/~t00fri/images/Celestia.exe


Careful, DON'T overwrite your original Celestia.exe!

Let me know whether it works for you. This is NOTHING for Daniel

Bye Fridger

[ANDREA]

Well, probably I lost my way among the many posts on the subject, but I didn't find any info about the cloud speed, so I think it should still be set =zero, correct?
But if it's so, I inform you that I'm looking just now at a very nice Don's 8k cloudmap, rotating at speed 10 around the Earth globe, showing correctly the cloud shadows, using 1.5.0pre5 with the Fridger's Celestia.exe executable from last posts.
Please confirm that this is the result due after Chris' enhancement.
Bye and thank you.

Andrea

[t00fri]

Well, probably I lost my way among the many posts on the subject, but I didn't find any info about the cloud speed, so I think it should still be set =zero, correct?
But if it's so, I inform you that I'm looking just now at a very nice Don's 8k cloudmap, rotating at speed 10 around the Earth globe, showing correctly the cloud shadows, using 1.5.0pre5 with the Fridger's Celestia.exe executable from last posts.
Please confirm that this is the result due after Chris' enhancement.
Bye and thank you.

Andrea

Andrea,

yes, it's all correct in your case: your are welcome to try the cloud shadows, of course, but since you own a fancy 8800 GTX card, the cloud shadows should work fine .

What we need as a much more critical test is someone with e.g. a FX 7900 or somewhat lower card.

Bye Fridger

[ANDREA]

Well, probably I lost my way among the many posts on the subject, but I didn't find any info about the cloud speed, so I think it should still be set =zero, correct?
But if it's so, I inform you that I'm looking just now at a very nice Don's 8k cloudmap, rotating at speed 10 around the Earth globe, showing correctly the cloud shadows, using 1.5.0pre5 with the Fridger's Celestia.exe executable from last posts.
Please confirm that this is the result due after Chris' enhancement.
Bye and thank you.
Andrea

Andrea,
yes, it's all correct in your case: your are welcome to try the cloud shadows, of course, but since you own a fancy 8800 GTX card, the cloud shadows should work fine .
What we need as a much more critical test is someone with e.g. a FX 7900 or somewhat lower card.
Bye Fridger

Fridger, obviously I well understood that the problem is NOT regarding the 8800 series (danielj's cry was sufficient to point out it), but I'm saying anyhow that I don't see written anywere that the cloud speed now must no more be set =0.
So, can someone please show me where it's told that now it can be set at any value, even if ONLY with 8800 cards, obviously?
Tks

Andrea

[t00fri]

Andrea,

since with your 8800 card the largest texture size is 8k (look in help-> OpenGl Info) , you don't have to set the speed for your 8k clouds to zero! People with older cards and smaller max texture sizes will have to.

Bye Fridger

[ANDREA]

Andrea, since with your 8800 card the largest texture size is 8k (look in help-> OpenGl Info) , you don't have to set the speed for your 8k clouds to zero! People with older cards and smaller max texture sizes will have to. Bye Fridger

Thank you for the confirmation, Fridger.
Mine was just a doubt, anyhow, no more.
BTW, I think the cloud movement itself useless, a couple tropical storms over Mediterranean Sea are a bit, hu, absurd!
Bye

Andrea

[ElChristou]

Zero speed for clouds wasn't for VTs only?

[t00fri]

Zero speed for clouds wasn't for VTs only?

No, it was as I said. I am using moving clouds with my VT's for example.

F.

[Johaen]

Thanks to Fridger's .exe, I have verified that both my 7900GT and my 6200 have the same super pixelated cloud shadows. I wish I could have compiled from code myself, but I had a hell of a time with the environment variables when trying to build it. Oh well. Maybe I'll get around to figuring it out sometime.

[t00fri]

Thanks to Fridger's .exe, I have verified that both my 7900GT and my 6200 have the same super pixelated cloud shadows. I wish I could have compiled from code myself, but I had a hell of a time with the environment variables when trying to build it. Oh well. Maybe I'll get around to figuring it out sometime.

Thanks, Johaen. Just to be sure: with "super pixelated" you mean the blocking effects we have displayed above from various cards that do NOT contain G80 chips.

Bye Fridger