Big bugs in 1.5.0 Pre 2

[selden]

Unfortunately, after rebuilding without the "complex cloud shaders", an eclipse with cloud shadows enabled still turns the Earth red.

Instructions to reproduce:

run Celestia, going to Earth
open eclipse finder
click on "compute"
double click on any of the resulting dates
toggle "Cloud shadows"
watch the color of the Earth change.

[chris]

Unfortunately, after rebuilding without the "complex cloud shaders", an eclipse with cloud shadows enabled still turns the Earth red.

Instructions to reproduce:

run Celestia, going to Earth
open eclipse finder
click on "compute"
double click on any of the resulting dates
toggle "Cloud shadows"
watch the color of the Earth change.

Not on my GeForce 8800, but I suspect that might because it has a lot more interpolators than older hardware. Can you paste or link to your shaders.log?

--Chris

[selden]

http://www.lepp.cornell.edu/~seb/celest ... haders.log
(29KB)

The error message at the end is


Error linking shader program:
Fragment info
-------------
(24) : error C5041: cannot locate suitable resource to bind parameter "cloudShadowTexCoord0"

[chris]

http://www.lepp.cornell.edu/~seb/celestia/files/shaders.log
(29KB)

The error message at the end is


Error linking shader program:
Fragment info
-------------
(24) : error C5041: cannot locate suitable resource to bind parameter "cloudShadowTexCoord0"

I just checked in a fix . . . However, there are still cases where you can exhaust all of the interpolators. A complete fix will have to wait a bit longer though.

--Chris

[t00fri]

Unfortunately, after rebuilding without the "complex cloud shaders", an eclipse with cloud shadows enabled still turns the Earth red.

Instructions to reproduce:

run Celestia, going to Earth
open eclipse finder
click on "compute"
double click on any of the resulting dates
toggle "Cloud shadows"
watch the color of the Earth change.

Not under Linux KDE (rebuilt ~20 hours ago)

Bye Fridger

[chris]

Unfortunately, after rebuilding without the "complex cloud shaders", an eclipse with cloud shadows enabled still turns the Earth red.

Instructions to reproduce:

run Celestia, going to Earth
open eclipse finder
click on "compute"
double click on any of the resulting dates
toggle "Cloud shadows"
watch the color of the Earth change.

Not under Linux KDE (rebuilt ~20 hours ago)

It depends on what textures you have enabled . . . Selden had night textures turned on, atmospheres, cloud shadows, and specular lighting. All of these things use interpolators:

Base texture: 1
Atmopheres: 1
Cloud shadows: 1/light source
Specular lighting: 1 + 1/light source
Eclipse shadows: 1
Night texture: 2
Additional interpolators: 1 for diffuse light

That's a total of 9 interpolants (some scalar, some vector): values that are generated by the vertex shader, perspective interpolated by the graphics chip, and passed on to the pixel shader. Unfortunately, graphics hardware prior to the GeForce 8800 only supports 8 interpolators (actually 10, but two seem to be reserved.) I worked around the bug by eliminating a scalar interpolator used for the night texture. As a result, there's a small amount of extra computation done in the pixel shader, but it should have a negligible performance impact.

Shading may be separated into multiple additive passes, with one pass per light. Thus, I'm not so worried about running out of interpolators due to extra light sources. However, the shaders do need to be constructed so that the most complex single light source shader does not run over the 8 interpolator limit common on most OpenGL 2.0 hardware. I have plan for accomplishing this in a future Celestia version.

--Chris

[t00fri]

Trying to compile your fix with Linux (gcc 4.1.x) gives this:

Code: Select all

shadermanager.cpp: In member function ?€?GLFragmentShader* ShaderManager::buildFragmentShader(const ShaderProperties&)?€™:
shadermanager.cpp:1564: error: name lookup of ?€?k?€™ changed for new ISO ?€?for?€™ scoping
shadermanager.cpp:1562: error:   using obsolete binding at ?€?k?€™
make[3]: *** [shadermanager.o] Error 1
make[3]: Leaving directory `/usr/local/cvs/celestia-kde/src/celengine'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/usr/local/cvs/celestia-kde/src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/usr/local/cvs/celestia-kde'
make: *** [all] Error 2


Bye Fridger

[t00fri]

It depends on what textures you have enabled . . . Selden had night textures turned on, atmospheres, cloud shadows, and specular lighting. All of these things use interpolators:

I use (fx 5900 Ultra): my 64k normalmap 1k x 1k tiles, 64k base+spec 1k x 1k tiles (RGBA), 32k nightlight 1k x 1k tiles, atmospheres, cloud shadows, specular lighting, eclipse shadows, my 2k clouds.

Bye Fridger

[chris]

Trying to compile your fix with Linux (gcc 4.1.x) gives this:

Code: Select all

shadermanager.cpp: In member function ?€?GLFragmentShader* ShaderManager::buildFragmentShader(const ShaderProperties&)?€™:
shadermanager.cpp:1564: error: name lookup of ?€?k?€™ changed for new ISO ?€?for?€™ scoping
shadermanager.cpp:1562: error:   using obsolete binding at ?€?k?€™
make[3]: *** [shadermanager.o] Error 1
make[3]: Leaving directory `/usr/local/cvs/celestia-kde/src/celengine'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/usr/local/cvs/celestia-kde/src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/usr/local/cvs/celestia-kde'
make: *** [all] Error 2


Bye Fridger

Damned MS compiler and it's non-ANSI for looping scoping! If you change the code so that k is declared outside the for loop, it will compile, i.e. change:

Code: Select all

for (int k = 0; k < ...; k++)


to:

Code: Select all

int k;
for (k = 0; k < ...; k++)


If you make that change, please go ahead and check it in; otherwise I'll do it later today. I think that there may be an MSVC compiler flag to enable ANSI for loop scoping conventions, which would help avoid breaks like this in the future.

--Chris

[t00fri]

Chris,

compilation went fine now, but that's the result:



Since a long time I get a red globe again. With the same settings everything was perfect before your "fix".


See my fading star labels

Bye Fridger

[chris]

Since a long time I get a red globe again. With the same settings everything was perfect before your "fix".

Ugh. Could you post your shaders.log?

--Chris

[t00fri]

Here it is (gzipped)

http://www.celestiaproject.net/~t00fri/images/ ... ers.log.gz

Bye Fridger

[t00fri]

Chris,

there are more things since your recent fixes that now go wrong. I just recompiled your scope-fix in shadermanager.cpp. I didn't commit it, since I think you will have more to fix...



See those dirty dark stripes that don't belong here??

Bye Fridger

[tech2000]

See my fading star labels

Fridger,
is your fading star labels code also went into the cvs?

Would all labels use this kind of fade?

bye, Anders

[t00fri]

See my fading star labels

Fridger,
is your fading star labels code also went into the cvs?

Would all labels use this kind of fade?

bye, Anders

Not yet, since I still want to do some checks and possible improvements. It's so long ago and I have to "warm up" again, trying to remember the quite tricky dependences ...

Bye Fridger

[chris]

Chris,

there are more things since your recent fixes that now go wrong. I just recompiled your scope-fix in shadermanager.cpp. I didn't commit it, since I think you will have more to fix...



See those dirty dark stripes that don't belong here??

I see the ring shadows, but I presume that's not what you're talking about. Are those gray bands above the ring shadows not supposed to be there?

I do know what's going wrong with the Earth shader; I'll have a fix late today.

Nice to see those faded star labels in the screenshot.

--Chris

[t00fri]

Chris,

I am confused. I did a /fresh/ compile and all those effects
above were gone. I am now missing the glowing red Earth globe...h?¶ h?¶

I have no idea what happened, but both the Saturn
artefacts and the red Earth are gone. Only my nice fading
labels remain

Sorry for causing possibly unnecessary work. But perhaps
you found a fix for something that noone has yet discovered


Bye Fridger

[chris]

Chris,

I am confused. I did a /fresh/ compile and all those effects
above were gone. I am now missing the glowing red Earth globe...h?¶ h?¶

I have no idea what happened, but both the Saturn
artefacts and the red Earth are gone. Only my nice fading
labels remain

Sorry for causing possibly unnecessary work. But perhaps
you found a fix for something that noone has yet discovered

It's not unnecessary work--you definitely found a bug with the red Earth. With my latest checkin, night lights+normal maps+specular will not work. Do you think that you may have night textures disabled now?

--Chris

[selden]

fyi:

Celestia as built this morning (6:35 EDT)
crashes when viewing Saturn
on my system at work.

System:
2GB, 1.86GHz Core2Duo, WinXP Pro SP2
Quadro FX 550, ForceWare 91.36

I'll update Forceware shortly to the current Quadro version, 91.85, but thought I'd let you know now, since that'll take a while.

[t00fri]

It's not unnecessary work--you definitely found a bug with the red Earth. With my latest checkin, night lights+normal maps+specular will not work. Do you think that you may have night textures disabled now?

--Chris

Beautiful!!! I got my red Earth back...Ahem. For some
reason the rendering path had shifted from OpenGL 2.0
and that was the reason for the nice natural (non-red)
display of mother Earth.

So indeed no vasted work. Good. As a cross-check: the
red globe is gone as soon as I switch off the night lights.


Bye Fridger