Shadows point light #3711

lilleyse · 2016-03-15T19:18:16Z

I'll finish up #3691 first.

Some of the changes here will help with the terrain artifacts in that PR. v_normalEC wasn't getting to the receive-shadows shader correctly because terrain lighting is off by default. This caused back faces to not be shadowed. Now those varyings are outputted if the globe receives shadows. There are still some other artifacts to deal with though.

First I organized ShadowMap to handle point lights. It now uses the concept of passes - point light shadows have 6 passes, cascaded shadows have 4 passes, single shadow has one pass. Each pass has a PassState to set the viewport, and a camera. From scene's perspective it doesn't care what type of shadowing is done.

The point light shadow texture is basically a cube map that is flattened into a 2d texture. I'm a little worried about the amount of shader operations to convert from a cube-map direction to a 2d uv, but overall the benefits are that we can still use a depth texture, avoid binding different faces to the framebuffer for each pass, and overall cleaner integration with the existing code.

lilleyse · 2016-03-15T19:20:26Z

pjcozzi · 2016-03-15T23:09:04Z

No OSX luck here either, but your screenshot looks awesome!

pjcozzi · 2016-03-15T23:11:45Z

point light shadows have 6 passes, cascaded shadows have 4 passes, single shadow has one pass.

So, as of now, the cube map faces are not cascaded? I am go with that to start as we discussed offline. I expect that we will need it at some point though.

pjcozzi · 2016-03-15T23:12:34Z

The point light shadow texture is basically a cube map that is flattened into a 2d texture.

Was this useful to you: http://webglinsights.github.io/downloads/WebGL-Insights-Chapter-16.pdf

pjcozzi · 2016-03-15T23:13:40Z

Apps/SampleData/models/ShadowTester/Shadow_Tester_Point.gltf

@@ -0,0 +1,270 @@
+{


Did you make this entire model? Do we need to credit anything?

I made this test model.

pjcozzi · 2016-03-15T23:20:28Z

In the future, we will also be interested in generating cube maps for relfection and refraction - and also rendering for picture-in-picture, multiple monitors, etc.

Is there anything you think we should do to the shadow map passes now in anticipation of this? Or just rework when we actually need to?

pjcozzi · 2016-03-15T23:23:18Z

Source/Scene/ShadowMap.js

+            this._passStates[2].viewport = new BoundingRectangle(0, size, size, size);
+            this._passStates[3].viewport = new BoundingRectangle(size, size, size, size);
+        } else if (numberOfPasses === 6) {
+            // +----+----+----+


For our default shadow map size, any concern with this exceeding the typically WebGL limit?

Also, as we discussed offline, we are going to have cases like spot lights where we know we won't need all 6 faces. At the least, we should be able to optimize the number of passes, but if we think it is worth it, I suppose we can also optimize the texture size (but not for dynamic field of views).

The texture size should be ok for now. The min texture size on most systems is 4096, so 1024*3 is fine.

Yes spotlights are the next type I'll support, it should be just one pass. Even point lights could remove one pass if we know their top face is facing the sky, but may not be worth handling.

pjcozzi · 2016-03-15T23:31:14Z

I'm a little worried about the amount of shader operations to convert from a cube-map direction to a 2d...

I agree this is a lot of instructions, but they are all cheap.

How much did you look into optimizing this (most likely involves changing the layout as in Jeff's WebGL chapter)? Could be a TODO for later...

pjcozzi · 2016-03-15T23:31:34Z

That's all my comments. Looks really cool. Can't wait to run this.

lilleyse · 2016-03-16T17:39:13Z

How much did you look into optimizing this (most likely involves changing the layout as in Jeff's WebGL chapter)? Could be a TODO for later...

I tried to keep optimization in mind. There are two other direction-to-uv functions I found (that use branching), so I'll check those out. I may also just get it working with cube maps and see how it compares.

I took a look at the chapter, but since it's using octahedral projection it doesn't apply as well. I think the idea is the same though, and looks like one of those functions I found.

lilleyse · 2016-03-16T17:42:53Z

So, as of now, the cube map faces are not cascaded? I am go with that to start as we discussed offline. I expect that we will need it at some point though.

No they don't use cascades. The results seems pretty good though even with low resolutions.

lilleyse · 2016-03-16T18:20:30Z

Updated to fix the Chrome OSX problem.

pjcozzi · 2016-03-17T16:54:22Z

OS X fix looks good.

Does this PR include everything in #3691? Should we close that?

It doesn't need to be in this PR, but do you plan to add an option to load different test models in the Sandcastle demo and perhaps disable parts of the UI when they don't work (maybe that's overkill); right now, a lot of options don't apply to point lights?

Otherwise, this is all good with me. @bagnell can you merge this (and #3691?) when you have reviewed.

lilleyse · 2016-03-17T18:22:56Z

Yes this includes all the comments in #3691, so it can be closed.

All those UI suggestions sound good to me. One thing I've wanted to fix for a while is rebuilding the terrain shaders when the shadow types change. Also fixing the models to turn shadowing on/off dynamically like primitives and terrain (using the derived commands from the 3d-tiles-styles branch).

lilleyse · 2016-03-17T18:26:20Z

#3691 Has a couple commits that aren't in this PR, so you could merge that branch into shadows or I can cherry-pick them here.

Use cube map for point light shadows

pjcozzi · 2016-03-17T18:29:23Z

Cherry pick them here and close #3691.

bagnell · 2016-03-17T19:41:42Z

Instead of enabling and disabling the UI, how about separate demos for each type of light source?

bagnell · 2016-03-17T20:15:29Z

Source/Scene/ShadowMap.js

@@ -791,7 +1055,20 @@ define([
        }
    };

+    ShadowMap.prototype.updatePass = function(context, pass) {
+        if (this._isPointLight && this._usesCubeMap) {
+            this._framebuffer._attachTexture(context, WebGLConstants.COLOR_ATTACHMENT0, this._passFaces[pass]);


It would be better to have a framebuffer for each face rather than changing color attachments. Binding a framebuffer should be faster than having to re-validate the framebuffer after a new texture is attached.

I can try that out and see. The one disadvantage is each framebuffer needs its own depth texture (in addition to the shadow texture).

Actually that isn't true... it should be fine.

bagnell · 2016-03-17T20:28:53Z

In the future, we will also be interested in generating cube maps for relfection and refraction - and also rendering for picture-in-picture, multiple monitors, etc.
Is there anything you think we should do to the shadow map passes now in anticipation of this? Or just rework when we actually need to?

I'm also interested in what you think about this. You may want to keep in mind that we will want to have multiple light sources. I see why we would have a shadow map dedicated to the Sun, but does czm_sunShadowMapTextureCube make sense?

bagnell · 2016-03-17T20:29:47Z

Other that those two comments, I think this is good to go.

lilleyse · 2016-03-18T14:36:19Z

This is updated. While changing the point light uniform names I figured I should just change the others as well since the shadow map is not necessarily tied to the sun.

bagnell · 2016-03-18T15:16:07Z

Source/Scene/ShadowMap.js

+        if (shadowMap._usesDepthTexture && (shadowMap._passFramebuffers[0].status !== WebGLConstants.FRAMEBUFFER_COMPLETE)) {
+            shadowMap._usesDepthTexture = false;
+            var colorMask = shadowMap._renderState.colorMask;
+            colorMask.red = colorMask.green = colorMask.blue = colorMask.alpha = true;


It's probably not documented well, but RenderStates are read only. This allows us to cache them and make minimal render state changes between draw calls.

lilleyse · 2016-03-18T16:20:15Z

Updated to use immutable render states and to fix some issues with destroying the cube-map fbo's.

Shadows point light

lilleyse added 6 commits March 9, 2016 11:20

Fit to scene near and far

5190927

Rename farPlane to distance

7c0cb2c

Change lamda to 1.0

5759fa3

Rename fitToScene to cascadesEnabled

9b25230

Reorganize for point lights

11bdd01

Support point lights

3da9583

pjcozzi reviewed Mar 15, 2016
View reviewed changes

lilleyse force-pushed the shadows-point-light branch from bfef910 to e1c61e8 Compare March 16, 2016 18:11

Fallback if depth FBO fails

4876a80

lilleyse force-pushed the shadows-point-light branch from e1c61e8 to 4876a80 Compare March 16, 2016 18:17

lilleyse mentioned this pull request Mar 16, 2016

Shadows use scene near/far #3691

Closed

lilleyse mentioned this pull request Mar 17, 2016

Use cube map for point light shadows #3718

Merged

lilleyse added 2 commits March 17, 2016 13:50

Use cube map for point light shadows

e2f05f6

Combine uniforms

692750b

Merge pull request #3718 from AnalyticalGraphicsInc/shadows-point-light2

9310f79

Use cube map for point light shadows

lilleyse added 2 commits March 17, 2016 14:35

Handle shadow near/far better

0fe2fd5

Add TODO

b4c2a3b

bagnell reviewed Mar 17, 2016
View reviewed changes

lilleyse added 3 commits March 18, 2016 10:05

Use separate fbo for each cubemap face

9726ff4

Rename sunShadowMap to shadowMap

7a5ec46

Only update point light cameras when point light moves

a98cbe8

Remove extra line

15c0362

bagnell reviewed Mar 18, 2016
View reviewed changes

lilleyse added 2 commits March 18, 2016 11:40

Destroy fbo resources correctly for cube maps

162e5d0

Make render states immutable

3ebb190

Small tweak

1acd59f

bagnell added a commit that referenced this pull request Mar 18, 2016

Merge pull request #3711 from AnalyticalGraphicsInc/shadows-point-light

a471f00

Shadows point light

bagnell merged commit a471f00 into shadows Mar 18, 2016

bagnell deleted the shadows-point-light branch March 18, 2016 19:22

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Shadows point light #3711

Shadows point light #3711

lilleyse commented Mar 15, 2016

lilleyse commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

pjcozzi Mar 15, 2016

lilleyse Mar 16, 2016

pjcozzi commented Mar 15, 2016

pjcozzi Mar 15, 2016

pjcozzi Mar 15, 2016

lilleyse Mar 16, 2016

lilleyse Mar 16, 2016

pjcozzi commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

lilleyse commented Mar 16, 2016

lilleyse commented Mar 16, 2016

lilleyse commented Mar 16, 2016

pjcozzi commented Mar 17, 2016

lilleyse commented Mar 17, 2016

lilleyse commented Mar 17, 2016

pjcozzi commented Mar 17, 2016

bagnell commented Mar 17, 2016

bagnell Mar 17, 2016

lilleyse Mar 17, 2016

lilleyse Mar 17, 2016

bagnell commented Mar 17, 2016

bagnell commented Mar 17, 2016

lilleyse commented Mar 18, 2016

bagnell Mar 18, 2016

lilleyse commented Mar 18, 2016

Shadows point light #3711

Shadows point light #3711

Conversation

lilleyse commented Mar 15, 2016

lilleyse commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

pjcozzi commented Mar 15, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

pjcozzi commented Mar 15, 2016

pjcozzi commented Mar 15, 2016

lilleyse commented Mar 16, 2016

lilleyse commented Mar 16, 2016

lilleyse commented Mar 16, 2016

pjcozzi commented Mar 17, 2016

lilleyse commented Mar 17, 2016

lilleyse commented Mar 17, 2016

pjcozzi commented Mar 17, 2016

bagnell commented Mar 17, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bagnell commented Mar 17, 2016

bagnell commented Mar 17, 2016

lilleyse commented Mar 18, 2016

Choose a reason for hiding this comment

lilleyse commented Mar 18, 2016