In-pass shadows (WIP) by shadowislord · Pull Request #723 · jMonkeyEngine/jmonkeyengine

shadowislord · 2017-09-09T20:36:16Z

This is the PR implementing in-pass shadows as shown in the April 2016 screenshot thread:
https://hub.jmonkeyengine.org/t/april-2016-monthly-wip-screenshot-thread/35512/128

How it works:
PreShadowArrayRenderer is responsible for generating a texture array containing shadow maps for every light that should be shadowed. Directional lights use PSSM with 1-4 slices, spot lights use 1 slice, and point lights will use 6 slices (not implemented yet). The information of which texture array to use and the slices is represented in the ShadowMap interface. This is then associated with the light so that later it can be retrieved.
When the scene is rendered, ShadowStaticPassLightingLogic is the TechniqueDefLogic implementation that is capable of rendering in-pass shadows. The light types are statically known which allows the shader to avoid control flow. It works similarly to SinglePass, except that the lights are first sorted by type and then their counts are known statically, so instead of this loop:

for each light do ...

It becomes

for each directional light do ...
for each point light do ...
for each spot light do ...

This then allows using different logic depending on the light type. The way shadow maps are fetched varies depending on the type of light, which is why this is required.

Theoretically, if we have control flow in the shader then we don't need to know the light type in advance and we can calculate the slice to fetch per each light. This is probably how in-pass shadows will be integrated with the rest of the lighting shaders.

TODO:

Point light support
Remove / fix the hacks from ShadowUtil
Remove all the static lighting crap

Optional:

Port the old shadow renderers to the new system
Implement and test PSSM-only in-pass shadows
Test on OpenGL ES 2.0

…adows

Nehon

Several remarks:
1- Don't you have a Test class to see the usage and well... test it.
2- It's seems that inpass shadows use a different technique and techniqueDefLogic so how could I combine them with the PBR shader?
3- It seems there is 1 shadow renderer by light type is that correct?
4- Could be nice that you explain the whole principle in a short summary.

shadowislord · 2017-09-10T17:31:23Z

Yes I have it... It kinda crappy though so I am going to clean it a bit and commit later
So the challenge is supporting any light type in the shader without knowing its type statically. Unfortunately this is more complicated so I took the easier route initially where all types are known ahead of time. Theoretically the shader is faster but due to the large number of shader permutations it causes micro-stuttering when the shader variants are compiled. Supporting arbitrary light types requires several passes where first the shadow slices to fetch are determined (based on NB_LIGHTS) and then fetching them. The next part is using those results in the lighting loop.
Not sure what you mean by renderer. Generating the shadow maps is handled by PreShadowArrayRenderer and then later they are used by ShadowStaticPassLightingLogic.
I will update the description ...

shadowislord · 2018-03-19T19:33:40Z

jme3-core/src/main/java/com/jme3/material/Material.java

    private void updateRenderState(RenderManager renderManager, Renderer renderer, TechniqueDef techniqueDef) {
        if (renderManager.getForcedRenderState() != null) {
-            renderer.applyRenderState(renderManager.getForcedRenderState());
+            if (techniqueDef.getForcedRenderState() != null) {


This could break compatibility... I think it fixes a bug, or changes behavior

shadowislord · 2018-03-19T19:34:48Z

jme3-core/src/main/java/com/jme3/material/logic/DefaultTechniqueDefLogic.java

+
+    protected float encodeLightType(Light light) {
+        switch (light.getType()) {
+            case Directional:


FYI, this breaks any custom lighting shaders

shadowislord · 2018-03-19T19:35:43Z

jme3-core/src/main/java/com/jme3/material/logic/ShadowStaticPassLightingLogic.java

+import com.jme3.texture.TextureArray;
+import java.util.EnumSet;
+
+public class ShadowStaticPassLightingLogic extends StaticPassLightingLogic {


Static pass should probably be removed. It doesn't add any value now

shadowislord · 2018-03-19T19:37:32Z

jme3-core/src/main/java/com/jme3/material/logic/SinglePassAndImageBasedLightingLogic.java

-        return super.makeCurrent(assetManager, renderManager, rendererCaps, lights, defines);
+        ambientLightColor.a = 1.0f;
+
+        filteredLightList.sort(new Comparator<Light>() {


Should refactor this comparator to avoid allocation

shadowislord · 2018-03-19T19:45:30Z

jme3-core/src/main/java/com/jme3/material/logic/StaticPassLightingLogic.java

 * @author Kirill Vainer
 */
-public final class StaticPassLightingLogic extends DefaultTechniqueDefLogic {
+public class StaticPassLightingLogic extends DefaultTechniqueDefLogic {


Remove this

shadowislord · 2018-03-19T19:49:24Z

jme3-core/src/main/java/com/jme3/shadow/next/PreShadowRenderer.java

+ *
+ * @author Kirill Vainer
+ */
+public class PreShadowRenderer implements SceneProcessor {


This needs a test... TestInPassShadows uses the regular PreShadowArrayRenderer

shadowislord · 2018-03-19T19:49:39Z

jme3-core/src/main/java/com/jme3/shadow/next/ShadowDebugControl.java

+ *
+ * @author Kirill Vainer
+ */
+final class ShadowDebugControl extends AbstractControl {


This needs to be used by the test

shadowislord · 2018-03-19T19:50:03Z

jme3-core/src/main/java/com/jme3/shadow/next/ShadowParameters.java

+/**
+ * @author Kirill Vainer
+ */
+public interface ShadowParameters {


What's the point of this?

DirectionalShadowParameters implement it... not sure why it's needed though.
There is no other type of Shadow parameter and directional parameters are used with the type DirectionalShadowParameters So this interface is IMO useless. I'll remove it if you agree.

shadowislord · 2018-03-19T19:52:53Z

jme3-core/src/main/java/com/jme3/shadow/next/array/SpotArrayShadowMapSlice.java

+        super(array, layer, textureSize, true);
+    }
+
+    private static boolean isParallelToYUp(Vector3f direction) {


Should this be used for directional lights too?

shadowislord · 2018-03-19T19:53:18Z

jme3-core/src/main/java/com/jme3/shadow/next/pssm/BaseShadowMapSlice.java

+    @Override
+    public Matrix4f getBiasedViewProjectionMatrix() {
+//        return shadowCamera.getViewProjectionMatrix();
+        throw new UnsupportedOperationException();


shadowislord · 2018-03-19T19:53:44Z

jme3-core/src/main/java/com/jme3/shadow/next/pssm/DirectionalShadowParameters.java

+    }
+
+    public void setNumSplits(int numSplits) {
+        // TODO: ensure it is 1 to 4


Add a check here

shadowislord · 2018-03-19T19:58:33Z

jme3-core/src/main/resources/Common/ShaderLib/Lighting.glsllib

    lightDir.w = clamp(lightDir.w, 1.0 - posLight, 1.0);
 #else
-    lightDir.w = clamp(1.0 - position.w * dist * posLight, 0.0, 1.0);
+    lightDir.w = 1.0; // clamp(1.0 - position.w * dist * posLight, 0.0, 1.0);


revert this

…. Also addressed some comments on the PR

empirephoenix · 2019-02-24T10:25:02Z

What is the state of this one btw? is this still alive?

shadowislord · 2019-03-26T03:14:04Z

@empirephoenix
It breaks backwards compatibility, mainly with custom lighting shaders. Also there are some significant core changes that require user testing. Might make sense to include in jME 3.3

shadowislord added 23 commits September 9, 2017 13:05

Use correct HALF_FLOAT constant for GLES

406c314

Support GLTracer and GL debug in Android

2596946

Support instancing in GLTracer

69139a1

Don't use unsupported TEXTURE_{BASE,MAX}_LEVEL in GLES

23700d5

Support ABGR8 using swizzle extension

7a22f8c

Move glFramebufferTextureLayer to GLFbo

af3a0c7

Avoid RGB111110F in filters unless its actually supported

b0316e4

Use the same depth format in all places in FPP

88aaa07

Print the object with the problem in NativeObjectManager

7441865

Merge remote-tracking branch 'origin/master' into opengles2-fixes

134c365

Don't upload NULL texture slices

bc50b09

Support shadow maps in GLES

16e4726

Merge remote-tracking branch 'origin/opengles2-fixes' into in-pass-sh…

4d60b2d

…adows

Don't reset lights that were never checked in light filter

2ce2995

Make tech logic impls responsible for light filtering

f5ad027

Remove scale from updateFrustumPoints since its always 1.0

55e9fd0

Add ShadowMap interface

cfcec44

StaticPass supports setting shadow map index on light

42051b0

Fix additional render state missing apply = false

ce28e35

Make sure apply = false for all forced render states

8a74727

Combine forced render state with the one in the technique

d159e17

Add PreShadowArrayRenderer

9c4fcac

Implement static pass in the lighting material

77e552f

shadowislord changed the base branch from master to opengles2-fixes September 9, 2017 20:36

Nehon reviewed Sep 10, 2017

View reviewed changes

shadowislord added 4 commits September 10, 2017 18:34

Fix syntax error

5aa2c72

Support dynamic number of shadow maps

a314588

optimize pssm split calculation

0a4a439

optimize pssm split calculation

2c38591

shadowislord added 10 commits September 30, 2017 18:08

update GLTracer

6487def

remove hacks from ShadowUtil

42432ed

Merge remote-tracking branch 'origin/master' into in-pass-shadows

5108f52

Merge remote-tracking branch 'origin/master' into in-pass-shadows

6fb2d02

in-pass-shadows: add spot / point light support

caad166

in-pass-shadows: fix syntax errors

5b80095

in-pass-shadows: use glsllib extension

5f66eea

in-pass-shadows: add test example

2e9996d

in-pass-shadows: add customized light sort

251511e

in-pass-shadows: show cfg dialog for test

cbf6ffa

shadowislord commented Mar 19, 2018

View reviewed changes

shadowislord assigned Nehon Mar 19, 2018

shadowislord requested a review from Nehon March 19, 2018 20:02

Nehon added 3 commits May 26, 2018 17:59

Merge master into branch

54ef6ec

Changes the name of the array shadow renderer to InPassShadowRenderer…

d444c28

…. Also addressed some comments on the PR

Properly remove static pass

dd2626c

stephengold added this to the Future Release milestone Mar 13, 2021

Uh oh!

Conversation

shadowislord commented Sep 9, 2017 • edited by Nehon Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Nehon left a comment

Choose a reason for hiding this comment

Uh oh!

shadowislord commented Sep 10, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

empirephoenix commented Feb 24, 2019

Uh oh!

shadowislord commented Mar 26, 2019

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

shadowislord commented Sep 9, 2017 •

edited by Nehon

Loading