Implement GPU Particle Collisions #42628

reduz · 2020-10-08T00:39:30Z

-Sphere Attractor
-Box Attractor
-Vector Field
-Sphere Collider
-Box Collider
-Baked SDF Collider
-Heightmap Collider

servers/rendering/rasterizer_rd/shaders/particles.glsl

bruvzg · 2020-10-08T05:38:28Z

editor/node_3d_editor_gizmos.cpp

+
+ Vector<Vector3> points;
+
+ for (int i = 0; i <= 360; i++) {


Last line 360..361 overlaps with the first 0..1:

Suggested change

for (int i = 0; i <= 360; i++) {

for (int i = 0; i < 360; i++) {

Ping @reduz.

We have similar code with < and with <= in multiple places, sometimes even in the same file, if some cases do need this overlapping line, it probably should have comment why it's required:

godot/editor/node_3d_editor_gizmos.cpp

Line 3653 in 2e68a28

for (int i = 0; i < 360; i++) {

godot/editor/node_3d_editor_gizmos.cpp

Line 1987 in 2e68a28

for (int i = 0; i <= 360; i += skip) {

Could you open an issue about it? That's likely something for @JFonS to have a look at.

Zireael07 · 2020-10-08T07:07:59Z

As usual: Vulkan only or potentially backportable to GLES 3?

Calinou · 2020-10-08T12:18:17Z

@Zireael07 This new functionality heavily relies on compute shaders. Most of this can't be done in GLES3.

reduz · 2020-10-08T12:51:04Z

@Zireael07 I think a good amount of the new particle work will probably work on GLES3, but will have to wait and see.

clayjohn

Looks mostly good! Most of my comments are just about cleaning up code.

I noticed 3 issues while testing:

If the GPUParticles3D is set to use local_coords then all collisions will be offset by the local transform of the GPUParticles3D.
If the GPUParticlesCollisionHeightField is following the camera, it doesn't follow in the y direction. so if the HeightField is rotated it no longer follows the camera.
when GPUParticlesCollisionHeightField is following the camera it is wasting a lot of resources by rendering the heightfield behind the camera, instead it should follow the position and rotation of the camera so that it is only covering a small area that is outside the camera's field of view.

Follow up tasks:

Marching cubes Gizmo to visualize SDF (@JFonS)
Debug mode for height_field_texture (Anyone)
Add NoiseTexture3D (Anyone)

clayjohn · 2020-10-09T05:09:48Z

scene/resources/particles_material.h

-Sub Emitters
-Attractors
-Emitter positions deformable by bones
+*Sub Emitters


Can delete these I guess now that they are done

clayjohn · 2020-10-09T05:14:47Z

servers/rendering/rasterizer_rd/shaders/particles.glsl

@@ -172,6 +196,11 @@ bool emit_particle(mat4 p_xform, vec3 p_velocity, vec4 p_color, vec4 p_custom, u
 return true;
 }

+float sd_round_box(vec3 p, vec3 b, float r) {


This looks unused.

clayjohn · 2020-10-09T05:15:29Z

servers/rendering/rasterizer_rd/shaders/particles.glsl

+
+ vec3 attractor_force = vec3(0.0);
+
+ if (PARTICLE.is_active) { //process collision


Looks like this can be deleted, unless you want to do more with collision here?

clayjohn · 2020-10-09T05:17:33Z

servers/rendering/rasterizer_rd/shaders/particles.glsl

+ depth = particle_size - s;
+ const float EPSILON = 0.001;
+ normal = mat3(FRAME.colliders[i].transform) * normalize(vec3(
+ texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(EPSILON, 0.0, 0.0)).r,


woooooah. Looks like we need to update our formatter tools

One needs to be a bit cleverer to avoid this. Should probably break manually before normalize.

clayjohn · 2020-10-09T05:17:47Z

servers/rendering/rasterizer_rd/shaders/particles.glsl

+ col = true;
+ //normal = mix(normal,vec3(0,1,0),smoothstep(uvw_pos.y,local_y,y));
+ depth = dot(normal, pos1) - dot(normal, local_pos_bottom);
+ //depth = -depth;


Can clean up comments here

clayjohn · 2020-10-09T05:18:48Z

servers/rendering/rasterizer_rd/shaders/particles.glsl

+ continue;
+ }
+
+ const float DELTA = 1.0 / 8192.0;


Shouldnt this be 1.0/ size_of_height_field_texture?

problem is that the texture is not really po2 or square, maybe should just pass this a a vec2

clayjohn · 2020-10-09T05:21:53Z

servers/rendering/rasterizer_rd/rasterizer_storage_rd.cpp

@@ -6871,6 +7255,14 @@ bool RasterizerStorageRD::free(RID p_rid) {
 _particles_free_data(particles);
 particles->instance_dependency.instance_notify_deleted(p_rid);
 particles_owner.free(p_rid);
+ } else if (particles_collision_owner.owns(p_rid)) {
+ ParticlesCollision *particles_collision = particles_collision_owner.getornull(p_rid);
+ //_particles_free_data(particles);


Can clean up

clayjohn · 2020-10-09T05:22:28Z

servers/rendering/rasterizer_rd/rasterizer_storage_rd.cpp

+ particles_collision->heightfield_fb = RD::get_singleton()->framebuffer_create(fb_tex);
+ particles_collision->heightfield_fb_size = size;
+
+ print_line("allooocc");


Can clean up

clayjohn · 2020-10-09T05:23:41Z

scene/3d/gpu_particles_collision_3d.cpp

+
+void GPUParticlesCollisionSDF::_find_closest_distance(const Vector3 &p_pos, const BVH *bvh, uint32_t p_bvh_cell, const Face3 *triangles, float thickness, float &closest_distance) {
+ if (p_bvh_cell & BVH::LEAF_BIT) {
+ //uint32_t triangle = p_bvh_cell & BVH::LEAF_MASK;


can clean up

reduz · 2020-10-09T11:27:31Z

@clayjohn some answers:

If the GPUParticles3D is set to use local_coords then all collisions will be offset by the local transform of the GPUParticles3D.

Seems like a bug, will give it a check.

If the GPUParticlesCollisionHeightField is following the camera, it doesn't follow in the y direction. so if the HeightField is rotated it no longer follows the camera.

This is intended, as the idea is that it works for open world, it only follows on XZ (local to the heighfield node). As this uses 32 bits floats for depth, you can specify a large vertical range anyway.

when GPUParticlesCollisionHeightField is following the camera it is wasting a lot of resources by rendering the heightfield behind the camera, instead it should follow the position and rotation of the camera so that it is only covering a small area that is outside the camera's field of view.

This is intended. The idea is that collision happens around the player and updates as you advance through the world.. If the lifetime of the particles is relatively long, you won t find collided particles when you turn around if this were the case.

akien-mga

Style nitpicks, but don't forget @bruvzg's comment which pinpoints a bug.

akien-mga · 2020-10-09T12:54:02Z

core/math/vector2.h

+ }
+ Vector2 max(const Vector2 &p_vector2) const {


Add empty line between method declarations, even if they're related.

akien-mga · 2020-10-09T12:54:08Z

core/thread_work_pool.h

+ bool is_working() const {
+ return current_work != nullptr;
+ }
+ uint32_t get_work_index() const {
+ return index;
+ }
+ void end_work() {


Add empty line between method declarations, even if they're related.

akien-mga · 2020-10-09T12:55:10Z

editor/node_3d_editor_gizmos.cpp

+//
+
+////


Suggested change

//

////

//

////

////////

or

Suggested change

//

////

//

////

//////

? ;)

We'll have to pick a style for this kind of separation comments eventually.

I think we should avoid using separation comments entirely, but that's just me. I feel like they don't help with readability significantly, while taking up vertical real estate.

akien-mga · 2020-10-09T12:55:44Z

editor/node_3d_editor_gizmos.cpp

+
+ Vector<Vector3> points;
+
+ for (int i = 0; i <= 360; i++) {


Ping @reduz.

akien-mga · 2020-10-09T12:56:04Z

editor/node_3d_editor_gizmos.cpp

+/////
+
 ////


for <=360, i just copied from other gizmos, so ideally @JFonS should give these a check and see if they are correct or not and decide to change for a < in a separate PR

akien-mga · 2020-10-09T12:57:39Z

scene/3d/gpu_particles_collision_3d.cpp

+}
+Ref<Texture3D> GPUParticlesCollisionSDF::get_texture() const {


Add empty line between method declarations, even if they're related.

akien-mga · 2020-10-09T12:58:04Z

scene/3d/gpu_particles_collision_3d.cpp

+}
+GPUParticlesAttractor3D::~GPUParticlesAttractor3D() {


Add empty line between method declarations, even if they're related.

akien-mga · 2020-10-09T12:58:10Z

scene/3d/gpu_particles_collision_3d.cpp

+}
+GPUParticlesAttractorSphere::GPUParticlesAttractorSphere() :


Add empty line between method declarations, even if they're related.

akien-mga · 2020-10-09T12:59:18Z

servers/rendering/rasterizer.h

+ virtual void particles_remove_collision(RID p_particles, RasterizerScene::InstanceBase *p_instance) = 0;
+
+ virtual void update_particles() = 0;
+ /* PARTICLES COLLISION */


Suggested change

/* PARTICLES COLLISION */

/* PARTICLES COLLISION */

akien-mga · 2020-10-09T13:00:47Z

servers/rendering/rasterizer_rd/shaders/particles.glsl

+ depth = particle_size - s;
+ const float EPSILON = 0.001;
+ normal = mat3(FRAME.colliders[i].transform) * normalize(vec3(
+ texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(EPSILON, 0.0, 0.0)).r,


One needs to be a bit cleverer to avoid this. Should probably break manually before normalize.

-Sphere Attractor -Box Attractor -Vector Field -Sphere Collider -Box Collider -Baked SDF Collider -Heightmap Collider

Changes done.

akien-mga · 2020-10-09T17:28:34Z

Thanks!

aaronfranke · 2020-10-11T17:09:16Z

Where is the documentation for these features?

reduz · 2020-10-11T18:37:25Z

@aaronfranke The fact a feature is merged does not mean its final. Most of what I PRd will still have changes over the subsequent months. I always document the features I add sometime before the stable release. This is often the case with large features that are developed over time with partial commits.

reduz · 2020-10-11T18:41:30Z

@aaronfranke In any case, as I suggested you earlier, you need to be patient. The 4.0 branch is still under development and things are changing too much, don't expect the same rules that apply for the stable branches for the time being. The exception is only features that are clearly final, or features that will be cherry picked or backported, which is not the case of this one.

bruvzg reviewed Oct 8, 2020

View reviewed changes

servers/rendering/rasterizer_rd/shaders/particles.glsl Outdated Show resolved Hide resolved

bruvzg mentioned this pull request Oct 8, 2020

[macOS] Fix GPU 3D particles shader compilation. #42570

Closed

bruvzg reviewed Oct 8, 2020

View reviewed changes

bruvzg added enhancement feature proposal topic:rendering labels Oct 8, 2020

bruvzg added this to the 4.0 milestone Oct 8, 2020

reduz force-pushed the particle-collisions branch from 66f8411 to 4e9425b Compare October 8, 2020 12:06

reduz force-pushed the particle-collisions branch from 4e9425b to c769723 Compare October 8, 2020 12:48

reduz force-pushed the particle-collisions branch from c769723 to a391dbe Compare October 8, 2020 12:56

clayjohn previously requested changes Oct 9, 2020

View reviewed changes

reduz force-pushed the particle-collisions branch from a391dbe to 57f6de4 Compare October 9, 2020 12:47

akien-mga reviewed Oct 9, 2020

View reviewed changes

Implement GPU Particle Collisions

26f5bd2

-Sphere Attractor -Box Attractor -Vector Field -Sphere Collider -Box Collider -Baked SDF Collider -Heightmap Collider

reduz force-pushed the particle-collisions branch from 57f6de4 to 26f5bd2 Compare October 9, 2020 16:26

akien-mga merged commit b724d38 into godotengine:master Oct 9, 2020

Calinou mentioned this pull request Oct 11, 2020

New particle system request for ideas! godotengine/godot-proposals#815

Closed

Calinou mentioned this pull request Apr 30, 2021

Particles bouncing off of collision objects? #3994

Closed

Mooses2k mentioned this pull request Mar 1, 2023

Make candles flicker when movement near them Mooses2k/SecretHistories#374

Open

	for (int i = 0; i <= 360; i++) {
	for (int i = 0; i < 360; i++) {


		vec3 attractor_force = vec3(0.0);

		if (PARTICLE.is_active) { //process collision

-//
-////
+//
+////
+////////

-//
-////
+//
+////
+//////

		}
		Ref<Texture3D> GPUParticlesCollisionSDF::get_texture() const {

		}
		GPUParticlesAttractorSphere::GPUParticlesAttractorSphere() :

		/////

		////

		}
		GPUParticlesAttractor3D::~GPUParticlesAttractor3D() {

Implement GPU Particle Collisions #42628

Implement GPU Particle Collisions #42628

Conversation

reduz commented Oct 8, 2020

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

Zireael07 commented Oct 8, 2020

Calinou commented Oct 8, 2020 • edited Loading

reduz commented Oct 8, 2020

clayjohn left a comment • edited Loading

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

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

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

reduz commented Oct 9, 2020

akien-mga left a comment

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

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

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

Choose a reason for hiding this comment

akien-mga commented Oct 9, 2020

aaronfranke commented Oct 11, 2020

reduz commented Oct 11, 2020 • edited Loading

reduz commented Oct 11, 2020

Calinou commented Oct 8, 2020 •

edited

Loading

clayjohn left a comment •

edited

Loading

reduz commented Oct 11, 2020 •

edited

Loading