first it worked, then i broke it and finally remembered to commit

This commit is contained in:
Kasper Sauramo
2025-08-01 19:18:36 +03:00
parent 6e04bf27ff
commit e2f1c8c13a
4 changed files with 134 additions and 58 deletions

View File

@@ -4,32 +4,66 @@
// Invocations in the (x, y, z) dimension // Invocations in the (x, y, z) dimension
layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
layout(r32f, set = 0, binding = 0) uniform restrict readonly image2D current_trail; layout(r32f, binding = 0) uniform restrict readonly image2D current_trail;
layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D output_trail; layout(r32f, binding = 1) uniform restrict writeonly image2D output_trail;
layout(set = 2, binding = 1) restrict buffer AgentBuffer { layout(binding = 2) restrict buffer AgentBuffer {
vec2 positions[]; vec2 positions[1024*1024];
float headings[1024*1024];
} agent_buffer; } agent_buffer;
layout(push_constant, std430) uniform Params { layout(push_constant, std430) uniform Params {
vec2 texture_size; ivec2 texture_size;
float sensor_angle;
float sensor_distance;
} params; } params;
const float SPEED = 1.0;
void main() { void main() {
ivec2 size = ivec2(params.texture_size.x - 1, params.texture_size.y - 1); ivec2 size = ivec2(params.texture_size.x - 1, params.texture_size.y - 1);
vec2 position = agent_buffer.positions[gl_GlobalInvocationID.x];
ivec2 uv = ivec2(gl_GlobalInvocationID.xy); uint id = gl_GlobalInvocationID.x;
vec2 position = agent_buffer.positions[id];
float heading = agent_buffer.headings[id];
ivec2 tl = ivec2(0, 0); float sa = params.sensor_angle;
float current_v = imageLoad(current_trail, clamp(uv - ivec2(0, 1), tl, size)).r; float left_heading = heading - sa;
float up_v = imageLoad(current_trail, clamp(uv - ivec2(0, 1), tl, size)).r; float right_heading = heading + sa;
float down_v = imageLoad(current_trail, clamp(uv + ivec2(0, 1), tl, size)).r;
float left_v = imageLoad(current_trail, clamp(uv - ivec2(1, 0), tl, size)).r;
float right_v = imageLoad(current_trail, clamp(uv + ivec2(1, 0), tl, size)).r;
float new_v = current_v + 0.05; vec2 left_detect_dir = vec2(cos(left_heading), sin(left_heading));
vec2 right_detect_dir = vec2(cos(right_heading), sin(right_heading));
vec2 forward_detect_dir = vec2(cos(heading), sin(heading));
float sd = params.sensor_distance;
vec2 left_detect_pos = position + left_detect_dir * sd;
vec2 right_detect_pos = position + right_detect_dir * sd;
vec2 forward_detect_pos = position + forward_detect_dir * sd;
left_detect_pos = mod(left_detect_pos, size);
right_detect_pos = mod(right_detect_pos, size);
forward_detect_pos = mod(forward_detect_pos, size);
float left = imageLoad(current_trail, ivec2(left_detect_pos)).r;
float right = imageLoad(current_trail, ivec2(right_detect_pos)).r;
float forward = imageLoad(current_trail, ivec2(forward_detect_pos)).r;
vec2 pos = position;
if (left > forward && left > right) {
pos += left_detect_dir * SPEED;
} else if (right > forward && right > left) {
pos += right_detect_dir * SPEED;
} else {
pos += forward_detect_dir * SPEED;
}
pos = mod(pos, size);
agent_buffer.positions[id] = pos;
ivec2 uv = ivec2(pos);
float current_v = imageLoad(current_trail, uv).r;
float new_v = current_v + 0.3;
vec4 result = vec4(new_v, new_v, new_v, 1.0); vec4 result = vec4(new_v, new_v, new_v, 1.0);
imageStore(output_trail, uv, result); imageStore(output_trail, uv, result);

View File

@@ -12,8 +12,17 @@ layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D output_ima
layout(push_constant, std430) uniform Params { layout(push_constant, std430) uniform Params {
vec2 texture_size; vec2 texture_size;
float decay_factor; float decay_factor;
float dissipation_factor;
} params; } params;
float get_dissipation_from(ivec2 uv) {
float v = imageLoad(current_image, uv).r;
return v * params.dissipation_factor;
}
void store_value(ivec2 uv, vec4 v) {
imageStore(output_image, uv, v);
}
// The code we want to execute in each invocation // The code we want to execute in each invocation
void main() { void main() {
@@ -21,24 +30,31 @@ void main() {
ivec2 size = ivec2(params.texture_size.x - 1, params.texture_size.y - 1); ivec2 size = ivec2(params.texture_size.x - 1, params.texture_size.y - 1);
ivec2 uv = ivec2(gl_GlobalInvocationID.xy); ivec2 uv = ivec2(gl_GlobalInvocationID.xy);
ivec2 tl = ivec2(0, 0);
ivec2 up_uv = clamp(uv - ivec2(0, 1), tl, size);
ivec2 down_uv = clamp(uv + ivec2(0, 1), tl, size);
ivec2 left_uv = clamp(uv - ivec2(1, 0), tl, size);
ivec2 right_uv = clamp(uv + ivec2(1, 0), tl, size);
// Just in case the texture size is not divisable by 8. // Just in case the texture size is not divisable by 8.
if ((uv.x > size.x) || (uv.y > size.y)) { if ((uv.x > size.x) || (uv.y > size.y)) {
return; return;
} }
ivec2 tl = ivec2(0, 0);
float current_v = imageLoad(current_image, uv).r; float current_v = imageLoad(current_image, uv).r;
float up_v = imageLoad(current_image, clamp(uv - ivec2(0, 1), tl, size)).r;
float down_v = imageLoad(current_image, clamp(uv + ivec2(0, 1), tl, size)).r;
float left_v = imageLoad(current_image, clamp(uv - ivec2(1, 0), tl, size)).r;
float right_v = imageLoad(current_image, clamp(uv + ivec2(1, 0), tl, size)).r;
float new_v = current_v * params.decay_factor; float new_v = current_v * params.decay_factor;
new_v += get_dissipation_from(up_uv);
new_v += get_dissipation_from(down_uv);
new_v += get_dissipation_from(right_uv);
new_v += get_dissipation_from(left_uv);
new_v = clamp(new_v, 0, 1);
if (new_v < 0.0) {
new_v = 0.0;
}
vec4 result = vec4(new_v, new_v, new_v, 1.0); vec4 result = vec4(new_v, new_v, new_v, 1.0);
imageStore(output_image, uv, result); imageStore(output_image, uv, result);
} }

View File

@@ -2,9 +2,18 @@ extends Node2D
@export var texture_size: Vector2i = Vector2i(512, 512) @export var texture_size: Vector2i = Vector2i(512, 512)
@export var decay_factor: float = 0.98 @export var decay_factor: float = 0.98
@export var dissipation_factor: float = 0.12
@export var sensor_angle: float = 32.0:
get():
return deg_to_rad(sensor_angle)
@export var sensor_distance: float = 3.86
@export var rng_seed: int = 80085
@onready var mesh := $MeshInstance2D @onready var mesh := $MeshInstance2D
## Needs to be changed from the shader as well
const AGENTS: int = 4096
var texture: Texture2DRD var texture: Texture2DRD
var next_texture: int = 0 var next_texture: int = 0
@@ -22,7 +31,7 @@ func _process(_delta: float) -> void:
texture.texture_rd_rid = texture_rds[next_texture] texture.texture_rd_rid = texture_rds[next_texture]
RenderingServer.call_on_render_thread( RenderingServer.call_on_render_thread(
_render_process.bind(next_texture, texture_size, decay_factor) _render_process.bind(next_texture, texture_size, decay_factor, dissipation_factor, sensor_angle, sensor_distance)
) )
@@ -50,9 +59,8 @@ var agent_pipeline: RID
var texture_rds: Array[RID] = [RID(), RID()] var texture_rds: Array[RID] = [RID(), RID()]
var texture_sets: Array[RID] = [RID(), RID()] var texture_sets: Array[RID] = [RID(), RID()]
var agent_uniform_sets: Array[RID] = [RID(), RID()]
var agent_buffer: RID var agent_buffer: RID
var agent_set: RID var agent_sets: Array[RID] = [RID(), RID()]
func _load_shader(rd_: RenderingDevice, path: String) -> RID: func _load_shader(rd_: RenderingDevice, path: String) -> RID:
@@ -69,7 +77,7 @@ func _create_trail_uniform(texture_rd: RID) -> RDUniform:
return uniform return uniform
func _initialize_compute_code(init_with_texture_size: Vector2i) -> void: func _initialize_compute_code(with_texture_size: Vector2i) -> void:
# As this becomes part of our normal frame rendering, # As this becomes part of our normal frame rendering,
# we use our main rendering device here. # we use our main rendering device here.
rd = RenderingServer.get_rendering_device() rd = RenderingServer.get_rendering_device()
@@ -83,26 +91,31 @@ func _initialize_compute_code(init_with_texture_size: Vector2i) -> void:
# Create agent buffer # Create agent buffer
var rng := RandomNumberGenerator.new() var rng := RandomNumberGenerator.new()
rng.seed = 80085 rng.seed = rng_seed
var agent_data := PackedVector2Array() var agent_positions := PackedVector2Array()
for n in 64 * 16: var agent_angles := PackedFloat32Array()
var new_position: Vector2i = Vector2( # This must match what is set in the shader, if fixed size
rng.randi_range(0, init_with_texture_size.x), for n in AGENTS:
rng.randi_range(0, init_with_texture_size.y) var new_position: Vector2 = Vector2(
rng.randf_range(0, with_texture_size.x),
rng.randf_range(0, with_texture_size.y)
) )
agent_data.push_back(new_position) agent_positions.push_back(new_position)
agent_angles.push_back(rng.randf_range(0, TAU))
var agent_size_in_bytes := 8 var agent_data := PackedByteArray(agent_positions.to_byte_array())
agent_data.append_array(agent_angles.to_byte_array())
agent_buffer = rd.storage_buffer_create( agent_buffer = rd.storage_buffer_create(
agent_data.size() * agent_size_in_bytes, agent_data.to_byte_array() agent_data.size(), agent_data, 0,
RenderingDevice.BUFFER_CREATION_AS_STORAGE_BIT
) )
# Create trail texture # Create trail texture
var tf: RDTextureFormat = RDTextureFormat.new() var tf: RDTextureFormat = RDTextureFormat.new()
tf.format = RenderingDevice.DATA_FORMAT_R32_SFLOAT tf.format = RenderingDevice.DATA_FORMAT_R32_SFLOAT
tf.texture_type = RenderingDevice.TEXTURE_TYPE_2D tf.texture_type = RenderingDevice.TEXTURE_TYPE_2D
tf.width = init_with_texture_size.x tf.width = with_texture_size.x
tf.height = init_with_texture_size.y tf.height = with_texture_size.y
tf.depth = 1 tf.depth = 1
tf.array_layers = 1 tf.array_layers = 1
tf.mipmaps = 1 tf.mipmaps = 1
@@ -115,7 +128,7 @@ func _initialize_compute_code(init_with_texture_size: Vector2i) -> void:
) )
var agent_uniforms = [] var agent_uniforms: Array[RDUniform] = []
for i in 2: for i in 2:
texture_rds[i] = rd.texture_create(tf, RDTextureView.new(), []) texture_rds[i] = rd.texture_create(tf, RDTextureView.new(), [])
# Make sure our textures are cleared. # Make sure our textures are cleared.
@@ -123,17 +136,28 @@ func _initialize_compute_code(init_with_texture_size: Vector2i) -> void:
var trail_uniform := _create_trail_uniform(texture_rds[i]) var trail_uniform := _create_trail_uniform(texture_rds[i])
texture_sets[i] = rd.uniform_set_create([trail_uniform], decay_shader, 0) texture_sets[i] = rd.uniform_set_create([trail_uniform], decay_shader, 0)
agent_uniform_sets[i] = rd.uniform_set_create([trail_uniform], agent_shader, 0)
var txtr_uniform_1 := RDUniform.new()
txtr_uniform_1.uniform_type = RenderingDevice.UNIFORM_TYPE_IMAGE
txtr_uniform_1.binding = 0
txtr_uniform_1.add_id(texture_rds[0])
agent_uniforms.append(txtr_uniform_1)
var txtr_uniform_2 := RDUniform.new()
txtr_uniform_2.uniform_type = RenderingDevice.UNIFORM_TYPE_IMAGE
txtr_uniform_2.binding = 1
txtr_uniform_2.add_id(texture_rds[1])
agent_uniforms.append(txtr_uniform_2)
var uniform := RDUniform.new() var uniform := RDUniform.new()
uniform.uniform_type = RenderingDevice.UNIFORM_TYPE_STORAGE_BUFFER uniform.uniform_type = RenderingDevice.UNIFORM_TYPE_STORAGE_BUFFER
uniform.binding = 1 uniform.binding = 2
uniform.add_id(agent_buffer) uniform.add_id(agent_buffer)
agent_set = rd.uniform_set_create([uniform], agent_shader, 0) agent_sets[0] = rd.uniform_set_create([txtr_uniform_1, txtr_uniform_2, uniform], agent_shader, 0)
agent_sets[1] = rd.uniform_set_create([txtr_uniform_2, txtr_uniform_1, uniform], agent_shader, 0)
func _render_process(with_next_texture: int, tex_size: Vector2i, decay_fctr: float) -> void: func _render_process(with_next_texture: int, tex_size: Vector2i, decay_fctr: float, dissipation_fctr: float, snsr_angle: float, snsr_distance: float) -> void:
var next_set := texture_sets[with_next_texture] var next_set := texture_sets[with_next_texture]
var current_set := texture_sets[(with_next_texture - 1) % 2] var current_set := texture_sets[(with_next_texture - 1) % 2]
@@ -141,19 +165,18 @@ func _render_process(with_next_texture: int, tex_size: Vector2i, decay_fctr: flo
var push_constant := PackedFloat32Array() var push_constant := PackedFloat32Array()
push_constant.push_back(tex_size.x) push_constant.push_back(tex_size.x)
push_constant.push_back(tex_size.y) push_constant.push_back(tex_size.y)
push_constant.push_back(0.0) push_constant.push_back(snsr_angle)
push_constant.push_back(0.0) push_constant.push_back(snsr_distance)
var agent_compute_list := rd.compute_list_begin() var agent_compute_list := rd.compute_list_begin()
rd.compute_list_bind_compute_pipeline(agent_compute_list, agent_pipeline) rd.compute_list_bind_compute_pipeline(agent_compute_list, agent_pipeline)
rd.compute_list_bind_uniform_set(agent_compute_list, current_set, 0) rd.compute_list_bind_uniform_set(agent_compute_list, agent_sets[with_next_texture], 0)
rd.compute_list_bind_uniform_set(agent_compute_list, next_set, 1)
rd.compute_list_bind_uniform_set(agent_compute_list, agent_set, 2)
rd.compute_list_set_push_constant( rd.compute_list_set_push_constant(
agent_compute_list, push_constant.to_byte_array(), push_constant.size() * 4 agent_compute_list, push_constant.to_byte_array(), push_constant.size() * 4
) )
rd.compute_list_dispatch(agent_compute_list, 16, 1, 1) @warning_ignore("integer_division")
rd.compute_list_dispatch(agent_compute_list, 64, 1, 1)
rd.compute_list_end() rd.compute_list_end()
# Decay and dissipate # Decay and dissipate
@@ -161,7 +184,7 @@ func _render_process(with_next_texture: int, tex_size: Vector2i, decay_fctr: flo
push_constant.push_back(tex_size.x) push_constant.push_back(tex_size.x)
push_constant.push_back(tex_size.y) push_constant.push_back(tex_size.y)
push_constant.push_back(decay_fctr) push_constant.push_back(decay_fctr)
push_constant.push_back(0.0) push_constant.push_back(dissipation_fctr)
@warning_ignore("integer_division") @warning_ignore("integer_division")
var x_groups := (tex_size.x - 1) / 8 + 1 var x_groups := (tex_size.x - 1) / 8 + 1
@@ -179,18 +202,15 @@ func _render_process(with_next_texture: int, tex_size: Vector2i, decay_fctr: flo
rd.compute_list_dispatch(compute_list, x_groups, y_groups, 1) rd.compute_list_dispatch(compute_list, x_groups, y_groups, 1)
rd.compute_list_end() rd.compute_list_end()
# We don't need to sync up here, Godots default barriers will do the trick.
# If you want the output of a compute shader to be used as input of
# another computer shader you'll need to add a barrier:
#rd.barrier(RenderingDevice.BARRIER_MASK_COMPUTE)
func _free_compute_resources() -> void: func _free_compute_resources() -> void:
# Note that our sets and pipeline are cleaned up automatically # Note that our sets and pipeline are cleaned up automatically
# as they are dependencies :P # as they are dependencies :P
for i in 3: for i in 2:
if texture_rds[i]: if texture_rds[i]:
rd.free_rid(texture_rds[i]) rd.free_rid(texture_rds[i])
if decay_shader: if decay_shader:
rd.free_rid(decay_shader) rd.free_rid(decay_shader)
if agent_shader:
rd.free_rid(agent_shader)

View File

@@ -13,9 +13,15 @@ shader_parameter/trail_tex = SubResource("Texture2DRD_pxe3a")
[node name="SlimeSimulation" type="Node2D"] [node name="SlimeSimulation" type="Node2D"]
script = ExtResource("1_pxe3a") script = ExtResource("1_pxe3a")
texture_size = Vector2i(1152, 648)
decay_factor = 0.75
dissipation_factor = 0.04
sensor_angle = 27.0
sensor_distance = 5.0
[node name="MeshInstance2D" type="MeshInstance2D" parent="."] [node name="MeshInstance2D" type="MeshInstance2D" parent="."]
self_modulate = Color(0.581022, 0.701395, 0.890748, 1)
material = SubResource("ShaderMaterial_pxe3a") material = SubResource("ShaderMaterial_pxe3a")
position = Vector2(256, 256) position = Vector2(576, 324)
scale = Vector2(512, 512) scale = Vector2(1152, 648)
mesh = SubResource("QuadMesh_pxe3a") mesh = SubResource("QuadMesh_pxe3a")