Skip to content

Various Solari improvements #21649

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Nov 2, 2025
Merged

Various Solari improvements #21649

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
c1a1710
Various Solari improvements
jasmine-nominal Oct 24, 2025
dac71e9
Fmt
JMS55 Oct 25, 2025
4e8ab80
Merge remote-tracking branch 'origin/main' into solari6-various-impro…
jasmine-nominal Oct 31, 2025
49c9f8f
Extract permute_pixel
jasmine-nominal Oct 31, 2025
191b794
Fix roughness cutoff
jasmine-nominal Oct 31, 2025
55d0d59
Merge pull request #40 from jasmine-nominal/solari6-various-improvements
JMS55 Oct 31, 2025
a504705
Jitter in tangent plane before looking up in world cache
JMS55 Nov 1, 2025
3669f59
Add imports
JMS55 Nov 1, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions crates/bevy_solari/src/realtime/gbuffer_utils.wgsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,3 +44,12 @@ fn pixel_dissimilar(depth: f32, world_position: vec3<f32>, other_world_position:

return tangent_plane_distance / view_z > 0.003 || dot(normal, other_normal) < 0.906;
}

fn permute_pixel(pixel_id: vec2<u32>, frame_index: u32, view_size: vec2<f32>) -> vec2<u32> {
let r = frame_index;
let offset = vec2(r & 3u, (r >> 2u) & 3u);
var shifted_pixel_id = pixel_id + offset;
shifted_pixel_id ^= vec2(3u);
shifted_pixel_id -= offset;
return min(shifted_pixel_id, vec2<u32>(view_size - 1.0));
}
3 changes: 2 additions & 1 deletion crates/bevy_solari/src/realtime/node.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,7 +204,8 @@ impl ViewNode for SolariLightingNode {
let bind_group_resolve_dlss_rr_textures = view_dlss_rr_textures.map(|d| {
render_context.render_device().create_bind_group(
"solari_lighting_bind_group_resolve_dlss_rr_textures",
&self.bind_group_layout_resolve_dlss_rr_textures,
&pipeline_cache
.get_bind_group_layout(&self.bind_group_layout_resolve_dlss_rr_textures),
&BindGroupEntries::sequential((
&d.diffuse_albedo.default_view,
&d.specular_albedo.default_view,
Expand Down
22 changes: 10 additions & 12 deletions crates/bevy_solari/src/realtime/restir_di.wgsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
#import bevy_render::maths::PI
#import bevy_render::view::View
#import bevy_solari::brdf::evaluate_brdf
#import bevy_solari::gbuffer_utils::{gpixel_resolve, pixel_dissimilar}
#import bevy_solari::gbuffer_utils::{gpixel_resolve, pixel_dissimilar, permute_pixel}
#import bevy_solari::presample_light_tiles::{ResolvedLightSamplePacked, unpack_resolved_light_sample}
#import bevy_solari::sampling::{LightSample, calculate_resolved_light_contribution, resolve_and_calculate_light_contribution, resolve_light_sample, trace_light_visibility}
#import bevy_solari::scene_bindings::{light_sources, previous_frame_light_id_translations, LIGHT_NOT_PRESENT_THIS_FRAME}
Expand All @@ -27,7 +27,7 @@
struct PushConstants { frame_index: u32, reset: u32 }
var<push_constant> constants: PushConstants;

const INITIAL_SAMPLES = 32u;
const INITIAL_SAMPLES = 8u;
Copy link
Contributor

@SparkyPotato SparkyPotato Oct 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What's the perf difference?

Copy link
Contributor

@jasmine-nominal jasmine-nominal Oct 27, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure, couldn't get bistro setup properly to test :(

const SPATIAL_REUSE_RADIUS_PIXELS = 30.0;
const CONFIDENCE_WEIGHT_CAP = 20.0;

Expand Down Expand Up @@ -73,7 +73,12 @@ fn spatial_and_shade(@builtin(global_invocation_id) global_id: vec3<u32>) {
let input_reservoir = load_reservoir_b(global_id.xy);
let spatial_reservoir = load_spatial_reservoir(global_id.xy, depth, surface.world_position, surface.world_normal, &rng);
let merge_result = merge_reservoirs(input_reservoir, spatial_reservoir, surface.world_position, surface.world_normal, diffuse_brdf, &rng);
let combined_reservoir = merge_result.merged_reservoir;
var combined_reservoir = merge_result.merged_reservoir;

if reservoir_valid(combined_reservoir) {
let resolved_light_sample = resolve_light_sample(combined_reservoir.sample, light_sources[combined_reservoir.sample.light_id >> 16u]);
combined_reservoir.unbiased_contribution_weight *= trace_light_visibility(surface.world_position, resolved_light_sample.world_position);
}

store_reservoir_a(global_id.xy, combined_reservoir);

Expand Down Expand Up @@ -133,7 +138,7 @@ fn generate_initial_reservoir(world_position: vec3<f32>, world_normal: vec3<f32>
fn load_temporal_reservoir(pixel_id: vec2<u32>, depth: f32, world_position: vec3<f32>, world_normal: vec3<f32>) -> Reservoir {
let motion_vector = textureLoad(motion_vectors, pixel_id, 0).xy;
let temporal_pixel_id_float = round(vec2<f32>(pixel_id) - (motion_vector * view.main_pass_viewport.zw));
let temporal_pixel_id = vec2<u32>(temporal_pixel_id_float);
let temporal_pixel_id = permute_pixel(vec2<u32>(temporal_pixel_id_float), constants.frame_index, view.viewport.zw);

// Check if the current pixel was off screen during the previous frame (current pixel is newly visible),
// or if all temporal history should assumed to be invalid
Expand Down Expand Up @@ -173,14 +178,7 @@ fn load_spatial_reservoir(pixel_id: vec2<u32>, depth: f32, world_position: vec3<
return empty_reservoir();
}

var spatial_reservoir = load_reservoir_b(spatial_pixel_id);

if reservoir_valid(spatial_reservoir) {
let resolved_light_sample = resolve_light_sample(spatial_reservoir.sample, light_sources[spatial_reservoir.sample.light_id >> 16u]);
spatial_reservoir.unbiased_contribution_weight *= trace_light_visibility(world_position, resolved_light_sample.world_position);
}

return spatial_reservoir;
return load_reservoir_b(spatial_pixel_id);
}

fn get_neighbor_pixel_id(center_pixel_id: vec2<u32>, rng: ptr<function, u32>) -> vec2<u32> {
Expand Down
80 changes: 24 additions & 56 deletions crates/bevy_solari/src/realtime/restir_gi.wgsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
#import bevy_render::maths::PI
#import bevy_render::view::View
#import bevy_solari::brdf::evaluate_diffuse_brdf
#import bevy_solari::gbuffer_utils::{gpixel_resolve, pixel_dissimilar}
#import bevy_solari::gbuffer_utils::{gpixel_resolve, pixel_dissimilar, permute_pixel}
#import bevy_solari::sampling::{sample_random_light, trace_point_visibility}
#import bevy_solari::scene_bindings::{trace_ray, resolve_ray_hit_full, RAY_T_MIN, RAY_T_MAX}
#import bevy_solari::world_cache::query_world_cache
Expand Down Expand Up @@ -67,7 +67,9 @@ fn spatial_and_shade(@builtin(global_invocation_id) global_id: vec3<u32>) {
let spatial = load_spatial_reservoir(global_id.xy, depth, surface.world_position, surface.world_normal, &rng);
let merge_result = merge_reservoirs(input_reservoir, surface.world_position, surface.world_normal, surface.material.base_color / PI,
spatial.reservoir, spatial.world_position, spatial.world_normal, spatial.diffuse_brdf, &rng);
let combined_reservoir = merge_result.merged_reservoir;
var combined_reservoir = merge_result.merged_reservoir;

combined_reservoir.radiance *= trace_point_visibility(surface.world_position, combined_reservoir.sample_point_world_position);

gi_reservoirs_a[pixel_index] = combined_reservoir;

Expand All @@ -76,10 +78,6 @@ fn spatial_and_shade(@builtin(global_invocation_id) global_id: vec3<u32>) {
var pixel_color = textureLoad(view_output, global_id.xy);
pixel_color += vec4(merge_result.selected_sample_radiance * combined_reservoir.unbiased_contribution_weight * view.exposure * brdf, 0.0);
textureStore(view_output, global_id.xy, pixel_color);

#ifdef VISUALIZE_WORLD_CACHE
textureStore(view_output, global_id.xy, vec4(query_world_cache(surface.world_position, surface.world_normal, view.world_position) * view.exposure, 1.0));
#endif
}

fn generate_initial_reservoir(world_position: vec3<f32>, world_normal: vec3<f32>, rng: ptr<function, u32>) -> Reservoir {
Expand Down Expand Up @@ -107,7 +105,7 @@ fn generate_initial_reservoir(world_position: vec3<f32>, world_normal: vec3<f32>
reservoir.radiance = direct_lighting.radiance;
reservoir.unbiased_contribution_weight = direct_lighting.inverse_pdf * uniform_hemisphere_inverse_pdf();
#else
reservoir.radiance = query_world_cache(sample_point.world_position, sample_point.geometric_world_normal, view.world_position);
reservoir.radiance = query_world_cache(sample_point.world_position, sample_point.geometric_world_normal, view.world_position, rng);
reservoir.unbiased_contribution_weight = uniform_hemisphere_inverse_pdf();
#endif

Expand All @@ -120,43 +118,28 @@ fn generate_initial_reservoir(world_position: vec3<f32>, world_normal: vec3<f32>
fn load_temporal_reservoir(pixel_id: vec2<u32>, depth: f32, world_position: vec3<f32>, world_normal: vec3<f32>) -> NeighborInfo {
let motion_vector = textureLoad(motion_vectors, pixel_id, 0).xy;
let temporal_pixel_id_float = round(vec2<f32>(pixel_id) - (motion_vector * view.main_pass_viewport.zw));
let temporal_pixel_id = permute_pixel(vec2<u32>(temporal_pixel_id_float), constants.frame_index, view.viewport.zw);

// Check if the current pixel was off screen during the previous frame (current pixel is newly visible),
// or if all temporal history should assumed to be invalid
if any(temporal_pixel_id_float < vec2(0.0)) || any(temporal_pixel_id_float >= view.main_pass_viewport.zw) || bool(constants.reset) {
return NeighborInfo(empty_reservoir(), vec3(0.0), vec3(0.0), vec3(0.0));
}

let temporal_pixel_id_base = vec2<u32>(round(temporal_pixel_id_float));
for (var i = 0u; i < 4u; i++) {
let temporal_pixel_id = permute_pixel(temporal_pixel_id_base, i);

// Check if the pixel features have changed heavily between the current and previous frame
let temporal_depth = textureLoad(previous_depth_buffer, temporal_pixel_id, 0);
let temporal_surface = gpixel_resolve(textureLoad(previous_gbuffer, temporal_pixel_id, 0), temporal_depth, temporal_pixel_id, view.main_pass_viewport.zw, previous_view.world_from_clip);
let temporal_diffuse_brdf = temporal_surface.material.base_color / PI;
if pixel_dissimilar(depth, world_position, temporal_surface.world_position, world_normal, temporal_surface.world_normal, view) {
continue;
}

let temporal_pixel_index = temporal_pixel_id.x + temporal_pixel_id.y * u32(view.main_pass_viewport.z);
var temporal_reservoir = gi_reservoirs_a[temporal_pixel_index];

temporal_reservoir.confidence_weight = min(temporal_reservoir.confidence_weight, CONFIDENCE_WEIGHT_CAP);

return NeighborInfo(temporal_reservoir, temporal_surface.world_position, temporal_surface.world_normal, temporal_diffuse_brdf);
// Check if the pixel features have changed heavily between the current and previous frame
let temporal_depth = textureLoad(previous_depth_buffer, temporal_pixel_id, 0);
let temporal_surface = gpixel_resolve(textureLoad(previous_gbuffer, temporal_pixel_id, 0), temporal_depth, temporal_pixel_id, view.main_pass_viewport.zw, previous_view.world_from_clip);
let temporal_diffuse_brdf = temporal_surface.material.base_color / PI;
if pixel_dissimilar(depth, world_position, temporal_surface.world_position, world_normal, temporal_surface.world_normal, view) {
return NeighborInfo(empty_reservoir(), vec3(0.0), vec3(0.0), vec3(0.0));
}

return NeighborInfo(empty_reservoir(), vec3(0.0), vec3(0.0), vec3(0.0));
}
let temporal_pixel_index = temporal_pixel_id.x + temporal_pixel_id.y * u32(view.main_pass_viewport.z);
var temporal_reservoir = gi_reservoirs_a[temporal_pixel_index];

fn permute_pixel(pixel_id: vec2<u32>, i: u32) -> vec2<u32> {
let r = constants.frame_index + i;
let offset = vec2(r & 3u, (r >> 2u) & 3u);
var shifted_pixel_id = pixel_id + offset;
shifted_pixel_id ^= vec2(3u);
shifted_pixel_id -= offset;
return min(shifted_pixel_id, vec2<u32>(view.main_pass_viewport.zw - 1.0));
temporal_reservoir.confidence_weight = min(temporal_reservoir.confidence_weight, CONFIDENCE_WEIGHT_CAP);

return NeighborInfo(temporal_reservoir, temporal_surface.world_position, temporal_surface.world_normal, temporal_diffuse_brdf);
}

fn load_spatial_reservoir(pixel_id: vec2<u32>, depth: f32, world_position: vec3<f32>, world_normal: vec3<f32>, rng: ptr<function, u32>) -> NeighborInfo {
Expand All @@ -170,9 +153,7 @@ fn load_spatial_reservoir(pixel_id: vec2<u32>, depth: f32, world_position: vec3<
}

let spatial_pixel_index = spatial_pixel_id.x + spatial_pixel_id.y * u32(view.main_pass_viewport.z);
var spatial_reservoir = gi_reservoirs_b[spatial_pixel_index];

spatial_reservoir.radiance *= trace_point_visibility(world_position, spatial_reservoir.sample_point_world_position);
let spatial_reservoir = gi_reservoirs_b[spatial_pixel_index];

return NeighborInfo(spatial_reservoir, spatial_surface.world_position, spatial_surface.world_normal, spatial_diffuse_brdf);
}
Expand Down Expand Up @@ -252,27 +233,19 @@ fn merge_reservoirs(
rng: ptr<function, u32>,
) -> ReservoirMergeResult {
// Radiances for resampling
let canonical_sample_radiance =
canonical_reservoir.radiance *
saturate(dot(normalize(canonical_reservoir.sample_point_world_position - canonical_world_position), canonical_world_normal));
let other_sample_radiance =
other_reservoir.radiance *
saturate(dot(normalize(other_reservoir.sample_point_world_position - canonical_world_position), canonical_world_normal));
let canonical_sample_radiance = canonical_reservoir.radiance * saturate(dot(normalize(canonical_reservoir.sample_point_world_position - canonical_world_position), canonical_world_normal));
let other_sample_radiance = other_reservoir.radiance * saturate(dot(normalize(other_reservoir.sample_point_world_position - canonical_world_position), canonical_world_normal));

// Target functions for resampling and MIS
let canonical_target_function_canonical_sample = luminance(canonical_sample_radiance * canonical_diffuse_brdf);
let canonical_target_function_other_sample = luminance(other_sample_radiance * canonical_diffuse_brdf);

// Extra target functions for MIS
let other_target_function_canonical_sample = luminance(
canonical_reservoir.radiance *
saturate(dot(normalize(canonical_reservoir.sample_point_world_position - other_world_position), other_world_normal)) *
other_diffuse_brdf
canonical_reservoir.radiance * saturate(dot(normalize(canonical_reservoir.sample_point_world_position - other_world_position), other_world_normal)) * other_diffuse_brdf
);
let other_target_function_other_sample = luminance(
other_reservoir.radiance *
saturate(dot(normalize(other_reservoir.sample_point_world_position - other_world_position), other_world_normal)) *
other_diffuse_brdf
other_reservoir.radiance * saturate(dot(normalize(other_reservoir.sample_point_world_position - other_world_position), other_world_normal)) * other_diffuse_brdf
);

// Jacobians for resampling and MIS
Expand All @@ -299,19 +272,14 @@ fn merge_reservoirs(
canonical_reservoir.confidence_weight * canonical_target_function_canonical_sample,
other_reservoir.confidence_weight * other_target_function_canonical_sample * other_target_function_canonical_sample_jacobian,
);
let canonical_sample_resampling_weight = canonical_sample_mis_weight *
canonical_target_function_canonical_sample *
canonical_reservoir.unbiased_contribution_weight;
let canonical_sample_resampling_weight = canonical_sample_mis_weight * canonical_target_function_canonical_sample * canonical_reservoir.unbiased_contribution_weight;

// Resampling weight for other sample
let other_sample_mis_weight = balance_heuristic(
other_reservoir.confidence_weight * other_target_function_other_sample,
canonical_reservoir.confidence_weight * canonical_target_function_other_sample * canonical_target_function_other_sample_jacobian,
);
let other_sample_resampling_weight = other_sample_mis_weight *
canonical_target_function_other_sample *
other_reservoir.unbiased_contribution_weight *
canonical_target_function_other_sample_jacobian;
let other_sample_resampling_weight = other_sample_mis_weight * canonical_target_function_other_sample * other_reservoir.unbiased_contribution_weight * canonical_target_function_other_sample_jacobian;

// Perform resampling
var combined_reservoir = empty_reservoir();
Expand Down
20 changes: 13 additions & 7 deletions crates/bevy_solari/src/realtime/specular_gi.wgsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ fn specular_gi(@builtin(global_invocation_id) global_id: vec3<u32>) {

var radiance: vec3<f32>;
var wi: vec3<f32>;
if surface.material.roughness > 0.04 {
if surface.material.roughness > 0.1 {
Copy link
Contributor

@SparkyPotato SparkyPotato Oct 25, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would it be worth subgroupAny-ing this (and flipping the condition)? One branch is a lot more expensive than the other.

Copy link
Contributor

@jasmine-nominal jasmine-nominal Oct 27, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe? Would have to test. I feel like for latency reasons, you'd want to trace the minimum amount of rays possible, even if some threads end up idle.

// Surface is very rough, reuse the ReSTIR GI reservoir
let gi_reservoir = gi_reservoirs_a[pixel_index];
wi = normalize(gi_reservoir.sample_point_world_position - surface.world_position);
Expand All @@ -59,25 +59,31 @@ fn specular_gi(@builtin(global_invocation_id) global_id: vec3<u32>) {
var pixel_color = textureLoad(view_output, global_id.xy);
pixel_color += vec4(radiance, 0.0);
textureStore(view_output, global_id.xy, pixel_color);

#ifdef VISUALIZE_WORLD_CACHE
textureStore(view_output, global_id.xy, vec4(query_world_cache(surface.world_position, surface.world_normal, view.world_position, &rng) * view.exposure, 1.0));
#endif
}

fn trace_glossy_path(initial_ray_origin: vec3<f32>, initial_wi: vec3<f32>, rng: ptr<function, u32>) -> vec3<f32> {
var ray_origin = initial_ray_origin;
var wi = initial_wi;

// Trace up to three bounces, getting the net throughput from them
var radiance = vec3(0.0);
var throughput = vec3(1.0);
for (var i = 0u; i < 3u; i += 1u) {
// Trace ray
let ray = trace_ray(ray_origin, wi, RAY_T_MIN, RAY_T_MAX, RAY_FLAG_NONE);
if ray.kind == RAY_QUERY_INTERSECTION_NONE { break; }
let ray_hit = resolve_ray_hit_full(ray);

// Add world cache contribution
let diffuse_brdf = ray_hit.material.base_color / PI;
radiance += throughput * diffuse_brdf * query_world_cache(ray_hit.world_position, ray_hit.geometric_world_normal, view.world_position, rng);

// Surface is very rough, terminate path in the world cache
if ray_hit.material.roughness > 0.04 || i == 2u {
let diffuse_brdf = ray_hit.material.base_color / PI;
return throughput * diffuse_brdf * query_world_cache(ray_hit.world_position, ray_hit.geometric_world_normal, view.world_position);
}
if ray_hit.material.roughness > 0.1 && i != 0u { break; }

// Sample new ray direction from the GGX BRDF for next bounce
let TBN = calculate_tbn_mikktspace(ray_hit.world_normal, ray_hit.world_tangent);
Expand All @@ -93,11 +99,11 @@ fn trace_glossy_path(initial_ray_origin: vec3<f32>, initial_wi: vec3<f32>, rng:
// Update throughput for next bounce
let pdf = ggx_vndf_pdf(wo_tangent, wi_tangent, ray_hit.material.roughness);
let brdf = evaluate_brdf(N, wo, wi, ray_hit.material);
let cos_theta = dot(wi, N);
let cos_theta = saturate(dot(wi, N));
throughput *= (brdf * cos_theta) / pdf;
}

return vec3(0.0);
return radiance;
}

// Don't adjust the size of this struct without also adjusting GI_RESERVOIR_STRUCT_SIZE.
Expand Down
Loading