Using min

gents83 · Feb 21, 2024 · 92d9567 · 92d9567
1 parent df659e1
commit 92d9567
Show file tree

Hide file tree

Showing 5 changed files with 25 additions and 21 deletions.
diff --git a/crates/graphics/src/common/passes/compute_culling.rs b/crates/graphics/src/common/passes/compute_culling.rs
@@ -4,7 +4,7 @@ use crate::{
  ArrayI32, ArrayU32, AsBinding, BVHBuffer, BindingData, BindingFlags, BindingInfo,
  CommandBuffer, ComputePass, ComputePassData, ConstantDataRw, DrawCommandType,
  DrawCommandsBuffer, GpuBuffer, Mesh, MeshFlags, MeshesBuffer, MeshletsBuffer, Pass,
- RenderContext, RenderContextRc, ShaderStage, TextureView, ATOMIC_SIZE,
+ RenderContext, RenderContextRc, ShaderStage, TextureView, ATOMIC_SIZE, MAX_LOD_LEVELS,
 };
 
 use inox_commands::CommandParser;
@@ -170,7 +170,7 @@ impl Pass for CullingPass {
  let num_meshlets = self.meshlets.read().unwrap().item_count();
  let mut max_meshlets_size = num_meshlets as u32 + ATOMIC_SIZE;
  max_meshlets_size += ATOMIC_SIZE - (max_meshlets_size % ATOMIC_SIZE);
- let processing_data = vec![-1; max_meshlets_size as usize];
+ let processing_data = vec![MAX_LOD_LEVELS as i32; max_meshlets_size as usize];
  self.processing_data.write().unwrap().set(processing_data);
 
  if self.update_meshes {

diff --git a/crates/graphics/src/common/render_context.rs b/crates/graphics/src/common/render_context.rs
@@ -17,6 +17,7 @@ use crate::{
 };
 
 const USE_FORCED_VULKAN: bool = false;
+const USE_FORCED_DX12: bool = false;
 
 #[derive(PartialEq, Eq, Copy, Clone)]
 pub enum RendererState {
@@ -144,12 +145,14 @@ impl RenderContext {
  let (instance, surface, adapter, device, queue) = {
  let backends = if USE_FORCED_VULKAN {
  wgpu::Backends::VULKAN
+ } else if USE_FORCED_DX12 {
+ wgpu::Backends::DX12
  } else {
  wgpu::Backends::all()
  };
  let instance = wgpu::Instance::new(wgpu::InstanceDescriptor {
  backends,
- flags: wgpu::InstanceFlags::advanced_debugging(),
+ flags: wgpu::InstanceFlags::empty(),
  ..Default::default()
  });
  let surface = Self::create_surface(&instance, handle.clone());

diff --git a/crates/graphics/src/data/gpu_data.rs b/crates/graphics/src/data/gpu_data.rs
@@ -40,7 +40,7 @@ pub struct DrawIndexedCommand {
  pub base_instance: u32,
 }
 
-#[repr(C, align(4))]
+#[repr(C)]
 #[derive(Default, PartialEq, Eq, Clone, Copy, Debug)]
 pub struct DrawCommand {
  pub vertex_count: u32,
@@ -91,7 +91,7 @@ impl GPUMesh {
  }
 }
 
-#[repr(C, align(16))]
+#[repr(C)]
 #[derive(Default, PartialEq, Clone, Copy, Debug)]
 pub struct GPUMeshlet {
  pub mesh_index_and_lod_level: u32, // 29 mesh + 3 lod bits
@@ -114,7 +114,7 @@ impl GPUMeshlet {
  }
 }
 
-#[repr(C, align(16))]
+#[repr(C)]
 #[derive(PartialEq, Clone, Copy, Debug)]
 pub struct GPUMaterial {
  pub roughness_factor: f32,
@@ -159,7 +159,7 @@ impl Default for GPUMaterial {
  }
 }
 
-#[repr(C, align(4))]
+#[repr(C)]
 #[derive(Default, PartialEq, Clone, Copy, Debug)]
 pub struct GPURuntimeVertexData {
  pub world_pos: [f32; 3],

diff --git a/crates/graphics/src/platform/pc.rs b/crates/graphics/src/platform/pc.rs
@@ -1,7 +1,7 @@
 #![cfg(not(target_arch = "wasm32"))]
 
 pub fn required_gpu_features() -> wgpu::Features {
- wgpu::Features::all_webgpu_mask()
+ wgpu::Features::default()
  | wgpu::Features::POLYGON_MODE_LINE
  | wgpu::Features::INDIRECT_FIRST_INSTANCE
  | wgpu::Features::MULTI_DRAW_INDIRECT

diff --git a/data_raw/shaders/wgsl/compute_culling.wgsl b/data_raw/shaders/wgsl/compute_culling.wgsl
@@ -123,7 +123,7 @@ fn main(
  let ncd_max = clip_mvp * vec4<f32>(max, 1.);
  let clip_max = ncd_max.xyz / ncd_max.w;
  let screen_max = clip_to_normalized(clip_max.xy);
- let screen_diff = max(screen_max, screen_min) - min(screen_max, screen_min);
+ let screen_diff = (max(screen_max, screen_min) - min(screen_max, screen_min)) * 2.;
  if clip_min.z > 1. && clip_max.z > 1. {
  desired_lod_level = 0;
  }
@@ -140,33 +140,34 @@ fn main(
  let meshlet_lod_level = meshlet.mesh_index_and_lod_level & 7u;
  let lod_level = u32(desired_lod_level);
  if(meshlet_lod_level < lod_level) { 
+ let max_lod_level = i32(MAX_LOD_LEVELS);
  if(meshlet.child_meshlets.x >= 0) { 
- let result = atomicCompareExchangeWeak(&processing_data[meshlet.child_meshlets.x], -1, desired_lod_level);
- if(result.exchanged) {
+ let v = atomicMin(&processing_data[meshlet.child_meshlets.x], desired_lod_level);
+ if (v == max_lod_level) {
  let child_index = atomicAdd(&global_count, 1u);
  atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.x));
  }
  } 
  if(meshlet.child_meshlets.y >= 0) { 
- let result = atomicCompareExchangeWeak(&processing_data[meshlet.child_meshlets.y], -1, desired_lod_level);
- if(result.exchanged) { 
+ let v = atomicMin(&processing_data[meshlet.child_meshlets.y], desired_lod_level);
+ if (v == max_lod_level) {
  let child_index = atomicAdd(&global_count, 1u); 
- atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.y));  
+ atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.y)); 
  }
  } 
  if(meshlet.child_meshlets.z >= 0) { 
- let result = atomicCompareExchangeWeak(&processing_data[meshlet.child_meshlets.z], -1, desired_lod_level);
- if(result.exchanged) { 
+ let v = atomicMin(&processing_data[meshlet.child_meshlets.z], desired_lod_level);
+ if (v == max_lod_level) {
  let child_index = atomicAdd(&global_count, 1u); 
- atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.z));  
+ atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.z)); 
  }
  } 
  if(meshlet.child_meshlets.w >= 0) { 
- let result = atomicCompareExchangeWeak(&processing_data[meshlet.child_meshlets.w], -1, desired_lod_level);
- if(result.exchanged) { 
+ let v = atomicMin(&processing_data[meshlet.child_meshlets.w], desired_lod_level);
+ if (v == max_lod_level) {
  let child_index = atomicAdd(&global_count, 1u); 
- atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.w));  
- }
+ atomicStore(&meshlet_culling_data[child_index], u32(meshlet.child_meshlets.w)); 
+ } 
  } 
  } 
  else if(meshlet_lod_level == lod_level)