Continuing cleanup

shared/src/core/geometry/ray.rs

@@ -1,7 +1,7 @@
 use super::{Normal3f, Point3f, Point3fi, Vector3f, VectorLike};
 use crate::core::medium::Medium;
 use crate::utils::math::{next_float_down, next_float_up};
-use crate::{gvec_with_capacity, Float, GVec, Ptr, SOA};
+use crate::{gvec_with_capacity, Float, GVec, Ptr};
 
 #[repr(C)]
 #[derive(Clone, Copy, Debug)]
@@ -123,56 +123,3 @@ pub struct RayDifferential {
     pub rx_direction: Vector3f,
     pub ry_direction: Vector3f,
 }
-
-#[derive(Clone)]
-pub struct RaySoA {
-    pub o: GVec<Point3f>,
-    pub d: GVec<Vector3f>,
-    pub time: GVec<Float>,
-    pub medium: GVec<Ptr<Medium>>,
-    pub has_differentials: GVec<bool>,
-    pub differential: GVec<RayDifferential>,
-}
-
-impl SoA for RaySoA {
-    type Item = Ray;
-
-    fn with_capacity(n: usize) -> Self {
-        Self {
-            o: gvec_with_capacity(n),
-            d: gvec_with_capacity(n),
-            time: gvec_with_capacity(n),
-            medium: gvec_with_capacity(n),
-            has_differentials: gvec_with_capacity(n),
-            differential: gvec_with_capacity(n),
-        }
-    }
-
-    fn len(&self) -> usize {
-        self.o.len()
-    }
-
-    unsafe fn get_unchecked(&self, i: usize) -> Ray {
-        Ray {
-            o: *self.o.get_unchecked(i),
-            d: *self.d.get_unchecked(i),
-            time: *self.time.get_unchecked(i),
-            medium: *self.medium.get_unchecked(i),
-            has_differentials: *self.has_differentials.get_unchecked(i),
-            differential: *self.differential.get_unchecked(i),
-        }
-    }
-
-    unsafe fn set_unchecked(&mut self, i: usize, v: Ray) {
-        *self.o.get_unchecked_mut(i) = v.o;
-        *self.d.get_unchecked_mut(i) = v.d;
-        *self.time.get_unchecked_mut(i) = v.time;
-        *self.medium.get_unchecked_mut(i) = v.medium;
-        *self.has_differentials.get_unchecked_mut(i) = v.has_differentials;
-        *self.differential.get_unchecked_mut(i) = v.differential;
-    }
-}
-
-impl SoAElement for Ray {
-    type SoA = RaySoA;
-}

shared/src/core/material.rs

@@ -178,6 +178,7 @@ pub trait MaterialTrait {
     fn has_subsurface_scattering(&self) -> bool;
 }
 
+#[repr(C)]
 #[derive(Clone, Copy, Debug)]
 #[enum_dispatch(MaterialTrait)]
 pub enum Material {

shared/src/core/primitive.rs

@@ -68,9 +68,6 @@ impl PrimitiveTrait for GeometricPrimitive {
             }
         }
 
-        if r.medium.is_null() {
-            return None;
-        }
         si.set_intersection_properties(
             self.material,
             self.area_light,

shared/src/lib.rs

@@ -20,5 +20,6 @@ pub mod wavefront;
 
 pub use core::pbrt::*;
 pub use utils::alloc::{gbox, gvec, gvec_from_slice, gvec_with_capacity, leak, GBox, GVec};
-pub use utils::{Array2D, PBRTOptions, Ptr, Transform, SOA};
-pub use wavefront::{WavefrontAggregate, WorkQueue};
+pub use utils::{Array2D, PBRTOptions, Ptr, Transform};
+pub use utils::soa::WorkQueue;
+pub use wavefront::{WavefrontAggregate};

shared/src/materials/coated.rs

@@ -334,7 +334,7 @@ impl MaterialTrait for CoatedConductorMaterial {
     }
 
     fn get_normal_map(&self) -> Option<&Image> {
-        Some(&*self.normal_map)
+        self.normal_map.get()
     }
 
     fn get_displacement(&self) -> Ptr<GPUFloatTexture> {

shared/src/utils/math.rs

@@ -544,7 +544,7 @@ pub fn next_float_up(v: Float) -> Float {
     if v >= 0.0 {
         ui = ui.wrapping_add(1);
     } else {
-        ui = ui.wrapping_sub(1);
+        ui.wrapping_sub(1);
     }
     bits_to_float(ui)
 }

shared/src/utils/mod.rs

@@ -1,4 +1,5 @@
 pub mod alloc;
+pub mod atomic;
 pub mod complex;
 pub mod containers;
 pub mod hash;
@@ -10,18 +11,18 @@ pub mod ptr;
 pub mod quaternion;
 pub mod rng;
 pub mod sampling;
+pub mod soa;
 pub mod sobol;
 pub mod splines;
 pub mod transform;
 
+pub use atomic::{AtomicFloat, AtomicU32};
+pub use containers::Array2D;
 pub use options::PBRTOptions;
 pub use ptr::Ptr;
 pub use transform::{AnimatedTransform, Transform, TransformGeneric};
-pub use containers::Array2D;
-pub use soa::SOA:
 
 use crate::Float;
-use core::sync::atomic::{AtomicU32, Ordering};
 
 #[inline]
 pub fn find_interval<F>(sz: u32, pred: F) -> u32
@@ -62,61 +63,6 @@ where
     i
 }
 
-#[repr(C)]
-#[derive(Debug)]
-pub struct AtomicFloat {
-    bits: AtomicU32,
-}
-
-impl Default for AtomicFloat {
-    fn default() -> Self {
-        Self::new(0.)
-    }
-}
-
-impl Clone for AtomicFloat {
-    fn clone(&self) -> Self {
-        Self::new(self.get())
-    }
-}
-
-impl AtomicFloat {
-    pub fn new(val: f32) -> Self {
-        Self {
-            bits: AtomicU32::new(val.to_bits()),
-        }
-    }
-
-    pub fn get(&self) -> Float {
-        Float::from_bits(self.bits.load(Ordering::Relaxed))
-    }
-
-    pub fn set(&self, val: Float) {
-        self.bits.store(val.to_bits(), Ordering::Relaxed);
-    }
-
-    /// Atomically adds `val` to the current value.
-    /// Compare-And-Swap loop.
-    pub fn add(&self, val: f32) {
-        let mut current_bits = self.bits.load(Ordering::Relaxed);
-        loop {
-            let current_val = f32::from_bits(current_bits);
-            let new_val = current_val + val;
-            let new_bits = new_val.to_bits();
-
-            match self.bits.compare_exchange_weak(
-                current_bits,
-                new_bits,
-                Ordering::Relaxed,
-                Ordering::Relaxed,
-            ) {
-                Ok(_) => break,
-                Err(x) => current_bits = x,
-            }
-        }
-    }
-}
-
 #[inline(always)]
 pub fn gpu_array_from_fn<T, const N: usize>(mut f: impl FnMut(usize) -> T) -> [T; N] {
     unsafe {
@@ -130,4 +76,3 @@ pub fn gpu_array_from_fn<T, const N: usize>(mut f: impl FnMut(usize) -> T) -> [T
         arr.assume_init()
     }
 }
-

shared/src/utils/soa.rs

@@ -1,15 +1,120 @@
-pub trait SoA: Clone {
+use crate::utils::AtomicU32;
+use crate::{Float, Ptr};
+
+#[repr(C)]
+#[derive(Debug, Clone, Copy)]
+pub struct SoABuffer<T: Copy> {
+    pub ptr: Ptr<T>,
+    pub capacity: u32,
+}
+
+impl<T: Copy> SoABuffer<T> {
+    pub fn null() -> Self {
+        Self {
+            ptr: Ptr::null(),
+            capacity: 0,
+        }
+    }
+
+    #[inline(always)]
+    pub fn get(&self, i: usize) -> T {
+        debug_assert!(i < self.capacity as usize);
+        unsafe { *self.ptr.as_raw().add(i) }
+    }
+
+    #[inline(always)]
+    pub fn set(&self, i: usize, val: T) {
+        debug_assert!(i < self.capacity as usize);
+        unsafe {
+            let p = self.ptr.as_raw() as *mut T;
+            *p.add(i) = val;
+        }
+    }
+}
+
+pub trait SoA {
     type Item: Copy;
-    fn with_capacity(n: usize) -> Self;
-    fn len(&self) -> usize;
-    fn is_empty(&self) -> bool {
-        self.len() == 0
+
+    fn allocate(n: u32, alloc: &dyn SoAAllocator) -> Self;
+
+    /// Read one element by scattering across all buffers.
+    unsafe fn get(&self, i: usize) -> Self::Item;
+
+    /// Write one element by scattering across all buffers.
+    /// Takes &self for GPU concurrent write (disjoint indices).
+    unsafe fn set(&self, i: usize, v: Self::Item);
 }
 
-    unsafe fn get_unchecked(&self, i: usize) -> Self::Item;
-    unsafe fn set_unchecked(&mut self, i: usize, v: Self::Item);
+pub trait SoAAllocator {
+    fn alloc_raw(&self, layout: core::alloc::Layout) -> *mut u8;
 }
 
-pub trait SoAElement: Copy {
-    type SoA: SoA<Item = Self>;
+pub fn alloc_soa_buffer<T: Copy>(n: u32, alloc: &dyn SoAAllocator) -> SoABuffer<T> {
+    if n == 0 {
+        return SoABuffer::null();
+    }
+    let layout = core::alloc::Layout::array::<T>(n as usize).unwrap();
+    let raw = alloc.alloc_raw(layout);
+    SoABuffer {
+        ptr: Ptr::from_raw(raw as *mut T),
+        capacity: n,
+    }
+}
+
+#[repr(C)]
+pub struct WorkQueue<S: SoA> {
+    pub storage: S,
+    pub count: AtomicU32,
+    pub capacity: u32,
+}
+
+impl<S: SoA> WorkQueue<S> {
+    pub fn new(storage: S, capacity: u32) -> Self {
+        Self {
+            storage,
+            count: AtomicU32::new(0),
+            capacity,
+        }
+    }
+
+    /// Number of items currently in the queue.
+    #[inline(always)]
+    pub fn size(&self) -> u32 {
+        self.count.load()
+    }
+
+    /// Reset the queue to empty. Call from host between kernel launches.
+    #[inline(always)]
+    pub fn reset(&self) {
+        self.count.store(0);
+    }
+
+    /// Push an item, returning its index. Returns None if the queue is full.
+    /// The atomic increment guarantees each thread gets a unique slot.
+    #[inline(always)]
+    pub fn push(&self, item: S::Item) -> Option<u32> {
+        let slot = self.count.fetch_add(1);
+        if slot >= self.capacity {
+            // Queue overflow — this shouldn't happen if capacity is
+            // sized correctly. In debug builds we want to know about it.
+            debug_assert!(false, "WorkQueue overflow: {} >= {}", slot, self.capacity);
+            return None;
+        }
+        unsafe {
+            self.storage.set(slot as usize, item);
+        }
+        Some(slot)
+    }
+
+    /// Read an item at the given index. Used by consumer kernels.
+    #[inline(always)]
+    pub unsafe fn get(&self, i: usize) -> S::Item {
+        debug_assert!(
+            (i as u32) < self.size(),
+            "WorkQueue::get out of bounds: {} >= {}",
+            i,
+            self.size()
+        );
+        self.storage.get(i)
+    }
 }

src/lib.rs

@@ -14,4 +14,4 @@ pub mod utils;
 pub mod wavefront;
 
 pub use utils::{Arena, FileLoc, ParameterDictionary, Upload, ArenaUpload};
-pub const MAX_TAGS = 16;
+pub const MAX_TAGS: u32 = 16;

src/utils/arena.rs

@@ -1,6 +1,7 @@
 use crate::utils::backend::GpuAllocator;
 use crate::utils::mipmap::MIPMap;
 use parking_lot::Mutex;
+use shared::utils::soa::SoAAllocator;
 use shared::Ptr;
 use std::alloc::Layout;
 use std::collections::HashMap;
@@ -161,6 +162,11 @@ impl<A: GpuAllocator> Arena<A> {
         (Ptr::from_raw(ptr), len)
     }
 
+    pub fn alloc_layout(&self, layout: Layout) -> *mut u8 {
+        let mut bump = self.bump.lock();
+        bump.alloc_layout(layout)
+    }
+
     pub fn get_texture_object(&self, mipmap: &Arc<MIPMap>) -> u64 {
         let key = Arc::as_ptr(mipmap) as usize;
         let mut cache = self.texture_cache.lock();
@@ -178,3 +184,9 @@ impl<A: GpuAllocator + Default> Default for Arena<A> {
         Self::new(A::default())
     }
 }
+
+impl<A: GpuAllocator> SoAAllocator for Arena<A> {
+    fn alloc_raw(&self, layout: core::alloc::Layout) -> *mut u8 {
+        self.alloc_layout(layout)
+    }
+}

src/wavefront/aggregate.rs

@@ -1,11 +0,0 @@
-use shared::core::geometry::Bounds3f;
-use super::{RayQueue, EscapedRayQueue, HitAreaLightQueue, MaterialEvalQueue, MediumSampleQueue, SubsurfaceScatterQueue};
-
-#[derive(Clone, Debug)]
-pub trait WavefrontAggregate {
-    fn bounds(&self) -> Bounds3f;
-    fn intersect_closest(max_rays: usize, ray_q: &mut RayQueue, hit_area_light_q: &mut HitAreaLightQueue, basic_mlt_q: &mut MaterialEvalQueue, universal_mtl_q: &mut MaterialEvalQueue, medium_sample_q: &mut MediumSampleQueue);
-    fn intersect_shadow(max_rays: usize, shadow_ray_q: &mut ShadowRayQueue, pixel_sample_state: &mut SOA<PixelSampleState>);
-    fn intersect_shadow_tr(max_rays: usize, shadow_ray_q: &mut ShadowRayQueue, pixel_sample_state: &mut SOA<PixelSampleState>);
-    fn intersect_one_random(max_rays: usize, subsurface_scatte_q: &mut SubsurfaceScatterQueue);
-}

src/wavefront/integrator.rs

@@ -1,38 +1,38 @@
-use crate::MAX_TAGS;
-use shared::{Ptr, GVec};
-use shared::core::film::Film;
-use shared::core::color::RGB;
-use shared::core::filter::Filter;
-use shared::core::light::Light;
-use shared::core::sampler::Filter;
-use shared::wavefront::{WavefrontAggregate, RayQueue, MediumSampleQueue, EscapedRayQueue, HitAreaLightQueue, MaterialEvalQueue, ShadowRayQueue, GetBSSRDFAndProbeRayQueue, SubsurfaceScatterQueue};
-
-pub struct WavefrontPathIntegrator {
-    pub init_visible_surface: bool,
-    pub have_subsurface: bool,
-    pub have_media: bool,
-    pub have_basic_eval_material: [bool; MAX_TAGS + 1],
-    pub have_universal_eval_material: [bool; MAX_TAGS + 1],
-    pub filter: Filter,
-    pub film: Film, 
-    pub sampler: Sampler,
-    pub camera: Camera,
-    pub infinite_lights: GVec<Light>,
-    pub max_depth: usize,
-    pub sampler_per_pixel: usize,
-    pub regularize: bool,
-    pub scanlines_per_pixel: usize,
-    pub max_queue_size: usize,
-    pub medium_sample_queue: Ptr<MediumSampleQueue>,
-    pub medium_scatter_queue: Ptr<MediumScatterQueue>,
-    pub escaped_ray_queue: Ptr<EscapedRayQueue>,
-    pub hit_area_light_queue: Ptr<HitAreaLightQueue>,
-    pub basic_eval_material_queue: Ptr<MaterialEvalQueue>,
-    pub universal_eval_material_queue: Ptr<MaterialEvalQueue>,
-    pub shadow_ray_queue: Ptr<ShadowRayQueue>,
-    pub bssrdf_eval_queue: PTr<GetBSSRDFAndProbeRayQueue>,
-    pub subsurface_scatter_queue: Ptr<SubsurfaceScatterQueue>,
-    pub display_rgb: Ptr<RGB>,
-    pub display_rgb_host: Ptr<RGB,
-
-}
+// use crate::MAX_TAGS;
+// use shared::{Ptr, GVec};
+// use shared::core::film::Film;
+// use shared::core::color::RGB;
+// use shared::core::filter::Filter;
+// use shared::core::light::Light;
+// use shared::core::sampler::Sampler;
+// use shared::wavefront::{WavefrontAggregate, RayQueue, MediumSampleQueue, EscapedRayQueue, HitAreaLightQueue, MaterialEvalQueue, ShadowRayQueue, GetBSSRDFAndProbeRayQueue, SubsurfaceScatterQueue};
+//
+// pub struct WavefrontPathIntegrator {
+//     pub init_visible_surface: bool,
+//     pub have_subsurface: bool,
+//     pub have_media: bool,
+//     pub have_basic_eval_material: [bool; MAX_TAGS + 1],
+//     pub have_universal_eval_material: [bool; MAX_TAGS + 1],
+//     pub filter: Filter,
+//     pub film: Film, 
+//     pub sampler: Sampler,
+//     pub camera: Camera,
+//     pub infinite_lights: GVec<Light>,
+//     pub max_depth: usize,
+//     pub sampler_per_pixel: usize,
+//     pub regularize: bool,
+//     pub scanlines_per_pixel: usize,
+//     pub max_queue_size: usize,
+//     pub medium_sample_queue: Ptr<MediumSampleQueue>,
+//     pub medium_scatter_queue: Ptr<MediumScatterQueue>,
+//     pub escaped_ray_queue: Ptr<EscapedRayQueue>,
+//     pub hit_area_light_queue: Ptr<HitAreaLightQueue>,
+//     pub basic_eval_material_queue: Ptr<MaterialEvalQueue>,
+//     pub universal_eval_material_queue: Ptr<MaterialEvalQueue>,
+//     pub shadow_ray_queue: Ptr<ShadowRayQueue>,
+//     pub bssrdf_eval_queue: PTr<GetBSSRDFAndProbeRayQueue>,
+//     pub subsurface_scatter_queue: Ptr<SubsurfaceScatterQueue>,
+//     pub display_rgb: Ptr<RGB>,
+//     pub display_rgb_host: Ptr<RGB>,
+//
+// }

src/wavefront/mod.rs

@@ -1,4 +1 @@
 pub mod integrator;
-pub mod aggregate;
-
-pub use aggregate::WavefrontAggregate;