Rename to Hikari.jl, refactor for Raycore.jl and more performance improvements

Project.toml

@@ -4,18 +4,35 @@ authors = ["Anton Smirnov <tonysmn97@gmail.com>"]
 version = "0.1.0"
 
 [deps]
+Atomix = "a9b6321e-bd34-4604-b9c9-b65b8de01458"
 FileIO = "5789e2e9-d7fb-5bc7-8068-2c6fae9b9549"
 GeometryBasics = "5c1252a2-5f33-56bf-86c9-59e7332b4326"
 ImageCore = "a09fc81d-aa75-5fe9-8630-4744c3626534"
 ImageIO = "82e4d734-157c-48bb-816b-45c225c6df19"
+KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 ProgressMeter = "92933f4c-e287-5a05-a399-4b506db050ca"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+RandomNumbers = "e6cf234a-135c-5ec9-84dd-332b85af5143"
+RayCaster = "afc56b53-c9a9-482a-a956-d1d800e05559"
 StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+StructArrays = "09ab397b-f2b6-538f-b94a-2f83cf4a842a"
 
 [compat]
+#Atomix = "0.1.0"
 FileIO = "1.16"
-GeometryBasics = "0.4"
+GeometryBasics = "0.4, 0.5"
 ImageCore = "0.10"
 ImageIO = "0.6"
+#KernelAbstractions = "0.9.24"
 ProgressMeter = "1.10"
-StaticArrays = "1.9"
+RandomNumbers = "1.6.0"
+StaticArrays = "1.9.7"
+StructArrays = "0.6.18"
+
+[extras]
+JET = "c3a54625-cd67-489e-a8e7-0a5a0ff4e31b"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+
+[targets]
+test = ["Test", "JET"]

docs/Project.toml

@@ -1,10 +1,15 @@
 [deps]
+BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
 DocumenterVitepress = "4710194d-e776-4893-9690-8d956a29c365"
 FileIO = "5789e2e9-d7fb-5bc7-8068-2c6fae9b9549"
 GeometryBasics = "5c1252a2-5f33-56bf-86c9-59e7332b4326"
 ImageCore = "a09fc81d-aa75-5fe9-8630-4744c3626534"
+ImageShow = "4e3cecfd-b093-5904-9786-8bbb286a6a31"
+KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
+Makie = "ee78f7c6-11fb-53f2-987a-cfe4a2b5a57a"
+Metal = "dde4c033-4e86-420c-a63e-0dd931031962"
 Trace = "afc56b53-c9a9-482a-a956-d1d800e05558"
 
 [compat]

docs/code/basic-scene.jl

@@ -0,0 +1,211 @@
+using GeometryBasics
+using Trace
+using FileIO
+using ImageCore
+# using BenchmarkTools
+# using FileIO, ImageShow
+
+function tmesh(prim, material)
+    prim =  prim isa Sphere ? Tesselation(prim, 64) : prim
+    mesh = normal_mesh(prim)
+    m = Trace.create_triangle_mesh(mesh, Trace.ShapeCore())
+    return Trace.GeometricPrimitive(m, material)
+end
+
+LowSphere(radius, contact=Point3f(0)) = Sphere(contact .+ Point3f(0, 0, radius), radius)
+
+begin
+
+    material_red = Trace.MatteMaterial(
+        Trace.ConstantTexture(Trace.RGBSpectrum(0.796f0, 0.235f0, 0.2f0)),
+        Trace.ConstantTexture(0.0f0),
+    )
+    material_blue = Trace.MatteMaterial(
+        Trace.ConstantTexture(Trace.RGBSpectrum(0.251f0, 0.388f0, 0.847f0)),
+        Trace.ConstantTexture(0.0f0),
+    )
+    material_white = Trace.MatteMaterial(
+        Trace.ConstantTexture(Trace.RGBSpectrum(1.0f0)),
+        Trace.ConstantTexture(0.0f0),
+    )
+    mirror = Trace.MirrorMaterial(Trace.ConstantTexture(Trace.RGBSpectrum(1.0f0)))
+    glass = Trace.GlassMaterial(
+        Trace.ConstantTexture(Trace.RGBSpectrum(1.0f0)),
+        Trace.ConstantTexture(Trace.RGBSpectrum(1.0f0)),
+        Trace.ConstantTexture(0.0f0),
+        Trace.ConstantTexture(0.0f0),
+        Trace.ConstantTexture(1.5f0),
+        true,
+    )
+
+    s1 = tmesh(LowSphere(0.5f0), material_white)
+    s2 = tmesh(LowSphere(0.3f0, Point3f(0.5, 0.5, 0)), material_white)
+    s3 = tmesh(LowSphere(0.3f0, Point3f(-0.5, 0.5, 0)), material_white)
+    s4 = tmesh(LowSphere(0.4f0, Point3f(0, 1.0, 0)), material_white)
+
+    ground = tmesh(Rect3f(Vec3f(-5, -5, 0), Vec3f(10, 10, 0.01)), material_white)
+    back = tmesh(Rect3f(Vec3f(-5, -3, 0), Vec3f(10, 0.01, 10)), material_white)
+    l = tmesh(Rect3f(Vec3f(-2, -5, 0), Vec3f(0.01, 10, 10)), material_white)
+    r = tmesh(Rect3f(Vec3f(2, -5, 0), Vec3f(0.01, 10, 10)), material_white)
+
+    bvh = Trace.no_material_bvh([s1, s2, s3, s4, ground, back, l, r]);
+
+    lights = (
+        # Trace.PointLight(Vec3f(0, -1, 2), Trace.RGBSpectrum(22.0f0)),
+        Trace.PointLight(Vec3f(0, 0, 2), Trace.RGBSpectrum(10.0f0)),
+        Trace.PointLight(Vec3f(0, 3, 3), Trace.RGBSpectrum(25.0f0)),
+    )
+    scene = Trace.Scene([lights...], bvh);
+    resolution = Point2f(1024)
+    f = Trace.LanczosSincFilter(Point2f(1.0f0), 3.0f0)
+
+    film = Trace.Film(resolution,
+        Trace.Bounds2(Point2f(0.0f0), Point2f(1.0f0)),
+        f, 1.0f0, 1.0f0,
+    )
+    screen_window = Trace.Bounds2(Point2f(-1), Point2f(1))
+    cam = Trace.PerspectiveCamera(
+        Trace.look_at(Point3f(0, 4, 2), Point3f(0, -4, -1), Vec3f(0, 0, 1)),
+        screen_window, 0.0f0, 1.0f0, 0.0f0, 1.0f6, 45.0f0, film,
+    )
+end
+
+using RayCaster
+using PProf, Profile
+using AMDGPU
+
+# begin
+#     Trace.clear!(film)
+#     integrator = Trace.WhittedIntegrator(cam, Trace.UniformSampler(8), 5)
+#     @time integrator(scene, film, cam)
+# end
+pres = []
+g_scene = Trace.to_gpu(ROCArray, scene; preserve=pres);
+g_film = Trace.to_gpu(ROCArray, film; preserve=pres);
+integrator = Trace.WhittedIntegrator(cam, Trace.UniformSampler(8), 5)
+AMDGPU.@device_code_warntype interactive = true integrator(g_scene, g_film, cam);
+@time integrator(scene, film, cam);
+@time Trace.integrator_threaded(integrator, scene, film, cam);
+using OpenCL
+
+cl_scene = Trace.to_gpu(CLArray, scene)
+cl_film = Trace.to_gpu(CLArray, film);
+
+@time integrator(cl_scene, cl_film, cam);
+OpenCL.@device_code_warntype interactive = true integrator(cl_scene, cl_film, cam)
+# Setup for profiling sample_kernel_inner!
+
+function profile_kernel(film, scene, cam)
+    tiles = film.tiles
+    tile_size = film.tile_size
+    filter_radius = film.filter_radius
+    filter_table = film.filter_table
+    pixels = film.pixels
+    tile_bounds = Trace.Bounds2(Point2f(1, 1), Point2f(16, 16))
+    tile_column = Int32(1)
+    max_depth = Int32(5)
+    sampler = Trace.UniformSampler(8)
+    camera = cam
+    spp_sqr = 1.0f0 / √Float32(sampler.samples_per_pixel)
+    @sync for i in 1:1024
+        for j in 1:1024
+            pixel = Point2f(i, j)
+            Threads.@spawn Trace.sample_kernel_inner!(
+                tiles, tile_bounds, tile_column, Point2f(size(pixels)),
+                max_depth, scene, sampler, camera,
+                pixel, spp_sqr, filter_table, filter_radius
+            )
+        end
+    end
+    return
+end
+@time Trace.integrator_threaded(integrator, scene,film, cam)
+
+begin
+    tiles = film.tiles
+    tile_size = film.tile_size
+    filter_radius = film.filter_radius
+    filter_table = film.filter_table
+    pixels = film.pixels
+    tile_bounds = Trace.Bounds2(Point2f(1, 1), Point2f(16, 16))
+    tile_column = Int32(1)
+    max_depth = Int32(5)
+    sampler = Trace.UniformSampler(8)
+    camera = cam
+    pixel = Point2f(8, 8)
+    spp_sqr = 1.0f0 / √Float32(sampler.samples_per_pixel)
+    Trace.sample_kernel_inner!(
+        tiles, tile_bounds, tile_column, Point2f(size(pixels)),
+        max_depth, scene, sampler, camera,
+        pixel, spp_sqr, filter_table, filter_radius
+    )
+end
+
+
+
+begin
+    Profile.Allocs.clear()
+    Profile.Allocs.@profview_allocs integrator(scene, film, cam)
+    PProf.Allocs.pprof()
+end
+
+@btime Trace.NoMaterial()
+
+bvh = scene.aggregate.bvh
+ray = RayCaster.Ray(o=Point3f(0, 4, 2), d=Point3f(0, -4, -1))
+args = Trace.intersect!(scene, ray)
+camera_sample = @inline get_camera_sample(sampler, campix)
+ray, ω = Trace.generate_ray_differential(cam, camera_sample)
+ray = scale_differentials(ray, spp_sqr)
+l = RGBSpectrum(0.0f0)
+
+
+function testo(sampler, max_depth, ray, scene)
+    l = Trace.RGBSpectrum(0.0f0)
+    for depth in 1:1000
+        # For demonstration, we just accumulate some dummy value
+        l += Trace.li_iterative(sampler, max_depth, ray, scene)
+    end
+    return l
+end
+
+@allocated testo(sampler, max_depth, ray, scene)
+
+@btime Trace.li_iterative(sampler, max_depth, ray, scene)
+t1 = scene.aggregate.bvh.primitives[1]
+si = Trace.triangle_to_surface_interaction(t1, ray, Point3f(0))
+m = scene.aggregate.materials[1]
+
+@edit m(si, false, Trace.UInt8(1))
+
+bsdf = Trace.matte_material(m, si, false, Trace.UInt8(1))
+m.type == Trace.MATTE_MATERIAL
+Trace.BSDF()
+
+
+@code_warntype Trace.specular(Trace.Transmit, bsdf, sampler, ray, si)
+using JET
+
+# begin
+#     resolution = Point2f(1024)
+#     Trace.clear!(film)
+#     @time render_scene(scene, film, cam)
+#     Trace.to_framebuffer!(film, 1.0f0)
+#     film.framebuffer
+# end
+
+
+# 6.296157 seconds (17.64 k allocations: 19.796 MiB, 0.13% gc time, 45 lock conflicts)
+# After more GPU optimizations
+# 4.169616 seconds (17.37 k allocations: 19.777 MiB, 0.14% gc time, 20 lock conflicts)
+# After first shading running on GPU
+# 3.835527 seconds (17.36 k allocations: 19.779 MiB, 0.16% gc time, 41 lock conflicts)
+# 4.191 s (4710 allocations: 18.36 MiB)
+# iterative_li: 5.2s -.-
+
+
+# begin
+#     integrator = Trace.SPPMIntegrator(cam, 0.075f0, 5, 1)
+#     integrator(scene)
+#     img = reverse(film.framebuffer, dims=1)
+# end

docs/code/caustic_glass.jl

@@ -1,8 +1,9 @@
-using GeometryBasics
-using LinearAlgebra
-using Trace
+using GeometryBasics, ImageShow
+using LinearAlgebra, Makie
+using Trace, FileIO, MeshIO
 
-function render()
+model = load(joinpath(@__DIR__, "..", "src", "assets", "models", "caustic-glass.ply"))
+begin
     glass = Trace.GlassMaterial(
         Trace.ConstantTexture(Trace.RGBSpectrum(1f0)),
         Trace.ConstantTexture(Trace.RGBSpectrum(1f0)),
@@ -17,82 +18,21 @@ function render()
         Trace.ConstantTexture(0.010408001f0),
         true,
     )
-
-    model = "./scenes/models/caustic-glass.ply"
-    triangle_meshes, triangles = Trace.load_triangle_mesh(
-        model, Trace.ShapeCore(Trace.translate(Vec3f(5, -1.49, -100)), false),
-    )
-    floor_triangles = Trace.create_triangle_mesh(
-        Trace.ShapeCore(Trace.translate(Vec3f(-10, 0, -87)), false),
-        2, UInt32[1, 2, 3, 1, 4, 3],
-        4,
-        [
-            Point3f(0, 0, 0), Point3f(0, 0, -30),
-            Point3f(30, 0, -30), Point3f(30, 0, 0),
-        ],
-        [
-            Trace.Normal3f(0, 1, 0), Trace.Normal3f(0, 1, 0),
-            Trace.Normal3f(0, 1, 0), Trace.Normal3f(0, 1, 0),
-        ],
-    )
-
-    primitives = Vector{Trace.GeometricPrimitive}(undef, 0)
-    for t in triangles
-        push!(primitives, Trace.GeometricPrimitive(t, glass))
-    end
-    for t in floor_triangles
-        push!(primitives, Trace.GeometricPrimitive(t, plastic))
+    scene = Scene(size=(1024, 1024); lights=[
+        AmbientLight(RGBf(1, 1, 1)),
+        PointLight(Vec3f(4, 4, 10), RGBf(150, 150, 150)),
+        PointLight(Vec3f(-3, 10, 2.5), RGBf(60, 60, 60)),
+        PointLight(Vec3f(0, 3, 0.5), RGBf(40, 40, 40))
+    ])
+    cam3d!(scene)
+    cm = scene.camera_controls
+    mesh!(scene, model, material=glass)
+    mini, maxi = extrema(Rect3f(decompose(Point, model)))
+    floorrect = Rect3f(Vec3f(-10, mini[2], -10), Vec3f(20, -1, 20))
+    mesh!(scene, floorrect, material=plastic)
+    center!(scene)
+    update_cam!(scene, Vec3f(-1.6, 6.2, 0.2), Vec3f(-3.6, 2.5, 2.4), Vec3f(0, 1, 0))
+    render_scene(scene) do camera
+        Trace.SPPMIntegrator(camera, 0.075f0, 5, 100)
     end
-
-    bvh = Trace.BVHAccel(primitives, 1)
-
-    from, to = Point3f(0, 2, 0), Point3f(-5, 0, 5)
-    cone_angle, cone_δ_angle = 30f0, 10f0
-    dir = normalize(Vec3f(to - from))
-    dir, du, dv = Trace.coordinate_system(dir)
-
-    dir_to_z = Trace.Transformation(transpose(Mat4f(
-        du[1], du[2], du[3], 0f0,
-        dv[1], dv[2], dv[3], 0f0,
-        dir[1], dir[2], dir[3], 0f0,
-        0f0, 0f0, 0f0, 1f0,
-    )))
-    light_to_world = (
-        Trace.translate(Vec3f(4.5, 0, -101))
-        * Trace.translate(Vec3f(from))
-        * inv(dir_to_z)
-    )
-
-    lights = [
-        Trace.SpotLight(
-            light_to_world, Trace.RGBSpectrum(60f0),
-            cone_angle, cone_angle - cone_δ_angle,
-        ),
-    ]
-
-    scene = Trace.Scene(lights, bvh)
-
-    resolution = Point2f(256)
-    n_samples = 8
-    ray_depth = 5
-
-    look_point = Point3f(-3, 0, -91)
-    screen = Trace.Bounds2(Point2f(-1f0), Point2f(1f0))
-    filter = Trace.LanczosSincFilter(Point2f(1f0), 3f0)
-
-    ir = Int64.(resolution)
-    film = Trace.Film(
-        resolution, Trace.Bounds2(Point2f(0), Point2f(1)),
-        filter, 1f0, 1f0,
-        "./scenes/caustics-sppm-$(ir[1])x$(ir[2]).png",
-    )
-    camera = Trace.PerspectiveCamera(
-        Trace.look_at(Point3f(0, 150, 150), look_point, Vec3f(0, 1, 0)),
-        screen, 0f0, 1f0, 0f0, 1f6, 90f0, film,
-    )
-
-    integrator = Trace.SPPMIntegrator(camera, 0.075f0, ray_depth, 100, -1)
-    integrator(scene)
 end
-
-render()