switch to using AK TaskPartitioner. Added logo

StellaOrg · Nov 3, 2024 · 23337b3 · 23337b3
1 parent 4f33412
commit 23337b3
Show file tree

Hide file tree

Showing 11 changed files with 121 additions and 211 deletions.
diff --git a/Project.toml b/Project.toml
@@ -17,7 +17,7 @@ AcceleratedKernels = "0.1"
 ArgCheck = "2.3.0"
 Atomix = "0.1"
 DocStringExtensions = "0.9"
-GPUArraysCore = "0.1.6"
+GPUArraysCore = "0.1.6, 0.2"
 KernelAbstractions = "0.9"
 LinearAlgebra = "1.6"
 julia = "1.6"
diff --git a/README.md b/README.md
@@ -1,13 +1,16 @@
-[![ImplicitBVH](https://github.com/StellaOrg/ImplicitBVH.jl/blob/main/docs/src/static/bunny.png?raw=true)](https://stellaorg.github.io/ImplicitBVH.jl/)
+[![ImplicitBVH](https://github.com/StellaOrg/ImplicitBVH.jl/blob/main/docs/src/assets/logo.png?raw=true)](https://stellaorg.github.io/ImplicitBVH.jl/)
+[![ImplicitBVH](https://github.com/StellaOrg/ImplicitBVH.jl/blob/main/docs/src/assets/bunny.png?raw=true)](https://stellaorg.github.io/ImplicitBVH.jl/)
 
 
-[![](https://img.shields.io/badge/docs-stable-blue.svg)](https://stellaorg.github.io/ImplicitBVH.jl/stable)
-[![](https://img.shields.io/badge/docs-dev-blue.svg)](https://stellaorg.github.io/ImplicitBVH.jl/dev)
+[![Docs Stable](https://img.shields.io/badge/docs-stable-blue.svg)](https://stellaorg.github.io/ImplicitBVH.jl/stable)
+[![Docs Dev](https://img.shields.io/badge/docs-dev-blue.svg)](https://stellaorg.github.io/ImplicitBVH.jl/dev)
+[![Build Status](https://github.com/StellaOrg/ImplicitBVH.jl/workflows/CI/badge.svg)](https://github.com/StellaOrg/ImplicitBVH.jl/actions?query=workflow%3ACI+branch%3Amaster)
+
 
 # ImplicitBVH.jl
-*High-Performance Cross-Architecture Bounding Volume Hierarchy for Collision Detection*
+*High-Performance Cross-Architecture Bounding Volume Hierarchy for Collision Detection and Ray Tracing*
 
-**New in v0.5.0: GPU acceleration via [AcceleratedKernels.jl](https://github.com/anicusan/AcceleratedKernels.jl)/[KernelAbstractions.jl](https://github.com/JuliaGPU/KernelAbstractions.jl) targeting all JuliaGPU backends, i.e. Nvidia CUDA, AMD ROCm, Intel oneAPI, Apple Metal.**
+**New in v0.5.0: Ray Tracing and GPU acceleration via [AcceleratedKernels.jl](https://github.com/anicusan/AcceleratedKernels.jl)/[KernelAbstractions.jl](https://github.com/JuliaGPU/KernelAbstractions.jl) targeting all JuliaGPU backends, i.e. Nvidia CUDA, AMD ROCm, Intel oneAPI, Apple Metal.**
 
 It uses an implicit bounding volume hierarchy constructed from an iterable of some geometric
 primitives' (e.g. triangles in a mesh) bounding volumes forming the `ImplicitTree` leaves. The leaves

diff --git a/docs/make.jl b/docs/make.jl
@@ -6,7 +6,8 @@ makedocs(
     sitename = "ImplicitBVH.jl",
     format = Documenter.HTML(
         # Only create web pretty-URLs on the CI
-        prettyurls = get(ENV, "CI", nothing) == "true",
+        prettyurls=get(ENV, "CI", nothing) == "true",
+        sidebar_sitename=false,
     ),
 )
 deploydocs(repo = "github.com/StellaOrg/ImplicitBVH.jl.git")
diff --git a/docs/src/static/bunny.png → docs/src/assets/bunny.png b/docs/src/static/bunny.png → docs/src/assets/bunny.png
diff --git a/docs/src/assets/logo.png b/docs/src/assets/logo.png
diff --git a/docs/src/utilities.md b/docs/src/utilities.md
@@ -3,5 +3,4 @@
 ```@docs
 ImplicitBVH.BVHOptions
 ImplicitBVH.get_index_type
-ImplicitBVH.TaskPartitioner
 ```
diff --git a/prototype/Project.toml b/prototype/Project.toml
@@ -1,6 +1,7 @@
 [deps]
 AcceleratedKernels = "6a4ca0a5-0e36-4168-a932-d9be78d558f1"
 Atomix = "a9b6321e-bd34-4604-b9c9-b65b8de01458"
+AtomixMetal = "fc0d88fc-5db5-40d7-bf35-da84fd759def"
 Cthulhu = "f68482b8-f384-11e8-15f7-abe071a5a75f"
 ImplicitBVH = "932a18dc-bb55-4cd5-bdd6-1368ec9cea29"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"

diff --git a/src/traverse/traverse.jl b/src/traverse/traverse.jl
@@ -1,5 +1,5 @@
 """
-    $(TYPEDEF)
+    const IndexPair{I} = Tuple{I, I}
 
 Alias for a tuple of two indices representing e.g. a contacting pair.
 """

diff --git a/src/traverse/traverse_pair_cpu.jl b/src/traverse/traverse_pair_cpu.jl
@@ -10,28 +10,27 @@ function traverse_nodes_pair!(bvh1, bvh2, src, dst, num_src, ::Nothing, level1,
 
     # Split computation into contiguous ranges of minimum 100 elements each; if only single thread
     # is needed, inline call
-    tp = TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
+    tp = AK.TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
     if tp.num_tasks == 1
         num_dst = traverse_nodes_pair_range!(
             bvh1, bvh2,
             src, dst, nothing,
             virtual_nodes_before1,
             virtual_nodes_before2,
-            (1, num_src),
+            1:num_src,
         )
     else
         # Keep track of tasks launched and number of elements written by each task in their unique
         # memory region. The unique region is equal to 4 dst elements per src element
         tasks = Vector{Task}(undef, tp.num_tasks)
         num_written = Vector{Int}(undef, tp.num_tasks)
         @inbounds for i in 1:tp.num_tasks
-            istart, iend = tp[i]
             tasks[i] = Threads.@spawn traverse_nodes_pair_range!(
                 bvh1, bvh2,
                 src, view(dst, 4istart - 3:4iend), view(num_written, i),
                 virtual_nodes_before1,
                 virtual_nodes_before2,
-                (istart, iend),
+                tp[i],
             )
         end
 
@@ -43,7 +42,7 @@ function traverse_nodes_pair!(bvh1, bvh2, src, dst, num_src, ::Nothing, level1,
 
             # Repack written contacts by the second, third thread, etc.
             if i > 1
-                istart, iend = tp[i]
+                istart = tp[i].start
                 for j in 1:task_num_written
                     dst[num_dst + j] = dst[4istart - 3 + j - 1]
                 end
@@ -63,7 +62,7 @@ function traverse_nodes_pair_range!(
     num_dst = 0
 
     # For each BVTT pair of nodes, check for contact
-    @inbounds for i in irange[1]:irange[2]
+    @inbounds for i in irange
         # Extract implicit indices of BVH nodes to test
         implicit1, implicit2 = src[i]
 
@@ -130,28 +129,27 @@ function traverse_nodes_left!(bvh1, bvh2, src, dst, num_src, ::Nothing, level1,
 
     # Split computation into contiguous ranges of minimum 100 elements each; if only single thread
     # is needed, inline call
-    tp = TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
+    tp = AK.TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
     if tp.num_tasks == 1
         num_dst = traverse_nodes_left_range!(
             bvh1, bvh2,
             src, dst, nothing,
             virtual_nodes_before1,
             virtual_nodes_before2,
-            (1, num_src),
+            1:num_src,
         )
     else
         # Keep track of tasks launched and number of elements written by each task in their unique
         # memory region. The unique region is equal to 2 dst elements per src element
         tasks = Vector{Task}(undef, tp.num_tasks)
         num_written = Vector{Int}(undef, tp.num_tasks)
         @inbounds for i in 1:tp.num_tasks
-            istart, iend = tp[i]
             tasks[i] = Threads.@spawn traverse_nodes_left_range!(
                 bvh1, bvh2,
                 src, view(dst, 2istart - 1:2iend), view(num_written, i),
                 virtual_nodes_before1,
                 virtual_nodes_before2,
-                (istart, iend),
+                tp[i],
             )
         end
 
@@ -163,7 +161,7 @@ function traverse_nodes_left!(bvh1, bvh2, src, dst, num_src, ::Nothing, level1,
 
             # Repack written contacts by the second, third thread, etc.
             if i > 1
-                istart, iend = tp[i]
+                istart = tp[i].start
                 for j in 1:task_num_written
                     dst[num_dst + j] = dst[2istart - 1 + j - 1]
                 end
@@ -184,7 +182,7 @@ function traverse_nodes_left_range!(
 
     # For each BVTT pair of nodes, check for contact. Only expand BVTT for BVH1, as BVH2 is already
     # one above leaf level
-    @inbounds for i in irange[1]:irange[2]
+    @inbounds for i in irange
         # Extract implicit indices of BVH nodes to test
         implicit1, implicit2 = src[i]
 
@@ -231,28 +229,27 @@ function traverse_nodes_right!(bvh1, bvh2, src, dst, num_src, ::Nothing, level1,
 
     # Split computation into contiguous ranges of minimum 100 elements each; if only single thread
     # is needed, inline call
-    tp = TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
+    tp = AK.TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
     if tp.num_tasks == 1
         num_dst = traverse_nodes_right_range!(
             bvh1, bvh2,
             src, dst, nothing,
             virtual_nodes_before1,
             virtual_nodes_before2,
-            (1, num_src),
+            1:num_src,
         )
     else
         # Keep track of tasks launched and number of elements written by each task in their unique
         # memory region. The unique region is equal to 2 dst elements per src element
         tasks = Vector{Task}(undef, tp.num_tasks)
         num_written = Vector{Int}(undef, tp.num_tasks)
         @inbounds for i in 1:tp.num_tasks
-            istart, iend = tp[i]
             tasks[i] = Threads.@spawn traverse_nodes_right_range!(
                 bvh1, bvh2,
                 src, view(dst, 2istart - 1:2iend), view(num_written, i),
                 virtual_nodes_before1,
                 virtual_nodes_before2,
-                (istart, iend),
+                tp[i],
             )
         end
 
@@ -264,7 +261,7 @@ function traverse_nodes_right!(bvh1, bvh2, src, dst, num_src, ::Nothing, level1,
 
             # Repack written contacts by the second, third thread, etc.
             if i > 1
-                istart, iend = tp[i]
+                istart = tp[i].start
                 for j in 1:task_num_written
                     dst[num_dst + j] = dst[2istart - 1 + j - 1]
                 end
@@ -285,7 +282,7 @@ function traverse_nodes_right_range!(
 
     # For each BVTT pair of nodes, check for contact. Only expand BVTT for BVH2, as BVH1 is already
     # one above leaf level
-    @inbounds for i in irange[1]:irange[2]
+    @inbounds for i in irange
         # Extract implicit indices of BVH nodes to test
         implicit1, implicit2 = src[i]
 
@@ -329,26 +326,25 @@ function traverse_nodes_leaves_left!(bvh1, bvh2, src, dst, num_src, ::Nothing, l
 
     # Split computation into contiguous ranges of minimum 100 elements each; if only single thread
     # is needed, inline call
-    tp = TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
+    tp = AK.TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
     if tp.num_tasks == 1
         num_dst = traverse_nodes_leaves_left_range!(
             bvh1, bvh2,
             src, dst, nothing,
             virtual_nodes_before1,
-            (1, num_src),
+            1:num_src,
         )
     else
         # Keep track of tasks launched and number of elements written by each task in their unique
         # memory region. The unique region is equal to 2 dst elements per src element
         tasks = Vector{Task}(undef, tp.num_tasks)
         num_written = Vector{Int}(undef, tp.num_tasks)
         @inbounds for i in 1:tp.num_tasks
-            istart, iend = tp[i]
             tasks[i] = Threads.@spawn traverse_nodes_leaves_left_range!(
                 bvh1, bvh2,
                 src, view(dst, 2istart - 1:2iend), view(num_written, i),
                 virtual_nodes_before1,
-                (istart, iend),
+                tp[i],
             )
         end
 
@@ -360,7 +356,7 @@ function traverse_nodes_leaves_left!(bvh1, bvh2, src, dst, num_src, ::Nothing, l
 
             # Repack written contacts by the second, third thread, etc.
             if i > 1
-                istart, iend = tp[i]
+                istart = tp[i].start
                 for j in 1:task_num_written
                     dst[num_dst + j] = dst[2istart - 1 + j - 1]
                 end
@@ -384,7 +380,7 @@ function traverse_nodes_leaves_left_range!(
 
     # For each BVTT pair of nodes, check for contact. Only expand BVTT for BVH1, as BVH2 is already
     # at leaf level
-    @inbounds for i in irange[1]:irange[2]
+    @inbounds for i in irange
         # Extract implicit indices of BVH nodes to test
         implicit1, implicit2 = src[i]
 
@@ -430,26 +426,25 @@ function traverse_nodes_leaves_right!(bvh1, bvh2, src, dst, num_src, ::Nothing,
 
     # Split computation into contiguous ranges of minimum 100 elements each; if only single thread
     # is needed, inline call
-    tp = TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
+    tp = AK.TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
     if tp.num_tasks == 1
         num_dst = traverse_nodes_leaves_right_range!(
             bvh1, bvh2,
             src, dst, nothing,
             virtual_nodes_before2,
-            (1, num_src),
+            1:num_src,
         )
     else
         # Keep track of tasks launched and number of elements written by each task in their unique
         # memory region. The unique region is equal to 2 dst elements per src element
         tasks = Vector{Task}(undef, tp.num_tasks)
         num_written = Vector{Int}(undef, tp.num_tasks)
         @inbounds for i in 1:tp.num_tasks
-            istart, iend = tp[i]
             tasks[i] = Threads.@spawn traverse_nodes_leaves_right_range!(
                 bvh1, bvh2,
                 src, view(dst, 2istart - 1:2iend), view(num_written, i),
                 virtual_nodes_before2,
-                (istart, iend),
+                tp[i],
             )
         end
 
@@ -461,7 +456,7 @@ function traverse_nodes_leaves_right!(bvh1, bvh2, src, dst, num_src, ::Nothing,
 
             # Repack written contacts by the second, third thread, etc.
             if i > 1
-                istart, iend = tp[i]
+                istart = tp[i].start
                 for j in 1:task_num_written
                     dst[num_dst + j] = dst[2istart - 1 + j - 1]
                 end
@@ -485,7 +480,7 @@ function traverse_nodes_leaves_right_range!(
 
     # For each BVTT pair of nodes, check for contact. Only expand BVTT for BVH2, as BVH1 is already
     # at leaf level
-    @inbounds for i in irange[1]:irange[2]
+    @inbounds for i in irange
         # Extract implicit indices of BVH nodes to test
         implicit1, implicit2 = src[i]
 
@@ -527,12 +522,12 @@ function traverse_leaves_pair!(bvh1, bvh2, src, contacts, num_src, ::Nothing, op
 
     # Split computation into contiguous ranges of minimum 100 elements each; if only single thread
     # is needed, inline call
-    tp = TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
+    tp = AK.TaskPartitioner(num_src, options.num_threads, options.min_traversals_per_thread)
     if tp.num_tasks == 1
         num_contacts = traverse_leaves_pair_range!(
             bvh1, bvh2,
             src, view(contacts, :), nothing,
-            (1, num_src),
+            1:num_src,
         )
     else
         num_contacts = 0
@@ -542,11 +537,10 @@ function traverse_leaves_pair!(bvh1, bvh2, src, contacts, num_src, ::Nothing, op
         tasks = Vector{Task}(undef, tp.num_tasks)
         num_written = Vector{Int}(undef, tp.num_tasks)
         @inbounds for i in 1:tp.num_tasks
-            istart, iend = tp[i]
             tasks[i] = Threads.@spawn traverse_leaves_pair_range!(
                 bvh1, bvh2,
                 src, view(contacts, istart:iend), view(num_written, i),
-                (istart, iend),
+                tp[i],
             )
         end
         @inbounds for i in 1:tp.num_tasks
@@ -555,7 +549,7 @@ function traverse_leaves_pair!(bvh1, bvh2, src, contacts, num_src, ::Nothing, op
 
             # Repack written contacts by the second, third thread, etc.
             if i > 1
-                istart, iend = tp[i]
+                istart = tp[i].start
                 for j in 1:task_num_written
                     contacts[num_contacts + j] = contacts[istart + j - 1]
                 end
@@ -579,7 +573,7 @@ function traverse_leaves_pair_range!(
     num_above2 = pow2(bvh2.tree.levels - 1) - 1
 
     # For each BVTT pair of nodes, check for contact
-    @inbounds for i in irange[1]:irange[2]
+    @inbounds for i in irange
         # Extract implicit indices of BVH leaves to test
         implicit1, implicit2 = src[i]