Move CPU/GPU agnostic code into pbat namespace

Q-Minh · Q-Minh · commit 6564535d791b · 2024-11-02T21:05:42.000-04:00
diff --git a/source/pbat/common/CMakeLists.txt b/source/pbat/common/CMakeLists.txt
@@ -8,9 +8,13 @@ target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit
         "Eigen.h"
         "Hash.h"
         "Indexing.h"
+        "Queue.h"
+        "Stack.h"
 )
 target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit
 PRIVATE
     "Eigen.cpp"
     "Indexing.cpp"
+    "Queue.cpp"
+    "Stack.cpp"
 )
diff --git a/source/pbat/common/Common.h b/source/pbat/common/Common.h
@@ -6,5 +6,7 @@
 #include "Eigen.h"
 #include "Hash.h"
 #include "Indexing.h"
+#include "Queue.h"
+#include "Stack.h"
 
 #endif // PBAT_COMMON_COMMON_H
diff --git a/source/pbat/common/Queue.cpp b/source/pbat/common/Queue.cpp
@@ -0,0 +1 @@
+#include "Queue.h"
diff --git a/source/pbat/common/Queue.h b/source/pbat/common/Queue.h
@@ -1,11 +1,9 @@
-#ifndef PBAT_GPU_COMMON_QUEUE_CUH
-#define PBAT_GPU_COMMON_QUEUE_CUH
+#ifndef PBAT_COMMON_QUEUE_H
+#define PBAT_COMMON_QUEUE_H
 
 #include "pbat/HostDevice.h"
-#include "pbat/gpu/Aliases.h"
 
 namespace pbat {
-namespace gpu {
 namespace common {
 
 template <class T, auto kCapacity = 64>
@@ -27,16 +25,15 @@ class Queue
     }
     PBAT_HOST_DEVICE bool IsFull() const { return n == kCapacity; }
     PBAT_HOST_DEVICE bool IsEmpty() const { return n == 0; }
-    PBAT_HOST_DEVICE GpuIndex Size() const { return n; }
+    PBAT_HOST_DEVICE auto Size() const { return n; }
     PBAT_HOST_DEVICE void Clear() { begin = end = n = 0; }
 
   private:
     T queue[kCapacity];
-    GpuIndex begin, end, n;
+    int begin, end, n;
 };
 
 } // namespace common
-} // namespace gpu
 } // namespace pbat
 
-#endif // PBAT_GPU_COMMON_QUEUE_CUH
+#endif // PBAT_COMMON_QUEUE_H
diff --git a/source/pbat/common/Stack.cpp b/source/pbat/common/Stack.cpp
@@ -0,0 +1 @@
+#include "Stack.h"
diff --git a/source/pbat/common/Stack.h b/source/pbat/common/Stack.h
@@ -1,11 +1,9 @@
-#ifndef PBAT_GPU_COMMON_STACK_CUH
-#define PBAT_GPU_COMMON_STACK_CUH
+#ifndef PBAT_COMMON_STACK_H
+#define PBAT_COMMON_STACK_H
 
 #include "pbat/HostDevice.h"
-#include "pbat/gpu/Aliases.h"
 
 namespace pbat {
-namespace gpu {
 namespace common {
 
 template <class T, auto kCapacity = 64>
@@ -16,18 +14,17 @@ class Stack
     PBAT_HOST_DEVICE void Push(T value) { stack[size++] = value; }
     PBAT_HOST_DEVICE T Pop() { return stack[--size]; }
     PBAT_HOST_DEVICE T const& Top() const { return stack[size - 1]; }
-    PBAT_HOST_DEVICE GpuIndex Size() const { return size; }
+    PBAT_HOST_DEVICE auto Size() const { return size; }
     PBAT_HOST_DEVICE bool IsEmpty() const { return size == 0; }
     PBAT_HOST_DEVICE bool IsFull() const { return size == kCapacity; }
     PBAT_HOST_DEVICE void Clear() { size = 0; }
 
   private:
     T stack[kCapacity];
-    GpuIndex size; ///< Serves as both the stack pointer and the number of elements in the stack
+    int size; ///< Serves as both the stack pointer and the number of elements in the stack
 };
 
 } // namespace common
-} // namespace gpu
 } // namespace pbat
 
-#endif // PBAT_GPU_COMMON_STACK_CUH
+#endif // PBAT_COMMON_STACK_H
diff --git a/source/pbat/geometry/CMakeLists.txt b/source/pbat/geometry/CMakeLists.txt
@@ -9,6 +9,7 @@ target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit
     "Geometry.h"
     "IntersectionQueries.h"
     "KdTree.h"
+    "Morton.h"
     "OverlapQueries.h"
     "TetrahedralAabbHierarchy.h"
     "TriangleAabbHierarchy.h"
@@ -21,6 +22,7 @@ target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit
     "DistanceQueries.cpp"
     "IntersectionQueries.cpp"
     "KdTree.cpp"
+    "Morton.cpp"
     "OverlapQueries.cpp"
     "TetrahedralAabbHierarchy.cpp"
     "TriangleAabbHierarchy.cpp"
diff --git a/source/pbat/geometry/Geometry.h b/source/pbat/geometry/Geometry.h
@@ -1,6 +1,15 @@
 #ifndef PBAT_GEOMETRY_GEOMETRY_H
 #define PBAT_GEOMETRY_GEOMETRY_H
 
+#include "AxisAlignedBoundingBox.h"
 #include "BoundingVolumeHierarchy.h"
+#include "ClosestPointQueries.h"
+#include "DistanceQueries.h"
+#include "IntersectionQueries.h"
+#include "KdTree.h"
+#include "Morton.h"
+#include "OverlapQueries.h"
+#include "TetrahedralAabbHierarchy.h"
+#include "TriangleAabbHierarchy.h"
 
 #endif // PBAT_GEOMETRY_GEOMETRY_H
diff --git a/source/pbat/geometry/Morton.cpp b/source/pbat/geometry/Morton.cpp
@@ -0,0 +1 @@
+#include "Morton.h"
diff --git a/source/pbat/geometry/Morton.h b/source/pbat/geometry/Morton.h
@@ -0,0 +1,47 @@
+#ifndef PBAT_GEOMETRY_MORTON_H
+#define PBAT_GEOMETRY_MORTON_H
+
+#include "pbat/HostDevice.h"
+
+#include <array>
+#include <cmath>
+#include <cstddef>
+
+namespace pbat {
+namespace geometry {
+
+using MortonCodeType = std::uint32_t;
+
+// NOTE:
+// We make these (otherwise non-templated) functions inline so that they
+// are compiled by nvcc whenever this header is included in cuda sources.
+
+// Expands a 10-bit integer into 30 bits
+// by inserting 2 zeros after each bit.
+PBAT_HOST_DEVICE inline MortonCodeType ExpandBits(MortonCodeType v)
+{
+    v = (v * 0x00010001u) & 0xFF0000FFu;
+    v = (v * 0x00000101u) & 0x0F00F00Fu;
+    v = (v * 0x00000011u) & 0xC30C30C3u;
+    v = (v * 0x00000005u) & 0x49249249u;
+    return v;
+}
+
+// Calculates a 30-bit Morton code for the
+// given 3D point located within the unit cube [0,1].
+[[maybe_unused]] PBAT_HOST_DEVICE inline MortonCodeType Morton3D(std::array<float, 3> x)
+{
+    using namespace std;
+    x[0]              = min(max(x[0] * 1024.0f, 0.0f), 1023.0f);
+    x[1]              = min(max(x[1] * 1024.0f, 0.0f), 1023.0f);
+    x[2]              = min(max(x[2] * 1024.0f, 0.0f), 1023.0f);
+    MortonCodeType xx = ExpandBits(static_cast<MortonCodeType>(x[0]));
+    MortonCodeType yy = ExpandBits(static_cast<MortonCodeType>(x[1]));
+    MortonCodeType zz = ExpandBits(static_cast<MortonCodeType>(x[2]));
+    return xx * 4 + yy * 2 + zz;
+}
+
+} // namespace geometry
+} // namespace pbat
+
+#endif // PBAT_GEOMETRY_MORTON_H
diff --git a/source/pbat/gpu/common/CMakeLists.txt b/source/pbat/gpu/common/CMakeLists.txt
@@ -4,9 +4,6 @@ target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit
     FILES
     "Buffer.cuh"
     "Cuda.cuh"
-    "Morton.cuh"
-    "Queue.cuh"
-    "Stack.cuh"
     "SynchronizedList.cuh"
     "Var.cuh"
 )
@@ -15,9 +12,6 @@ target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit
     PRIVATE
     "Buffer.cu"
     "Cuda.cu"
-    "Morton.cu"
-    "Queue.cu"
-    "Stack.cu"
     "SynchronizedList.cu"
     "Var.cu"
 )
diff --git a/source/pbat/gpu/common/Morton.cu b/source/pbat/gpu/common/Morton.cu
diff --git a/source/pbat/gpu/common/Morton.cuh b/source/pbat/gpu/common/Morton.cuh
diff --git a/source/pbat/gpu/common/Queue.cu b/source/pbat/gpu/common/Queue.cu
diff --git a/source/pbat/gpu/common/Stack.cu b/source/pbat/gpu/common/Stack.cu
diff --git a/source/pbat/gpu/geometry/BvhImpl.cuh b/source/pbat/gpu/geometry/BvhImpl.cuh
@@ -2,9 +2,9 @@
 #define PBAT_GPU_BVH_IMPL_CUH
 
 #include "PrimitivesImpl.cuh"
+#include "pbat/geometry/Morton.h"
 #include "pbat/gpu/Aliases.h"
 #include "pbat/gpu/common/Buffer.cuh"
-#include "pbat/gpu/common/Morton.cuh"
 #include "pbat/gpu/common/SynchronizedList.cuh"
 #include "pbat/gpu/common/Var.cuh"
 
@@ -28,7 +28,7 @@ class BvhImpl
 {
   public:
     using OverlapType    = cuda::std::pair<GpuIndex, GpuIndex>;
-    using MortonCodeType = pbat::gpu::common::MortonCodeType;
+    using MortonCodeType = pbat::geometry::MortonCodeType;
 
     friend class BvhQueryImpl;
 
diff --git a/source/pbat/gpu/geometry/BvhImplKernels.cuh b/source/pbat/gpu/geometry/BvhImplKernels.cuh
@@ -3,8 +3,8 @@
 
 #include "BvhImpl.cuh"
 #include "pbat/HostDevice.h"
+#include "pbat/common/Stack.h"
 #include "pbat/gpu/Aliases.h"
-#include "pbat/gpu/common/Stack.cuh"
 #include "pbat/gpu/common/SynchronizedList.cuh"
 
 #include <array>
@@ -58,7 +58,8 @@ struct FComputeMortonCode
             auto cd = GpuScalar{0.5} * (b[d][bs] + e[d][bs]);
             c[d]    = (cd - sb[d]) / sbe[d];
         }
-        morton[s] = common::Morton3D(c);
+        using pbat::geometry::Morton3D;
+        morton[s] = Morton3D(c);
     }
 
     std::array<GpuScalar, 3> sb;
@@ -223,7 +224,8 @@ struct FDetectSelfOverlaps
     PBAT_DEVICE void operator()(auto leaf)
     {
         // Traverse nodes depth-first starting from the root=0 node
-        common::Stack<GpuIndex, 64> stack{};
+        using pbat::common::Stack;
+        Stack<GpuIndex, 64> stack{};
         stack.Push(0);
         do
         {
diff --git a/source/pbat/gpu/geometry/BvhQueryImplKernels.cuh b/source/pbat/gpu/geometry/BvhQueryImplKernels.cuh
diff --git a/source/pbat/math/linalg/LinAlg.h b/source/pbat/math/linalg/LinAlg.h

Original file line number	Diff line number	Diff line change
`@@ -8,9 +8,13 @@ target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit`
`8`	`8`	`"Eigen.h"`
`9`	`9`	`"Hash.h"`
`10`	`10`	`"Indexing.h"`
	`11`	`+ "Queue.h"`
	`12`	`+ "Stack.h"`
`11`	`13`	`)`
`12`	`14`	`target_sources(PhysicsBasedAnimationToolkit_PhysicsBasedAnimationToolkit`
`13`	`15`	`PRIVATE`
`14`	`16`	`"Eigen.cpp"`
`15`	`17`	`"Indexing.cpp"`
	`18`	`+ "Queue.cpp"`
	`19`	`+ "Stack.cpp"`
`16`	`20`	`)`