Array refactor (#796)

Internal arrays using macro generated functions instead of stretchy
buffers.
Added `b2Body_GetWorld`
Added `b2CosSin` and `b2ComputeCosSin`

.github/workflows/build.yml

@@ -52,8 +52,8 @@ jobs:
     - uses: actions/checkout@v4
 
     - name: Configure CMake
-      # run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBOX2D_SANITIZE=ON -DBUILD_SHARED_LIBS=OFF
-      run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBUILD_SHARED_LIBS=OFF
+      run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBOX2D_SANITIZE=ON -DBUILD_SHARED_LIBS=OFF
+      # run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBUILD_SHARED_LIBS=OFF
 
     - name: Build
       run: cmake --build ${{github.workspace}}/build --config ${{env.BUILD_TYPE}}
@@ -76,8 +76,8 @@ jobs:
 
     - name: Configure CMake
       # enkiTS is failing ASAN on windows
-      # run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBOX2D_SANITIZE=ON -DBUILD_SHARED_LIBS=OFF
-      run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBUILD_SHARED_LIBS=OFF
+      run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBOX2D_SANITIZE=ON -DBUILD_SHARED_LIBS=OFF
+      # run: cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DBOX2D_SAMPLES=OFF -DBUILD_SHARED_LIBS=OFF
 
     - name: Build
       run: cmake --build ${{github.workspace}}/build --config ${{env.BUILD_TYPE}}

CMakeLists.txt

@@ -17,6 +17,9 @@ message(STATUS "CMake C++ compiler: ${CMAKE_CXX_COMPILER_ID}")
 message(STATUS "CMake system name: ${CMAKE_SYSTEM_NAME}")
 message(STATUS "CMake host system processor: ${CMAKE_HOST_SYSTEM_PROCESSOR}")
 
+# static link
+set(CMAKE_MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>")
+
 if (MSVC OR APPLE)
 	option(BOX2D_SANITIZE "Enable sanitizers for some builds" OFF)
 	if(BOX2D_SANITIZE)

benchmark/amd7950x_avx2/joint_grid.csv

@@ -1,9 +1,9 @@
 threads,fps
-1,333.07
-2,641.434
-3,944.038
-4,1206.21
-5,1496.76
-6,1722.19
-7,1956.35
-8,2146.81
+1,335.038
+2,638.117
+3,938.253
+4,1211.86
+5,1495.94
+6,1719.76
+7,1964.31
+8,2148.41

benchmark/amd7950x_avx2/large_pyramid.csv

@@ -1,9 +1,9 @@
 threads,fps
-1,337.452
-2,623.273
-3,889.504
-4,1124.83
-5,1339.88
-6,1493.98
-7,1689.74
-8,1726.49
+1,334.105
+2,617.266
+3,895.838
+4,1131.52
+5,1338.68
+6,1503.01
+7,1686.74
+8,1721.57

benchmark/amd7950x_avx2/many_pyramids.csv

@@ -1,9 +1,9 @@
 threads,fps
-1,87.7948
-2,165.138
-3,244.692
-4,310.302
-5,379.072
-6,436.226
-7,503.71
-8,554.489
+1,85.2506
+2,163.857
+3,241.226
+4,309.927
+5,369.713
+6,431.47
+7,496.292
+8,547.372

benchmark/amd7950x_avx2/smash.csv

@@ -1,9 +1,9 @@
 threads,fps
-1,178.234
-2,282.589
-3,360.22
-4,431.746
-5,485.852
-6,529.01
-7,570.548
-8,599.725
+1,180.404
+2,284.615
+3,364.465
+4,441.008
+5,495.098
+6,538.398
+7,578.614
+8,610.615

benchmark/amd7950x_avx2/tumbler.csv

@@ -1,9 +1,9 @@
 threads,fps
-1,379.168
-2,593.692
-3,768.849
-4,908.277
-5,1050.74
-6,1145.49
-7,1231.01
-8,1298.58
+1,385.465
+2,604.521
+3,790.813
+4,934.896
+5,1088.09
+6,1192.74
+7,1276.24
+8,1346.88

benchmark/main.c

@@ -15,11 +15,11 @@
 #include <string.h>
 
 #if defined( _WIN64 )
-	#include <windows.h>
+#include <windows.h>
 #elif defined( __APPLE__ )
-	#include <unistd.h>
+#include <unistd.h>
 #elif defined( __linux__ )
-	#include <unistd.h>
+#include <unistd.h>
 #endif
 
 #define ARRAY_COUNT( A ) (int)( sizeof( A ) / sizeof( A[0] ) )
@@ -114,10 +114,23 @@ static void FinishTask( void* userTask, void* userContext )
 	enkiWaitForTaskSet( scheduler, task );
 }
 
+// Box2D benchmark application. On Windows I recommend running this in an administrator command prompt. Don't use Windows Terminal.
 int main( int argc, char** argv )
 {
+	Benchmark benchmarks[] = {
+		{ "joint_grid", JointGrid, 500 },
+		{ "large_pyramid", LargePyramid, 500 },
+		{ "many_pyramids", ManyPyramids, 200 },
+		{ "smash", Smash, 300 },
+		{ "tumbler", Tumbler, 750 },
+	};
+
+	int benchmarkCount = ARRAY_COUNT( benchmarks );
+
 	int maxThreadCount = GetNumberOfCores();
 	int runCount = 4;
+	int singleBenchmark = -1;
+	int singleWorkerCount = -1;
 	b2Counters counters = { 0 };
 	bool enableContinuous = true;
 
@@ -129,30 +142,47 @@ int main( int argc, char** argv )
 		if ( strncmp( arg, "-t=", 3 ) == 0 )
 		{
 			int threadCount = atoi( arg + 3 );
-			maxThreadCount = b2MinInt( maxThreadCount, threadCount );
+			maxThreadCount = b2ClampInt( threadCount, 1, maxThreadCount );
+		}
+		else if ( strncmp( arg, "-b=", 3 ) == 0 )
+		{
+			singleBenchmark = atoi( arg + 3 );
+			singleBenchmark = b2ClampInt( singleBenchmark, 0, benchmarkCount - 1 );
+		}
+		else if ( strncmp( arg, "-w=", 3 ) == 0 )
+		{
+			singleWorkerCount = atoi( arg + 3 );
+		}
+		else if ( strncmp( arg, "-r=", 3 ) == 0 )
+		{
+			runCount = b2ClampInt(atoi( arg + 3 ), 1, 16);
 		}
 		else if ( strcmp( arg, "-h" ) == 0 )
 		{
 			printf( "Usage\n"
-					"-t=<thread count>: the maximum number of threads to use\n" );
+					"-t=<integer>: the maximum number of threads to use\n"
+					"-b=<integer>: run a single benchmark\n"
+					"-w=<integer>: run a single worker count\n"
+					"-r=<integer>: number of repeats (default is 4)\n" );
+			exit( 0 );
 		}
 	}
 
-	Benchmark benchmarks[] = {
-		{ "joint_grid", JointGrid, 500 },
-		{ "large_pyramid", LargePyramid, 500 },
-		{ "many_pyramids", ManyPyramids, 200 },
-		{ "smash", Smash, 300 },
-		{ "tumbler", Tumbler, 750 },
-	};
-
-	int benchmarkCount = ARRAY_COUNT( benchmarks );
+	if ( singleWorkerCount != -1 )
+	{
+		singleWorkerCount = b2ClampInt( singleWorkerCount, 1, maxThreadCount );
+	}
 
 	printf( "Starting Box2D benchmarks\n" );
 	printf( "======================================\n" );
 
 	for ( int benchmarkIndex = 0; benchmarkIndex < benchmarkCount; ++benchmarkIndex )
 	{
+		if ( singleBenchmark != -1 && benchmarkIndex != singleBenchmark )
+		{
+			continue;
+		}
+
 #ifdef NDEBUG
 		int stepCount = benchmarks[benchmarkIndex].stepCount;
 #else
@@ -167,6 +197,11 @@ int main( int argc, char** argv )
 
 		for ( int threadCount = 1; threadCount <= maxThreadCount; ++threadCount )
 		{
+			if ( singleWorkerCount != -1 && singleWorkerCount != threadCount )
+			{
+				continue;
+			}
+
 			printf( "thread count: %d\n", threadCount );
 
 			for ( int runIndex = 0; runIndex < runCount; ++runIndex )
@@ -195,6 +230,7 @@ int main( int argc, char** argv )
 
 				// Initial step can be expensive and skew benchmark
 				b2World_Step( worldId, timeStep, subStepCount );
+				taskCount = 0;
 
 				b2Timer timer = b2CreateTimer();
 
@@ -242,9 +278,9 @@ int main( int argc, char** argv )
 		}
 
 		fprintf( file, "threads,fps\n" );
-		for ( int threadCount = 1; threadCount <= maxThreadCount; ++threadCount )
+		for ( int threadIndex = 1; threadIndex <= maxThreadCount; ++threadIndex )
 		{
-			fprintf( file, "%d,%g\n", threadCount, maxFps[threadCount - 1] );
+			fprintf( file, "%d,%g\n", threadIndex, maxFps[threadIndex - 1] );
 		}
 
 		fclose( file );

include/box2d/box2d.h

@@ -399,6 +399,9 @@ B2_API bool b2Body_IsBullet( b2BodyId bodyId );
 ///	@see b2ShapeDef::enableHitEvents
 B2_API void b2Body_EnableHitEvents( b2BodyId bodyId, bool enableHitEvents );
 
+/// Get the world that owns this body
+B2_API b2WorldId b2Body_GetWorld( b2BodyId bodyId );
+
 /// Get the number of shapes on this body
 B2_API int b2Body_GetShapeCount( b2BodyId bodyId );
 

include/box2d/math_functions.h

@@ -26,6 +26,15 @@ typedef struct b2Vec2
 	float x, y;
 } b2Vec2;
 
+/// Cosine and sine pair
+/// This uses a custom implementation designed for cross platform determinism
+typedef struct b2CosSin
+{
+	/// cosine and sine
+	float cosine;
+	float sine;
+} b2CosSin;
+
 /// 2D rotation
 /// This is similar to using a complex number for rotation
 typedef struct b2Rot
@@ -323,7 +332,14 @@ B2_INLINE float b2DistanceSquared( b2Vec2 a, b2Vec2 b )
 }
 
 /// Make a rotation using an angle in radians
-B2_API b2Rot b2MakeRot( float angle );
+B2_API b2CosSin b2ComputeCosSin( float angle );
+
+/// Make a rotation using an angle in radians
+B2_INLINE b2Rot b2MakeRot( float angle )
+{
+	b2CosSin cs = b2ComputeCosSin( angle );
+	return B2_LITERAL( b2Rot ){ cs.cosine, cs.sine };
+}
 
 /// Is this rotation normalized?
 B2_INLINE bool b2IsNormalized( b2Rot q )
@@ -346,7 +362,6 @@ B2_INLINE b2Rot b2NLerp( b2Rot q1, b2Rot q2, float t )
 	return b2NormalizeRot( q );
 }
 
-
 /// Compute the angular velocity necessary to rotate between two rotations over a give time
 ///	@param q1 initial rotation
 ///	@param q2 final rotation

samples/draw.cpp

@@ -12,16 +12,18 @@
 #include <vector>
 
 #if defined( _WIN32 )
-	#define _CRTDBG_MAP_ALLOC
-	#include <crtdbg.h>
-	#include <stdlib.h>
+#define _CRTDBG_MAP_ALLOC
+#include <crtdbg.h>
+#include <stdlib.h>
 #else
-	#include <stdlib.h>
+#include <stdlib.h>
 #endif
 
+// clang-format off
 #include <glad/glad.h>
-// Keep glad.h before glfw3.h
 #include <GLFW/glfw3.h>
+// clang-format on
+
 #include <imgui.h>
 
 #define BUFFER_OFFSET( x ) ( (const void*)( x ) )

samples/main.cpp

@@ -13,23 +13,25 @@
 #include "box2d/box2d.h"
 #include "box2d/math_functions.h"
 
+// clang-format off
 #include <glad/glad.h>
-// Keep glad.h before glfw3.h
 #include <GLFW/glfw3.h>
+// clang-format on
+
 #include <imgui.h>
 #include <imgui_impl_glfw.h>
 #include <imgui_impl_opengl3.h>
 #include <stdio.h>
 #include <stdlib.h>
 
 #ifdef BOX2D_PROFILE
-	#include <tracy/Tracy.hpp>
+#include <tracy/Tracy.hpp>
 #else
-	#define FrameMark
+#define FrameMark
 #endif
 
 #if defined( _WIN32 )
-	#include <crtdbg.h>
+#include <crtdbg.h>
 
 static int MyAllocHook( int allocType, void* userData, size_t size, int blockType, long requestNumber,
 						const unsigned char* filename, int lineNumber )
@@ -515,7 +517,6 @@ static void UpdateUI()
 	}
 }
 
-//
 int main( int, char** )
 {
 #if defined( _WIN32 )
@@ -707,9 +708,9 @@ int main( int, char** )
 
 		// if (g_draw.m_showUI)
 		{
-			 snprintf( buffer, 128, "%.1f ms - step %d - camera (%g, %g, %g)", 1000.0f * frameTime, s_sample->m_stepCount,
+			snprintf( buffer, 128, "%.1f ms - step %d - camera (%g, %g, %g)", 1000.0f * frameTime, s_sample->m_stepCount,
 					  g_camera.m_center.x, g_camera.m_center.y, g_camera.m_zoom );
-			//snprintf( buffer, 128, "%.1f ms", 1000.0f * frameTime );
+			// snprintf( buffer, 128, "%.1f ms", 1000.0f * frameTime );
 
 			ImGui::Begin( "Overlay", nullptr,
 						  ImGuiWindowFlags_NoTitleBar | ImGuiWindowFlags_NoInputs | ImGuiWindowFlags_AlwaysAutoResize |

samples/sample_benchmark.cpp

@@ -671,8 +671,8 @@ class BenchmarkManyPyramids : public Sample
 		m_extent = 0.5f;
 		m_round = 0.0f;
 		m_baseCount = 10;
-		m_rowCount = g_sampleDebug ? 4 : 13;
-		m_columnCount = g_sampleDebug ? 4 : 14;
+		m_rowCount = g_sampleDebug ? 4 : 20;
+		m_columnCount = g_sampleDebug ? 4 : 20;
 		m_groundId = b2_nullBodyId;
 		m_bodyIds = nullptr;
 		m_bodyCount = 0;