Merge branch '106-gpu-poisson-benchmarks' into 'master' (54094c40) · Commits · Jan Šmíd / Analyst 2

GPU/src/main/java/cz/fidentis/analyst/opencl/CLProgramDef.java

+11 −1

Original line number	Diff line number	Diff line
		@@ -34,7 +34,6 @@ public enum CLProgramDef {
		List.of(CLSourceDef.SPATIAL_DATA_UTILS,
		CLSourceDef.BOUNDING_BOX_UTILS,
		CLSourceDef.TRIANGLE_STACK_STRUCTURES,
		CLSourceDef.OCTREE_CONSTRUCTION_UTILS,
		CLSourceDef.OCTREE_BUILDER),
		List.of(CLProgram.CompilerOptions.ENABLE_MAD,
		CLProgram.CompilerOptions.FAST_RELAXED_MATH,
		@@ -51,6 +50,17 @@ public enum CLProgramDef {
		List.of(CLProgram.CompilerOptions.ENABLE_MAD,
		CLProgram.CompilerOptions.FAST_RELAXED_MATH,
		CLProgram.CompilerOptions.NO_SIGNED_ZEROS)
		),

		/**
		* Kernels specific to GPU-based Poisson sub-sampling
		*/
		POISSON_GPU_UTILS(
		List.of(CLSourceDef.SPATIAL_DATA_UTILS,
		CLSourceDef.POISSON_GPU_UTILS),
		List.of(CLProgram.CompilerOptions.ENABLE_MAD,
		CLProgram.CompilerOptions.FAST_RELAXED_MATH,
		CLProgram.CompilerOptions.NO_SIGNED_ZEROS)
		);

		private final List<CLSourceDef> sources = new ArrayList<>();

+2 −2

Original line number	Diff line number	Diff line
		@@ -13,11 +13,11 @@ public enum CLSourceDef {
		RAY_INTERSECTION("RayIntersection.c"),
		COMPUTE_INTERSECTIONS("ComputeIntersections.c"),
		OCTREE_TRAVERSAL("OctreeTraversal.c"),
		OCTREE_CONSTRUCTION_UTILS("OctreeConstruction.c"),
		OCTREE_BUILDER("OctreeBuilder.c"),
		COMMON_KERNEL_SERVICES("CommonKernelServices.c"),
		TRIANGLE_STACK_STRUCTURES("TriangleStackStructures.c"),
		OCTREE_TRAVERSAL_STRUCTURES("OctreeTraversalStructures.c");
		OCTREE_TRAVERSAL_STRUCTURES("OctreeTraversalStructures.c"),
		POISSON_GPU_UTILS("ProjectorUtils.c");
		private final String file;

		CLSourceDef(String file) {

deleted100644 → 0

+0 −39

Original line number	Diff line number	Diff line
		package cz.fidentis.analyst.opencl;

		import com.jogamp.opencl.CLContext;
		import com.jogamp.opencl.CLPlatform;

		/**
		* Provides access to OpenCL Context
		*
		* @author Marek Horský
		*/
		public interface OpenCLProvider {

		/**
		* Checks if any present device supports OpenCL or returns false straight away, if initialization is impossible
		*
		* @return true, if OpenCL is available, false otherwise
		*/
		static boolean isOpenCLAvailable() {
		try {
		return CLPlatform.listCLPlatforms().length != 0;
		} catch (Exception e) {
		return false;
		}
		}

		/**
		* Creates OpenCL Context using most powerful available device
		* The decision of the device relies on inner calculations of JogAmp's implementation,
		* which can mistake number of processing units on cards RTX 4xxx as of version 2.5.0
		*
		* @return OpenCL context
		*/

		static CLContext createContext() {
		return isOpenCLAvailable()
		? CLContext.create() //CLPlatform.listCLPlatforms()[0] //Toggle index to manually change device
		: null;
		}
		}

+34 −1

Original line number	Diff line number	Diff line
		package cz.fidentis.analyst.opencl;

		import com.jogamp.opencl.CLContext;
		import com.jogamp.opencl.CLPlatform;
		import cz.fidentis.analyst.opencl.impl.SingletonGpuManagerCL;

		/**
		@@ -8,11 +9,43 @@ import cz.fidentis.analyst.opencl.impl.SingletonGpuManagerCL;
		*
		* @author Ondrej Simecek
		* @author Radek Oslejsek
		* @author Pavol Kycina
		* @author Marek Horský
		*/
		public interface OpenCLServices {

		/**
		* Checks if any present device supports OpenCL or returns false straight away, if initialization is impossible
		* Also enforces minimal supported work group size to 256 as this value needs to be present as constant in .c source files
		*
		* @return true, if OpenCL is available, false otherwise
		*/
		static boolean isOpenCLAvailable() {
		try {
		if (CLPlatform.listCLPlatforms().length != 0) {
		CLContext clContext = CLContext.create();
		int maxSupportedWorkGroupSize = clContext.getMaxFlopsDevice().getMaxWorkGroupSize();
		clContext.release();
		return maxSupportedWorkGroupSize >= 256;
		}
		} catch (Exception ignored) {
		}
		return false;
		}

		/**
		* Creates OpenCL Context using most powerful available device
		* The decision of the device relies on inner calculations of JogAmp's implementation,
		* which can mistake number of processing units on cards RTX 4xxx as of version 2.5.0
		*
		* @return OpenCL context
		*/

		static CLContext createContext() {
		return isOpenCLAvailable()
		? CLContext.create() //Toggle index to manually change device
		: null;
		}

		/**
		* Compiles the OpenCL program, re-uses already compiled programs.
		*

+13 −1

Original line number	Diff line number	Diff line
		@@ -2,12 +2,15 @@ package cz.fidentis.analyst.opencl.memory;

		import com.jogamp.opencl.CLContext;
		import cz.fidentis.analyst.data.mesh.MeshTriangle;
		import cz.fidentis.analyst.opencl.memory.impl.IntegerBuffer;
		import cz.fidentis.analyst.opencl.memory.impl.MeshTriangleBuffer;
		import cz.fidentis.analyst.opencl.memory.impl.RayIntersectionBuffer;
		import cz.fidentis.analyst.opencl.memory.impl.VoxelBuffer;

		import javax.vecmath.Point3d;
		import javax.vecmath.Vector3d;
		import java.nio.FloatBuffer;
		import java.nio.IntBuffer;

		/**
		* Factory for OpenCL buffers with additional functionality
		@@ -15,6 +18,15 @@ import javax.vecmath.Vector3d;
		* @author Marek Horský
		*/
		public interface BufferFactory {
		/**
		* Creates resizable {@link IntegerBuffer}
		*
		* @return IntegerBuffer
		*/
		static BufferGPU<IntBuffer> getIntBuffer(CLContext clContext) {
		return new IntegerBuffer(clContext);
		}

		/**
		* Creates resizable {@link VoxelBuffer}
		*
		@@ -38,7 +50,7 @@ public interface BufferFactory {
		*
		* @return RayIntersectionBuffer
		*/
		static BufferGPU getRayIntersectionBuffer(CLContext context) {
		static BufferGPU<FloatBuffer> getRayIntersectionBuffer(CLContext context) {
		return new RayIntersectionBuffer(context);
		}