diff --git a/README.md b/README.md index 98dd9a8..5386558 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,27 @@ -**University of Pennsylvania, CIS 565: GPU Programming and Architecture, -Project 1 - Flocking** +CIS 565 Project 1 - Flocking +============================ +* Richard Lee +* Tested on: Windows 7, i7-3720QM @ 2.60GHz 8GB, GT 650M 4GB (Personal Computer) -* (TODO) YOUR NAME HERE -* Tested on: (TODO) Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab) +![](images/flocking.gif) -### (TODO: Your README) +## Performance analysis -Include screenshots, analysis, etc. (Remember, this is public, so don't put -anything here that you don't want to share with the world.) +Performance testing was done by measuring the time taken to run 1000 frames of the simulation on each implementation with the different variables. + +**Number of boids** +Testing the number of boids was run with a block size of 128. +![](images/chart1.PNG) +The time elapsed took longer as the number of boids increased, which was to be expected due to the increased number of comparisons. + +**Block size** +Testing the block size was run with 5000 boids. +![](images/chart2.PNG) +Changing the block size caused fluctuations in the performance of all three methods, but did not lead to any significant increases or decreases in their performance. This could be because the increased number of available threads may not have been efficiently utilized to improve the performance. + +**Coherent uniform grid analysis** +For the coherent uniform grid, the performance was about equal at 5000 agents, with the scattered uniform grid being slightly faster than the coherent grid. However, as the number of agents increased, the time taken for the coherent grid increased at a slower rate and outperformed the scattered grid. + +I expected the coherent grid to outperform the scattered grid at all numbers of agents, and I think this result was because less agents meant that there wasn't as much chance to take advantage of the contiguous boid data, and led to a small difference in performance due to the extra pre-processing step to shuffle the boid data. + +However, the coherent grid allowed for faster direct access of all the boids after the pre-processing shuffle as the number of boids increased, while the scattered grid still had to perform lookups for each boid in the position and velocity arrays, which led to performance advantages as the number of boids scaled up. \ No newline at end of file diff --git a/images/chart1.PNG b/images/chart1.PNG new file mode 100755 index 0000000..15b1376 Binary files /dev/null and b/images/chart1.PNG differ diff --git a/images/chart2.PNG b/images/chart2.PNG new file mode 100755 index 0000000..977477f Binary files /dev/null and b/images/chart2.PNG differ diff --git a/images/flocking.gif b/images/flocking.gif new file mode 100755 index 0000000..976a3b3 Binary files /dev/null and b/images/flocking.gif differ diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index fdd636d..750f0cb 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -10,5 +10,5 @@ set(SOURCE_FILES cuda_add_library(src ${SOURCE_FILES} - OPTIONS -arch=sm_20 + OPTIONS -arch=sm_30 ) diff --git a/src/kernel.cu b/src/kernel.cu index 30356b9..a8148d2 100644 --- a/src/kernel.cu +++ b/src/kernel.cu @@ -21,14 +21,14 @@ * Check for CUDA errors; print and exit if there was a problem. */ void checkCUDAError(const char *msg, int line = -1) { - cudaError_t err = cudaGetLastError(); - if (cudaSuccess != err) { - if (line >= 0) { - fprintf(stderr, "Line %d: ", line); - } - fprintf(stderr, "Cuda error: %s: %s.\n", msg, cudaGetErrorString(err)); - exit(EXIT_FAILURE); - } + cudaError_t err = cudaGetLastError(); + if (cudaSuccess != err) { + if (line >= 0) { + fprintf(stderr, "Line %d: ", line); + } + fprintf(stderr, "Cuda error: %s: %s.\n", msg, cudaGetErrorString(err)); + exit(EXIT_FAILURE); + } } @@ -85,11 +85,14 @@ int *dev_gridCellEndIndices; // to this cell? // TODO-2.3 - consider what additional buffers you might need to reshuffle // the position and velocity data to be coherent within cells. +glm::vec3 *dev_pos_new; +glm::vec3 *dev_vel1_new; +glm::vec3 *dev_vel2_new; // LOOK-2.1 - Grid parameters based on simulation parameters. // These are automatically computed for you in Boids::initSimulation int gridCellCount; -int gridSideCount; +int gridSideCount; float gridCellWidth; float gridInverseCellWidth; glm::vec3 gridMinimum; @@ -99,13 +102,13 @@ glm::vec3 gridMinimum; ******************/ __host__ __device__ unsigned int hash(unsigned int a) { - a = (a + 0x7ed55d16) + (a << 12); - a = (a ^ 0xc761c23c) ^ (a >> 19); - a = (a + 0x165667b1) + (a << 5); - a = (a + 0xd3a2646c) ^ (a << 9); - a = (a + 0xfd7046c5) + (a << 3); - a = (a ^ 0xb55a4f09) ^ (a >> 16); - return a; + a = (a + 0x7ed55d16) + (a << 12); + a = (a ^ 0xc761c23c) ^ (a >> 19); + a = (a + 0x165667b1) + (a << 5); + a = (a + 0xd3a2646c) ^ (a << 9); + a = (a + 0xfd7046c5) + (a << 3); + a = (a ^ 0xb55a4f09) ^ (a >> 16); + return a; } /** @@ -113,10 +116,10 @@ __host__ __device__ unsigned int hash(unsigned int a) { * Function for generating a random vec3. */ __host__ __device__ glm::vec3 generateRandomVec3(float time, int index) { - thrust::default_random_engine rng(hash((int)(index * time))); - thrust::uniform_real_distribution unitDistrib(-1, 1); + thrust::default_random_engine rng(hash((int)(index * time))); + thrust::uniform_real_distribution unitDistrib(-1, 1); - return glm::vec3((float)unitDistrib(rng), (float)unitDistrib(rng), (float)unitDistrib(rng)); + return glm::vec3((float)unitDistrib(rng), (float)unitDistrib(rng), (float)unitDistrib(rng)); } /** @@ -124,52 +127,83 @@ __host__ __device__ glm::vec3 generateRandomVec3(float time, int index) { * CUDA kernel for generating boids with a specified mass randomly around the star. */ __global__ void kernGenerateRandomPosArray(int time, int N, glm::vec3 * arr, float scale) { - int index = (blockIdx.x * blockDim.x) + threadIdx.x; - if (index < N) { - glm::vec3 rand = generateRandomVec3(time, index); - arr[index].x = scale * rand.x; - arr[index].y = scale * rand.y; - arr[index].z = scale * rand.z; - } + int index = (blockIdx.x * blockDim.x) + threadIdx.x; + if (index < N) { + glm::vec3 rand = generateRandomVec3(time, index); + arr[index].x = scale * rand.x; + arr[index].y = scale * rand.y; + arr[index].z = scale * rand.z; + } } /** * Initialize memory, update some globals */ void Boids::initSimulation(int N) { - numObjects = N; - dim3 fullBlocksPerGrid((N + blockSize - 1) / blockSize); - - // LOOK-1.2 - This is basic CUDA memory management and error checking. - // Don't forget to cudaFree in Boids::endSimulation. - cudaMalloc((void**)&dev_pos, N * sizeof(glm::vec3)); - checkCUDAErrorWithLine("cudaMalloc dev_pos failed!"); - - cudaMalloc((void**)&dev_vel1, N * sizeof(glm::vec3)); - checkCUDAErrorWithLine("cudaMalloc dev_vel1 failed!"); - - cudaMalloc((void**)&dev_vel2, N * sizeof(glm::vec3)); - checkCUDAErrorWithLine("cudaMalloc dev_vel2 failed!"); - - // LOOK-1.2 - This is a typical CUDA kernel invocation. - kernGenerateRandomPosArray<<>>(1, numObjects, - dev_pos, scene_scale); - checkCUDAErrorWithLine("kernGenerateRandomPosArray failed!"); - - // LOOK-2.1 computing grid params - gridCellWidth = 2.0f * std::max(std::max(rule1Distance, rule2Distance), rule3Distance); - int halfSideCount = (int)(scene_scale / gridCellWidth) + 1; - gridSideCount = 2 * halfSideCount; - - gridCellCount = gridSideCount * gridSideCount * gridSideCount; - gridInverseCellWidth = 1.0f / gridCellWidth; - float halfGridWidth = gridCellWidth * halfSideCount; - gridMinimum.x -= halfGridWidth; - gridMinimum.y -= halfGridWidth; - gridMinimum.z -= halfGridWidth; - - // TODO-2.1 TODO-2.3 - Allocate additional buffers here. - cudaThreadSynchronize(); + numObjects = N; + dim3 fullBlocksPerGrid((N + blockSize - 1) / blockSize); + + // LOOK-1.2 - This is basic CUDA memory management and error checking. + // Don't forget to cudaFree in Boids::endSimulation. + cudaMalloc((void**)&dev_pos, N * sizeof(glm::vec3)); + checkCUDAErrorWithLine("cudaMalloc dev_pos failed!"); + + cudaMalloc((void**)&dev_vel1, N * sizeof(glm::vec3)); + checkCUDAErrorWithLine("cudaMalloc dev_vel1 failed!"); + + cudaMalloc((void**)&dev_vel2, N * sizeof(glm::vec3)); + checkCUDAErrorWithLine("cudaMalloc dev_vel2 failed!"); + + cudaMalloc((void**)&dev_particleArrayIndices, N * sizeof(int)); + checkCUDAErrorWithLine("cudaMalloc dev_particleArrayIndices failed!"); + + cudaMalloc((void**)&dev_particleGridIndices, N * sizeof(int)); + checkCUDAErrorWithLine("cudaMalloc dev_particleGridIndices failed!"); + + cudaMalloc((void**)&dev_pos_new, N * sizeof(glm::vec3)); + checkCUDAErrorWithLine("cudaMalloc dev_pos_new failed!"); + + cudaMalloc((void**)&dev_vel1_new, N * sizeof(glm::vec3)); + checkCUDAErrorWithLine("cudaMalloc dev_vel1_new failed!"); + + cudaMalloc((void**)&dev_vel2_new, N * sizeof(glm::vec3)); + checkCUDAErrorWithLine("cudaMalloc dev_vel2_new failed!"); + + //Initialize thrust pointers + dev_thrust_particleArrayIndices = thrust::device_pointer_cast(dev_particleArrayIndices); + dev_thrust_particleGridIndices = thrust::device_pointer_cast(dev_particleGridIndices); + + // LOOK-1.2 - This is a typical CUDA kernel invocation. + kernGenerateRandomPosArray << > >(1, numObjects, + dev_pos, scene_scale); + checkCUDAErrorWithLine("kernGenerateRandomPosArray failed!"); + + // LOOK-2.1 computing grid params + gridCellWidth = 2.0f * std::max(std::max(rule1Distance, rule2Distance), rule3Distance); + int halfSideCount = (int)(scene_scale / gridCellWidth) + 1; + gridSideCount = 2 * halfSideCount; + + gridCellCount = gridSideCount * gridSideCount * gridSideCount; + gridInverseCellWidth = 1.0f / gridCellWidth; + float halfGridWidth = gridCellWidth * halfSideCount; + gridMinimum.x -= halfGridWidth; + gridMinimum.y -= halfGridWidth; + gridMinimum.z -= halfGridWidth; + + cudaMalloc((void**)&dev_gridCellStartIndices, gridCellCount * sizeof(int)); + checkCUDAErrorWithLine("cudaMalloc dev_gridCellStartIndices failed!"); + + cudaMalloc((void**)&dev_gridCellEndIndices, gridCellCount * sizeof(int)); + checkCUDAErrorWithLine("cudaMalloc dev_gridCellEndIndices failed!"); + + /*kernResetIntBuffer << > >(N, dev_gridCellStartIndices, -1); + checkCUDAErrorWithLine("kernResetIntBuffer failed!"); + + kernResetIntBuffer << > >(N, dev_gridCellEndIndices, -1); + checkCUDAErrorWithLine("kernResetIntBuffer failed!");*/ + + // TODO-2.1 TODO-2.3 - Allocate additional buffers here. + cudaThreadSynchronize(); } @@ -181,41 +215,41 @@ void Boids::initSimulation(int N) { * Copy the boid positions into the VBO so that they can be drawn by OpenGL. */ __global__ void kernCopyPositionsToVBO(int N, glm::vec3 *pos, float *vbo, float s_scale) { - int index = threadIdx.x + (blockIdx.x * blockDim.x); + int index = threadIdx.x + (blockIdx.x * blockDim.x); - float c_scale = -1.0f / s_scale; + float c_scale = -1.0f / s_scale; - if (index < N) { - vbo[4 * index + 0] = pos[index].x * c_scale; - vbo[4 * index + 1] = pos[index].y * c_scale; - vbo[4 * index + 2] = pos[index].z * c_scale; - vbo[4 * index + 3] = 1.0f; - } + if (index < N) { + vbo[4 * index + 0] = pos[index].x * c_scale; + vbo[4 * index + 1] = pos[index].y * c_scale; + vbo[4 * index + 2] = pos[index].z * c_scale; + vbo[4 * index + 3] = 1.0f; + } } __global__ void kernCopyVelocitiesToVBO(int N, glm::vec3 *vel, float *vbo, float s_scale) { - int index = threadIdx.x + (blockIdx.x * blockDim.x); - - if (index < N) { - vbo[4 * index + 0] = vel[index].x + 0.3f; - vbo[4 * index + 1] = vel[index].y + 0.3f; - vbo[4 * index + 2] = vel[index].z + 0.3f; - vbo[4 * index + 3] = 1.0f; - } + int index = threadIdx.x + (blockIdx.x * blockDim.x); + + if (index < N) { + vbo[4 * index + 0] = vel[index].x + 0.3f; + vbo[4 * index + 1] = vel[index].y + 0.3f; + vbo[4 * index + 2] = vel[index].z + 0.3f; + vbo[4 * index + 3] = 1.0f; + } } /** * Wrapper for call to the kernCopyboidsToVBO CUDA kernel. */ void Boids::copyBoidsToVBO(float *vbodptr_positions, float *vbodptr_velocities) { - dim3 fullBlocksPerGrid((numObjects + blockSize - 1) / blockSize); + dim3 fullBlocksPerGrid((numObjects + blockSize - 1) / blockSize); - kernCopyPositionsToVBO << > >(numObjects, dev_pos, vbodptr_positions, scene_scale); - kernCopyVelocitiesToVBO << > >(numObjects, dev_vel1, vbodptr_velocities, scene_scale); + kernCopyPositionsToVBO << > >(numObjects, dev_pos, vbodptr_positions, scene_scale); + kernCopyVelocitiesToVBO << > >(numObjects, dev_vel1, vbodptr_velocities, scene_scale); - checkCUDAErrorWithLine("copyBoidsToVBO failed!"); + checkCUDAErrorWithLine("copyBoidsToVBO failed!"); - cudaThreadSynchronize(); + cudaThreadSynchronize(); } @@ -230,10 +264,34 @@ void Boids::copyBoidsToVBO(float *vbodptr_positions, float *vbodptr_velocities) * in the `pos` and `vel` arrays. */ __device__ glm::vec3 computeVelocityChange(int N, int iSelf, const glm::vec3 *pos, const glm::vec3 *vel) { - // Rule 1: boids fly towards their local perceived center of mass, which excludes themselves - // Rule 2: boids try to stay a distance d away from each other - // Rule 3: boids try to match the speed of surrounding boids - return glm::vec3(0.0f, 0.0f, 0.0f); + glm::vec3 center(0.0); + float neighborCount = 0; + glm::vec3 cohesion(0.0); + glm::vec3 separation(0.0); + glm::vec3 alignment(0.0); + for (int i = 0; i < N; i++) { + if (i == iSelf) continue; + float distance = glm::length(pos[i] - pos[iSelf]); + // Rule 1: boids fly towards their local perceived center of mass, which excludes themselves + if (distance < rule1Distance) { + center += pos[i]; + neighborCount += 1.0; + } + // Rule 2: boids try to stay a distance d away from each other + if (distance < rule2Distance) { + separation -= pos[i] - pos[iSelf]; + } + // Rule 3: boids try to match the speed of surrounding boids + if (distance < rule3Distance) { + alignment += vel[i]; + } + } + if (neighborCount > 0) { + center /= neighborCount; + cohesion = center - pos[iSelf]; + } + + return cohesion * rule1Scale + separation * rule2Scale + alignment * rule3Scale; } /** @@ -241,10 +299,20 @@ __device__ glm::vec3 computeVelocityChange(int N, int iSelf, const glm::vec3 *po * For each of the `N` bodies, update its position based on its current velocity. */ __global__ void kernUpdateVelocityBruteForce(int N, glm::vec3 *pos, - glm::vec3 *vel1, glm::vec3 *vel2) { - // Compute a new velocity based on pos and vel1 - // Clamp the speed - // Record the new velocity into vel2. Question: why NOT vel1? + glm::vec3 *vel1, glm::vec3 *vel2) { + int index = threadIdx.x + (blockIdx.x * blockDim.x); + if (index >= N) { + return; + } + + // Compute a new velocity based on pos and vel1 + glm::vec3 newVel = vel1[index] + computeVelocityChange(N, index, pos, vel1); + // Clamp the speed + if (glm::length(newVel) > maxSpeed) { + newVel = glm::normalize(newVel) * maxSpeed; + } + // Record the new velocity into vel2. Question: why NOT vel1? + vel2[index] = newVel; } /** @@ -252,24 +320,24 @@ __global__ void kernUpdateVelocityBruteForce(int N, glm::vec3 *pos, * For each of the `N` bodies, update its position based on its current velocity. */ __global__ void kernUpdatePos(int N, float dt, glm::vec3 *pos, glm::vec3 *vel) { - // Update position by velocity - int index = threadIdx.x + (blockIdx.x * blockDim.x); - if (index >= N) { - return; - } - glm::vec3 thisPos = pos[index]; - thisPos += vel[index] * dt; - - // Wrap the boids around so we don't lose them - thisPos.x = thisPos.x < -scene_scale ? scene_scale : thisPos.x; - thisPos.y = thisPos.y < -scene_scale ? scene_scale : thisPos.y; - thisPos.z = thisPos.z < -scene_scale ? scene_scale : thisPos.z; - - thisPos.x = thisPos.x > scene_scale ? -scene_scale : thisPos.x; - thisPos.y = thisPos.y > scene_scale ? -scene_scale : thisPos.y; - thisPos.z = thisPos.z > scene_scale ? -scene_scale : thisPos.z; - - pos[index] = thisPos; + // Update position by velocity + int index = threadIdx.x + (blockIdx.x * blockDim.x); + if (index >= N) { + return; + } + glm::vec3 thisPos = pos[index]; + thisPos += vel[index] * dt; + + // Wrap the boids around so we don't lose them + thisPos.x = thisPos.x < -scene_scale ? scene_scale : thisPos.x; + thisPos.y = thisPos.y < -scene_scale ? scene_scale : thisPos.y; + thisPos.z = thisPos.z < -scene_scale ? scene_scale : thisPos.z; + + thisPos.x = thisPos.x > scene_scale ? -scene_scale : thisPos.x; + thisPos.y = thisPos.y > scene_scale ? -scene_scale : thisPos.y; + thisPos.z = thisPos.z > scene_scale ? -scene_scale : thisPos.z; + + pos[index] = thisPos; } // LOOK-2.1 Consider this method of computing a 1D index from a 3D grid index. @@ -279,181 +347,386 @@ __global__ void kernUpdatePos(int N, float dt, glm::vec3 *pos, glm::vec3 *vel) { // for(y) // for(z)? Or some other order? __device__ int gridIndex3Dto1D(int x, int y, int z, int gridResolution) { - return x + y * gridResolution + z * gridResolution * gridResolution; + return x + y * gridResolution + z * gridResolution * gridResolution; } __global__ void kernComputeIndices(int N, int gridResolution, - glm::vec3 gridMin, float inverseCellWidth, - glm::vec3 *pos, int *indices, int *gridIndices) { - // TODO-2.1 - // - Label each boid with the index of its grid cell. - // - Set up a parallel array of integer indices as pointers to the actual - // boid data in pos and vel1/vel2 + glm::vec3 gridMin, float inverseCellWidth, + glm::vec3 *pos, int *indices, int *gridIndices) { + int index = threadIdx.x + (blockIdx.x * blockDim.x); + if (index >= N) { + return; + } + // TODO-2.1 + // - Label each boid with the index of its grid cell. + glm::ivec3 gridIndex3D = (pos[index] - gridMin) * inverseCellWidth; + int gridIndex = gridIndex3Dto1D(gridIndex3D.x, gridIndex3D.y, gridIndex3D.z, gridResolution); + gridIndices[index] = gridIndex; + + // - Set up a parallel array of integer indices as pointers to the actual + // boid data in pos and vel1/vel2 + indices[index] = index; } // LOOK-2.1 Consider how this could be useful for indicating that a cell // does not enclose any boids __global__ void kernResetIntBuffer(int N, int *intBuffer, int value) { - int index = (blockIdx.x * blockDim.x) + threadIdx.x; - if (index < N) { - intBuffer[index] = value; - } + int index = (blockIdx.x * blockDim.x) + threadIdx.x; + if (index < N) { + intBuffer[index] = value; + } } __global__ void kernIdentifyCellStartEnd(int N, int *particleGridIndices, - int *gridCellStartIndices, int *gridCellEndIndices) { - // TODO-2.1 - // Identify the start point of each cell in the gridIndices array. - // This is basically a parallel unrolling of a loop that goes - // "this index doesn't match the one before it, must be a new cell!" + int *gridCellStartIndices, int *gridCellEndIndices) { + int index = (blockIdx.x * blockDim.x) + threadIdx.x; + if (index < N) { + // TODO-2.1 + // Identify the start point of each cell in the gridIndices array. + // This is basically a parallel unrolling of a loop that goes + // "this index doesn't match the one before it, must be a new cell!" + int gridIndex = particleGridIndices[index]; + + if (index == 0 || gridIndex != particleGridIndices[index - 1]) { + gridCellStartIndices[gridIndex] = index; + } + + if (index == N - 1 || gridIndex != particleGridIndices[index + 1]) { + gridCellEndIndices[gridIndex] = index; + } + } +} + +__global__ void kernShufflePosAndVel(int N, int *particleArrayIndices, + glm::vec3 *pos, glm::vec3 *vel1, glm::vec3 *vel2, glm::vec3 *pos_new, glm::vec3 *vel1_new, glm::vec3 *vel2_new) { + int index = (blockIdx.x * blockDim.x) + threadIdx.x; + if (index < N) { + //Rearrange pos and vel1 into pos_new and vel1_new + int boidIndex = particleArrayIndices[index]; + pos_new[index] = pos[boidIndex]; + vel1_new[index] = vel1[boidIndex]; + vel2_new[index] = vel2[boidIndex]; + } } __global__ void kernUpdateVelNeighborSearchScattered( - int N, int gridResolution, glm::vec3 gridMin, - float inverseCellWidth, float cellWidth, - int *gridCellStartIndices, int *gridCellEndIndices, - int *particleArrayIndices, - glm::vec3 *pos, glm::vec3 *vel1, glm::vec3 *vel2) { - // TODO-2.1 - Update a boid's velocity using the uniform grid to reduce - // the number of boids that need to be checked. - // - Identify the grid cell that this particle is in - // - Identify which cells may contain neighbors. This isn't always 8. - // - For each cell, read the start/end indices in the boid pointer array. - // - Access each boid in the cell and compute velocity change from - // the boids rules, if this boid is within the neighborhood distance. - // - Clamp the speed change before putting the new speed in vel2 + int N, int gridResolution, glm::vec3 gridMin, + float inverseCellWidth, float cellWidth, + int *gridCellStartIndices, int *gridCellEndIndices, + int *particleArrayIndices, + glm::vec3 *pos, glm::vec3 *vel1, glm::vec3 *vel2) { + int index = threadIdx.x + (blockIdx.x * blockDim.x); + if (index >= N) { + return; + } + // TODO-2.1 - Update a boid's velocity using the uniform grid to reduce + // the number of boids that need to be checked. + // - Identify the grid cell that this particle is in + glm::ivec3 gridIndex3D = (pos[index] - gridMin) * inverseCellWidth; + int xIndex = gridIndex3D.x; + int yIndex = gridIndex3D.y; + int zIndex = gridIndex3D.z; + + // - For each cell, read the start/end indices in the boid pointer array. + // - Access each boid in the cell and compute velocity change from + // the boids rules, if this boid is within the neighborhood distance. + glm::vec3 center(0.0); + float neighborCount = 0; + glm::vec3 cohesion(0.0); + glm::vec3 separation(0.0); + glm::vec3 alignment(0.0); + + for (int z = imax(zIndex - 1, 0); z <= imin(zIndex + 1, N-1); z++) { + for (int y = imax(yIndex - 1, 0); y <= imin(yIndex + 1, N - 1); y++) { + for (int x = imax(xIndex - 1, 0); x <= imin(xIndex + 1, N - 1); x++) { + if (x < 0 || x >= gridResolution + || y < 0 || y >= gridResolution + || z < 0 || z >= gridResolution) { + continue; + } + + int neighborCellIndex = gridIndex3Dto1D(x, y, z, gridResolution); + int startIndex = gridCellStartIndices[neighborCellIndex]; + int endIndex = gridCellEndIndices[neighborCellIndex]; + + for (int i = startIndex; i < endIndex; i++) { + int thatBoidIndex = particleArrayIndices[i]; + if (index == thatBoidIndex) continue; + float distance = glm::length(pos[thatBoidIndex] - pos[index]); + if (distance < rule1Distance) { + center += pos[thatBoidIndex]; + neighborCount += 1.0; + } + if (distance < rule2Distance) { + separation -= pos[thatBoidIndex] - pos[index]; + } + if (distance < rule3Distance) { + alignment += vel1[thatBoidIndex]; + } + } + } + } + } + + if (neighborCount > 0) { + center /= neighborCount; + cohesion = center - pos[index]; + } + + glm::vec3 newVel = vel1[index] + cohesion * rule1Scale + separation * rule2Scale + alignment * rule3Scale; + // - Clamp the speed change before putting the new speed in vel2 + if (glm::length(newVel) > maxSpeed) { + newVel = glm::normalize(newVel) * maxSpeed; + } + vel2[index] = newVel; } __global__ void kernUpdateVelNeighborSearchCoherent( - int N, int gridResolution, glm::vec3 gridMin, - float inverseCellWidth, float cellWidth, - int *gridCellStartIndices, int *gridCellEndIndices, - glm::vec3 *pos, glm::vec3 *vel1, glm::vec3 *vel2) { - // TODO-2.3 - This should be very similar to kernUpdateVelNeighborSearchScattered, - // except with one less level of indirection. - // This should expect gridCellStartIndices and gridCellEndIndices to refer - // directly to pos and vel1. - // - Identify the grid cell that this particle is in - // - Identify which cells may contain neighbors. This isn't always 8. - // - For each cell, read the start/end indices in the boid pointer array. - // DIFFERENCE: For best results, consider what order the cells should be - // checked in to maximize the memory benefits of reordering the boids data. - // - Access each boid in the cell and compute velocity change from - // the boids rules, if this boid is within the neighborhood distance. - // - Clamp the speed change before putting the new speed in vel2 + int N, int gridResolution, glm::vec3 gridMin, + float inverseCellWidth, float cellWidth, + int *gridCellStartIndices, int *gridCellEndIndices, + glm::vec3 *pos, glm::vec3 *vel1, glm::vec3 *vel2) { + int index = threadIdx.x + (blockIdx.x * blockDim.x); + if (index >= N) { + return; + } + // TODO-2.3 - This should be very similar to kernUpdateVelNeighborSearchScattered, + // except with one less level of indirection. + // This should expect gridCellStartIndices and gridCellEndIndices to refer + // directly to pos and vel1. + // - Identify the grid cell that this particle is in + glm::ivec3 gridIndex3D = (pos[index] - gridMin) * inverseCellWidth; + int xIndex = gridIndex3D.x; + int yIndex = gridIndex3D.y; + int zIndex = gridIndex3D.z; + + // - Identify which cells may contain neighbors. This isn't always 8. + // - For each cell, read the start/end indices in the boid pointer array. + // DIFFERENCE: For best results, consider what order the cells should be + // checked in to maximize the memory benefits of reordering the boids data. + // - Access each boid in the cell and compute velocity change from + // the boids rules, if this boid is within the neighborhood distance. + glm::vec3 center(0.0); + float neighborCount = 0; + glm::vec3 cohesion(0.0); + glm::vec3 separation(0.0); + glm::vec3 alignment(0.0); + + for (int z = imax(zIndex - 1, 0); z <= imin(zIndex + 1, N - 1); z++) { + for (int y = imax(yIndex - 1, 0); y <= imin(yIndex + 1, N - 1); y++) { + for (int x = imax(xIndex - 1, 0); x <= imin(xIndex + 1, N - 1); x++) { + if (x < 0 || x >= gridResolution + || y < 0 || y >= gridResolution + || z < 0 || z >= gridResolution) { + continue; + } + + int neighborCellIndex = gridIndex3Dto1D(x, y, z, gridResolution); + int startIndex = gridCellStartIndices[neighborCellIndex]; + int endIndex = gridCellEndIndices[neighborCellIndex]; + + for (int i = startIndex; i < endIndex; i++) { + if (index == i) continue; + float distance = glm::length(pos[i] - pos[index]); + if (distance < rule1Distance) { + center += pos[i]; + neighborCount += 1.0; + } + if (distance < rule2Distance) { + separation -= pos[i] - pos[index]; + } + if (distance < rule3Distance) { + alignment += vel1[i]; + } + } + } + } + } + + if (neighborCount > 0) { + center /= neighborCount; + cohesion = center - pos[index]; + } + + glm::vec3 newVel = vel1[index] + cohesion * rule1Scale + separation * rule2Scale + alignment * rule3Scale; + // - Clamp the speed change before putting the new speed in vel2 + if (glm::length(newVel) > maxSpeed) { + newVel = glm::normalize(newVel) * maxSpeed; + } + vel2[index] = newVel; } /** * Step the entire N-body simulation by `dt` seconds. */ void Boids::stepSimulationNaive(float dt) { - // TODO-1.2 - use the kernels you wrote to step the simulation forward in time. - // TODO-1.2 ping-pong the velocity buffers + dim3 fullBlocksPerGrid((numObjects + blockSize - 1) / blockSize); + // TODO-1.2 - use the kernels you wrote to step the simulation forward in time. + kernUpdateVelocityBruteForce << > >(numObjects, dev_pos, dev_vel1, dev_vel2); + checkCUDAErrorWithLine("kernUpdateVelocityBruteForce failed!"); + + kernUpdatePos << > >(numObjects, dt, dev_pos, dev_vel2); + checkCUDAErrorWithLine("kernUpdatePos failed!"); + + // TODO-1.2 ping-pong the velocity buffers + std::swap(dev_vel1, dev_vel2); } void Boids::stepSimulationScatteredGrid(float dt) { - // TODO-2.1 - // Uniform Grid Neighbor search using Thrust sort. - // In Parallel: - // - label each particle with its array index as well as its grid index. - // Use 2x width grids. - // - Unstable key sort using Thrust. A stable sort isn't necessary, but you - // are welcome to do a performance comparison. - // - Naively unroll the loop for finding the start and end indices of each - // cell's data pointers in the array of boid indices - // - Perform velocity updates using neighbor search - // - Update positions - // - Ping-pong buffers as needed + dim3 fullBlocksPerGrid((numObjects + blockSize - 1) / blockSize); + // TODO-2.1 + // Uniform Grid Neighbor search using Thrust sort. + // In Parallel: + // - label each particle with its array index as well as its grid index. + // Use 2x width grids. + kernComputeIndices << > >(numObjects, gridSideCount, gridMinimum, + gridInverseCellWidth, dev_pos, dev_particleArrayIndices, dev_particleGridIndices); + checkCUDAErrorWithLine("kernComputeIndices failed!"); + + thrust::sort_by_key(dev_thrust_particleGridIndices, dev_thrust_particleGridIndices + numObjects, dev_thrust_particleArrayIndices); + + // - Naively unroll the loop for finding the start and end indices of each + // cell's data pointers in the array of boid indices + kernIdentifyCellStartEnd << > >(numObjects, dev_particleGridIndices, + dev_gridCellStartIndices, dev_gridCellEndIndices); + checkCUDAErrorWithLine("kernIdentifyCellStartEnd failed!"); + + // - Perform velocity updates using neighbor search + kernUpdateVelNeighborSearchScattered << > >(numObjects, gridSideCount, + gridMinimum, gridInverseCellWidth, gridCellWidth, dev_gridCellStartIndices, dev_gridCellEndIndices, + dev_particleArrayIndices, dev_pos, dev_vel1, dev_vel2); + checkCUDAErrorWithLine("kernUpdateVelNeighborSearchScattered failed!"); + + // - Update positions + kernUpdatePos << > >(numObjects, dt, dev_pos, dev_vel2); + checkCUDAErrorWithLine("kernUpdatePos failed!"); + + // - Ping-pong buffers as needed + std::swap(dev_vel1, dev_vel2); + } void Boids::stepSimulationCoherentGrid(float dt) { - // TODO-2.3 - start by copying Boids::stepSimulationNaiveGrid - // Uniform Grid Neighbor search using Thrust sort on cell-coherent data. - // In Parallel: - // - Label each particle with its array index as well as its grid index. - // Use 2x width grids - // - Unstable key sort using Thrust. A stable sort isn't necessary, but you - // are welcome to do a performance comparison. - // - Naively unroll the loop for finding the start and end indices of each - // cell's data pointers in the array of boid indices - // - BIG DIFFERENCE: use the rearranged array index buffer to reshuffle all - // the particle data in the simulation array. - // CONSIDER WHAT ADDITIONAL BUFFERS YOU NEED - // - Perform velocity updates using neighbor search - // - Update positions - // - Ping-pong buffers as needed. THIS MAY BE DIFFERENT FROM BEFORE. + dim3 fullBlocksPerGrid((numObjects + blockSize - 1) / blockSize); + // TODO-2.3 - start by copying Boids::stepSimulationNaiveGrid + // Uniform Grid Neighbor search using Thrust sort on cell-coherent data. + // In Parallel: + // - Label each particle with its array index as well as its grid index. + // Use 2x width grids + kernComputeIndices << > >(numObjects, gridSideCount, gridMinimum, + gridInverseCellWidth, dev_pos, dev_particleArrayIndices, dev_particleGridIndices); + checkCUDAErrorWithLine("kernComputeIndices failed!"); + + // - Unstable key sort using Thrust. A stable sort isn't necessary, but you + // are welcome to do a performance comparison. + thrust::sort_by_key(dev_thrust_particleGridIndices, dev_thrust_particleGridIndices + numObjects, dev_thrust_particleArrayIndices); + + // - Naively unroll the loop for finding the start and end indices of each + // cell's data pointers in the array of boid indices + kernIdentifyCellStartEnd << > >(numObjects, dev_particleGridIndices, + dev_gridCellStartIndices, dev_gridCellEndIndices); + checkCUDAErrorWithLine("kernIdentifyCellStartEnd failed!"); + + // - BIG DIFFERENCE: use the rearranged array index buffer to reshuffle all + // the particle data in the simulation array. + // CONSIDER WHAT ADDITIONAL BUFFERS YOU NEED + kernShufflePosAndVel << > >(numObjects, dev_particleArrayIndices, + dev_pos, dev_vel1, dev_vel2, dev_pos_new, dev_vel1_new, dev_vel2_new); + checkCUDAErrorWithLine("kernShufflePosAndVel failed!"); + + // - Perform velocity updates using neighbor search + kernUpdateVelNeighborSearchCoherent << > >(numObjects, gridSideCount, + gridMinimum, gridInverseCellWidth, gridCellWidth, dev_gridCellStartIndices, dev_gridCellEndIndices, + dev_pos_new, dev_vel1_new, dev_vel2_new); + checkCUDAErrorWithLine("kernUpdateVelNeighborSearchCoherent failed!"); + + // - Update positions + kernUpdatePos << > >(numObjects, dt, dev_pos_new, dev_vel2_new); + checkCUDAErrorWithLine("kernUpdatePos failed!"); + + // - Ping-pong buffers as needed. THIS MAY BE DIFFERENT FROM BEFORE. + std::swap(dev_pos, dev_pos_new); + std::swap(dev_vel1, dev_vel2_new); + } void Boids::endSimulation() { - cudaFree(dev_vel1); - cudaFree(dev_vel2); - cudaFree(dev_pos); - - // TODO-2.1 TODO-2.3 - Free any additional buffers here. + cudaFree(dev_vel1); + cudaFree(dev_vel2); + cudaFree(dev_pos); + + // TODO-2.1 TODO-2.3 - Free any additional buffers here. + cudaFree(dev_particleArrayIndices); + cudaFree(dev_particleGridIndices); + cudaFree(dev_gridCellStartIndices); + cudaFree(dev_gridCellEndIndices); + cudaFree(dev_vel1_new); + cudaFree(dev_vel2_new); + cudaFree(dev_pos_new); } void Boids::unitTest() { - // LOOK-1.2 Feel free to write additional tests here. - - // test unstable sort - int *dev_intKeys; - int *dev_intValues; - int N = 10; - - int *intKeys = new int[N]; - int *intValues = new int[N]; - - intKeys[0] = 0; intValues[0] = 0; - intKeys[1] = 1; intValues[1] = 1; - intKeys[2] = 0; intValues[2] = 2; - intKeys[3] = 3; intValues[3] = 3; - intKeys[4] = 0; intValues[4] = 4; - intKeys[5] = 2; intValues[5] = 5; - intKeys[6] = 2; intValues[6] = 6; - intKeys[7] = 0; intValues[7] = 7; - intKeys[8] = 5; intValues[8] = 8; - intKeys[9] = 6; intValues[9] = 9; - - cudaMalloc((void**)&dev_intKeys, N * sizeof(int)); - checkCUDAErrorWithLine("cudaMalloc dev_intKeys failed!"); - - cudaMalloc((void**)&dev_intValues, N * sizeof(int)); - checkCUDAErrorWithLine("cudaMalloc dev_intValues failed!"); - - dim3 fullBlocksPerGrid((N + blockSize - 1) / blockSize); - - std::cout << "before unstable sort: " << std::endl; - for (int i = 0; i < N; i++) { - std::cout << " key: " << intKeys[i]; - std::cout << " value: " << intValues[i] << std::endl; - } - - // How to copy data to the GPU - cudaMemcpy(dev_intKeys, intKeys, sizeof(int) * N, cudaMemcpyHostToDevice); - cudaMemcpy(dev_intValues, intValues, sizeof(int) * N, cudaMemcpyHostToDevice); - - // Wrap device vectors in thrust iterators for use with thrust. - thrust::device_ptr dev_thrust_keys(dev_intKeys); - thrust::device_ptr dev_thrust_values(dev_intValues); - // LOOK-2.1 Example for using thrust::sort_by_key - thrust::sort_by_key(dev_thrust_keys, dev_thrust_keys + N, dev_thrust_values); - - // How to copy data back to the CPU side from the GPU - cudaMemcpy(intKeys, dev_intKeys, sizeof(int) * N, cudaMemcpyDeviceToHost); - cudaMemcpy(intValues, dev_intValues, sizeof(int) * N, cudaMemcpyDeviceToHost); - checkCUDAErrorWithLine("memcpy back failed!"); - - std::cout << "after unstable sort: " << std::endl; - for (int i = 0; i < N; i++) { - std::cout << " key: " << intKeys[i]; - std::cout << " value: " << intValues[i] << std::endl; - } - - // cleanup - delete(intKeys); - delete(intValues); - cudaFree(dev_intKeys); - cudaFree(dev_intValues); - checkCUDAErrorWithLine("cudaFree failed!"); - return; + // LOOK-1.2 Feel free to write additional tests here. + + // test unstable sort + int *dev_intKeys; + int *dev_intValues; + int N = 10; + + int *intKeys = new int[N]; + int *intValues = new int[N]; + + intKeys[0] = 0; intValues[0] = 0; + intKeys[1] = 1; intValues[1] = 1; + intKeys[2] = 0; intValues[2] = 2; + intKeys[3] = 3; intValues[3] = 3; + intKeys[4] = 0; intValues[4] = 4; + intKeys[5] = 2; intValues[5] = 5; + intKeys[6] = 2; intValues[6] = 6; + intKeys[7] = 0; intValues[7] = 7; + intKeys[8] = 5; intValues[8] = 8; + intKeys[9] = 6; intValues[9] = 9; + + cudaMalloc((void**)&dev_intKeys, N * sizeof(int)); + checkCUDAErrorWithLine("cudaMalloc dev_intKeys failed!"); + + cudaMalloc((void**)&dev_intValues, N * sizeof(int)); + checkCUDAErrorWithLine("cudaMalloc dev_intValues failed!"); + + dim3 fullBlocksPerGrid((N + blockSize - 1) / blockSize); + + std::cout << "before unstable sort: " << std::endl; + for (int i = 0; i < N; i++) { + std::cout << " key: " << intKeys[i]; + std::cout << " value: " << intValues[i] << std::endl; + } + + // How to copy data to the GPU + cudaMemcpy(dev_intKeys, intKeys, sizeof(int) * N, cudaMemcpyHostToDevice); + cudaMemcpy(dev_intValues, intValues, sizeof(int) * N, cudaMemcpyHostToDevice); + + // Wrap device vectors in thrust iterators for use with thrust. + thrust::device_ptr dev_thrust_keys(dev_intKeys); + thrust::device_ptr dev_thrust_values(dev_intValues); + // LOOK-2.1 Example for using thrust::sort_by_key + thrust::sort_by_key(dev_thrust_keys, dev_thrust_keys + N, dev_thrust_values); + + // How to copy data back to the CPU side from the GPU + cudaMemcpy(intKeys, dev_intKeys, sizeof(int) * N, cudaMemcpyDeviceToHost); + cudaMemcpy(intValues, dev_intValues, sizeof(int) * N, cudaMemcpyDeviceToHost); + checkCUDAErrorWithLine("memcpy back failed!"); + + std::cout << "after unstable sort: " << std::endl; + for (int i = 0; i < N; i++) { + std::cout << " key: " << intKeys[i]; + std::cout << " value: " << intValues[i] << std::endl; + } + + // cleanup + delete(intKeys); + delete(intValues); + cudaFree(dev_intKeys); + cudaFree(dev_intValues); + checkCUDAErrorWithLine("cudaFree failed!"); + return; } diff --git a/src/main.cpp b/src/main.cpp index e416836..a226db1 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -14,11 +14,11 @@ // LOOK-2.1 LOOK-2.3 - toggles for UNIFORM_GRID and COHERENT_GRID #define VISUALIZE 1 -#define UNIFORM_GRID 0 -#define COHERENT_GRID 0 +#define UNIFORM_GRID 1 +#define COHERENT_GRID 1 // LOOK-1.2 - change this to adjust particle count in the simulation -const int N_FOR_VIS = 5000; +const int N_FOR_VIS = 10000; const float DT = 0.2f; /** @@ -219,14 +219,21 @@ void initShaders(GLuint * program) { double fps = 0; double timebase = 0; int frame = 0; + int totalFrames = 0; Boids::unitTest(); // LOOK-1.2 We run some basic example code to make sure // your CUDA development setup is ready to go. + cudaEvent_t start, stop; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + cudaEventRecord(start); while (!glfwWindowShouldClose(window)) { glfwPollEvents(); frame++; + totalFrames++; double time = glfwGetTime(); if (time - timebase > 1.0) { @@ -237,6 +244,15 @@ void initShaders(GLuint * program) { runCUDA(); + //printf("totalFrames %d\n", totalFrames); + if (totalFrames == 500) { + cudaEventRecord(stop); + cudaEventSynchronize(stop); + float milliseconds = 0; + cudaEventElapsedTime(&milliseconds, start, stop); + printf("%f seconds elapsed.\n", milliseconds * .001); + } + std::ostringstream ss; ss << "["; ss.precision(1);