serial_no
int64
1
24.2k
cuda_source
stringlengths
11
9.01M
5,001
#include "includes.h" __global__ void cuda_dot(int N, double *a, double *b, double *c) { // __shared__ double localDot[threadsPerBlock]; /* Statically defined */ extern __shared__ double localDot[]; int ix = threadIdx.x + blockIdx.x * blockDim.x; int localIndex = threadIdx.x; double localSum = 0; while (ix < N) { localSum += a[ix] * b[ix]; /* Reduction is here */ ix += blockDim.x * gridDim.x; } /* Store sum computed by this thread */ localDot[localIndex] = localSum; /* Wait for all threads to get to this point */ __syncthreads(); /* Every block should add up sum computed on threads in the block */ int i = blockDim.x/2; while (i != 0) { if (localIndex < i) { localDot[localIndex] += localDot[localIndex + i]; } __syncthreads(); i /= 2; } /* Each block stores local dot product */ if (localIndex == 0) c[blockIdx.x] = localDot[0]; }
5,002
__global__ void add2( double * v1, const double * v2 ) { int idx = threadIdx.x; v1[idx] += v2[idx]; }
5,003
#include<cuda_runtime.h> #include<device_launch_parameters.h> #include<stdio.h> #include<stdlib.h> #include<string.h> __global__ void add(char * d_resbuffer , char * d_buffer, int * d_length) { int id = threadIdx.x ; int start = id * (*d_length); for(int i = 0 ; i<=(*d_length)-1;i++) { d_resbuffer[start] = d_buffer[i]; start ++ ; } } int main(void) { char buffer[100] ="Hello"; char res[100] = ""; int length = strlen(buffer) ; int duplicate = 4 ; char * d_resbuffer ; char * d_buffer ; int * d_length ; cudaMalloc((void **)&d_resbuffer,(length*duplicate+1) * sizeof(char)); cudaMalloc((void **)&d_buffer,(length+1)*sizeof(char)); cudaMalloc((void **)&d_length , sizeof(int)); cudaMemcpy(d_buffer,buffer,(length+1)*sizeof(char),cudaMemcpyHostToDevice); cudaMemcpy(d_length , &length , sizeof(int),cudaMemcpyHostToDevice); add<<<1,duplicate>>>(d_resbuffer,d_buffer,d_length); cudaError_t error = cudaGetLastError(); if(error!= cudaSuccess) { printf("%s\n",cudaGetErrorString(error)); } cudaMemcpy(res,d_resbuffer,(length*duplicate)*sizeof(char),cudaMemcpyDeviceToHost); int location = length * duplicate ; res[location]='\0'; printf("Result :%s",res); }
5,004
#include <cuda_runtime.h> #include <iostream> class CUDAdem { public: __device__ void add(int igdx) { printf("hello GPU = %d\n",igdx); } }; __global__ void add() { int igdx = threadIdx.x; CUDAdem cdmo; cdmo.add(igdx); } int main() { add<<<1,4>>>(); cudaDeviceReset(); printf("hello world!\n"); return 0; }
5,005
#include <iostream> #include <string> #include <stdio.h> #include <cuda.h> #include <fstream> using namespace std; __global__ void MatrixMulKernel(float *d_M, float *d_N, float *d_P,int width){ int Row = blockIdx.y*blockDim.y + threadIdx.y; int Col = blockIdx.x*blockDim.x + threadIdx.x; if ((Row < width)&&(Col < width)){ float Pvalue = 0; for (int i = 0; i < width; ++i){ Pvalue += d_M[Row*width+i]*d_N[i*width+Col]; } d_P[Row*width + Col] = Pvalue; } } int matrixMulHost(float *h_M, float *h_N, float *h_P, int width){ int Pvalue; for(int row = 0; row < width ; ++row){ for(int col = 0; col < width ; ++col){ Pvalue = 0; for(int k = 0; k < width ; ++k){ Pvalue += h_M[row*width+k] * h_N[k*width+col]; } h_P[row*width+col] = Pvalue; } } return 0; } int initValues(float *data, int width){ for(int i = 0; i < width*width; i++) data[i] = 2; return 0; } int main(int argc, char const *argv[]) { float *h_M, *h_N, *h_P,*h_P_d; float *d_M, *d_N,*d_P; std::string num = argv[1]; int width = std::stoi(num); int size = width * width * sizeof(float); clock_t start, end, startGPU, endGPU; double cpu_time_used, gpu_time_used, aceleration; for (int times = 0; times < 20; times++){ h_M = (float*)malloc(size); h_N = (float*)malloc(size); h_P = (float*)malloc(size); h_P_d = (float*)malloc(size); initValues(h_M, width); initValues(h_N, width); /////////Algoritmo Secuencial//////////////////////////////////////////// start = clock(); matrixMulHost(h_M, h_N, h_P, width); end = clock(); cpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC; printf("Tiempo algoritmo secuencial: %.10f\n", cpu_time_used); /////////Algoritmo Secuencial///////////////////////////////////////////// cudaMalloc((void**)&d_M,size); cudaMalloc((void**)&d_N,size); cudaMalloc((void**)&d_P,size); //////////////////////Algoritmo Paralelo/////////////////////////// startGPU = clock(); cudaMemcpy(d_M, h_M, size, cudaMemcpyHostToDevice); cudaMemcpy(d_N, h_N, size, cudaMemcpyHostToDevice); int blockSize = 32; dim3 dimBlock(blockSize,blockSize,1); dim3 dimGrid(ceil(width/float(blockSize)),ceil(width/float(blockSize)),1); MatrixMulKernel<<<dimGrid,dimBlock>>>(d_M,d_N,d_P,width); cudaDeviceSynchronize(); cudaMemcpy(h_P_d,d_P,size,cudaMemcpyDeviceToHost); endGPU = clock(); gpu_time_used = ((double) (endGPU - startGPU)) / CLOCKS_PER_SEC; aceleration = cpu_time_used/gpu_time_used; printf("Tiempo algoritmo paralelo: %.10f\n", gpu_time_used); printf("La aceleración obtenida es de %.10fX\n",aceleration); std::string name = "TimesMult.txt"+num; ofstream outfile(name,ios::binary | ios::app); outfile << gpu_time_used<<" "<< cpu_time_used <<" "<< aceleration << "\n"; outfile.close(); free(h_M); free(h_N); free(h_P); cudaFree(d_M); cudaFree(d_N); cudaFree(d_P); } return 0; }
5,006
#include <stdio.h> __device__ void decipher(unsigned int, unsigned int*, unsigned int const*); __global__ void decrypt_bytes(unsigned int *decrypted, unsigned int *encrypted, unsigned char *key) { //Get thread const int tx = threadIdx.x + (blockIdx.x * blockDim.x); unsigned int deciphered[2]; deciphered[0] = encrypted[0]; deciphered[1] = encrypted[1]; decipher(32, deciphered, (unsigned int*)key); if (tx == 0) { decrypted[0] = deciphered[0] ^ (unsigned int)1; decrypted[1] = deciphered[1] ^ (unsigned int)2; } //divide work on threads int i = (tx + 1) * 2; deciphered[0] = encrypted[i]; deciphered[1] = encrypted[i+1]; decipher(32, deciphered, (unsigned int*)key); decrypted[i] = deciphered[0] ^ encrypted[i-2]; decrypted[i+1] = deciphered[1] ^ encrypted[i-1]; } __global__ void reconstruct_secret(unsigned char *result, unsigned int *decrypted) { /* decrypted: pointer to the decrypted data result: pointer to where to store the unshuffled data */ //Get thread const int tx = threadIdx.x + (blockIdx.x * blockDim.x); //Divide work on each thread, max 10000 threads if (tx < 10000) { unsigned int element = decrypted[tx]; result[(element >> 8) % 10000] = element & 0xff; } } __device__ void decipher(unsigned int num_rounds, unsigned int v[2], unsigned int const key[4]) { /* num_rounds -- the number of iterations in the algorithm, 32 is reccomended input_data -- the input data to use, 32 bits of the first 2 elements are used key -- 128-bit key to use */ unsigned int i; unsigned int v0=v[0], v1=v[1], delta=0x9E3779B9, sum=delta*num_rounds; for (i=0; i < num_rounds; i++) { v1 -= (((v0 << 4) ^ (v0 >> 5)) + v0) ^ (sum + key[(sum>>11) & 3]); sum -= delta; v0 -= (((v1 << 4) ^ (v1 >> 5)) + v1) ^ (sum + key[sum & 3]); } v[0]=v0; v[1]=v1; }
5,007
#include <stdio.h> #include <stdlib.h> #include <time.h> #include <assert.h> #define BLOCK_SIZE 256 #define STR_SIZE 256 #define DEVICE 0 #define HALO 1 // halo width along one direction when advancing to the next iteration #define BENCH_PRINT void run(int argc, char** argv); int rows, cols; int* data; int** wall; int* result; #define M_SEED 9 int pyramid_height; //#define BENCH_PRINT void init(int argc, char** argv) { if(argc==4){ cols = atoi(argv[1]); rows = atoi(argv[2]); pyramid_height=atoi(argv[3]); }else{ printf("Usage: dynproc row_len col_len pyramid_height\n"); exit(0); } data = new int[rows*cols]; wall = new int*[rows]; for(int n=0; n<rows; n++) wall[n]=data+cols*n; result = new int[cols]; int seed = M_SEED; srand(seed); for (int i = 0; i < rows; i++) { for (int j = 0; j < cols; j++) { wall[i][j] = rand() % 10; } } #ifdef BENCH_PRINT for (int i = 0; i < rows; i++) { for (int j = 0; j < cols; j++) { printf("%d ",wall[i][j]) ; } printf("\n") ; } #endif } void fatal(char *s) { fprintf(stderr, "error: %s\n", s); } #define IN_RANGE(x, min, max) ((x)>=(min) && (x)<=(max)) #define CLAMP_RANGE(x, min, max) x = (x<(min)) ? min : ((x>(max)) ? max : x ) #define MIN(a, b) ((a)<=(b) ? (a) : (b)) __global__ void dynproc_kernel( int iteration, int *gpuWall, int *gpuSrc, int *gpuResults, int cols, int rows, int startStep, int border) { __shared__ int prev[BLOCK_SIZE]; __shared__ int result[BLOCK_SIZE]; int bx = blockIdx.x; int tx=threadIdx.x; // each block finally computes result for a small block // after N iterations. // it is the non-overlapping small blocks that cover // all the input data // calculate the small block size int small_block_cols = BLOCK_SIZE-iteration*HALO*2; // calculate the boundary for the block according to // the boundary of its small block int blkX = small_block_cols*bx-border; int blkXmax = blkX+BLOCK_SIZE-1; // calculate the global thread coordination int xidx = blkX+tx; // effective range within this block that falls within // the valid range of the input data // used to rule out computation outside the boundary. int validXmin = (blkX < 0) ? -blkX : 0; int validXmax = (blkXmax > cols-1) ? BLOCK_SIZE-1-(blkXmax-cols+1) : BLOCK_SIZE-1; int W = tx-1; int E = tx+1; W = (W < validXmin) ? validXmin : W; E = (E > validXmax) ? validXmax : E; bool isValid = IN_RANGE(tx, validXmin, validXmax); if(IN_RANGE(xidx, 0, cols-1)){ prev[tx] = gpuSrc[xidx]; } __syncthreads(); // [Ronny] Added sync to avoid race on prev Aug. 14 2012 bool computed; for (int i=0; i<iteration ; i++){ computed = false; if( IN_RANGE(tx, i+1, BLOCK_SIZE-i-2) && \ isValid){ computed = true; int left = prev[W]; int up = prev[tx]; int right = prev[E]; int shortest = MIN(left, up); shortest = MIN(shortest, right); int index = cols*(startStep+i)+xidx; result[tx] = shortest + gpuWall[index]; } __syncthreads(); if(i==iteration-1) break; if(computed) //Assign the computation range prev[tx]= result[tx]; __syncthreads(); // [Ronny] Added sync to avoid race on prev Aug. 14 2012 } // update the global memory // after the last iteration, only threads coordinated within the // small block perform the calculation and switch on ``computed'' if (computed){ gpuResults[xidx]=result[tx]; } } /* compute N time steps */ int calc_path(int *gpuWall, int *gpuResult[2], int rows, int cols, \ int pyramid_height, int blockCols, int borderCols) { dim3 dimBlock(BLOCK_SIZE); dim3 dimGrid(blockCols); int src = 1, dst = 0; for (int t = 0; t < rows-1; t+=pyramid_height) { int temp = src; src = dst; dst = temp; dynproc_kernel<<<dimGrid, dimBlock>>>( MIN(pyramid_height, rows-t-1), gpuWall, gpuResult[src], gpuResult[dst], cols,rows, t, borderCols); } return dst; } int main(int argc, char** argv) { int num_devices; cudaGetDeviceCount(&num_devices); if (num_devices > 1) cudaSetDevice(DEVICE); run(argc,argv); return EXIT_SUCCESS; } void run(int argc, char** argv) { init(argc, argv); /* --------------- pyramid parameters --------------- */ int borderCols = (pyramid_height)*HALO; int smallBlockCol = BLOCK_SIZE-(pyramid_height)*HALO*2; int blockCols = cols/smallBlockCol+((cols%smallBlockCol==0)?0:1); printf("pyramidHeight: %d\ngridSize: [%d]\nborder:[%d]\nblockSize: %d\nblockGrid:[%d]\ntargetBlock:[%d]\n",\ pyramid_height, cols, borderCols, BLOCK_SIZE, blockCols, smallBlockCol); int *gpuWall, *gpuResult[2]; int size = rows*cols; cudaMalloc((void**)&gpuResult[0], sizeof(int)*cols); cudaMalloc((void**)&gpuResult[1], sizeof(int)*cols); cudaMemcpy(gpuResult[0], data, sizeof(int)*cols, cudaMemcpyHostToDevice); cudaMalloc((void**)&gpuWall, sizeof(int)*(size-cols)); cudaMemcpy(gpuWall, data+cols, sizeof(int)*(size-cols), cudaMemcpyHostToDevice); int final_ret = calc_path(gpuWall, gpuResult, rows, cols, \ pyramid_height, blockCols, borderCols); cudaMemcpy(result, gpuResult[final_ret], sizeof(int)*cols, cudaMemcpyDeviceToHost); #ifdef BENCH_PRINT for (int i = 0; i < cols; i++) printf("%d ",data[i]) ; printf("\n") ; for (int i = 0; i < cols; i++) printf("%d ",result[i]) ; printf("\n") ; #endif cudaFree(gpuWall); cudaFree(gpuResult[0]); cudaFree(gpuResult[1]); delete [] data; delete [] wall; delete [] result; }
5,008
#include "includes.h" __global__ void Sum(float * A, float *B, float *C, int size) { int id = blockDim.x*blockIdx.y*gridDim.x + blockDim.x*blockIdx.x + threadIdx.x; if (id < size) { C[id] = A[id] + B[id]; } }
5,009
__global__ void thinEdgesGPU(int *mag, int *dir, int width, int height){ int y = blockIdx.y*blockDim.y + threadIdx.y + 1; int x = blockIdx.x*blockDim.x + threadIdx.x + 1; // Check whether thread is within image boundary if (x > width-2 || y > height-2) return; // Get gradient direction for current thread int tDir = dir[y*width + x]; // Transform offsets so we can find adjacent pixels in direction of gradient mag // 0 <= dir <= 3 3 2 1 // 0 P 0 // 1 2 3 int xOff = 1; int yOff = 0; if (tDir > 0){ xOff = tDir - 2; yOff = 1; } int adjPixel1 = mag[ (y+yOff)*width + x + xOff ]; int curPixel = mag[ y*width + x ]; int adjPixel2 = mag[ (y-yOff)*width + x - xOff ]; if ( adjPixel1 > curPixel || adjPixel2 > curPixel){ mag[y*width + x] = 0; } }
5,010
// System includes #include <stdio.h> #include <assert.h> #include <iostream> #include <numeric> #include <stdlib.h> // CUDA runtime #include <cuda.h> #include <cuda_runtime.h> #define CUDA_ERROR_CHECK #define CudaSafeCall( err ) __cudaSafeCall( err, __FILE__, __LINE__ ) #define CudaCheckError() __cudaCheckError( __FILE__, __LINE__ ) inline void __cudaSafeCall( cudaError err, const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK if ( cudaSuccess != err ) { fprintf( stderr, "cudaSafeCall() failed at %s:%i : %s\n", file, line, cudaGetErrorString( err ) ); exit( -1 ); } #endif return; } inline void __cudaCheckError( const char *file, const int line ) { #ifdef CUDA_ERROR_CHECK cudaError err = cudaGetLastError(); if ( cudaSuccess != err ) { fprintf( stderr, "cudaCheckError() failed at %s:%i : %s\n", file, line, cudaGetErrorString( err ) ); exit( -1 ); } // More careful checking. However, this will affect performance. // Comment away if needed. err = cudaDeviceSynchronize(); if( cudaSuccess != err ) { fprintf( stderr, "cudaCheckError() with sync failed at %s:%i : %s\n", file, line, cudaGetErrorString( err ) ); exit( -1 ); } #endif return; } // C = AB template <int BLOCK_SIZE_X, int BLOCK_SIZE_Y> __global__ void kernel_batch_matmul( const float * __restrict__ matA, const float * __restrict__ matB, float * __restrict__ matC, int dim0, int dim1A, int dim2A, int dim1B, int dim2B, int dim1C, int dim2C){ extern __shared__ float smem[]; const unsigned int len_subA = BLOCK_SIZE_Y * dim2A, len_subB = BLOCK_SIZE_X * dim1B; //len of sub matrices of A and B. const unsigned long len_A = dim0*dim1A*dim2A, len_B = dim0*dim1B*dim2B, len_C = dim0*dim1C*dim2C; const unsigned long len_A_signleBatch = dim1A*dim2A, len_B_signleBatch = dim1B*dim2B, len_C_signleBatch = dim1C*dim2C; const unsigned int BLOCKSIZE_P2 = BLOCK_SIZE_X*BLOCK_SIZE_Y; //smemA = smem + 0; //smemB = smem + len_subA; // Block index unsigned int bx = blockIdx.x; // mapped to the sub-matrices of output unsigned int by = blockIdx.y; // mapped to the sub-matrices of output unsigned int bz = blockIdx.z; // batch index // Thread index unsigned int tx = threadIdx.x; unsigned int ty = threadIdx.y; unsigned int c_pos_x, c_pos_y; c_pos_x = bx*BLOCK_SIZE_X + tx; c_pos_y = by*BLOCK_SIZE_Y + ty; unsigned long gidx1,gidx2; unsigned int _d1,_d2; //printf("## bx:%u, by:%u, tx:%u, ty:%u, c_pos_x:%u, c_pos_y:%u\n",bx,by,tx,ty,c_pos_x,c_pos_y); unsigned long offsetA = (by * BLOCK_SIZE_Y) * dim2A; unsigned long offsetB = (bx * BLOCK_SIZE_X); //first row (d1=0) // Load sub matrices from global memory into shared memory unsigned long idxA, idxB; idxA = ty* BLOCK_SIZE_X + tx; idxB = ty* BLOCK_SIZE_X + tx; //printf("*** bx:%u, by:%u, tx:%u, ty:%u ,idxA:%ld, idxB:%ld\n",bx,by,tx,ty,idxA,idxB); while(idxA < len_subA){//Block-stride loop gidx1 = offsetA + idxA; if(idxA < len_subA && gidx1 < len_A) { smem[idxA] = matA[bz * len_A_signleBatch + gidx1]; /*printf("bx:%u, by:%u, tx:%u, ty:%u ,idxA:%ld, gidx1:%ld\n",bx,by,tx,ty,idxA,gidx1);*/ }else{ smem[idxA] = 0; } idxA += BLOCKSIZE_P2; } ///TODO: It might be better to store transposed subMatB in shared memory to avoid shared memory read conflict. /// But then we might get shared memory write conflict. (?) while(idxB < len_subB ){//Block-stride loop //gidx2 = offsetB + (bx*BLOCK_SIZE)*dim2B + (idxB % dim2B); _d2 = idxB%BLOCK_SIZE_X; _d1 = (idxB/BLOCK_SIZE_X); gidx2 = offsetB + _d1*dim2B + _d2; if(idxB < len_subB && _d1<dim1B && _d2<dim2B){ smem[len_subA+idxB] = matB[bz * len_B_signleBatch +gidx2]; /*printf("* bx:%u, by:%u ,tx:%u, ty:%u ,idxB:%ld, _d1:%d, _d2:%d, gidx2:%ld\n",bx,by,tx,ty,idxB,_d1,_d2,gidx2);*/ }else{ smem[len_subA+idxB] = 0; } idxB += BLOCKSIZE_P2; } __syncthreads(); // Multiply and add each result to produce output element of current thread in the thread block. if(c_pos_x<dim2C && c_pos_y<dim1C){ float output_element = 0.0f; //dim2A=dim1B is common equal dimension of 2 matrices --- block-stride loop for (int k = 0; k < dim2A; k++) { output_element += smem[ty*dim2A+k] * smem[len_subA+ k*BLOCK_SIZE_X + tx]; /*printf("###bz:%d, c_pos_x:%d, c_pos_y:%d, smem[%d]=%f, smem[%d]=%f\n", bz,c_pos_x,c_pos_y, ty*dim2A+k,smem[ty*dim2A+k], len_subA+ k*BLOCK_SIZE+tx,smem[len_subA+ k*BLOCK_SIZE+tx]);*/ } ///TODO: Check matC index to not to exceed the len of matC! matC[bz * len_C_signleBatch + c_pos_y*dim2C + c_pos_x] = output_element; } } void batch_matmul( const float * matA, //row-major device ptr (batch, hA, wA) == (dim0A, dim1A , *dim2A* ) const float * matB, //row-major device ptr (batch, hB, wB) == (dim0B, *dim1B* , dim2B ) float * matC, //row-major device ptr (batch, hB, wB) == (dim0B, dim1A , dim2B ) int dim0A, int dim1A, int dim2A, int dim0B, int dim1B, int dim2B){ if(dim2A != dim1B){printf("ERR@batched_matmul: BAD SHAPE.\n"); return;} if(dim0B != dim0A){printf("ERR@batched_matmul: BAD BATCH SIZES.\n"); return;} const int BLOCK_DIM_X = 4; const int BLOCK_DIM_Y = 4; dim3 blocksize(BLOCK_DIM_X,BLOCK_DIM_Y,1); dim3 gridsize(0,0,0); gridsize.x = (dim2B + BLOCK_DIM_X-1)/BLOCK_DIM_X; gridsize.y = (dim1A + BLOCK_DIM_Y-1)/BLOCK_DIM_Y; gridsize.z = (dim0A); unsigned long sharedmemsize = (BLOCK_DIM_Y*dim2A + BLOCK_DIM_X* dim1B)*sizeof(float); //printf("@batched_matmul:\n"); //printf("\tBLOCK:(%d, %d)\n",blocksize.x,blocksize.y); //printf("\t GRID:(%d, %d, %d)\n",gridsize.x,gridsize.y,gridsize.z); //printf("\t SHARED: %d Bytes\n",sharedmemsize); if(BLOCK_DIM_X==4 && BLOCK_DIM_Y==4){ kernel_batch_matmul<4,4> <<<gridsize, blocksize, sharedmemsize>>>( matA, matB, matC, dim0A, dim1A, //hA dim2A, //wA dim1B, //hA dim2B, //wA dim1A, dim2B); CudaCheckError(); }else{ printf("ERR@batched_matmul: UNDEFINED BLOCK_DIM.\n"); return; } }
5,011
#include "includes.h" __global__ void reduceUnrollWarps8 (int *g_idata, int *g_odata, unsigned int n) { // set thread ID unsigned int tid = threadIdx.x; unsigned int idx = blockIdx.x * blockDim.x * 8 + threadIdx.x; // convert global data pointer to the local pointer of this block int *idata = g_idata + blockIdx.x * blockDim.x * 8; // unrolling 8 if (idx + 7 * blockDim.x < n) { int a1 = g_idata[idx]; int a2 = g_idata[idx + blockDim.x]; int a3 = g_idata[idx + 2 * blockDim.x]; int a4 = g_idata[idx + 3 * blockDim.x]; int b1 = g_idata[idx + 4 * blockDim.x]; int b2 = g_idata[idx + 5 * blockDim.x]; int b3 = g_idata[idx + 6 * blockDim.x]; int b4 = g_idata[idx + 7 * blockDim.x]; g_idata[idx] = a1 + a2 + a3 + a4 + b1 + b2 + b3 + b4; } __syncthreads(); // in-place reduction in global memory for (int stride = blockDim.x / 2; stride > 32; stride >>= 1) { if (tid < stride) { idata[tid] += idata[tid + stride]; } // synchronize within threadblock __syncthreads(); } // unrolling warp if (tid < 32) { volatile int *vmem = idata; vmem[tid] += vmem[tid + 32]; vmem[tid] += vmem[tid + 16]; vmem[tid] += vmem[tid + 8]; vmem[tid] += vmem[tid + 4]; vmem[tid] += vmem[tid + 2]; vmem[tid] += vmem[tid + 1]; } // write result for this block to global mem if (tid == 0) g_odata[blockIdx.x] = idata[0]; }
5,012
//////////////////////////////////////////////////////////////////////////// // // Copyright 1993-2015 NVIDIA Corporation. All rights reserved. // // Please refer to the NVIDIA end user license agreement (EULA) associated // with this source code for terms and conditions that govern your use of // this software. Any use, reproduction, disclosure, or distribution of // this software and related documentation outside the terms of the EULA // is strictly prohibited. // //////////////////////////////////////////////////////////////////////////// /* Template project which demonstrates the basics on how to setup a project * example application. * Host code. */ #include <thrust/host_vector.h> #include <thrust/device_vector.h> #include <thrust/generate.h> #include <thrust/sort.h> #include <thrust/copy.h> #include <algorithm> #include <cstdlib> int main(void) { // generate 32M random numbers serially thrust::host_vector<int> h_vec(32 << 20); std::generate(h_vec.begin(), h_vec.end(), rand); // transfer data to the device thrust::device_vector<int> d_vec = h_vec; // sort data on the device (846M keys per second on GeForce GTX 480) thrust::sort(d_vec.begin(), d_vec.end()); // transfer data back to host thrust::copy(d_vec.begin(), d_vec.end(), h_vec.begin()); return 0; }
5,013
//#include<iostream> //#include<cstring> //#include<algorithm> //#include<string> //#include<cassert> //#include<iomanip> //using namespace std; // //#define MAX 100 //#define for(i,a,b) for(i=a;i<b; i++) // //string gram[MAX][MAX]; //to store entered grammar //string dpr[MAX]; //int p, np; //np-> number of productions // //inline string concat(string a, string b) //concatenates unique non-terminals //{ // int i; // string r = a; // for (i, 0, b.length()) // if (r.find(b[i]) > r.length()) // r += b[i]; // return (r); //} // //inline void break_gram(string a) //seperates right hand side of entered grammar //{ // int i; // p = 0; // while (a.length()) // { // i = a.find("|"); // if (i>a.length()) // { // dpr[p++] = a; // a = ""; // } // else // { // dpr[p++] = a.substr(0, i); // a = a.substr(i + 1, a.length()); // } // } //} // //inline int lchomsky(string a) //checks if LHS of entered grammar is in CNF //{ // if (a.length() == 1 && a[0] >= 'A' && a[0] <= 'Z') // return 1; // return 0; //} // //inline int rchomsky(string a) //checks if RHS of grammar is in CNF //{ // if (a.length() == 1 && a[0] >= 'a' && a[0] <= 'z') // return 1; // if (a.length() == 2 && a[0] >= 'A' && a[0] <= 'Z' && a[1] >= 'A' && a[1] <= 'Z') // return 1; // return 0; //} // //inline string search_prod(string p) //returns a concatenated string of variables which can produce string p //{ // int j, k; // string r = ""; // for (j, 0, np) // { // k = 1; // while (gram[j][k] != "") // { // if (gram[j][k] == p) // { // r = concat(r, gram[j][0]); // } // k++; // } // } // return r; //} // //inline string gen_comb(string a, string b) //creates every combination of variables from a and b . For eg: BA * AB = {BA, BB, AA, BB} //{ // int i, j; // string pri = a, re = ""; // for (i, 0, a.length()) // for (j, 0, b.length()) // { // pri = ""; // pri = pri + a[i] + b[j]; // re = re + search_prod(pri); //searches if the generated productions can be created or not // } // return re; //} // //int main() //{ // int i, pt, j, l, k; // string a, str, r, pr, start; // cout << "\nEnter the start Variable "; // cin >> start; // cout << "\nNumber of productions "; // cin >> np; // for (i, 0, np) // { // cin >> a; // pt = a.find("->"); // gram[i][0] = a.substr(0, pt); // if (lchomsky(gram[i][0]) == 0) // { // cout << "\nGrammar not in Chomsky Form"; // abort(); // } // a = a.substr(pt + 2, a.length()); // break_gram(a); // for (j, 0, p) // { // gram[i][j + 1] = dpr[j]; // if (rchomsky(dpr[j]) == 0) // { // cout << "\nGrammar not in Chomsky Form"; // abort(); // } // } // } // string matrix[MAX][MAX], st; // cout << "\nEnter string to be checked : "; // cin >> str; // for (i, 0, str.length()) //Assigns values to principal diagonal of matrix // { // r = ""; // st = ""; // st += str[i]; // for (j, 0, np) // { // k = 1; // while (gram[j][k] != "") // { // if (gram[j][k] == st) // { // r = concat(r, gram[j][0]); // } // k++; // } // } // matrix[i][i] = r; // } // int ii, kk; // for (k, 1, str.length()) //Assigns values to upper half of the matrix // { // for (j, k, str.length()) // { // r = ""; // for (l, j - k, j) // { // pr = gen_comb(matrix[j - k][l], matrix[l + 1][j]); // r = concat(r, pr); // } // matrix[j - k][j] = r; // } // } // // for (i, 0, str.length()) //Prints the matrix // { // k = 0; // l = str.length() - i - 1; // for (j, l, str.length()) // { // cout << setw(5) << matrix[k++][j] << " "; // } // cout << endl; // } // // int f = 0; // for (i, 0, start.length()) // if (matrix[0][str.length() - 1].find(start[i]) <= matrix[0][str.length() - 1].length()) //Checks if last element of first row contains a Start variable // { // cout << "String can be generated\n"; // getchar(); // // } // cout << "String cannot be generated\n"; // getchar(); // return 0; //}
5,014
#include "model.cuh" float Model::train_batch(std::vector<Matrix>::iterator X_i, std::vector<Matrix>::iterator Y_i, const unsigned int batch_size, const float lr, const float momentum) { float loss = 0.0f; for (unsigned int i = 0; i < batch_size; i++) { // create the inputs Matrix tmp = *X_i; // feedforward through the neural network for (auto layer = layers.begin(); layer != layers.end(); ++layer) { (*layer)->feedforward_update(tmp); tmp = (*layer)->neurons; } // calculate the loss loss += loss_function.func(*Y_i, tmp); // calculate the derivative for the loss function // reuse temporary variable, as it is no longer needed tmp = loss_function.deriv(*Y_i, tmp); // now backpropogate through the layers for (auto layer = layers.rbegin(); layer != layers.rend(); ++layer) tmp = (*layer)->backpropogate(tmp); ++X_i; ++Y_i; // advance the iterators } for (auto layer = layers.rbegin(); layer != layers.rend(); ++layer) (*layer)->gradient_update(lr, momentum); return loss / (float) batch_size; } Model::Model(Loss_Function loss_function) { this->loss_function = loss_function; } void Model::operator+=(Layer* layer) { //if (layers.empty() or layers.back()->get_out_shape() == layer.get_in_shape()) layers.push_back(layer); //else //yeet "This layer's input shape is not compatible"; } Matrix Model::feed (const Matrix& input) const { Matrix tmp = (*(layers.begin()))->feedforward(input); // feedforward through the neural network auto layer = layers.begin(); for (++layer; layer != layers.end(); ++layer) tmp = (*layer)->feedforward(tmp); return tmp; }
5,015
/****************************************************************************80 Array element addition using CUDA on GPUs Note: changes from the C++ / CPU-only file marked intentionally w/ "CUDA" supposed to be *annoyingly* commented for (self-)educational purposes "host" is assumed to be the CPU running the C program. "device" is assumed to be the GPU running the kernel. MIT License Copyright (c) 2017 Marcelo Novaes ******************************************************************************/ #include <chrono> #include <iostream> #include <math.h> #include <cuda_runtime.h> /****************************************************************************80 PRINT_TIME calculates how many seconds since the start of the program ******************************************************************************/ using namespace std::chrono; const system_clock::time_point start = system_clock::now(); float print_time() { // elapsed time system_clock::time_point end = system_clock::now(); duration<double> elapsed_seconds = end - start; std::time_t end_time = system_clock::to_time_t(end); float elapsed = elapsed_seconds.count(); // print printf("@ %06.3fs > ", elapsed); return elapsed; } /****************************************************************************80 ADD_ARRAYS adds elements of two arrays on GPU ******************************************************************************/ // CUDA: kernel defined by __global__ declaration specifier, accessed by host & device // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#kernels __global__ void add_arrays(int n, float *x, float *y, float *z) { // loop over the array index adding elements *in parallel* // CUDA: threads have an ID, reside within a block, reside within a grid // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#thread-hierarchy int i = blockDim.x * blockIdx.x + threadIdx.x; if (i < n) { z[i] = x[i] + y[i]; } } /****************************************************************************80 MAIN is the main program for array element addition on GPU example ******************************************************************************/ int main(void) { print_time(); // set up array dimensions and byte size printf("Setting up array dimension and size...\n"); int N = pow(10, 8); // 100M elements per array size_t sizeN = N * sizeof(float); print_time(); // allocate memory for arrays // CUDA: this allocation is happening in the host, thus renamed as `h?` printf("Allocating memory for arrays...\n"); float *hx = (float *)malloc(sizeN); float *hy = (float *)malloc(sizeN); float *hz = (float *)malloc(sizeN); // CUDA: this allocation is happening in the device, thus renamed as `d?` // CUDA: allocate Unified Memory to be bridged between the CPU and GPU // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#heterogeneous-programming // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#um-explicit-allocation float *dx = NULL; cudaMalloc(&dx, sizeN); float *dy = NULL; cudaMalloc(&dy, sizeN); float *dz = NULL; cudaMalloc(&dz, sizeN); print_time(); // initialize x, y, z arrays // CUDA: initializing in the host printf("Initializing arrays...\n"); for (int i = 0; i < N; i++) { hx[i] = 2.0f; hy[i] = 3.0f; hz[i] = 0.0f; //printf("i = %d, hx = %0.6f, hy = %0.6f, hz = %0.6f\n", i, hx[i], hy[i], hz[i]); } print_time(); // CUDA: copies the arrays from host to device // CUDA: cudaMemcpy(from, to, size, direction); printf("Copying arrays from host to device...\n"); cudaMemcpy(dx, hx, sizeN, cudaMemcpyHostToDevice); cudaMemcpy(dy, hy, sizeN, cudaMemcpyHostToDevice); cudaMemcpy(dz, hz, sizeN, cudaMemcpyHostToDevice); print_time(); // add elements (on GPU) // CUDA: run the kernel call, <<<...>>> notation defines # of CUDA blocks/threads // e.g.: <<<16, 256>>> means 16 blocks of 256 threads // this alone doesn't parallelize it, must make changes in the kernel function // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#kernels // see: http://users.wfu.edu/choss/CUDA/docs/Lecture%205.pdf int nThreads = 256; // CUDA: Threads per block int nBlocks = N / nThreads; // CUDA: Number of blocks on total grid printf("Adding %d elements from both arrays...\n", N); printf("Kernel using %d threads for each of %d blocks.\n", nThreads, nBlocks); add_arrays<<<nBlocks, nThreads>>>(N, dx, dy, dz); // CUDA: kernel call<<<blocks, threads>>>(args); // CUDA: wait the GPU to finish commands in all streams of all host threads // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#explicit-synchronization cudaDeviceSynchronize(); print_time(); // CUDA: copies the arrays from device back to host // CUDA: cudaMemcpy(from, to, size, direction); printf("Copying arrays from device back to host...\n"); cudaMemcpy(hz, dz, sizeN, cudaMemcpyDeviceToHost); print_time(); // sum errors to check for problems printf("Aggregating deltas over all elements to check for errors...\n"); float sumOfErrors = 0.0f; for (int i = 0; i < N; i++) { sumOfErrors += fabs(hx[i] + hy[i] - hz[i]); } printf("Sum of errors: %.6f\n", sumOfErrors); // print first and last z array contents printf("hz[0] = %.6f\n", hz[0]); printf("hz[N-1] = %.6f\n", hz[N-1]); print_time(); // free mem when done printf("Freeing memory...\n"); delete [] hx; delete [] hy; delete [] hz; // CUDA: device memory can be allocated as linear memory or as CUDA arrays // linear memory usually allocated using cudaMalloc(). freed using cudaFree() // see: http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#device-memory cudaFree(dx); cudaFree(dy); cudaFree(dz); // C'est fini print_time(); printf("Done.\n"); return 0; }
5,016
#include<iostream> #include<cstdlib> #include<cmath> #include<time.h> #include <assert.h> #include <cuda.h> #include <cuda_runtime.h> #define N 10000000 #define MAX_ERR 1e-6 using namespace std; __global__ void vector_add(float *out, float *a, float *b, int n) { int i = threadIdx.x + blockIdx.x * blockDim.x; if(i<N){ out[i]=a[i]+b[i]; } } int main(){ float *a, *b, *out,*cpu_out; float *d_a, *d_b, *d_out; // Allocate host memory a = (float*)malloc(sizeof(float) * N); b = (float*)malloc(sizeof(float) * N); out = (float*)malloc(sizeof(float) * N); cpu_out = (float*)malloc(sizeof(float) * N); // Initialize host arrays for(int i = 0; i < N; i++){ a[i] = i*1.0f; b[i] = i*1.0f; } // Allocate device memory cudaMalloc((void**)&d_a, sizeof(float) * N); cudaMalloc((void**)&d_b, sizeof(float) * N); cudaMalloc((void**)&d_out, sizeof(float) * N); // Transfer data from host to device memory cudaMemcpy(d_a, a, sizeof(float) * N, cudaMemcpyHostToDevice); cudaMemcpy(d_b, b, sizeof(float) * N, cudaMemcpyHostToDevice); // Executing kernel int block_size = 256; int grid_size = ((N + block_size) / block_size); vector_add<<<grid_size,block_size>>>(d_out, d_a, d_b, N); // Transfer data back to host memory cudaMemcpy(out, d_out, sizeof(float) * N, cudaMemcpyDeviceToHost); clock_t t=clock(); for(int i=0;i<N;i++){ cpu_out[i] = a[i]+b[i]; } t=clock()-t; cout<<"\nCPU Time Elapsed: "<<((double)t)<<"\n"; // Verification for(int i = 0; i < N; i++){ assert(fabs(out[i] - a[i] - b[i]) < MAX_ERR); } printf("PASSED\n"); // for(int i=0;i<N;i++) // printf("%lf ",out[i]); // Deallocate device memory cudaFree(d_a); cudaFree(d_b); cudaFree(d_out); // Deallocate host memory free(a); free(b); free(out); }
5,017
#include<stdio.h> #include<stdlib.h> #include<math.h> #include<string.h> #include<cuda.h> #define INPUT_SIZE 100000000 #define PRIME_RANGE 1000000 #define BLOCK_SIZE 32 typedef unsigned long long int uint64_c; void initializeInput(char* , int ); int generate_seed_primes(char*, int*, uint64_c); void copy_seed_primes(uint64_c *,int *,int); void print_primelist(uint64_c *, uint64_c); void print_inputlist(uint64_c *input_list,uint64_c range); void initializing_inputlist(uint64_c *input_list, uint64_c start,uint64_c range); void calculatePrime(uint64_c* , uint64_c* , uint64_c ,uint64_c); void appending_prime(uint64_c* input_list, uint64_c* prime_list, uint64_c range, uint64_c prev_number_of_primes, uint64_c number_of_primes); uint64_c counting_primes(uint64_c*, uint64_c, uint64_c); //KERNAL CODE GOES HERE!! __global__ void prime_generator(uint64_c* d_input_list, uint64_c* d_prime_list, uint64_c* d_range,uint64_c* d_number_of_primes) { int p= blockIdx.x * blockDim.x + threadIdx.x; int prime = d_prime_list[p]; for(uint64_c i=0;i<d_range[0];i++){ if(d_input_list[i] % prime ==0) { d_input_list[i]=0; } } } //KERNAL CODE ENDS HERE!!! int main() { cudaSetDevice(1); // This code is just to generate the seed prime numbers int input_size=100; char *input; uint64_c n= 10 ;// seed prime list. int *seed_primelist; input=(char *)malloc(input_size*sizeof(char)); initializeInput(input, input_size); seed_primelist=(int *)malloc(input_size*sizeof(int)); int num_of_seed = generate_seed_primes(input,seed_primelist,n); uint64_c* input_list; uint64_c* prime_list; uint64_c number_of_primes= num_of_seed; prime_list=(uint64_c *)malloc(number_of_primes*sizeof(uint64_c)); copy_seed_primes(prime_list,seed_primelist,num_of_seed); uint64_c* d_input_list; uint64_c* d_prime_list; uint64_c* d_number_of_primes; uint64_c* d_range; for(int i=0;i<3;i++){ uint64_c start=n; uint64_c end=pow(n,2); printf("CALCULATING PRIMES FROM 0 - %llu\n",end); uint64_c range=end-start; input_list=(uint64_c *)malloc(range*sizeof(uint64_c)); initializing_inputlist(input_list,start,range); if(cudaMalloc((void **)&d_input_list,range*sizeof(uint64_c))!=cudaSuccess) { printf("Error: 1\n"); } if(cudaMemcpy(d_input_list,input_list,range*sizeof(uint64_c),cudaMemcpyHostToDevice)!=cudaSuccess) { printf("copy Error: 1\n"); } if(cudaMalloc((void **)&d_prime_list,number_of_primes*sizeof(uint64_c))!=cudaSuccess) { printf("Error: 2\n"); } if(cudaMemcpy(d_prime_list,prime_list,number_of_primes*sizeof(uint64_c),cudaMemcpyHostToDevice)!=cudaSuccess) { printf("copy Error: 2\n"); } if(cudaMalloc((void **)&d_range,sizeof(uint64_c))!=cudaSuccess) { printf("Error: 3\n"); } if(cudaMemcpy(d_range,&range,sizeof(uint64_c),cudaMemcpyHostToDevice)!=cudaSuccess) { printf("copy Error: 3\n"); } if(cudaMalloc((void **)&d_number_of_primes,sizeof(uint64_c))!=cudaSuccess) { printf("Error: 4\n"); } if(cudaMemcpy(d_number_of_primes,&number_of_primes,sizeof(uint64_c),cudaMemcpyHostToDevice)!=cudaSuccess) { printf("copy Error: 4\n"); } // calculatePrime(input_list,prime_list,range,number_of_primes); prime_generator<<<43,32>>>(d_input_list,d_prime_list,d_range,d_number_of_primes); if(cudaMemcpy(input_list,d_input_list,range*sizeof(uint64_c),cudaMemcpyDeviceToHost)!=cudaSuccess) { printf("copy Host Error: 1\n"); exit(0); } //print_inputlist(input_list,range); uint64_c previous_number_of_primes= number_of_primes; number_of_primes = counting_primes(input_list, range, number_of_primes)+previous_number_of_primes; printf("THE NUMBER OF PRIMES ARE: %llu\n",number_of_primes); prime_list=(uint64_c *)realloc(prime_list,number_of_primes*sizeof(uint64_c)); appending_prime(input_list, prime_list, range, previous_number_of_primes, number_of_primes); print_primelist(prime_list,number_of_primes); n=pow(n,2); printf("******************************************\n\n"); } return 0; } int generate_seed_primes(char *input,int *primelist, uint64_c n) { for (int p=2; p*p<=n; p++) { if (input[p] == 'P') { for (int i=p*2; i<=n; i += p) input[i] = 'N'; } } int i=0; for (int p=2; p<=n; p++){ if (input[p]=='P') { primelist[i]=p; i++; } } return i; } void initializeInput(char *input, int input_size) { for(int i=0;i<input_size;i++) { input[i]='P'; } } void initializing_inputlist(uint64_c *input_list, uint64_c start,uint64_c range){ for(uint64_c i=0;i<range;i++) { input_list[i]=start+i; } } void print_inputlist(uint64_c *input_list,uint64_c range) { for(uint64_c i=0;i<range;i++) { printf("%llu\t--->\t%llu\n", i,input_list[i]); } } void print_primelist(uint64_c *prime_list,uint64_c number_of_primes) { for(uint64_c i=0;i<number_of_primes;i++) { printf("%llu\n",prime_list[i]); } } void copy_seed_primes(uint64_c *prime_list,int * seed_primelist,int num_of_seed) { for(int i=0;i<num_of_seed;i++) { prime_list[i]=seed_primelist[i]; } } void calculatePrime(uint64_c* input_list, uint64_c* prime_list, uint64_c range,uint64_c number_of_primes) { // print_primelist(prime_list,number_of_primes); for(uint64_c i=0;i<range;i++) { for(uint64_c j=0;j<number_of_primes;j++){ if(input_list[i] % prime_list[j]==0) { input_list[i]=0; } } } } void appending_prime(uint64_c* input_list, uint64_c* prime_list, uint64_c range, uint64_c prev_number_of_primes, uint64_c number_of_primes) { for(uint64_c i=0;i<range;i++) { if(input_list[i]>0){ //printf("XXXXXXXXX>>> %llu\n", input_list[i]); prime_list[prev_number_of_primes]=input_list[i]; prev_number_of_primes++; } } // printf("the number he he ha ha... %llu\n",prev_number_of_primes); //exit(0); } uint64_c counting_primes(uint64_c* input_list,uint64_c range,uint64_c number_of_primes) { int prime=0; for(uint64_c i=0;i<range;i++) { if(input_list[i]>0) { prime++; } } return prime; }
5,018
#include <cuda_runtime.h> #include <device_launch_parameters.h> #include <thrust/scan.h> #include <thrust/device_vector.h> #include <stdio.h> __global__ void voxelOccupancy(int* occupancy, int granularity) { int x = blockIdx.x * blockDim.x + threadIdx.x; int y = blockIdx.y * blockDim.y + threadIdx.y; int z = blockIdx.z * blockDim.z + threadIdx.z; // Detect out of bounds if (x >= granularity || y >= granularity || z >= granularity) { return; } // Determine if this thread's block is solid int base = z * granularity * granularity + y * granularity + x; int solid = 1 - (threadIdx.y % 2); if (threadIdx.x % 2 == threadIdx.z % 2) { solid = threadIdx.y % 2; } occupancy[base] = solid; } __global__ void compactVoxels(int* compact, int* occupancy, int* scanned, int granularity) { int x = blockIdx.x * blockDim.x + threadIdx.x; int y = blockIdx.y * blockDim.y + threadIdx.y; int z = blockIdx.z * blockDim.z + threadIdx.z; // Detect out of bounds if (x >= granularity || y >= granularity || z >= granularity) { return; } // Resolve elements of the compacted array int base = z * granularity * granularity + y * granularity + x; if (occupancy[base] == 1) { compact[scanned[base]] = base; } } __global__ void generateVoxels(int* dCompact, float3* pos, float3* norm, int granularity, int maxVert) { int x = blockIdx.x * blockDim.x + threadIdx.x; int y = blockIdx.y * blockDim.y + threadIdx.y; int z = blockIdx.z * blockDim.z + threadIdx.z; int i = z * granularity * granularity + y * granularity + x; // Detect out of bounds if (x >= granularity || y >= granularity || z >= granularity || i > maxVert) { return; } // Compute relative spatial locations of this voxel int spatialIndex = dCompact[i]; int relz = spatialIndex / (granularity * granularity); int rely = (spatialIndex - relz * (granularity * granularity)) / granularity; int relx = spatialIndex - relz * (granularity * granularity) - rely * granularity; float length = 1.0f / granularity; float half = 0.5f * granularity; float3 rel = { (relx - (half - 0.5f)) * 2.0f * length, (rely - (half - 0.5f)) * 2.0f * length, (relz - (half - 0.5f)) * 2.0f * length, }; // Hardcoded cube data (should be replaced using instancing) int base = 36 * i; pos[base] = { -length, -length, -length }; pos[base + 1] = { -length, -length, length }; pos[base + 2] = { -length, length, length }; pos[base + 3] = { length, length, -length }; pos[base + 4] = { -length, -length, -length }; pos[base + 5] = { -length, length, -length }; pos[base + 6] = { length, -length, length }; pos[base + 7] = { -length, -length, -length }; pos[base + 8] = { length, -length, -length }; pos[base + 9] = { length, length, -length }; pos[base + 10] = { length, -length, -length }; pos[base + 11] = { -length, -length, -length }; pos[base + 12] = { -length, -length, -length }; pos[base + 13] = { -length, length, length }; pos[base + 14] = { -length, length, -length }; pos[base + 15] = { length, -length, length }; pos[base + 16] = { -length, -length, length }; pos[base + 17] = { -length, -length, -length }; pos[base + 18] = { -length, length, length }; pos[base + 19] = { -length, -length, length }; pos[base + 20] = { length, -length, length }; pos[base + 21] = { length, length, length }; pos[base + 22] = { length, -length, -length }; pos[base + 23] = { length, length, -length }; pos[base + 24] = { length, -length, -length }; pos[base + 25] = { length, length, length }; pos[base + 26] = { length, -length, length }; pos[base + 27] = { length, length, length }; pos[base + 28] = { length, length, -length }; pos[base + 29] = { -length, length, -length }; pos[base + 30] = { length, length, length }; pos[base + 31] = { -length, length, -length }; pos[base + 32] = { -length, length, length }; pos[base + 33] = { length, length, length }; pos[base + 34] = { -length, length, length }; pos[base + 35] = { length, -length, length }; norm[base] = { -1, 0, 0 }; norm[base + 1] = { -1, 0, 0 }; norm[base + 2] = { -1, 0, 0 }; norm[base + 3] = { 0, 0, -1 }; norm[base + 4] = { 0, 0, -1 }; norm[base + 5] = { 0, 0, -1 }; norm[base + 6] = { 0, -1, 0 }; norm[base + 7] = { 0, -1, 0 }; norm[base + 8] = { 0, -1, 0 }; norm[base + 9] = { 0, 0, -1 }; norm[base + 10] = { 0, 0, -1 }; norm[base + 11] = { 0, 0, -1 }; norm[base + 12] = { -1, 0, 0 }; norm[base + 13] = { -1, 0, 0 }; norm[base + 14] = { -1, 0, 0 }; norm[base + 15] = { 0, -1, 0 }; norm[base + 16] = { 0, -1, 0 }; norm[base + 17] = { 0, -1, 0 }; norm[base + 18] = { 0, 0, 1 }; norm[base + 19] = { 0, 0, 1 }; norm[base + 20] = { 0, 0, 1 }; norm[base + 21] = { 1, 0, 0 }; norm[base + 22] = { 1, 0, 0 }; norm[base + 23] = { 1, 0, 0 }; norm[base + 24] = { 1, 0, 0 }; norm[base + 25] = { 1, 0, 0 }; norm[base + 26] = { 1, 0, 0 }; norm[base + 27] = { 0, 1, 0 }; norm[base + 28] = { 0, 1, 0 }; norm[base + 29] = { 0, 1, 0 }; norm[base + 30] = { 0, 1, 0 }; norm[base + 31] = { 0, 1, 0 }; norm[base + 32] = { 0, 1, 0 }; norm[base + 33] = { 0, 0, 1 }; norm[base + 34] = { 0, 0, 1 }; norm[base + 35] = { 0, 0, 1 }; for (int j = 0; j < 36; j++) { pos[base + j].x += rel.x; pos[base + j].y += rel.y; pos[base + j].z += rel.z; } } void launchVoxelOccupancy(dim3 grid, dim3 block, int *occupancy, int granularity) { voxelOccupancy<<<grid, block>>>(occupancy, granularity); } void launchOccupancyScan(int numVoxels, int* scanned, int* occupancy) { thrust::exclusive_scan(thrust::device_ptr<int>(occupancy), thrust::device_ptr<int>(occupancy + numVoxels), thrust::device_ptr<int>(scanned)); } void launchCompactVoxels(dim3 grid, dim3 block, int* dCompact, int* dOccupancy, int* dScanned, int granularity) { compactVoxels<<<grid, block>>>(dCompact, dOccupancy, dScanned, granularity); } void launchGenerateVoxels(dim3 grid, dim3 block, int* dCompact, float3* dVoxels, float3* dNormals, int granularity, int maxVert) { generateVoxels<<<grid, block>>>(dCompact, dVoxels, dNormals, granularity, maxVert); }
5,019
#include <stdio.h> #include <stdlib.h> #include <math.h> #include <cuda.h> #include <cuda_runtime.h> #define BLOCKSIZE 256 __global__ void MatrixAddI(int *matrix1, int *matrix2, int *matrix3, int m, int n) { int x = blockIdx.x * blockDim.x + threadIdx.x; if (x < m*n) { matrix3[x] = matrix1[x] + matrix2[x]; } } __global__ void MatrixAddF(float *matrix1, float *matrix2, float *matrix3, int m, int n) { int x = blockIdx.x * blockDim.x + threadIdx.x; if (x < m*n) { matrix3[x] = matrix1[x] + matrix2[x]; } } void read_imatrix(char *filename, int *m, int *n, int **values) { FILE* name; int i, j, k; int t1, t2, t3; name = fopen(filename, "r+"); if(name != NULL) { k = 0; fscanf(name, "%d %d\n", &t1, &t2); *m = t1; *n = t2; *values = (int *)calloc(t1*t2, sizeof(int)); for(i = 1; i <= t1; i++) { for(j = 1; j <= t2; j++) { if(j < t2) { fscanf(name, "%d,", &t3); *(*values+k) = t3; k++; } else { fscanf(name, "%d\n", &t3); *(*values+k) = t3; k++; } } } fclose(name); } else { printf("File read failed\n"); exit(1); } } void read_fmatrix(char *filename, int *m, int *n, float **values) { FILE* name; int i, j, k; int t1, t2; float t3; name = fopen(filename, "r+"); if(name != NULL) { k = 0; fscanf(name, "%d %d\n", &t1, &t2); *m = t1; *n = t2; *values = (float *)calloc(t1*t2, sizeof(float)); for(i = 1; i <= t1; i++) { for(j = 1; j <= t2; j++) { if(j < t2) { fscanf(name, "%f,", &t3); *(*values+k) = t3; k++; } else { fscanf(name, "%f\n", &t3); *(*values+k) = t3; k++; } } } fclose(name); } else { printf("File read failed\n"); exit(1); } } void write_imatrix(char *filename, int *m, int *n, int **values) { FILE* name; int i, j, k; int t1, t2, t3; name = fopen(filename, "w+"); if(name != NULL) { k = 0; t1 = *m; t2 = *n; fprintf(name, "%d %d\n", t1, t2); for(i = 1; i <= t1; i++) { for(j = 1; j <= t2; j++) { if(j < t2) { t3 = *(*values+k); fprintf(name, "%d,", t3); k++; } else { t3 = *(*values+k); fprintf(name, "%d\n", t3); k++; } } } fclose(name); } else { printf("File write failed\n"); exit(1); } } void write_fmatrix(char *filename, int *m, int *n, float **values) { FILE* name; int i, j, k; int t1, t2; float t3; name = fopen(filename, "w+"); if(name != NULL) { k = 0; t1 = *m; t2 = *n; fprintf(name, "%d %d\n", t1, t2); for(i = 1; i <= t1; i++) { for(j = 1; j <= t2; j++) { if(j < t2) { t3 = *(*values+k); fprintf(name, "%f,", t3); k++; } else { t3 = *(*values+k); fprintf(name, "%f\n", t3); k++; } } } fclose(name); } else { printf("File write failed\n"); exit(1); } } void matrix_check(int m1, int n1, int m2, int n2) { if ((m1-m2)+(n1-n2) != 0) { printf("Matrix dimensions must be PxQ and PxQ respectively\n"); exit(1); } } int main(int argc, char *argv[]) { int m1, n1, m2, n2; if (argc != 5) { printf("Usage: ./matrix-addition matrix1.mat matrix2.mat matrix3.mat float/int \n"); exit(1); } if (strcmp(argv[4], "float") == 0) { float *hostmatrix1, *hostmatrix2, *hostmatrix3; float *devicematrix1, *devicematrix2, *devicematrix3; int GRIDSIZE; read_fmatrix(argv[1], &m1, &n1, &hostmatrix1); read_fmatrix(argv[2], &m2, &n2, &hostmatrix2); matrix_check(m1, n1, m2, n2); size_t matrix_size = m1*n1*sizeof(float); hostmatrix3 = (float *)calloc(matrix_size, sizeof(float)); cudaMalloc(&devicematrix1, matrix_size); cudaMalloc(&devicematrix2, matrix_size); cudaMalloc(&devicematrix3, matrix_size); cudaMemcpy(devicematrix1, hostmatrix1, matrix_size, cudaMemcpyHostToDevice); cudaMemcpy(devicematrix2, hostmatrix2, matrix_size, cudaMemcpyHostToDevice); GRIDSIZE = (int)ceil((float)(m1*n1)/BLOCKSIZE); dim3 dimGrid(GRIDSIZE, 1, 1); dim3 dimBlock(BLOCKSIZE, 1, 1); MatrixAddF <<< dimGrid, dimBlock >>> (devicematrix1, devicematrix2, devicematrix3, m1, n1); cudaMemcpy(hostmatrix3, devicematrix3, matrix_size, cudaMemcpyDeviceToHost); write_fmatrix(argv[3], &m1, &n1, &hostmatrix3); cudaFree(devicematrix1); cudaFree(devicematrix2); cudaFree(devicematrix3); free(hostmatrix1); free(hostmatrix2); free(hostmatrix3); } if (strcmp(argv[4], "int") == 0) { int *hostmatrix1, *hostmatrix2, *hostmatrix3; int *devicematrix1, *devicematrix2, *devicematrix3; int GRIDSIZE; read_imatrix(argv[1], &m1, &n1, &hostmatrix1); read_imatrix(argv[2], &m2, &n2, &hostmatrix2); matrix_check(m1, n1, m2, n2); size_t matrix_size = m1*n1*sizeof(int); hostmatrix3 = (int *)calloc(m1*n1, sizeof(int)); cudaMalloc(&devicematrix1, matrix_size); cudaMalloc(&devicematrix2, matrix_size); cudaMalloc(&devicematrix3, matrix_size); cudaMemcpy(devicematrix1, hostmatrix1, matrix_size, cudaMemcpyHostToDevice); cudaMemcpy(devicematrix2, hostmatrix2, matrix_size, cudaMemcpyHostToDevice); GRIDSIZE = (int)ceil((float)(m1*n1)/BLOCKSIZE); dim3 dimGrid(GRIDSIZE, 1, 1); dim3 dimBlock(BLOCKSIZE, 1, 1); MatrixAddI <<< dimGrid, dimBlock >>> (devicematrix1, devicematrix2, devicematrix3, m1, n1); cudaMemcpy(hostmatrix3, devicematrix3, matrix_size, cudaMemcpyDeviceToHost); write_imatrix(argv[3], &m1, &n1, &hostmatrix3); cudaFree(devicematrix1); cudaFree(devicematrix2); cudaFree(devicematrix3); free(hostmatrix1); free(hostmatrix2); free(hostmatrix3); } return 0; }
5,020
#include <stdio.h> #include "cuda_runtime.h" #include "device_launch_parameters.h" __global__ void print_threadIds_blockIds_gridDim() { printf("threadIdx.x: %d, threadIdx.y: %d, threadIdx.z: %d,\ blockIdx.x: %d, blockIdx.y: %d, blockIdx.z: %d,\ gridDim.x: %d, gridDim.y: %d, gridDim.z: %d \n", threadIdx.x, threadIdx.y, threadIdx.z, blockIdx.x, blockIdx.y, blockIdx.z, gridDim.x, gridDim.y, gridDim.z); } int main() { int nx, ny, nz; nx = 4; ny = 4; nz = 4; dim3 block(2, 2, 2); dim3 grid(nx / block.x, ny / block.y, nz / block.z); print_threadIds_blockIds_gridDim<<<grid, block>>>(); cudaDeviceSynchronize(); cudaDeviceReset(); return 0; }
5,021
#include <cuda.h> #include <cuda_runtime_api.h> #include <device_launch_parameters.h> #include <iostream> __global__ void RankSortKernel(float* DataIn, float* DataOut, int* rank, int size) { // Retrieve our coordinates in the block int tx = (blockIdx.x * 512) + threadIdx.x; rank[tx] = 0; if(tx < size) { for(int i=0;i<size;i++) { if(DataIn[tx]>=DataIn[i]) { rank[tx]++; } } DataOut[(rank[tx]-1)] = DataIn[tx]; } } __global__ void OddEvenSortKernel(float* Array, int size, bool Odd_Phase) { int tx = (blockIdx.x * 512) + threadIdx.x; float temp; int index = 2*tx; if(tx < (size/2)) { if(Odd_Phase == false) { if((index+1) < size) { if(Array[index]>Array[index+1]) { temp = Array[index]; Array[index] = Array[index+1]; Array[index+1] = temp; } } } else { if((index+2) < size) { if(Array[index+1]>Array[index+2]) { temp = Array[index+1]; Array[index+1] = Array[index+2]; Array[index+2] = temp; } } } } } bool RankSortGPU( float* InputArray, float* SortedArray, int size) { int blocksize, gridsize; // Error return value cudaError_t status; // Number of bytes int bytes = size * sizeof(float); // Pointers to the device arrays float *DataIn, *DataOut; int *rank; int bytes1 = size * sizeof(float); // Allocate memory on the device cudaMalloc((void**) &DataIn, bytes); cudaMalloc((void**) &DataOut, bytes); cudaMalloc((void**) &rank, bytes1); // Copy the host input data to the device cudaMemcpy(DataIn, InputArray, bytes, cudaMemcpyHostToDevice); // Specify the size of the grid and the size of the block dim3 dimBlock(512, 1); dim3 dimGrid((int)ceil((float)size/512), 1); // Launch the kernel on a size-by-size block of threads RankSortKernel<<<dimGrid, dimBlock>>>(DataIn, DataOut, rank, size); // Wait for completion cudaThreadSynchronize(); // Check for errors status = cudaGetLastError(); if (status != cudaSuccess) { std::cout << "Kernel failed 1: " << cudaGetErrorString(status) << std::endl; cudaFree(DataIn); cudaFree(DataOut); return false; } // Retrieve the result matrix cudaMemcpy(SortedArray, DataOut, bytes, cudaMemcpyDeviceToHost); // Free device memory cudaFree(DataIn); cudaFree(DataOut); // Success return true; } bool OddEvenSortGPU( float* InputArray, float* SortedArray, int size) { int blocksize, gridsize; // Error return value cudaError_t status; // Number of bytes int bytes = size * sizeof(float); // Pointers to the device arrays float *Array; bool Odd_Phase; // Allocate memory on the device cudaMalloc((void**) &Array, bytes); // Copy the host input data to the device cudaMemcpy(Array, InputArray, bytes, cudaMemcpyHostToDevice); int new_size = size/2; // Specify the size of the grid and the size of the block dim3 dimBlock(512, 1); dim3 dimGrid((int)ceil((float)new_size/512), 1); for(int i=0;i<size;i++) { //even phase Odd_Phase = false; // Launch the kernel on a size-by-size block of threads OddEvenSortKernel<<<dimGrid, dimBlock>>>(Array, size, Odd_Phase); // Wait for completion cudaThreadSynchronize(); // Check for errors status = cudaGetLastError(); if (status != cudaSuccess) { std::cout << "Kernel failed 2: " << cudaGetErrorString(status) << std::endl; cudaFree(Array); return false; } //odd phase Odd_Phase = true; // Launch the kernel on a size-by-size block of threads OddEvenSortKernel<<<dimGrid, dimBlock>>>(Array, size, Odd_Phase); // Wait for completion cudaThreadSynchronize(); // Check for errors status = cudaGetLastError(); if (status != cudaSuccess) { std::cout << "Kernel failed 3: " << cudaGetErrorString(status) << std::endl; cudaFree(Array); return false; } } // Retrieve the result matrix cudaMemcpy(SortedArray, Array, bytes, cudaMemcpyDeviceToHost); // Free device memory cudaFree(Array); // Success return true; }
5,022
#include "CNextStateLookupTable.cuh" #include "CStateLookupTable.cuh" CNextStateLookupTable::CNextStateLookupTable(unsigned int const p_cnK) : CStateLookupTable(p_cnK #ifdef _USE_CUDA_ , LookupTableType_Next // Set table type #endif ) { } CNextStateLookupTable::~CNextStateLookupTable(void) { } unsigned int CNextStateLookupTable::Shift(unsigned int const p_cnState, unsigned int const p_cnInput) const { // Shift right once and put the input in the MSB return (p_cnState >> 1) + ((p_cnInput == 0) ? (0) : (1 << (m_cnMemory - 1))); } CNextStateLookupTable* CNextStateLookupTable::Create(unsigned int const p_cnK) { CNextStateLookupTable* l_pcInstance = new CNextStateLookupTable(p_cnK); l_pcInstance->InitializeStateLookupTable(); return l_pcInstance; }
5,023
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> #include <stdlib.h> #include <math.h> __global__ void convolution_1D(float *N , float *M , float *P , int Mask_width,int width) { int i = blockIdx.x*blockDim.x + threadIdx.x; float pvalue = 0.0; int N_start_point = i - ((Mask_width)/2); for(int j =0;j<Mask_width;j++) { if(((N_start_point+j)>=0)&& ((N_start_point + j)<width)) { pvalue += N[N_start_point + j] * M[j]; } } P[i] = pvalue; } int main() { float *a , *m , *t; float *d_a, *d_m, *d_t; int width , mask_width; printf("Enter the size of array \n"); scanf("%d",&width); a = (float*)malloc(sizeof(float)*width); t = (float*)malloc(sizeof(float)*width); printf("Enter the array \n"); int i = 0; for(i = 0;i<width;i++) { scanf("%f",&a[i]); } printf("Enter the size of mask \n"); scanf("%d",&mask_width); m = (float*)malloc(sizeof(float)*mask_width); printf("Enter the mask \n"); for(i = 0;i<mask_width;i++) { scanf("%f",&m[i]); } int size1 = sizeof(float)*width; int size2 = sizeof(float)*mask_width; cudaMalloc((void**)&d_a,size1); cudaMalloc((void**)&d_m,size2); cudaMalloc((void**)&d_t,size1); cudaMemcpy(d_a,a,size1,cudaMemcpyHostToDevice); cudaMemcpy(d_m,m,size2,cudaMemcpyHostToDevice); dim3 dimGrid((width-1)/mask_width + 1,1,1); dim3 dimBlock(mask_width,1,1); convolution_1D<<<dimGrid,dimBlock>>>(d_a,d_m,d_t,mask_width,width); cudaMemcpy(t,d_t,size1,cudaMemcpyDeviceToHost); printf("The result array is \n"); for(int i = 0;i<width;i++) { printf("%f ",t[i]); } printf("\n"); cudaFree(d_a); cudaFree(d_m); cudaFree(d_t); }
5,024
#include <iostream> #include <fstream> #include <stdlib.h> #include <cstring> #include <limits> // radi definiranja beskonačnosti #include <ctime> // radi mjerenja vremena izvršavanja #include <cmath> // radi "strop" funkcije using namespace std; /* Definiramo beskonačnost kao najveći mogući integer broj. */ #define infty std::numeric_limits<int>::max() void printMatrix (int* G, unsigned int dim) { cout << "\r\n"; for (int i = 0; i < dim*dim; i++) { if (G[i] < infty) { cout << G[i] << "\t"; } else { cout << "∞" << "\t"; } /* Ako je ispisao sve za jedan vrh, prijeđi na sljedeći u novi redak. */ if ((i+1)%dim == 0) { cout << "\r\n"; } } } /* Kernel za device koji paralelizira unutarnje dvije for petlje Floyd-Warshall algoritma. */ __global__ void FW_Cuda(int k, int* D, int* PI, unsigned int dim) { int i = blockDim.y * blockIdx.y + threadIdx.y; int j = blockDim.x * blockIdx.x + threadIdx.x; if (i < dim && j < dim && D[i*dim+k] < INT_MAX && D[k*dim+j] < INT_MAX) { if (D[i*dim+j] > D[i*dim+k] + D[k*dim+j]) { D[i*dim+j] = D[i*dim+k] + D[k*dim+j]; PI[i*dim+j] = PI[k*dim+j]; } } } void Floyd_Warshall_Cuda (int* W, int* D, int* PI, unsigned int dim) { unsigned int n = dim*dim; /* Error varijabla za handleanje CUDA errora. */ cudaError_t err = cudaSuccess; /* Alociranje device varijabli matrica D i PI. */ int* d_D = NULL; err = cudaMalloc((void**) &d_D, n*sizeof(int)); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno alociranje matrice D (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } int* d_PI = NULL; err = cudaMalloc((void**) &d_PI, n*sizeof(int)); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno alociranje matrice PI (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } /* Kopiranje podataka iz host matrica u device. */ err = cudaMemcpy(d_D, D, n*sizeof(int), cudaMemcpyHostToDevice); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno kopiranje matrice D iz hosta u device (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } err = cudaMemcpy(d_PI, PI, n*sizeof(int), cudaMemcpyHostToDevice); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno kopiranje matrice PI iz hosta u device (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } /* Pozivanje CUDA kernela. */ int blockDim = 32; /* Ukoliko je dimenzija bloka 32, imamo 1024 threadova po bloku. */ int gridDim = ceil((float)dim/(float)blockDim); /* Računa se kolika mora biti dimenzija grida ovisno o dimenziji blokova. */ cout << "CUDA kernel se pokreće sa " << gridDim*gridDim << " blokova i " << blockDim*blockDim << " threadova po bloku.\r\n"; /* Vanjsku petlju Floyd-Warshall algoritma vrtimo na CPU, unutarnje dvije paraleliziramo. */ for (int k = 0; k < dim; k++) { FW_Cuda<<<dim3(gridDim, gridDim, 1), dim3(blockDim, blockDim, 1)>>> (k, d_D, d_PI, dim); err = cudaGetLastError(); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno pokrenuta kernel metoda (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } /* Sinkronizacija threadova kako bi se završila k-ta iteracija, te kako bi se prešlo na (k+1). iteraciju. */ cudaThreadSynchronize(); } /* Kopiranje podataka iz device matrica u host. */ err = cudaMemcpy(D, d_D, n*sizeof(int), cudaMemcpyDeviceToHost); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno kopiranje matrice D iz devicea u host (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } err = cudaMemcpy(PI, d_PI, n*sizeof(int), cudaMemcpyDeviceToHost); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno kopiranje matrice PI iz devicea u host (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } /* Dealociranje device varijabli matrica D i PI. */ err = cudaFree(d_D); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno dealociranje matrice D (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } err = cudaFree(d_PI); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno dealociranje matrice PI (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } /* Reset CUDA devicea i završavanje CUDA Floyd-Warshalla. */ err = cudaDeviceReset(); if (err != cudaSuccess) { fprintf(stderr, "Neuspješno resetiranje devicea (završavanje sa CUDA FW, priprema za sljedeće pokretanje)! error=%s\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } } /* Metoda koja rekonstruira težinu najkraćeg puta za dani par vrhova koristeći matricu prethodnika PI i matricu inicijalnih težina W. */ int getPath (int* W, int* PI, int i, int j, unsigned int dim) { if (i == j) { return 0; } else if (PI[i*dim+j] == -1) { return infty; } else { int recursivePath = getPath(W, PI, i, PI[i*dim+j], dim); if (recursivePath < infty) { return recursivePath + W[PI[i*dim+j]*dim+j]; } else { return infty; } } } /* Za svaki par vrhova pokreće getPath metodu koja rekonstruira težinu najkraćeg puta između njih koristeći matricu prethodnika PI. Tu težinu onda uspoređuje sa dobivenom težinom za isti par vrhova u matrici najkraćih putova D. */ bool checkSolutionCorrectness (int* W, int* D, int* PI, unsigned int dim) { for (int i = 0; i < dim; i++) { for (int j = 0; j < dim; j++) { if (getPath(W, PI, i, j, dim) != D[i*dim+j]) { return false; } } } return true; } int main() { /* V - broj vrhova E - broj bridova u - prvi vrh pri učitavanju grafa iz datoteke v - drugi vrh pri učitavanju grafa iz datoteke w - težina između v1 i v2 pri učitavanju grafa iz datoteke */ unsigned int V, E; int u, v, w; ifstream inputGraphFile; inputGraphFile.open("graphFile.txt"); ofstream outputFile; outputFile.open("output_cuda.txt"); inputGraphFile >> V >> E; cout << "V = " << V << ", E = " << E << "\r\n"; unsigned int n = V*V; /* Inicijalizacija grafova u memoriji. */ int* W = (int*)malloc(n*sizeof(int)); int* D = (int*)malloc(n*sizeof(int)); int* PI = (int*)malloc(n*sizeof(int)); /* Postavljanje inicijalnih vrijednosti za matricu prethodnika PI(0), matricu težina W i matricu najkraćih putova D(0). */ fill_n(W, n, infty); fill_n(PI, n, -1); for (int i = 0; i < E; i++) { inputGraphFile >> u >> v >> w; //cout << u << " <-- " << w << " --> " << v << "\r\n"; W[u*V+v] = w; if (u != v) { PI[u*V+v] = u; } } for (int i = 0; i < V; i++) { W[i*V+i] = 0; } /* D(0) = W na početku. */ memcpy (D, W, n*sizeof(int)); // printMatrix(W, V); printMatrix(D, V); printMatrix(PI, V); /* Početak mjerenja izvršavanja Floyd-Warshall algoritma. */ clock_t begin = clock(); /* Pozivamo Floyd-Warshall CPU algoritam nad učitanim grafom. */ Floyd_Warshall_Cuda(W, D, PI, V); /* Kraj mjerenja izvršavanja Floyd-Warshall algoritma. */ clock_t end = clock(); double elapsedTime = double(end - begin) / CLOCKS_PER_SEC; //printMatrix(W, V); printMatrix(D, V); printMatrix(PI, V); /* Ispis rezultata u datoteku. */ outputFile << "|V| = " << V << ", |E| = " << E << "\r\n\r\n"; for (int i = 0; i < n; i++) { if (i%V==0) outputFile << "\r\n"; if (D[i] < infty) outputFile << D[i] << "\t"; else outputFile << "∞" << "\t"; } outputFile << "\r\n\r\n"; for (int i = 0; i < n; i++) { if (i%V==0) outputFile << "\r\n"; outputFile << PI[i] << "\t"; } cout << "Vrijeme izvršavanja Floyd-Warshall algoritma: " << elapsedTime << "s.\r\n"; if (checkSolutionCorrectness(W, D, PI, V) == true) cout << "Svi najkraći putevi su točno izračunati!\r\n"; else cout << "Najkraći putevi nisu točno izračunati.\r\n"; inputGraphFile.close(); outputFile.close(); free(W); free(D); free(PI); return 0; }
5,025
// compile with -std=c++11 -O3 -lcurand #include <iostream> #include <cstdio> #include <curand.h> using std::cout; using std::endl; #define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); } inline void gpuAssert(cudaError_t code, const char *file, int line, bool abort=true) { if (code != cudaSuccess) { fprintf(stderr,"GPUassert: %s %s %d\n", cudaGetErrorString(code), file, line); if (abort) exit(code); } } // Utility class used to avoid linker errors with extern // unsized shared memory arrays with templated type template <class T> struct SharedMemory { __device__ inline operator T*() { extern __shared__ int __smem[]; return (T*) __smem; } __device__ inline operator const T*() const { extern __shared__ int __smem[]; return (T*) __smem; } }; // specialize for double to avoid unaligned memory // access compile errors template <> struct SharedMemory<double> { __device__ inline operator double*() { extern __shared__ double __smem_d[]; return (double*) __smem_d; } __device__ inline operator const double*() const { extern __shared__ double __smem_d[]; return (double*) __smem_d; } }; template <class T> __device__ void reduce(T* g_idata, T* sdata) { // load shared mem unsigned int tid = threadIdx.x; sdata[tid] = g_idata[tid]; __syncthreads(); // do reduction in shared mem for (unsigned int s = blockDim.x / 2; s > 0; s >>= 1) { if (tid < s) { sdata[tid] += sdata[tid + s]; } __syncthreads(); } // result is now in sdata[0] } // This kernel requires blockDim * sizeof(T) Bytes of shared memory. // Each block processes `c * blockDim` elements. // The buffers, which should be affected by the call to `__threadfence_system` // must be volatile, as it is described in the CUDA C programming guide. template <class T, bool COMMUNICATION_ON> __global__ void producer_kernel(T* data, volatile T* partial_data, volatile unsigned* counter, const unsigned c) { const unsigned global_start = blockIdx.x * blockDim.x * c; T* sdata = SharedMemory<T>(); for (unsigned i = 0; i < c; ++i) { const unsigned offset = i * blockDim.x; const auto curr_start = data + global_start + offset; reduce(curr_start, sdata); // now we have the sum of blockDim elements in sdata[0] if (threadIdx.x == 0) { // save the mean of recently processed elements partial_data[blockIdx.x * c + i] = sdata[0] / (T) blockDim.x; if (COMMUNICATION_ON) { __threadfence_system(); ++counter[blockIdx.x]; // mark this block as processed } } } } template<class T, bool COMMUNICATION_ON> __global__ void consumer_kernel(T* data, const volatile T* partial_data, const volatile unsigned* counter, const unsigned c) { __shared__ T mean; const unsigned global_start = blockIdx.x * blockDim.x * c; for (unsigned i = 0; i < c; ++i) { const unsigned offset = i * blockDim.x; if (COMMUNICATION_ON) { if (threadIdx.x == 0) { while (counter[blockIdx.x] < blockDim.x) {} mean = partial_data[blockIdx.x * c + i]; } __syncthreads(); data[offset + global_start + threadIdx.x] = mean; } else { // no communication data[offset + global_start + threadIdx.x] = threadIdx.x; } } } int runMeasurement( const unsigned num_threads, const unsigned num_blocks, const unsigned c) { curandGenerator_t gen; curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT); using DataT = float; //const unsigned num_threads = 32; //const unsigned num_blocks = 2; const size_t sh_mem_size = num_threads * sizeof(DataT); //const unsigned c = 2; // number of results per block //const size_t N = num_threads * num_blocks * c; // total number of elements const size_t size = c * num_threads * num_blocks * sizeof(DataT); // total size in Bytes //cout << " - size = " << size / 1e9 << " GB" << endl; //cout << " - num_threads = " << num_threads << endl; //cout << " - num_blocks = " << num_blocks << endl; //cout << " - sh_mem_size = " << sh_mem_size << endl; //cout << " - N = " << N << endl; //cout << " - c = " << c << endl; DataT* in_data; DataT* out_data; DataT* partial_data; unsigned* counter; cudaSetDevice(0); cudaMalloc(&in_data, size); cudaMallocManaged(&partial_data, num_blocks * c * sizeof(DataT)); curandGenerateUniform(gen, (float*) in_data, size / sizeof(float)); // fill with random bits cudaDeviceSynchronize(); cudaSetDevice(1); cudaMallocManaged(&counter, num_blocks * sizeof(unsigned)); cudaMemAdvise(counter, num_blocks * sizeof(unsigned), cudaMemAdviseSetPreferredLocation, cudaCpuDeviceId); cudaMalloc(&out_data, size); cudaMemset(out_data, 0, size); cudaSetDevice(0); // cout << " - Going to start the kernel" << endl; cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start, 0); producer_kernel<DataT, false><<<num_blocks, num_threads, sh_mem_size>>>(in_data, partial_data, counter, c); cudaSetDevice(1); consumer_kernel<DataT, false><<<num_blocks, num_threads>>>(out_data, partial_data, counter, c); cudaSetDevice(0); gpuErrchk(cudaEventRecord(stop, 0)); gpuErrchk(cudaEventSynchronize(stop)); float time_in_ms; gpuErrchk(cudaEventElapsedTime(&time_in_ms, start, stop)); gpuErrchk(cudaSetDevice(0)); gpuErrchk(cudaDeviceSynchronize()); cudaEventDestroy(start); cudaEventDestroy(stop); cudaFree(in_data); cudaFree(partial_data); cudaFree(counter); curandDestroyGenerator(gen); //cout << "time:" << time_in_ms << endl; cout << time_in_ms; return 0; } int main() { cout << "np.array("; int tArray[] = {32,256,512,1024}; int tLength = 4; int bArray[] = {1,2,4,8}; int bLength = 4; int cArray[] = {256, 1024, 4096, 16384, 65536}; int cLength = 5; cout << "("; for (int t = 0; t < tLength; t++) { cout << "("; for (int b = 0; b < bLength; b++) { cout << "("; for (int c = 0; c < cLength; c++) { //run kernel runMeasurement(tArray[t], bArray[b], cArray[c]); if(c < cLength - 1) { cout << ","; } } cout << ")"; if(b < bLength - 1) { cout << ","; } } cout << ")"; if(t < tLength - 1) { cout << ","; } } cout << ")"; // runMeasurement(512,4,4096); //b, t, c cout << ")" << std::endl; }
5,026
#include <thrust/device_vector.h> #include <thrust/host_vector.h> #include <thrust/random/linear_congruential_engine.h> #include <thrust/random/uniform_real_distribution.h> #include <iostream> // nvcc -std=c++14 -O3 tarefa2.cu -o t2 && ./t2 struct fillRand { thrust::uniform_real_distribution<double> dist; thrust::minstd_rand rng; fillRand(thrust::uniform_real_distribution<double> dist, thrust::minstd_rand rng) : dist(dist), rng(rng) {} __host__ __device__ double operator()(const double &i){ rng.seed(i*100000); return dist(rng); } }; int main(){ thrust::minstd_rand rng; thrust::uniform_real_distribution<double> dist(25, 40); thrust::device_vector<double> vetor(10, 0); thrust::counting_iterator<int> iter(0); thrust::transform(iter, iter + vetor.size(), vetor.begin(), fillRand(dist, rng)); thrust::host_vector<double> host(vetor); for (auto i = host.begin(); i != host.end(); i++) std::cout << *i << " "; printf("\n"); }
5,027
#include <iostream> #include <string.h> #include <stdio.h> #include <math.h> using namespace std; namespace myNamespace_00_01{ static double* hmem_i; static double* hmem_o; static double* dmem_i; static double* dmem_o; static cudaStream_t stream; static int nb = 1; //1024*1024*64*2; // max 1024*1024*64*2 static int nthre = 1; // max 65535 static int nthre_total = nb*nthre; static int nword = 1024*1024; static int mem_size = sizeof(double) * nword; static int mem_size_o = nthre_total*sizeof(double); __device__ double myDeviceFunc(double* in, int nword) { double z=0.0; for(int i=0; i<nword; i++) z += in[i]; return (z); } __global__ void kernel(double* in, double* out, int nword) { int tid = threadIdx.x; int bid = blockIdx.x; int index = blockDim.x*bid + tid; double z = myDeviceFunc(in, nword); out[index] = z; } void initialize() { static bool is_first = true; if(false == is_first) return; // setup stream cudaStreamCreate(&stream); // input buffer (Host) hmem_i = (double*) malloc(mem_size); for(int i=0; i<nword; i++) hmem_i[i] = (double)i; // input buffer (GPU) cudaMalloc( (void**) &dmem_i, mem_size); cudaMemcpyAsync(dmem_i, hmem_i, mem_size, cudaMemcpyHostToDevice, stream); // output buffer (Host/GPU) cudaMalloc( (void**) &dmem_o, mem_size_o); hmem_o = (double*) malloc(mem_size_o); printf("stream #: %d\n", stream); printf("# threads: %d \n", nthre_total); printf("mem_size: %d MB\n", mem_size >> 20); printf("mem_size_o: %d kB\n", mem_size_o >> 10); is_first = false; } void run(int n_run) { kernel<<< nb, nthre, 0, stream >>>(dmem_i, dmem_o, nword); cudaMemcpyAsync(hmem_o, dmem_o, mem_size_o, cudaMemcpyDeviceToHost, stream); /* for(int i=0; i<nthre_total; i++){ double z = hmem_o[i]; if(i>(nthre_total-4)) printf("%d, %f\n", i, z); } */ printf("%d: %d, %f\n", stream, nthre_total-1, hmem_o[nthre_total-1]); // if(n_run % 32 == 31) cudaStreamSynchronize(stream); return; } void finalize(){ free(hmem_i); free(hmem_o); cudaFree(dmem_i); cudaFree(dmem_o); cudaStreamSynchronize(stream); cudaStreamDestroy(stream); } }
5,028
#include "includes.h" using namespace std; //Check for edges valid to be part of augmented path //Update frontier __global__ void k2(const int N, bool* visited, int* frontier, bool* new_frontier) { int count = 0; for(int i=0;i<N;i++) { if(new_frontier[i]) { new_frontier[i] = false; frontier[++count] = i; visited[i] = true; } } frontier[0] = count; }
5,029
#include "csv_data.cuh" CSV_Data::CSV_Data(string fileName, bool printInfo) { (this->resultFile).open(fileName, ios::out); (this->resultFile) << "Target,#Threads,#ThreadBlks,ExecTime\n"; this->printInfo = printInfo; } CSV_Data::~CSV_Data() { resultFile.close(); } void CSV_Data::AddData(string Target, int numThreads, int numThreadBlks, float ExecTime) { if (printInfo) { printf("[INFO] Target: %10s, numThreads: %5d, numThreadBlks: %5d, ExecTime: %.5f\n", Target.c_str(), numThreads, numThreadBlks, ExecTime); } (this->resultFile) << Target << "," << numThreads << "," << numThreadBlks << "," << ExecTime << "\n"; } bool CSV_Data::CompareDeviceData(void *device_result, size_t device_result_size) { if (this->host_result == NULL) { printf("[WARN] Host result data is empty.\n"); return false; } if (this->host_result_size == device_result_size) { if (memcmp(device_result, this->host_result, device_result_size) == 0) { return true; } else { printf("[WARN] Data is not matching with host's and device's.\n"); return false; } } else { printf("[WARN] Data size is not matching with host's and device's.\n" " Device: %lu, Host: %lu\n", device_result_size, this->host_result_size); return false; } } void CSV_Data::AddHostData(float ExecTime, void *host_result, size_t host_result_size) { this->host_exec_time = ExecTime; if (this->host_result) { free(this->host_result); } this->host_result = (void *)malloc(host_result_size); this->host_result_size = host_result_size; memcpy(this->host_result, host_result, host_result_size); this->AddData("host", 1, 1, ExecTime); } void CSV_Data::AddDeviceData(int numThreads, int numThreadBlks, float ExecTime, void *device_result, size_t device_result_size) { this->CompareDeviceData(device_result, device_result_size); this->AddData("device", numThreads, numThreadBlks, ExecTime); }
5,030
#include <stdio.h> #include <iostream> #include <vector> #define CUDA_CHECK(condition) \ /* Code block avoids redefinition of cudaError_t error */ \ do { \ cudaError_t error = condition; \ if (error != cudaSuccess) { \ std::cout << cudaGetErrorString(error) << std::endl; \ } \ } while (0) #define CUDA_1D_KERNEL_LOOP(i, n) \ for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; \ i += blockDim.x * gridDim.x) const int block_num = 512; #define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0)) const int threadsPerBlock = sizeof(unsigned long long) * 8; __global__ void cumsumKernel(int b,int n,const float * __restrict__ inp,float * __restrict__ out){ const int BlockSize=2048; const int paddingLevel=5; __shared__ float buffer4[BlockSize*4]; __shared__ float buffer[BlockSize+(BlockSize>>paddingLevel)]; for (int i=blockIdx.x;i<b;i+=gridDim.x){ float runningsum=0,runningsum2=0; for (int j=0;j<n;j+=BlockSize*4){ int n24_i=min(n-j,BlockSize*4); int n24=(n24_i+3)&~3; int n2=n24>>2; for (int k=threadIdx.x*4;k<n24_i;k+=blockDim.x*4){ if (k+3<n24_i){ float v1=inp[i*n+j+k]; float v2=inp[i*n+j+k+1]; v2+=v1; float v3=inp[i*n+j+k+2]; float v4=inp[i*n+j+k+3]; v4+=v3; v3+=v2; v4+=v2; buffer4[k]=v1; buffer4[k+1]=v2; buffer4[k+2]=v3; buffer4[k+3]=v4; buffer[(k>>2)+(k>>(2+paddingLevel))]=v4; }else{ float v=0; for (int k2=k;k2<n24_i;k2++){ v+=inp[i*n+j+k2]; buffer4[k2]=v; } for (int k2=n24_i;k2<n24;k2++){ buffer4[k2]=v; } buffer[(k>>2)+(k>>(2+paddingLevel))]=v; } } int u=0; for (;(2<<u)<=n2;u++){ __syncthreads(); for (int k=threadIdx.x;k<int(n2>>(u+1));k+=blockDim.x){ int i1=(((k<<1)+2)<<u)-1; int i2=(((k<<1)+1)<<u)-1; i1+=i1>>paddingLevel; i2+=i2>>paddingLevel; buffer[i1]+=buffer[i2]; } } u--; for (;u>=0;u--){ __syncthreads(); for (int k=threadIdx.x;k<int((n2-(1<<u))>>(u+1));k+=blockDim.x){ int i1=(((k<<1)+3)<<u)-1; int i2=(((k<<1)+2)<<u)-1; i1+=i1>>paddingLevel; i2+=i2>>paddingLevel; buffer[i1]+=buffer[i2]; } } __syncthreads(); for (int k=threadIdx.x*4;k<n24;k+=blockDim.x*4){ if (k!=0){ int k2=((k>>2)-1)+(((k>>2)-1)>>paddingLevel); buffer4[k]+=buffer[k2]; buffer4[k+1]+=buffer[k2]; buffer4[k+2]+=buffer[k2]; buffer4[k+3]+=buffer[k2]; } } __syncthreads(); for (int k=threadIdx.x;k<n24_i;k+=blockDim.x){ out[i*n+j+k]=buffer4[k]+runningsum; } float t=buffer[(n2-1)+((n2-1)>>paddingLevel)]+runningsum2; float r2=runningsum+t; runningsum2=t-(r2-runningsum); runningsum=r2; __syncthreads(); } } } __global__ void binarysearchKernel(int b,int n,int m,const float * __restrict__ dataset,const float * __restrict__ query, int * __restrict__ result){ int base=1; while (base<n) base<<=1; for (int i=blockIdx.x;i<b;i+=gridDim.x){ for (int j=blockIdx.y*blockDim.x+threadIdx.x;j<m;j+=blockDim.x*gridDim.y){ float q=query[i*m+j]*dataset[i*n+n-1]; int r=n-1; for (int k=base;k>=1;k>>=1) if (r>=k && dataset[i*n+r-k]>=q) r-=k; result[i*m+j]=r; } } } template <unsigned int BlockSize> __global__ void farthestpointsamplingKernel(int b,int n,int c,int m,const float * __restrict__ dataset,float * __restrict__ temp,int * __restrict__ idxs){ if (m<=0) return; // const int BlockSize=512; __shared__ float dists[BlockSize]; __shared__ int dists_i[BlockSize]; for (int i=blockIdx.x;i<b;i+=gridDim.x){ int old=0; if (threadIdx.x==0) idxs[i*m+0]=old; //initialize temp for (int j=threadIdx.x;j<n;j+=blockDim.x){ temp[blockIdx.x*n+j]=1e38; } __syncthreads(); for (int j=1;j<m;j++){ int besti=0; float best=-1; for (int k=threadIdx.x;k<n;k+=blockDim.x){ float td=temp[blockIdx.x*n+k]; float d = 0; float p1, p2; for (int l=0;l<c;l++){ p1 = dataset[i*n*c+old*c+l]; p2 = dataset[i*n*c+k*c+l]; d += (p2-p1) * (p2-p1); } float d2=min(d,td); if (d2!=td) temp[blockIdx.x*n+k]=d2; if (d2>best){ best=d2; besti=k; } } dists[threadIdx.x]=best; dists_i[threadIdx.x]=besti; for (int u=0;(1<<u)<blockDim.x;u++){ __syncthreads(); if (threadIdx.x<(blockDim.x>>(u+1))){ int i1=(threadIdx.x*2)<<u; int i2=(threadIdx.x*2+1)<<u; if (dists[i1]<dists[i2]){ dists[i1]=dists[i2]; dists_i[i1]=dists_i[i2]; } } } __syncthreads(); old=dists_i[0]; if (threadIdx.x==0) idxs[i*m+j]=old; } } } template <unsigned int BlockSize> __global__ void farthestpointsamplingwithdistKernel(int b,int n,int m,const float * __restrict__ dataset,float * __restrict__ temp,int * __restrict__ idxs){ if (m<=0) return; // const int BlockSize=512; __shared__ float dists[BlockSize]; __shared__ int dists_i[BlockSize]; for (int i=blockIdx.x;i<b;i+=gridDim.x){ int old=0; if (threadIdx.x==0) idxs[i*m+0]=old; //initialize temp for (int j=threadIdx.x;j<n;j+=blockDim.x){ temp[blockIdx.x*n+j]=1e38; } __syncthreads(); for (int j=1;j<m;j++){ int besti=0; float best=-1; for (int k=threadIdx.x;k<n;k+=blockDim.x){ float td=temp[blockIdx.x*n+k]; float d = 0; d = dataset[i * n * n + old * n + k]; float d2=min(d,td); if (d2!=td) temp[blockIdx.x*n+k]=d2; if (d2>best){ best=d2; besti=k; } } dists[threadIdx.x]=best; dists_i[threadIdx.x]=besti; for (int u=0;(1<<u)<blockDim.x;u++){ __syncthreads(); if (threadIdx.x<(blockDim.x>>(u+1))){ int i1=(threadIdx.x*2)<<u; int i2=(threadIdx.x*2+1)<<u; if (dists[i1]<dists[i2]){ dists[i1]=dists[i2]; dists_i[i1]=dists_i[i2]; } } } __syncthreads(); old=dists_i[0]; if (threadIdx.x==0) idxs[i*m+j]=old; } } } template <unsigned int BlockSize> __global__ void farthestpointsamplingwithpreidxKernel(int b,int n,int c,int m,int m1,const float * __restrict__ dataset,const int * __restrict__ preidx,float * __restrict__ temp,int * __restrict__ idxs){ // b: batch_size, n: ndataset, c: channel_num, m: points_num after fps, m1: preidx number // dataset: [b, n, c] preidx: [b, m1], temp: [b, n], idxs: [b, m] if (m<=0) return; // const int BlockSize=512; __shared__ float dists[BlockSize]; __shared__ int dists_i[BlockSize]; for (int i=blockIdx.x;i<b;i+=gridDim.x){ for (int j=threadIdx.x;j<n;j+=blockDim.x){ temp[blockIdx.x*n+j]=1e38; } int pre_idx; for (int j=threadIdx.x;j<n;j+=blockDim.x){ // update temp metrics float pre_best = 1e38; float pre_p1, pre_p2; for (int k=0; k<m1; k++){ pre_idx = preidx[i * m1 + k]; float pre_d = 0; for (int l=0; l < c; l++){ pre_p1 = dataset[i * n * c + pre_idx * c + l]; pre_p2 = dataset[i * n * c + j * c + l]; pre_d += (pre_p2 - pre_p1) * (pre_p2 - pre_p1); } pre_best = min(pre_best, pre_d); } temp[blockIdx.x*n+j] = pre_best; } // then find current smallest distance as current old __syncthreads(); int old=0; float pre_best = -1; for (int j=0; j<n; j++){ if (pre_best < temp[blockIdx.x*n+j]){ pre_best = temp[blockIdx.x*n+j]; old = j; } } if (threadIdx.x==0) idxs[i*m+0]=old; //initialize temp __syncthreads(); for (int j=1;j<m;j++){ int besti=0; float best=-1; for (int k=threadIdx.x;k<n;k+=blockDim.x){ float td=temp[blockIdx.x*n+k]; float d = 0; float p1, p2; for (int l=0;l<c;l++){ p1 = dataset[i*n*c+old*c+l]; p2 = dataset[i*n*c+k*c+l]; d += (p2-p1) * (p2-p1); } float d2=min(d,td); if (d2!=td) temp[blockIdx.x*n+k]=d2; if (d2>best){ best=d2; besti=k; } } dists[threadIdx.x]=best; dists_i[threadIdx.x]=besti; for (int u=0;(1<<u)<blockDim.x;u++){ __syncthreads(); if (threadIdx.x<(blockDim.x>>(u+1))){ int i1=(threadIdx.x*2)<<u; int i2=(threadIdx.x*2+1)<<u; if (dists[i1]<dists[i2]){ dists[i1]=dists[i2]; dists_i[i1]=dists_i[i2]; } } } __syncthreads(); old=dists_i[0]; if (threadIdx.x==0) idxs[i*m+j]=old; } } } // inp: [b, n, c] idx: [b, m] // out: [b, m, c] __global__ void gatherpointKernel(int b,int n,int m,int c,const float * __restrict__ inp,const int * __restrict__ idx,float * __restrict__ out){ int loop_time = b * m * c; CUDA_1D_KERNEL_LOOP(index, loop_time){ int cur_batch_size = index / (m * c); int cur_point_idx = index / c; int cur_channel = index % c; int a=idx[cur_point_idx]; int current_idx = cur_batch_size * (n * c) + a * c + cur_channel; out[index] = inp[current_idx]; } } // out_g: [b, m, c] idx: [b, m] // inp_g: [b, n, c] __global__ void scatteraddpointKernel(int b,int n,int m,int c,const float * __restrict__ out_g,const int * __restrict__ idx,float * __restrict__ inp_g){ int loop_time = b * m * c; CUDA_1D_KERNEL_LOOP(index, loop_time){ int cur_batch_size = index / (m * c); int cur_point_idx = index / c; int cur_channel = index % c; int a = idx[cur_point_idx]; int current_idx = cur_batch_size * n * c + a * c + cur_channel; atomicAdd(&inp_g[current_idx],out_g[index]); } } // inp: [b, n, c] mask: [b, n] // out: [b, proposal_num, c] __global__ void GatherByMaskKernel(int b,int n,int c,int proposal_num,const float *inp,const float *mask,float *out){ for (int cur_batch=blockIdx.x; cur_batch<b; cur_batch+=gridDim.x){ const float *cur_inp = inp + cur_batch * n * c; const float *cur_mask = mask + cur_batch * n; float* cur_out = out + cur_batch * proposal_num * c; int proposal_cnt = 0; int loop_time, tmp_channel_idx; for (int cur_pts=0; cur_pts<n; cur_pts++){ if(int(cur_mask[cur_pts]) == 0) continue; if(proposal_cnt == proposal_num) break; // a valid proposal if (proposal_cnt == 0){ loop_time = proposal_num * c; for (int i=threadIdx.x; i<loop_time; i+=blockDim.x){ tmp_channel_idx = i % c; cur_out[i] = cur_inp[cur_pts * c + tmp_channel_idx]; } __syncthreads(); } else { loop_time = c; for (int i=threadIdx.x; i<loop_time; i+=blockDim.x){ cur_out[proposal_cnt * c + i] = cur_inp[cur_pts * c + i]; } __syncthreads(); } proposal_cnt += 1; } } } void cumsumLauncher(int b,int n,const float * inp,float * out){ cumsumKernel<<<32,512>>>(b,n,inp,out); } //require b*n working space void probsampleLauncher(int b,int n,int m,const float * inp_p,const float * inp_r,float * temp,int * out){ cumsumKernel<<<32,512>>>(b,n,inp_p,temp); binarysearchKernel<<<dim3(32,8,1),512>>>(b,n,m,temp,inp_r,out); } //require 32*n working space void farthestpointsamplingLauncher(int b,int n,int c,int m,const float * inp,float * temp,int * out){ farthestpointsamplingKernel<1024><<<b,1024>>>(b,n,c,m,inp,temp,out); } //require 32*n working space void farthestpointsamplingwithdistLauncher(int b,int n,int m,const float * inp,float * temp,int * out){ farthestpointsamplingwithdistKernel<1024><<<b,1024>>>(b,n,m,inp,temp,out); } //require 32*n working space void farthestpointsamplingwithpreidxLauncher(int b,int n,int c,int m,int m1,const float * inp, const int* preidx,float * temp,int * out){ farthestpointsamplingwithpreidxKernel<1024><<<b,1024>>>(b,n,c,m,m1,inp,preidx,temp,out); } void gatherpointLauncher(int b,int n,int m,int c,const float * inp,const int * idx,float * out){ gatherpointKernel<<<block_num,threadsPerBlock>>>(b,n,m,c,inp,idx,out); //int thread_num = 512 / b; // gatherpointKernel<<<dim3(256,8,1),512>>>(b,n,m,inp,idx,out); } void scatteraddpointLauncher(int b,int n,int m,int c,const float * out_g,const int * idx,float * inp_g){ scatteraddpointKernel<<<block_num,threadsPerBlock>>>(b,n,m,c,out_g,idx,inp_g); } void GatherByMaskLauncher(int b,int n,int c,int proposal_num,const float *inp,const float *mask,float *out){ GatherByMaskKernel<<<block_num,threadsPerBlock>>>(b,n,c,proposal_num,inp,mask,out); }
5,031
#include <stdio.h> #include "time.h" #include <stdlib.h> #include <limits.h> /* The old-fashioned CPU-only way to add two vectors */ void add_vectors_host(int *result, int *a, int *b, int n) { for (int i=0; i<n; i++) result[i] = a[i] + b[i]; } /* The kernel that will execute on the GPU */ __global__ void add_vectors_kernel(int *result, int *a, int *b, int n) { int idx = blockDim.x * blockIdx.x + threadIdx.x; // If we have more threads than the magnitude of our vector, we need to // make sure that the excess threads don't try to save results into // unallocated memory. if (idx < n) result[idx] = a[idx] + b[idx]; } /* This function encapsulates the process of creating and tearing down the * environment used to execute our vector addition kernel. The steps of the * process are: * 1. Allocate memory on the device to hold our vectors * 2. Copy the vectors to device memory * 3. Execute the kernel * 4. Retrieve the result vector from the device by copying it to the host * 5. Free memory on the device */ void add_vectors_dev(int *result, int *a, int *b, int n) { // Step 1: Allocate memory int *a_dev, *b_dev, *result_dev; // Since cudaMalloc does not return a pointer like C's traditional malloc // (it returns a success status instead), we provide as it's first argument // the address of our device pointer variable so that it can change the // value of our pointer to the correct device address. cudaMalloc((void **) &a_dev, sizeof(int) * n); cudaMalloc((void **) &b_dev, sizeof(int) * n); cudaMalloc((void **) &result_dev, sizeof(int) * n); // Step 2: Copy the input vectors to the device cudaMemcpy(a_dev, a, sizeof(int) * n, cudaMemcpyHostToDevice); cudaMemcpy(b_dev, b, sizeof(int) * n, cudaMemcpyHostToDevice); // Step 3: Invoke the kernel // We allocate enough blocks (each 512 threads long) in the grid to // accommodate all `n` elements in the vectors. The 512 long block size // is somewhat arbitrary, but with the constraint that we know the // hardware will support blocks of that size. dim3 dimGrid((n + 512 - 1) / 512, 1, 1); dim3 dimBlock(512, 1, 1); add_vectors_kernel<<<dimGrid, dimBlock>>>(result_dev, a_dev, b_dev, n); // Step 4: Retrieve the results cudaMemcpy(result, result_dev, sizeof(int) * n, cudaMemcpyDeviceToHost); // Step 5: Free device memory cudaFree(a_dev); cudaFree(b_dev); cudaFree(result_dev); } void print_vector(int *array, int n) { int i; for (i=0; i<n; i++) printf("%d ", array[i]); printf("\n"); } int main(void) { const int CONST_VEC = 4; // It looks like the crossover is just above this. At 2.6e8, GPU calculation time cuts in half (roughly). int* a = (int *)malloc(CONST_VEC * sizeof(int)); int* b = (int *)malloc(CONST_VEC * sizeof(int)); int* host_result = (int *)malloc(CONST_VEC * sizeof(int)); int* device_result = (int *)malloc(CONST_VEC * sizeof(int)); clock_t ar = clock(); srand((unsigned)time(0)); for(int i=0; i<CONST_VEC; i++){ a[i] = 1; b[i] = 1; } ar = clock() - ar; printf ("It took me %ld clicks (%f seconds).\n",ar,((float)ar)/CLOCKS_PER_SEC); printf("The CPU's answer: "); clock_t t = clock(); add_vectors_host(host_result, a, b, CONST_VEC); t = clock() - t; //print_vector(host_result, CONST_VEC); printf ("It took me %ld clicks (%f seconds).\n",t,((float)t)/CLOCKS_PER_SEC); printf("The GPU's answer: "); clock_t t2 = clock(); add_vectors_dev(device_result, a, b, CONST_VEC); t2 = clock() - t2; //print_vector(device_result, CONST_VEC); printf ("It took me %ld clicks (%f seconds).\n",t2,((float)t2)/CLOCKS_PER_SEC); for(int i=0; i<CONST_VEC; i++) { printf("%i", device_result[i]); printf(" "); } printf("\n\n"); for(int i=0; i<CONST_VEC; i++) { printf("%i",host_result[i]); printf(" "); } return 0; }
5,032
#include <stdio.h> #include <stdlib.h> #include <math.h> #include <cuda.h> #define PI 3.1415 void gauss (int sigma, int gauss_matrix[][5]); void gpuComputing(int gauss_matrix[][5], int** image_matrix, int** final_matrix, int height, int width); __global__ void kernel(int* image, int* final, int* gauss, int pitch, int pitch_i, int pitch_f, int height, int width) { int abs_Pos, c, r; int x, aux, aux2; int gauss_element; int image_element; int tid = threadIdx.x; float result; /** Todos los threads de un bloque comparten la misma fila **/ __shared__ int fila; fila = blockIdx.x + 2; /** Los threads van recorriendo la fila cada 512 casilleros **/ while(tid < width - 4) { result = 0; /** Posicion absoluta del thread en el arreglo con la imagen **/ abs_Pos = fila * width + (tid +2); /** Posicion x en la matriz **/ x = abs_Pos % width; /** Recorre la matriz de gauss y simultaneamente copia los valores necesarios de ** la matriz de imagen para realizar el calculo **/ for (r = 0; r < 5; ++r) { aux = r - 2; int* gauss_row = (int*)((char*)gauss + r * pitch); int* image_row = (int*)((char*)image + ((fila + aux) * pitch)); for (c = 0; c < 5; ++c) { aux2 = c - 2; gauss_element = gauss_row[c]; image_element = image_row[x + aux2]; result += (gauss_element * image_element)/273; } } /** Guarda el valor a la celda correspondiente en la imagen final **/ int* final_row = (int*)((char*)final + blockIdx.x * pitch_f); final_row[tid] = result; /** Incremento para realizar todas las operaciones de una fila en un bloque **/ tid += 512; } return; } int main(int argc, char *argv[]) { int width, height, i, j; int **image_matrix, **final_matrix, **auxiliar_matrix; int gauss_matrix[5][5]; int sigma = (int)strtod(argv[2], NULL); int *temp, *temp2, *temp3; /** Lectura del archivo contenedor de los datos **/ FILE *in_f = fopen(argv[1], "r"); /** Lee el ancho de la matriz **/ fscanf(in_f, "%d", &width); /** Lee el alto de la matriz **/ fscanf(in_f, "%d", &height); /** Arreglo dinamico 2D para almacenar la matriz **/ image_matrix = (int**)malloc(width * sizeof(int*)); temp = (int*)malloc(width * height * sizeof(int)); for(i = 0; i < width; i++) image_matrix[i] = temp + (i * height); /** Arrelgo dinamico 2D con 4 filas y columnas adicionales **/ auxiliar_matrix = (int**)malloc((width + 4) * sizeof(int*)); temp2 = (int*)malloc((width + 4) * (height + 4) * sizeof(int)); for(i = 0; i < (width + 4); i++) auxiliar_matrix[i] = temp2 + (i * (height + 4)); /** Arreglo dinamico 2D para almacenar el resultado final **/ final_matrix = (int**)malloc(width * sizeof(int*)); temp3 = (int*)malloc(width * height * sizeof(int)); for(i = 0; i < width; i++) final_matrix[i] = temp3 + (i * height ); /** Lee los valores del archivo y los almacena en image_matrix **/ for (i = 0; i < width; i++) { for(j = 0; j < height; j++) fscanf(in_f, "%d", &image_matrix[i][j]); } fclose(in_f); /** Calculo de la matriz con los valores del filtro **/ gauss(sigma, gauss_matrix); /** Desplazamiento de la matriz 2 filas hacia arriba y 2 hacia la izquierda **/ for(i = 2; i < width + 2; i++) { for(j = 2; j < height + 2; j++){ auxiliar_matrix[i][j] = image_matrix[i-2][j-2]; } } /** Reflexion de filas **/ for(i = 2; i < width + 2; i++) { auxiliar_matrix[i][0] = auxiliar_matrix[i][4]; auxiliar_matrix[i][1] = auxiliar_matrix[i][3]; auxiliar_matrix[i][height + 2] = auxiliar_matrix[i][height]; auxiliar_matrix[i][height + 3] = auxiliar_matrix[i][height-1]; } /** Reflexion de columnas **/ for(i = 0; i < height + 4; i++) { auxiliar_matrix[0][i] = auxiliar_matrix[4][i]; auxiliar_matrix[1][i] = auxiliar_matrix[3][i]; auxiliar_matrix[width + 2][i] = auxiliar_matrix[width][i]; auxiliar_matrix[width + 3][i] = auxiliar_matrix[width-1][i]; } gpuComputing(gauss_matrix, auxiliar_matrix, final_matrix, height, width); /** Abre archivo que contendra los valores finales **/ FILE *out = fopen("salida", "w"); for (i = 0; i < height; i++) { for (j = 0; j < width; j++) { if ( j != width - 1) fprintf(out, "%d\t", final_matrix[i][j]); else if (i != height-1) fprintf(out, "%d\n", final_matrix[i][j]); else fprintf(out, "%d", final_matrix[i][j]); } } fclose(out); /** Liberacion de memoria correspondiente a las 3 matrices **/ free(temp); free(temp2); free(temp3); return 0; } void gauss(int sigma, int gauss_matrix[][5]) { int i, j; int x = -2, y = 2; double u, v, varianza; double operation; for(i = 0; i < 5; i++) { for(j = 0; j < 5; j++) { u = pow(x,2); v = pow(y,2); varianza = pow(sigma,2); operation = 273.0*exp((-u-v)/(2.0*varianza))/(2.0*PI*varianza); gauss_matrix[i][j] = (int)ceil(operation); x++; } y--; x = -2; } return; } void gpuComputing(int gauss_matrix[][5], int** image_matrix, int** final_matrix, int height, int width) { int *d_image, *d_final, *d_gauss; int r_height = height + 4; int r_width = width + 4; size_t pitch, pitch_i, pitch_f; /** Metodo que entrega la direccion del arreglo en el dispositivo **/ cudaMallocPitch(&d_gauss, &pitch, 5 * sizeof(int), 5); cudaMallocPitch(&d_image, &pitch_i, r_width * sizeof(int), r_height); cudaMallocPitch(&d_final, &pitch_f, width * sizeof(int), height); /** Traspaso de los datos desde el host al dispositivo **/ cudaMemcpy2D(d_gauss, pitch, *gauss_matrix, 5 * sizeof(int), 5 * sizeof(int), 5, cudaMemcpyHostToDevice); cudaMemcpy2D(d_image, pitch_i, *image_matrix, r_width * sizeof(int), r_width * sizeof(int), r_height, cudaMemcpyHostToDevice); /** Llamada al kernel, si bien su implementacion pudo ser mas eficiente, los tiempos permanecen bajos **/ kernel<<<height, 512>>>(d_image, d_final, d_gauss, pitch, pitch_i, pitch_f, r_height, r_width); /** Traspaso del resultado final **/ cudaMemcpy2D(*final_matrix, width*sizeof(int), d_final, pitch_f, width*sizeof(int), height, cudaMemcpyDeviceToHost); cudaFree(d_final); cudaFree(d_image); cudaFree(d_gauss); return; }
5,033
extern "C" __global__ void kernelFunction(int *input) { input[threadIdx.x] = 32 - threadIdx.x; }
5,034
#include "lsystem.cuh"
5,035
#include "includes.h" #define ITER 10000000000 // Number of bins #define NUMBLOCKS 13 // Number of thread blocks #define NUMTHREADS 192 // Number of threads per block int tid; float pi; // Kernel // Main __global__ void pic(float *sum, int nbin, float step, int nthreads, int nblocks) { int i; float x; int idx = blockIdx.x*blockDim.x+threadIdx.x; // Sequential thread index across the blocks for (i=idx; i< nbin; i+=nthreads*nblocks) { x = (i+0.5)*step; sum[idx] += 4.0/(1.0+x*x); } }
5,036
#include <stdio.h> #include <stdlib.h> #include <time.h> #include <assert.h> #include <sys/time.h> #define THREADS 512 #ifdef __cplusplus extern "C" { #endif int cuda_sort(int number_of_elements, float *a) { return 0; } #ifdef __cplusplus } #endif
5,037
// // Created by kindr on 2021/5/8. // #include "multiKernelConcurrent.cuh" #include "../../common/utils.cuh" #include <cstdio> const int N = 1 << 25; __global__ void math_kernel1(int n) { double sum = 0; for (int i = 0; i < n; i++) sum += tan(0.1) * tan(0.1); printf("sum=%g\n", sum); } __global__ void math_kernel2(int n) { double sum = 0; for (int i = 0; i < n; i++) sum += tan(0.1) * tan(0.1); printf("sum=%g\n", sum); } void multiKernelConcurrent() { int n_stream = 4; size_t nStreamBytes = n_stream * sizeof(cudaStream_t); auto *stream = static_cast<cudaStream_t *>(malloc(nStreamBytes)); for (int i = 0; i < n_stream; i++) { cudaStreamCreate(&stream[i]); } CHECK(cudaGetLastError()); for (int i = 0; i < n_stream; i++) { math_kernel1<<<1, 1, 0, stream[i]>>>(N); math_kernel2<<<1, 1, 0, stream[i]>>>(N); } for (int i = 0; i < n_stream; i++) { math_kernel1<<<1, 1>>>(N); math_kernel2<<<1, 1>>>(N); } cudaDeviceSynchronize(); CHECK(cudaGetLastError()); for (int i = 0; i < n_stream; i++) { cudaStreamDestroy(stream[i]); } free(stream); }
5,038
/* CUDA finite difference wave equation solver, written by * Jeff Amelang, 2012 * * Modified by Kevin Yuh, 2013-14 */ #include <cstdio> #include <cuda_runtime.h> #include "Cuda1DFDWave_cuda.cuh" /* kernel to calculate new displacements */ __global__ void GenerateDisplacements(float* dev_Data, int oldStart, int currentStart, int newStart, float courantSquared, int numberOfNodes) { unsigned int index = blockDim.x * blockIdx.x + threadIdx.x; // ignore both left and right boundary conditions (already set on GPU) while (index < numberOfNodes - 1) { if (index > 0) { // compute next displacement: // y(x,t+1) = 2 * y(x,t) - y(x,t-1) + courantSquared(y(x+1,t) // - 2 * y(x,t) + y(x-1,t)) dev_Data[newStart + index] = 2 * dev_Data[currentStart + index] - dev_Data[oldStart + index] + courantSquared * (dev_Data[currentStart + index + 1] - 2 * dev_Data[currentStart + index] + dev_Data[currentStart + index - 1]); } index += blockDim.x * gridDim.x; // increment index to next value } } /* helper function to call the kernel */ void kernelCall(float* dev_Data, int oldStart, int currentStart, int newStart, float courantSquared, int numberOfNodes, unsigned int blocks, unsigned int threadsPerBlock) { GenerateDisplacements <<<blocks, threadsPerBlock>>> (dev_Data, oldStart, currentStart, newStart, courantSquared, numberOfNodes); }
5,039
// CUDA kernel in C extern "C" __global__ void sincos_kernel(int nx, int ny, int nz, float* x, float* y, float* xy) { int i = threadIdx.x + blockIdx.x * blockDim.x; int j = threadIdx.y + blockIdx.y * blockDim.y; int k = threadIdx.z + blockIdx.z * blockDim.z; if ((i >= nx) || (j >= ny) || (k >= nz)) return; int index = i + j * nx + k * nx * ny; xy[index] = sin(x[index]) + cos(y[index]); }
5,040
#ifndef uint32_t #define uint32_t unsigned int #endif #define H0 0x6a09e667 #define H1 0xbb67ae85 #define H2 0x3c6ef372 #define H3 0xa54ff53a #define H4 0x510e527f #define H5 0x9b05688c #define H6 0x1f83d9ab #define H7 0x5be0cd19 __device__ uint rotr(uint x, int n) { if (n < 32) return (x >> n) | (x << (32 - n)); return x; } __device__ uint ch(uint x, uint y, uint z) { return (x & y) ^ (~x & z); } __device__ uint maj(uint x, uint y, uint z) { return (x & y) ^ (x & z) ^ (y & z); } __device__ uint sigma0(uint x) { return rotr(x, 2) ^ rotr(x, 13) ^ rotr(x, 22); } __device__ uint sigma1(uint x) { return rotr(x, 6) ^ rotr(x, 11) ^ rotr(x, 25); } __device__ uint gamma0(uint x) { return rotr(x, 7) ^ rotr(x, 18) ^ (x >> 3); } __device__ uint gamma1(uint x) { return rotr(x, 17) ^ rotr(x, 19) ^ (x >> 10); } __constant__ uint K[64]={ 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5, 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174, 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da, 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967, 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85, 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070, 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3, 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2 }; __device__ uint get_global_id() { uint blockId, threadsPerBlock; blockId = blockIdx.z * gridDim.x * gridDim.y + blockIdx.y * gridDim.x + blockIdx.x; threadsPerBlock = blockDim.x; return threadIdx.x + threadsPerBlock * blockId; } __global__ void crypt_kernel(ulong start, uint *prefix, ulong plen, uint mask, uint *match){ int t; uint W[80], rnd, id, A,B,C,D,E,F,G,H,T1,T2; uint Ws[16]; id = get_global_id(); //if (id == 0) { // printf("%08x\n", start); //} // brutforce is build up as: prefix | thr_id:04x | <rnd>:04x | start:08x for (t = 0; t < plen; ++t) { Ws[t] = prefix[t]; // printf("%04x", prefix[t]); } // printf("%04x\n", id); T1 = (id & 0xf) | (((id >> 4) & 0xf) << 8) | (((id >> 8) & 0xf) << 16) | (((id >> 12) & 0xf) << 24); T2 = (T1 & 0xe0e0e0e); T2 = ((((T2 >> 1) & T2) >> 2) | (((T2 >> 2) & T2) >> 1)) & 0x1010101; Ws[plen] = T1 + 0x30303030 + T2 * 0x27; T1 = (uint)(start >> 32); T1 = (T1 & 0xf) | (((T1 >> 4) & 0xf) << 8) | (((T1 >> 8) & 0xf) << 16) | (((T1 >> 12) & 0xf) << 24); T2 = (T1 & 0xe0e0e0e); T2 = ((((T2 >> 1) & T2) >> 2) | (((T2 >> 2) & T2) >> 1)) & 0x1010101; Ws[plen + 2] = T1 + 0x30303030 + T2 * 0x27; T1 = (uint)start; T1 = (T1 & 0xf) | (((T1 >> 4) & 0xf) << 8) | (((T1 >> 8) & 0xf) << 16) | (((T1 >> 12) & 0xf) << 24); T2 = (T1 & 0xe0e0e0e); T2 = ((((T2 >> 1) & T2) >> 2) | (((T2 >> 2) & T2) >> 1)) & 0x1010101; Ws[plen + 3] = T1 + 0x30303030 + T2 * 0x27; Ws[plen + 4] = 0x80000000; for (t = plen + 5; t < 15; ++t) { Ws[t] = 0; } Ws[15] = 128 + 32 * plen; // preparing buffer done /* if (id == 0) { printf("%016x: ", start); for (t = 0; t < 16; ++t) { printf("%08x", Ws[t]); } printf(" - %u\n", Ws[15]); } */ for (rnd = 0; rnd < 0x10000; ++rnd) { uint digest[8] = {H0, H1, H2, H3, H4, H5, H6, H7}; #pragma unroll for (t = 0; t < 16; ++t) { W[t] = Ws[t]; } T1 = (rnd & 0xf) | (((rnd >> 4) & 0xf) << 8) | (((rnd >> 8) & 0xf) << 16) | (((rnd >> 12) & 0xf) << 24); T2 = (T1 & 0xe0e0e0e); T2 = ((((T2 >> 1) & T2) >> 2) | (((T2 >> 2) & T2) >> 1)) & 0x1010101; W[plen + 1] = T1 + 0x30303030 + T2 * 0x27; A = digest[0] = H0; B = digest[1] = H1; C = digest[2] = H2; D = digest[3] = H3; E = digest[4] = H4; F = digest[5] = H5; G = digest[6] = H6; H = digest[7] = H7; for (t = 16; t < 64; t++) { W[t] = gamma1(W[t - 2]) + W[t - 7] + gamma0(W[t - 15]) + W[t - 16]; } for (t = 0; t < 64; t++) { T1 = H + sigma1(E) + ch(E, F, G) + K[t] + W[t]; T2 = sigma0(A) + maj(A, B, C); H = G; G = F; F = E; E = D + T1; D = C; C = B; B = A; A = T1 + T2; } digest[0] += A; if ((digest[0] & mask) == 0) { /* for (t = 0; t < 16; ++t) { printf("%08x", Ws[t]); } printf(" - %u\n", Ws[15]); */ match[0] = 1; match[1] = id; match[2] = rnd; } } }
5,041
extern "C" { __device__ int KerSobel(int a1, int a2, int a3, int a4, int a5, int a6) { return(a1 + 2 * a2 + a3 - (a4 + 2 * a5 + a6)); } __global__ void laplacian_filter(unsigned int *lpSrc,unsigned int *lpDst, int width, int height,int* gc_weight, int amplitude) { int x = blockIdx.x * blockDim.x + threadIdx.x; int y = blockIdx.y * blockDim.y + threadIdx.y; int idx = width * y + x; int xy[9]; int dr, dg, db; int powR, powG, powB; if(x >= 1 && x < width - 1 && y >= 1 && y < height - 1) { /*lpDst[idx] = 8 * lpSrc[idx] - lpSrc[idx - 1] - lpSrc[idx + 1] - lpSrc[idx - width] - lpSrc[idx + width] - lpSrc[idx - width - 1] - lpSrc[idx + width + 1] - lpSrc[idx - width + 1] - lpSrc[idx + width - 1]; */ xy[0]= lpSrc[idx - width - 1]; xy[1]= lpSrc[idx - width]; xy[2]= lpSrc[idx - width + 1]; xy[3]= lpSrc[idx - 1]; xy[4]= lpSrc[idx]; xy[5]= lpSrc[idx + 1]; xy[6]= lpSrc[idx + width - 1]; xy[7]= lpSrc[idx + width]; xy[8]= lpSrc[idx + width + 1]; dr = dg = db = 0; for(int i = 0; i < 9; i ++) { dr += gc_weight[i] * (0xFF & (xy[i] >> 16)); dg += gc_weight[i] * (0xFF & (xy[i] >> 8)); db += gc_weight[i] * (0xFF & (xy[i])); } /* Calculate power */ powR = amplitude * dr * dr >> 10; // * amplitude / 1024 powG = amplitude * dg * dg >> 10; powB = amplitude * db * db >> 10; if(powR > 255) { powR = 255; } if(powG > 255) { powG = 255; } if(powB > 255) { powB = 255; } lpDst[y * width + x] = (powR << 16) | (powG << 8) | (powB); } } }
5,042
#include "includes.h" __global__ void forwardDifference2DKernel(const int cols, const int rows, const float* data, float* dx, float* dy) { for (auto idy = blockIdx.y * blockDim.y + threadIdx.y + 1; idy < cols - 1; idy += blockDim.y * gridDim.y) { for (auto idx = blockIdx.x * blockDim.x + threadIdx.x + 1; idx < rows - 1; idx += blockDim.x * gridDim.x) { const auto index = idx + rows * idy; dx[index] = data[index + 1] - data[index]; dy[index] = data[index + rows] - data[index]; } } }
5,043
#include "cuda_runtime.h" #include "device_launch_parameters.h" #define _USE_MATH_DEFINES #include <math.h> __global__ void kernel(unsigned char* src) { __shared__ float temp[16][16]; int x = threadIdx.x + blockIdx.x * blockDim.x; int y = threadIdx.y + blockIdx.y * blockDim.y; int offset = x + y * blockDim.x * gridDim.x; const float period = 128.0f; temp[threadIdx.x][threadIdx.y] = 255 * (sinf(x * 2.0f * M_PI/period) + 1.0f) * (sinf(y * 2.0f * M_PI/period) + 1.0f) / 4.0f; __syncthreads(); src[offset * 4] = 0; src[offset * 4 + 1] = temp[15 - threadIdx.x][15 - threadIdx.y]; src[offset * 4 + 2] = 0; src[offset * 4 + 3] = 255; } extern "C" void GenerateBitmap(unsigned char* dst, int dimension) { int size = dimension * dimension * 4; cudaError_t status; unsigned char* src; status = cudaMalloc(&src, size); dim3 blocks(dimension / 16, dimension / 16); dim3 threads(16, 16); kernel<<<blocks, threads>>>(src); cudaMemcpy(dst, src, size, cudaMemcpyDeviceToHost); cudaFree(src); }
5,044
#include <stdio.h> #include <stdlib.h> #include <sys/time.h> #include <math.h> /* struct CDP { char name[256]; size_t totalGlobalMem; size_t sharedMemPerBlock; int regsPerBlock; int warpSize; size_t memPitch; int maxThreadsPerBlock; int maxThreadsDim[3]; int maxGridSize[3]; size_t totalConstMem; int major; int minor; int clockRate; size_t textureAlignment; int deviceOverlap; int multiProcessorCount; int kernelExecTimeoutEnabled; int integrated; int canMapHostMemory; int computeMode; int maxTexture1D; int maxTexture2D[2]; int maxTexture3D[3]; int maxTexture2DArray[3]; int concurrentKernels; }; */ int main( void ) { cudaDeviceProp prop; int i; int count; cudaGetDeviceCount (&count); for(i =0; i<count; ++i) { cudaGetDeviceProperties( &prop, i ); printf(" ---- General Information for Device %d ---\n", i); printf("Name : %s\n", prop.name); printf("Compute Capability : %d.%d/n", prop.major, prop.minor); printf("Clock Rate : %d\n", prop.clockRate); printf("Device Copy Overlap: " ); if(prop.deviceOverlap) printf( " Enabled\n"); else printf( " Disabled\n"); printf("Kernel Execution Timeout : "); if(prop.kernelExecTimeoutEnabled) printf( " Enabled\n"); else printf( " Disabled\n"); printf("\n ---- Memory Information for Device %d ---\n", i); printf("Total Global Memory: %ld\n", prop.totalGlobalMem); printf("Total Constant Memory: %ld\n", prop.totalConstMem); printf("Max mem pitch: %ld\n", prop.memPitch); printf("Texture Alignment: %ld\n", prop.textureAlignment); printf("\n --- MP Information for Device %d ---\n", i); printf("Multiprocessor Count: %d\n", prop.multiProcessorCount); printf("Shared mem per mp: %ld\n", prop.sharedMemPerBlock); printf("Registers per mp : %d\n", prop.regsPerBlock); printf("Threads in warp: %d\n", prop.warpSize); printf("Max threads per block: %d\n", prop.maxThreadsPerBlock); printf("Max thread dimensions: (%d %d %d)\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2]); printf("Max Grid Dimensions: (%d %d %d)\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2]); printf("\n"); } }
5,045
__device__ void left_to_right(int j0, int j1, int *d_rows_mp, int *d_aux_mp, int *d_low, int m, int p){ // Compute symmetric difference of supp(j0) and supp(j1) and store in d_aux // If rows are initially sorted, this returns a sorted list int idx0 = j0*p; int idx1 = j1*p; int idx0_MAX = (j0+1)*p; int idx1_MAX = (j1+1)*p; int idx = idx1; bool idx0_ok = d_rows_mp[idx0] != -1 && idx0 < idx0_MAX; bool idx1_ok = d_rows_mp[idx1] != -1 && idx1 < idx1_MAX; while (idx0_ok || idx1_ok){ if (idx0_ok && idx1_ok){ if (d_rows_mp[idx0] < d_rows_mp[idx1]){ d_aux_mp[idx++] = d_rows_mp[idx0++]; }else if (d_rows_mp[idx1] < d_rows_mp[idx0]){ d_aux_mp[idx++] = d_rows_mp[idx1++]; }else{ idx0++; idx1++; if (idx0 == idx0_MAX || idx1 == idx1_MAX){ //printf("WARNING: Reached memalloc limit\n"); // needs thurst import //asm("trap;"); } } }else{ if (idx0_ok){ d_aux_mp[idx++] = d_rows_mp[idx0++]; } if (idx1_ok){ d_aux_mp[idx++] = d_rows_mp[idx1++]; } } idx0_ok = d_rows_mp[idx0] != -1 && idx0 < idx0_MAX; idx1_ok = d_rows_mp[idx1] != -1 && idx1 < idx1_MAX; } int low_j1 = -1; // At least one value was written in d_aux_mp idx1 = j1*p; while(d_aux_mp[idx1] != -1 && idx1 < idx1_MAX){ d_rows_mp[idx1] = d_aux_mp[idx1]; d_aux_mp[idx1] = -1; if (d_rows_mp[idx1] > -1) low_j1 = d_rows_mp[idx1]; idx1++; } while(d_rows_mp[idx1] != -1 && idx1 < idx1_MAX){ d_rows_mp[idx1] = -1; idx1++; } d_low[j1] = low_j1; } __device__ void clear_column(int j, int *d_rows_mp, int p){ int idx = j*p; int idx_MAX = (j+1)*p; while (idx < idx_MAX && d_rows_mp[idx] != -1){ d_rows_mp[idx++] = -1; } } __global__ void reduce_col(int j, int *d_rows_mp, int *d_aux_mp, int *d_low, int *d_arglow, int m, int p){ int tid = threadIdx.x + blockDim.x*blockIdx.x; if(j == tid && j < m){ int j0 = -1; int low_j = d_low[j]; // low_j = -1, 0, 1, ..., m-1 while (low_j > -1 && d_arglow[low_j] != -1){ j0 = d_arglow[low_j]; left_to_right(j0, j, d_rows_mp, d_aux_mp, d_low, m, p); low_j = d_low[j]; } low_j = d_low[j]; if (low_j > -1){ d_arglow[low_j] = j; } } } __global__ void update_classes_final(int *d_classes, int *d_low, int *d_ess, int m){ int tid = threadIdx.x + blockDim.x*blockIdx.x; if(tid < m){ if (d_low[tid] > -1){ d_classes[tid] = -1; d_classes[d_low[tid]] = 1; }else{ if (d_ess[tid] == 1){ d_classes[tid] = 2; } } } } __global__ void update_classes(int *d_classes, int *d_low, int *d_arglow, int m){ int tid = threadIdx.x + blockDim.x*blockIdx.x; if(tid < m){ if (d_arglow[tid] > -1){ d_classes[d_arglow[tid]] = -1; d_classes[tid] = 1; } } } __global__ void ess_hat_final(int *d_essential_hat, int *d_low, int *d_arglow, int m){ int j = threadIdx.x + blockDim.x*blockIdx.x; if(j < m){ if (d_low[j] > -1){ d_essential_hat[j] = 0; d_essential_hat[d_low[j]] = 0; } } } __global__ void ess_hat(int *d_essential_hat, int *d_low, int *d_arglow, int m){ int j = threadIdx.x + blockDim.x*blockIdx.x; if(j < m){ if (d_low[j] > -1){ d_essential_hat[d_low[j]] = 0; } if (d_arglow[j] > -1){ d_essential_hat[d_arglow[j]] = 0; d_essential_hat[j] = 0; } } }
5,046
#include <stdio.h> // Exmple doesn't work __global__ void print_kernel() { if (threadIdx.x == 1) { printf("Hello from block %d, thread %d\n", blockIdx.x, threadIdx.x); } } int main() { print_kernel<<<100, 10>>>(); //cudaDeviceSynchronize(); }
5,047
#include "includes.h" __global__ void axpb_y_i32 (int a, int* x, int b, int* y, int len) { int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx < len) { y[idx] *= a * x[idx] + b; } }
5,048
#include "includes.h" __global__ void MultiplyAdd(float *d_Result, float *d_Data, int width, int height) { const int x = __mul24(blockIdx.x, 16) + threadIdx.x; const int y = __mul24(blockIdx.y, 16) + threadIdx.y; int p = __mul24(y, width) + x; if (x<width && y<height) d_Result[p] = d_ConstantA[0]*d_Data[p] + d_ConstantB[0]; __syncthreads(); }
5,049
__global__ void kernel_add(float *proj1, float *proj, int iv, int na, int nb, float weight){ int ia = 16 * blockIdx.x + threadIdx.x; int ib = 16 * blockIdx.y + threadIdx.y; if (ia >= na || ib >= nb) return; proj1[ia + ib * na] += proj[ia + ib * na + iv * na * nb] * weight; } // __global__ void kernel_add(cudaArray *proj1, cudaArray *proj, int iv, int na, int nb, float weight){ // int ia = 16 * blockIdx.x + threadIdx.x; // int ib = 16 * blockIdx.y + threadIdx.y; // if (ia >= na || ib >= nb) // return; // proj1[ia + ib * na] += proj[ia + ib * na + iv * na * nb] * weight; // }
5,050
/* * Device code */ __global__ void GaussSolve( int const Nsize, double* d_Aug, double* d_Piv) { // Assign matrix elements to blocks and threads int i = blockDim.x*blockIdx.x + threadIdx.x; // Parallel forward elimination for (int k = 0; k < Nsize-1; k++) { d_Piv[i] = d_Aug[i%Nsize+k*Nsize]/d_Aug[k*(Nsize+1)]; __syncthreads(); if (((i%Nsize)>k) && ((i/Nsize/*+1*/)>=k) && ((i/Nsize/*+1*/)<=Nsize)) d_Aug[i] -= d_Piv[i]*d_Aug[i-(i%Nsize)+k]; __syncthreads(); } }
5,051
#include<stdlib.h> #include<math.h> #include<iostream> #include<time.h> #define omega 1.5 using namespace std; __global__ void calculateU(double* u, double* f, double* pu, int N, double h2, int rb, double * e) { __shared__ double s_u[10][10]; e[0]=0; int i = blockIdx.x*blockDim.x + threadIdx.x; // "" int j = blockIdx.y*blockDim.y + threadIdx.y; // "" int tx = threadIdx.x+1; int ty = threadIdx.y+1; int k = j*N + i; // "" if(i<N && j<N) { s_u[ty][tx] = u[k]; if(ty==1 && j>0) s_u[ty-1][tx] = u[k-N]; if(ty==8 && j<N-1) s_u[ty+1][tx] = u[k+N]; if(tx==1 && i>0) s_u[ty][tx-1] = u[k-1]; if(tx==8 && i<N-1) s_u[ty][tx+1] = u[k+1]; if(i==0) s_u[ty][tx-1] = u[k+N-2]; if(i==N-1) s_u[ty][tx+1] = u[k-N+2]; } __syncthreads(); if((i>0)&&(i<N-1)&&(j>0)&&(j<N-1)&&(k<N*N)&&(i+j)%2==rb){ u[k] = omega*0.25*(h2*f[k] +s_u[ty-1][tx] + s_u[ty+1][tx] + s_u[ty][tx-1] + s_u[ty][tx+1] - 4*s_u[ty][tx])+ s_u[ty][tx]; if(e[0]<abs(pu[k]-u[k])) e[0] = abs(pu[k]-u[k]); } } int main( int argc, char * argv [] ) { double h =(double)1/(double)1024; double h2=h*h; int numBytes = sizeof(double)*1024*1024; double* U; double* f; double* pU; double* eps; U = (double*)malloc(numBytes); f = (double*)malloc(numBytes); pU = (double*)malloc(numBytes); eps = (double*)malloc(sizeof(double)); eps[0]=1; // for(int i=0; i<1024; i++) { double x = i*h; for(int j=0; j<1024; j++) { double y = j*h; f[i*1024+j] = 4; U[i*1024+j] = 0; pU[i*1024+j] = -x*x -y*y +x +y; } } // for(int i=0; i<1024 ; i++) { double x = i*h; U[i*1024] = - x*x + x; U[i] = - x*x + x; U[i*1024+1023] = - x*x + x; U[1023*1024+i] = - x*x + x; pU[i*1024] = - x*x + x; pU[i] = - x*x + x; pU[i*1024+1023] = - x*x + x; pU[1024*1023+i] = - x*x + x; } // allocate device memory double * devU = NULL; double * devpU = NULL; double * devF = NULL; double * devE = NULL; cudaMalloc ( (void**)&devU, numBytes ); cudaMalloc ( (void**)&devpU, numBytes ); cudaMalloc ( (void**)&devF, numBytes ); cudaMalloc ( (void**)&devE, sizeof(double)); // set kernel launch configuration dim3 threads = dim3(128, 128,1); dim3 blocks = dim3(8, 8,1); cudaMemcpy(devU,U,numBytes,cudaMemcpyHostToDevice); cudaMemcpy(devpU,pU,numBytes,cudaMemcpyHostToDevice); cudaMemcpy(devF,f,numBytes,cudaMemcpyHostToDevice); cudaMemcpy(devE,eps,sizeof(double),cudaMemcpyHostToDevice); // int count = 0; clock(); for(int i = 0; (i<50000)&&(eps[0]>0.0005); i++){ //eps[0]=0; calculateU<<<threads,blocks,0>>>(devU,devF,devpU,1024,h2,0,devE); calculateU<<<threads,blocks,0>>>(devU,devF,devpU,1024,h2,1,devE); cudaMemcpy(eps,devE,sizeof(double),cudaMemcpyDeviceToHost); count = i; } // cudaMemcpy(U,devU,numBytes,cudaMemcpyDeviceToHost); cudaMemcpy(pU,devpU,numBytes,cudaMemcpyDeviceToHost); cudaMemcpy(f,devF,numBytes,cudaMemcpyDeviceToHost); clock_t c2 = clock(); cerr << "Время: " << ((double)c2)/CLOCKS_PER_SEC << "мс" << endl << "Погрешность: " << eps[0] << ", кол-во итераций:" <<count << endl;; for(int i=0; i<1024 ; i++) { for(int j=0; j<1024 ; j++){ cout << i*h << " " << j*h << " " << U[i*1024+j]<< "\n"; } } return 0; }
5,052
#include "shared.cuh" struct ParticleRef { Point pos; Point dir; double nextdist; }; inline __device__ ParticleRef make_ref(const ParticleView &view, int i) { return {view.get_pos(i), view.get_dir(i), view.get_nextdist(i)}; } __device__ inline void saxpy(double *__restrict__ x, double *__restrict__ y, double *__restrict__ z, const double *__restrict__ u, const double *__restrict__ v, const double *__restrict__ w, double distance) { *x += *u * distance; *y += *v * distance; *z += *w * distance; } __device__ inline void move_impl(const ParticleRef ref) { saxpy(ref.pos.x, ref.pos.y, ref.pos.z, ref.dir.x, ref.dir.y, ref.dir.z, ref.nextdist); } __global__ void move(ParticleView view) { int i = thread_id(); if (i >= view.size) return; move_impl(make_ref(view, i)); }
5,053
#include "includes.h" __global__ void histDupeKernel(const float* data1, const float* data2, const float* confidence1, const float* confidence2, int* ids1, int* ids2, int* results_id1, int* results_id2, float* results_similarity, int* result_count, const int N1, const int N2, const int max_results) { const unsigned int thread = threadIdx.x; // Thread index within block const unsigned int block = blockIdx.x; // Block index const unsigned int block_size = blockDim.x; // Size of each block const unsigned int block_start = block_size * block; // Index of the start of the block const unsigned int index = block_start + thread; // Index of this thread //__shared__ float conf[64]; // Shared array of confidence values for all histograms owned by this block //conf[thread] = confidence1[index]; // Coalesced read of confidence values float conf = confidence1[index]; int id = ids1[index]; __shared__ float hists[128 * 64]; // Shared array of all histograms owned by this block for (unsigned int i = 0; i < 64; i++) { hists[i * 128 + thread] = data1[(block_start + i) * 128 + thread]; // Coalesced read of first half of histogram hists[i * 128 + thread + 64] = data1[(block_start + i) * 128 + 64 + thread]; // Coalesced read of second half of histogram } __shared__ float other[128]; // Histogram to compare all owned histograms against parallely for (unsigned int i = 0; i < N2 && *result_count < max_results; i++) { other[thread] = data2[i * 128 + thread]; // Coalesced read of first half of other histogram other[thread + 64] = data2[i * 128 + thread + 64]; // Second half __syncthreads(); // Ensure all values read if (index < N1) { float d = 0; for (unsigned int k = 0; k < 128; k++) { // Compute sum of distances between thread-owned histogram and shared histogram d += fabsf(hists[thread * 128 + k] - other[k]); } d = 1 - (d / 8); // Massage the difference into a nice % similarity number, between 0 and 1 int other_id = ids2[i]; if (other_id != id && d > fmaxf(conf, confidence2[i])) { // Don't compare against self, only compare using highest confidence int result_index = atomicAdd(result_count, 1); // Increment result count by one atomically (returns value before increment) if (result_index < max_results) { // Store resulting pair results_similarity[result_index] = d; results_id1[result_index] = id; results_id2[result_index] = other_id; } } } __syncthreads(); // Ensure all threads have finished before looping and reading new shared histogram } }
5,054
/* * GPUKernels.cu * * Created on: Oct 19, 2010 * Author: yiding */ #include <cuda_runtime.h> #include <math.h> #include <cuda.h> #define NOID 0xFFFFFFFF #define CUDA_MAJOR_VER 1 #define CUDA_MINOR_VER 3 typedef unsigned int CoordType; typedef unsigned int IdType; typedef unsigned short CapType; typedef unsigned int SizeType; typedef float CostType; //Define CUDA Kernels in this file
5,055
#include <pthread.h> #include <stdio.h> #include <string.h> #include <math.h> #ifndef D #define D 10000 #endif #ifndef N_FILES #define N_FILES 21000 #endif #define ARG_COUNT 4 #define MAX_FILE_NAME 100 #define GMEM_GRANULARITY 128 #define INV_DICT_WIDTH ((unsigned int)(ceil(N_FILES / (float)(sizeof(int)*8)))) #define INV_DICT_WIDTH_PAD ((INV_DICT_WIDTH/GMEM_GRANULARITY+1)*GMEM_GRANULARITY) #define INV_DICT_SIZE (D*INV_DICT_WIDTH_PAD) #define BLOCK_SIZE 64 #define N_BLOCKS ((unsigned int)(ceil(N_FILES/ (float)(sizeof(int)*8*BLOCK_SIZE)))) #define N_THREADS (BLOCK_SIZE*N_BLOCKS) unsigned int *inv_dictionary; char files[N_FILES][MAX_FILE_NAME]; char query[D]; unsigned int query_ones[D]; unsigned int matches[N_FILES]; // inv_dict_width = width of inv dictionary in number of elements // int_size = size of an integer in bytes __global__ void queryKernel(unsigned int * __restrict__ inv_dictionary, unsigned int inv_dict_width, unsigned int inv_dict_width_pad, unsigned int * __restrict__ query_ones, unsigned int ones_cnt, unsigned int * matches, unsigned int int_size, unsigned int n_threads) { unsigned short curr_bit = 0; unsigned int match_pos; unsigned int match_cnt = 0; // unsigned int match_idx; __shared__ unsigned int match_idx_s[BLOCK_SIZE]; unsigned int index = threadIdx.x + blockIdx.x * blockDim.x; if(index < inv_dict_width){ //inv_dict_width = 657 match_idx_s[threadIdx.x] = 0xFFFFFFFF; for(int i = 0; i < ones_cnt; i++) { match_idx_s[threadIdx.x] = match_idx_s[threadIdx.x] & *(inv_dictionary + query_ones[i]*inv_dict_width_pad + index); } match_cnt = 0; for(int j=1; j<=(int_size*8) && match_idx_s[threadIdx.x]>0; j++) { curr_bit = match_idx_s[threadIdx.x] & 1; match_idx_s[threadIdx.x] = match_idx_s[threadIdx.x] >> 1; if(curr_bit==1) { match_pos = (index + 1)*int_size*8 - j + 1; // starts indexing at 1 not 0 matches[index*int_size*8+match_cnt] = match_pos; match_cnt++; } } } } __host__ void load_inv_dictionary(char file[]){ unsigned int num_read; FILE *fp = NULL; fp = fopen(file, "r"); if (fp == NULL){ printf("Error while opening %s\n", file); } int i; for(int j = 0; j<D; j++) { for(i=0; i<INV_DICT_WIDTH-1; i++){ // printf("%d:\t%d:\t", j, i); num_read = fscanf(fp, "%u ", inv_dictionary + j*INV_DICT_WIDTH_PAD + i); // printf("%u\n", *(inv_dictionary + j*INV_DICT_WIDTH_PAD + i)); if (num_read != 1) printf("ERROR!\t fscanf did not fill all arguments\n"); } // printf("%d:\t%d:\t", j, i); num_read = fscanf(fp, "%u\n", inv_dictionary + j*INV_DICT_WIDTH_PAD + i); // printf("%u\n", *(inv_dictionary + j*INV_DICT_WIDTH_PAD + i)); if (num_read != 1) printf("ERROR!\t fscanf did not fill all arguments\n"); } fclose(fp); } __host__ void load_query(char file[]){ unsigned int num_read; FILE *fp = NULL; fp = fopen(file, "r"); if (fp == NULL){ printf("Error while opening %s\n", file); } for (int i=0; i<D; i++) { num_read = fscanf(fp, "%c", &query[i]); if (num_read != 1) printf("ERROR!\t fscanf did not fill all arguments\n"); } // printf("Loaded query: "); // for (int i=0; i<D; i++) { // printf("%c", query[i]); // } // printf("\n"); fclose(fp); } __host__ void printMatches(){ for (int i=0; i<N_FILES; i++) { printf("Match %d:\t%u\n", i, matches[i]); } } __host__ void load_files(char file[]) { FILE *fp = NULL; fp = fopen(file, "r"); int num_read = 0; if (fp == NULL){ printf("Error while opening %s\n", file); } for(int i = 0; i<N_FILES; i++) { num_read = fscanf(fp, "%s", files[i]); if (num_read != 1) printf("Error reading file: %s\n", file); // printf("%d: %s\n", i, files[i]); } fclose(fp); } unsigned int findQueryOnes(char* query){ unsigned int ones_cnt = 0; for(unsigned int i=0; i<D; i++) { if(query[i] == '1') { query_ones[ones_cnt] = i; printf("One position:\t%d\n", i); ones_cnt++; } } return ones_cnt; } __host__ void reportQuery(char* report_f){ FILE *rep = fopen(report_f, "w+"); for (int i=0; i < N_FILES; i++){ if (matches[i] != 0) { fprintf(rep, "%s\n", files[matches[i]-1]); } } fclose(rep); } int main(int argc, char** argv){ // Command-line arguments: // D, itemmem input file, dictionary input file, query if(argc != (ARG_COUNT+1)){ printf("Requires arguments: <files file> <dictionary input file> <query input file> <output directory>\n"); return 1; } char dict_file[MAX_FILE_NAME]; char files_file[MAX_FILE_NAME]; char query_file[MAX_FILE_NAME]; char output_dir[MAX_FILE_NAME]; char report_file[MAX_FILE_NAME]; sprintf(files_file, "%s", argv[1]); sprintf(dict_file, "%s", argv[2]); sprintf(query_file, "%s", argv[3]); sprintf(output_dir, "%s", argv[4]); printf("INV_DICT_SIZE:\t%u\nINV_DICT_WIDTH:\t%u\nINV_DICT_WIDTH_PAD:\t%u\nN_BLOCKS:\t%u\nBLOCK_SIZE:\t%d\n", INV_DICT_SIZE, INV_DICT_WIDTH, INV_DICT_WIDTH_PAD, N_BLOCKS, BLOCK_SIZE); sprintf(report_file, "%s/cu_query_report.txt", output_dir); unsigned int *d_inv_dict, *d_matches; unsigned int *d_query_ones; unsigned int ones_cnt; inv_dictionary = (unsigned int *)malloc(INV_DICT_SIZE*sizeof(unsigned int)); load_inv_dictionary(dict_file); load_query(query_file); ones_cnt = findQueryOnes(query); #ifdef PROFILING for (int i = 0; i < 100; ++i) { #endif // const int num_streams = 8; // cudaStream_t streams[num_streams]; cudaMalloc((void **)&d_inv_dict, INV_DICT_SIZE*sizeof(unsigned int)); cudaMalloc((void **)&d_query_ones, ones_cnt*sizeof(unsigned int)); cudaMalloc((void **)&d_matches, N_FILES*sizeof(unsigned int)); cudaMemset(d_matches, 0, N_FILES*sizeof(unsigned int)); // printf("Allocated arrays...\n"); cudaMemcpy(d_inv_dict, inv_dictionary, INV_DICT_SIZE*sizeof(unsigned int), cudaMemcpyHostToDevice); cudaError_t err = cudaMemcpy(d_query_ones, query_ones, ones_cnt*sizeof(unsigned int), cudaMemcpyHostToDevice); printf("Err: %d\n", err); // printf("Query...\n"); // for(int i=0; i < num_streams; i++) { // cudaStreamCreate(&streams[i]); // } // for(int i=0; i < num_streams; i++) { queryKernel<<<N_BLOCKS, BLOCK_SIZE>>>((unsigned int *)d_inv_dict, INV_DICT_WIDTH, INV_DICT_WIDTH_PAD, d_query_ones, ones_cnt, d_matches, sizeof(int), N_THREADS); cudaMemcpy(matches, d_matches, N_FILES*sizeof(unsigned int), cudaMemcpyDeviceToHost); // } // printMatches(); load_files(files_file); reportQuery(report_file); cudaFree(d_inv_dict); cudaFree(d_query_ones); cudaFree(d_matches); #ifdef PROFILING } #endif free(inv_dictionary); printf("Ended! :)\n"); return 0; }
5,056
#include "includes.h" __global__ void VecAdd(int *a, int *b, int *c, int n) { int i = blockIdx.x * blockDim.x + threadIdx.x; if(i < n) { c[i] = a[i] + b[i]; } }
5,057
/** * Copyright (c) 2017 Darius Rückert * Licensed under the MIT License. * See LICENSE file for more information. */ #include <iostream> #include <vector> #include <cuda_runtime.h> #include <thrust/device_vector.h> template <unsigned int BLOCK_SIZE> __global__ static void oob(int* data, int size) { auto id = threadIdx.x + BLOCK_SIZE * blockIdx.x; if (id >= size) return; data[id + 1] = 0; } static void intraBlockTest(int N) { // Compute launch arguments const unsigned int BLOCK_SIZE = 128; const unsigned int BLOCKS = N / BLOCK_SIZE; thrust::device_vector<int> d_data(N); oob<BLOCK_SIZE><<<BLOCKS, BLOCK_SIZE>>>(d_data.data().get(), N); cudaDeviceSynchronize(); } // nvcc -arch=sm_61 -g main.cu -o segfault int main(int argc, char* argv[]) { intraBlockTest(1024 * 2); std::cout << "Done." << std::endl; }
5,058
#include <stdio.h> #include <stdlib.h> inline void check_cuda_errors(const char *filename, const int line_number){ #ifdef DEBUG cudaThreadSynchronize(); cudaError_t error = cudaGetLastError(); if(error != cudaSuccess){ printf("CUDA error at %s:%i: %s\n", filename, line_number, cudaGetErrorString(error)); exit(-1); } #endif } __global__ void foo(int *ptr){ *ptr = 7; } int main(){ foo<<<1,1>>>(0); check_cuda_errors(__FILE__, __LINE__); return 0; }
5,059
#include<stdio.h> #define NUM_THREADS_PER_BLOCK 256 __global__ void print_hello() { int idx = threadIdx.x; printf("Hello World! My threadId is %d\n", idx); } int main() { print_hello<<<1, NUM_THREADS_PER_BLOCK>>>(); cudaDeviceSynchronize(); return 0; }
5,060
// Array reversing in CUDA using shared memory. #include <cuda.h> #include <cuda_runtime.h> #include <device_launch_parameters.h> #include <chrono> #include <cstdlib> #include <iostream> __global__ void reverseKernel(float* A, int N) { int i = blockDim.x * blockIdx.x + threadIdx.x; if (i < N) { extern __shared__ float B[]; int r = N - i - 1; B[i] = A[i]; __syncthreads(); A[i] = B[r]; } } void displayArray(float* a, int N) { for (int i = 0; i < N; i++) { std::cout << a[i]; if (i + 1 != N) std::cout << " "; } std::cout << "\n"; } int main(int argc, char** argv) { if (argc == 2) { const int N = atoi(argv[1]); size_t size = N * sizeof(float); cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); float* h_A = (float*)malloc(size); if (h_A == NULL) { std::cerr << "Failed malloc for h_A!\n"; return 1; } float* h_B = (float*)malloc(size); if (h_B == NULL) { std::cerr << "Failed malloc for h_B!\n"; return 2; } for (int i = 0; i < N; i++) { h_A[i] = i + 1; } auto begin = std::chrono::high_resolution_clock::now(); for (int i = 0; i < N; i++) { h_B[i] = h_A[N - i - 1]; } auto end = std::chrono::high_resolution_clock::now(); std::chrono::duration<double, std::milli> cputime = end - begin; std::cout << "CPU Elapsed Time: " << cputime.count() << " ms" << std::endl; // displayArray(h_A, N); // displayArray(h_B, N); float* d_A = NULL; cudaMalloc((void**)&d_A, size); cudaMemcpy(d_A, h_A, size, cudaMemcpyHostToDevice); const int BLOCK_SIZE = 1024; const int GRID_SIZE = (N - 1) / BLOCK_SIZE + 1; cudaEventRecord(start); reverseKernel<<<GRID_SIZE, BLOCK_SIZE, size>>>(d_A, N); cudaDeviceSynchronize(); cudaEventRecord(stop); cudaEventSynchronize(stop); float msecs = 0; cudaEventElapsedTime(&msecs, start, stop); std::cout << "GPU Elapsed Time: " << msecs << " ms.\n"; cudaMemcpy(h_A, d_A, size, cudaMemcpyDeviceToHost); // displayArray(h_A, N); for (int i = 0; i < N; i++) { if (h_A[i] != h_B[i]) { std::cerr << "TEST FAILED...\n"; return 5; } } std::cout << "TEST PASSED!\n"; cudaFree(d_A); free(h_A); } return 0; }
5,061
/*------------vec_add.cu------------------------------------------------------// * * Purpose: This is a simple cuda file for vector addition * *-----------------------------------------------------------------------------*/ #include <iostream> #include <math.h> __global__ void vecAdd(double *a, double *b, double *c, int n){ // First we need to find our global threadID int id = blockIdx.x*blockDim.x + threadIdx.x; // Make sure we are not out of range if (id < n){ c[id] = a[id] + b[id]; } } int main(){ // size of vectors int n = 1000; // Host vectors double *h_a, *h_b, *h_c; // Device vectors double *d_a, *d_b, *d_c; // allocating space on host and device h_a = (double*)malloc(sizeof(double)*n); h_b = (double*)malloc(sizeof(double)*n); h_c = (double*)malloc(sizeof(double)*n); // Allocating space on GPU cudaMalloc(&d_a, sizeof(double)*n); cudaMalloc(&d_b, sizeof(double)*n); cudaMalloc(&d_c, sizeof(double)*n); //initializing host vectors for (int i = 0; i < n; ++i){ h_a[i] = 1; h_b[i] = 1; } // copying these components to the GPU cudaMemcpy(d_a, h_a, sizeof(double)*n, cudaMemcpyHostToDevice); cudaMemcpy(d_b, h_b, sizeof(double)*n, cudaMemcpyHostToDevice); // Creating blocks and grid ints int threads, grid; threads = 64; grid = (int)ceil((float)n/threads); vecAdd<<<grid, threads>>>(d_a, d_b, d_c, n); // Now to copy c back cudaMemcpy(h_c, d_c, sizeof(double)*n, cudaMemcpyDeviceToHost); double sum = 0; for (int i = 0; i < n; ++i){ sum += h_c[i]; } std::cout << "Sum is: " << sum << '\n'; // Release memory cudaFree(d_a); cudaFree(d_b); cudaFree(d_c); free(h_a); free(h_b); free(h_c); }
5,062
#include <stdio.h> #include <stdlib.h> //#include <sys/time.h> #define NUM_PARTICLES 10000 // Third argument #define NUM_ITERATIONS 100 // Second argument #define BLOCK_SIZE 16 // First argument typedef struct { float3 position; float3 velocity; } Particle; __global__ void timeStep(Particle *particles, int time) { const int i = blockIdx.x * blockDim.x + threadIdx.x; if (i < NUM_PARTICLES) { particles[i].velocity.x = (time % 3 == 0) ? time % 5 / 1.7 : -time % 5 / 2.8; particles[i].velocity.y = (time % 7 == 0) ? time % 4 / 4.5: -time % 3 / 2.3; particles[i].velocity.z = (time % 2 == 0) ? time % 3 * 1.6 : -time % 7 / 1.2; particles[i].position.x += particles[i].velocity.x; particles[i].position.y += particles[i].velocity.y; particles[i].position.z += particles[i].velocity.z; } } void timeStepCPU(Particle *particles, int time) { for (int i = 0; i < NUM_PARTICLES; ++i) { particles[i].velocity.x = (time % 3 == 0) ? time % 5 / 1.7 : -time % 5 / 2.8; particles[i].velocity.y = (time % 7 == 0) ? time % 4 / 4.5: -time % 3 / 2.3; particles[i].velocity.z = (time % 2 == 0) ? time % 3 * 1.6 : -time % 7 / 1.2; particles[i].position.x += particles[i].velocity.x; particles[i].position.y += particles[i].velocity.y; particles[i].position.z += particles[i].velocity.z; } } int main(int argc, char *argv[]){ int numParticles, numIterations, blockSize; if (argc == 1) { numParticles = NUM_PARTICLES; numIterations = NUM_ITERATIONS; blockSize = BLOCK_SIZE; } else if (argc == 2) { numParticles = NUM_PARTICLES; numIterations = NUM_ITERATIONS; blockSize = atoi(argv[1]); } else if (argc == 3) { numParticles = NUM_PARTICLES; numIterations = atoi(argv[2]); blockSize = atoi(argv[1]); } else { numParticles = atoi(argv[3]); numIterations = atoi(argv[2]); blockSize = atoi(argv[1]); } //struct timeval start, end; Particle *particles = (Particle*)malloc(numParticles * sizeof(Particle)); Particle *particles_GPU = (Particle*)malloc(numParticles * sizeof(Particle)); Particle *d_particles; for (int i = 0; i < numParticles; ++i) { particles[i].velocity.x = 1; particles[i].velocity.y = 1; particles[i].velocity.z = 1; particles[i].position.x = 0; particles[i].position.y = 0; particles[i].position.z = 0; } cudaMalloc(&d_particles, numParticles * sizeof(Particle)); cudaMemcpy(d_particles, particles, numParticles * sizeof(Particle), cudaMemcpyHostToDevice); //gettimeofday(&start, NULL); for (int t = 0; t < numIterations; ++t){ timeStepCPU(particles, t); } //gettimeofday(&end, NULL); //printf("Computing %d iterations on the CPU... Done in %ld microseconds!\n", numIterations, ((end.tv_sec * 1000000 + end.tv_usec) - (start.tv_sec * 1000000 + start.tv_usec))); //gettimeofday(&start, NULL); for (int t = 0; t < numIterations; ++t){ timeStep<<<blockSize, numParticles / blockSize + 1>>>(d_particles, t); } //gettimeofday(&end, NULL); //printf("Computing %d iterations on the GPU... Done in %ld microseconds!\n", numIterations, ((end.tv_sec * 1000000 + end.tv_usec) - (start.tv_sec * 1000000 + start.tv_usec))); cudaMemcpy(particles_GPU, d_particles, numParticles * sizeof(Particle), cudaMemcpyDeviceToHost); int sameArray = 1; for (int i = 0; i < numParticles; ++i) { if ((particles[i].position.x != particles_GPU[i].position.x) || (particles[i].position.y != particles_GPU[i].position.y)|| (particles[i].position.z != particles_GPU[i].position.z)) { printf("Comparing the output for each implementation… Wrong at %d!\n", i); printf("GPU.x: %f, CPU.x: %f\n", particles[i].position.x, particles_GPU[i].position.x); sameArray = 0; break; } } if (sameArray == 1) { printf("Comparing the output for each implementation… Correct!\n"); } free(particles); free(particles_GPU); cudaFree(d_particles); }
5,063
// #include <bits/stdc++.h> #include<stdio.h> #include<stdlib.h> #include<iostream> #include<vector> #include<algorithm> #include <climits> #include <thrust/swap.h> #include <thrust/extrema.h> #include <thrust/functional.h> #include <thrust/host_vector.h> #include <thrust/device_vector.h> using namespace std; typedef long long ll; #define CUDA_ERROR(err) { \ if (err != cudaSuccess) { \ fprintf(stderr, "ERROR: CUDA failed in %s:%d: %s\n", __FILE__, __LINE__, cudaGetErrorString(err)); \ return(1); \ } \ } \ struct abs_comparator{ __host__ __device__ bool operator()(double x, double y) { return fabs(x) < fabs(y); } }; __global__ void kernel_compute_L (double* data, int n, int i) { int idx = blockIdx.x * blockDim.x + threadIdx.x; int offsetX = gridDim.x * blockDim.x; for (int j = idx + i + 1; j < n; j += offsetX) { data[j + i * n] /= data[i + i * n]; } } __global__ void kernel_compute_U (double* data, int n, int i) { int idx = blockIdx.x * blockDim.x + threadIdx.x; int idy = blockIdx.y * blockDim.y + threadIdx.y; int offsetX = gridDim.x * blockDim.x; int offsetY = gridDim.y * blockDim.y; for (int j = idx + i + 1; j < n; j += offsetX) { for (int q = idy + i + 1; q < n; q += offsetY) { data[j + q * n] -= data[j + i * n] * data[i + q * n]; } } } // for (int j = i; j < n; ++j) { // for (int q = 0; q < i; ++q) { // data[i + j * n] -= data[i + q * n] * data[q + j * n]; // } // } __global__ void kernel_swap(double* data, int n, int i, int max_idx) { int idx = blockIdx.x * blockDim.x + threadIdx.x; int offsetX = gridDim.x * blockDim.x; for (int j = idx; j < n; j += offsetX) { thrust::swap(data[i + j * n], data[max_idx + j * n]); } } int main(int argc, char const *argv[]) { ios_base::sync_with_stdio(false); cin.tie(nullptr); cout.tie(nullptr); int n; cin >> n; double* data, *dev_data; CUDA_ERROR(cudaMalloc((void**)&dev_data, sizeof(double) * n * n)); int* p = (int*)malloc(sizeof(int) * n); data = (double*)malloc(sizeof(double) * n * n); // Data inputing + p initializing for (int i = 0; i < n; ++i) { p[i] = i; for (int j = 0; j < n; ++j) { cin >> data[i + j * n]; } } // cudaEvent_t start, stop; // float gpu_time = 0.0; // cudaEventCreate(&start); // cudaEventCreate(&stop); // cudaEventRecord(start, 0); CUDA_ERROR(cudaMemcpy(dev_data, data, sizeof(double) * n * n, cudaMemcpyHostToDevice)); // fprintf(stderr, "Got data\n"); dim3 BLOCKS(32, 32); dim3 THREADS(32, 32); int max_idx; // double mx; abs_comparator cmp; thrust::device_ptr<double> data_ptr; thrust::device_ptr<double> max_ptr; for (int i = 0; i < n - 1; ++i) { max_idx = i; // Find max in column from i to n (cast to data_ptr == start) data_ptr = thrust::device_pointer_cast(dev_data + i * n); // Pointer e.x. largest == data + 3 max_ptr = thrust::max_element(data_ptr + i, data_ptr + n, cmp); max_idx = max_ptr - data_ptr; // mx = fabs(*max_ptr); // fprintf(stderr, "Find max idx=%d\n", max_idx); // fprintf(stderr, "MAX=%f\n", mx); p[i] = max_idx; if (max_idx != i) { kernel_swap<<<32, 32>>>(dev_data, n, i, max_idx); CUDA_ERROR(cudaGetLastError()); } kernel_compute_L<<<32, 32>>>(dev_data, n, i); CUDA_ERROR(cudaGetLastError()); CUDA_ERROR(cudaThreadSynchronize()); kernel_compute_U<<<BLOCKS, THREADS>>>(dev_data, n, i); CUDA_ERROR(cudaGetLastError()); CUDA_ERROR(cudaThreadSynchronize()); fprintf(stderr, "Iter=%d\n", i); } CUDA_ERROR(cudaMemcpy(data, dev_data, sizeof(double) * n * n, cudaMemcpyDeviceToHost)); CUDA_ERROR(cudaFree(dev_data)); for (int i = 0; i < n; ++i) { for (int j = 0; j < n; ++j) { // cout << data[i + j * n] << " "; printf("%.10e ", data[i + j * n]); } printf("\n"); } for (int j = 0; j < n; ++j) { // cout << p[j] << " "; printf("%d ", p[j]); } printf("\n"); free(data); free(p); // cudaEventRecord(stop, 0); // cudaEventSynchronize(stop); // cudaEventElapsedTime(&gpu_time, start, stop); // fprintf(stderr, "Time %f\n", gpu_time); return 0; }
5,064
#include "includes.h" __global__ void kRotate180(float* filters, float* targets, const int filterSize) { // __shared__ float shFilter[16][16]; const int filtIdx = blockIdx.x; const int readStart = MUL24(MUL24(filterSize, filterSize), filtIdx); filters += readStart; targets += readStart; for(int y = threadIdx.y; y < filterSize; y += 16) { for(int x = threadIdx.x; x < filterSize; x += 16) { const int writeX = filterSize - 1 - x; const int writeY = filterSize - 1 - y; targets[MUL24(writeY, filterSize) + writeX] = filters[MUL24(y, filterSize) + x]; } } }
5,065
#include "includes.h" __global__ void forwardDifference2DAdjointKernel(const int cols, const int rows, const float* dx, const float* dy, float* target) { for (auto idy = blockIdx.y * blockDim.y + threadIdx.y + 1; idy < cols - 1; idy += blockDim.y * gridDim.y) { for (auto idx = blockIdx.x * blockDim.x + threadIdx.x + 1; idx < rows - 1; idx += blockDim.x * gridDim.x) { const auto index = idx + rows * idy; target[index] = -dx[index] + dx[index - 1] - dy[index] + dy[index - rows]; } } }
5,066
#include "cuda_runtime.h" #include "stdio.h" __device__ float devData[5]; __global__ void checkGlobalVariable(){ devData[threadIdx.x] += 2.0f; } int main(void){ float value[5] = {3.14, 3.14, 3.14, 3.14, 3.14}; cudaMemcpyToSymbol(devData, &value, sizeof(float)*5); printf("Copy \n"); checkGlobalVariable<<<1, 5>>>(); cudaMemcpyFromSymbol(&value, devData, sizeof(float)*5); for(int i=0; i<5; i++){ printf("%d num is %f \n", i, value[i]); } cudaDeviceReset(); return EXIT_SUCCESS; }
5,067
#include <assert.h> #include <stdio.h> #include <stdlib.h> __device__ int iterate_pixel(float x, float y, float c_re, float c_im) { int c=0; float z_re=x; float z_im=y; while (c<255) { float re2=z_re*z_re; float im2=z_im*z_im; if ((re2+im2) > 4) break; z_im=2*z_re*z_im + c_im; z_re=re2-im2 + c_re; c++; } return c; } __global__ void calc_fractal(int width, int height, float c_re, float c_im, unsigned char* dest) { int x; int y; x = blockIdx.x * blockDim.x + threadIdx.x ; y = blockIdx.y * blockDim.y + threadIdx.y; float f_x=(float)(x*0.8)/(float)(width)-0.8; float f_y=(float)(y*0.8)/(float)(height)-0.8; dest[y*width+x]=iterate_pixel(f_x,f_y,c_re,c_im); } /*void calc_fractal(int width, int height, float c_re, float c_im, unsigned char* dest) { int x; int y; for (y=0;y<height;y++) for (x=0;x<width;x++) { float f_x=(float)(x*0.8)/(float)(width)-0.8; float f_y=(float)(y*0.8)/(float)(height)-0.8; dest[x+y*width]=iterate_pixel(f_x,f_y,c_re,c_im); } }*/ // Write a width by height 8-bit color image into File "filename" void write_ppm(unsigned char* data,unsigned int width,unsigned int height,char* filename) { if (data == NULL) { printf("Provide a valid data pointer!\n"); return; } if (filename == NULL) { printf("Provide a valid filename!\n"); return; } if ( (width>4096) || (height>4096)) { printf("Only pictures upto 4096x4096 are supported!\n"); return; } FILE *f=fopen(filename,"wb"); if (f == NULL) { printf("Opening File %s failed!\n",filename); return; } if (fprintf(f,"P6 %i %i 255\n",width,height) <= 0) { printf("Writing to file failed!\n"); return; }; int i; for (i=0;i<height;i++) { unsigned char buffer[4096*3]; int j; for (j=0;j<width;j++) { int v=data[i*width+j]; int s; s= v << 0; s=s > 255? 255 : s; buffer[j*3+0]=s; s= v << 1; s=s > 255? 255 : s; buffer[j*3+1]=s; s= v << 2; s=s > 255? 255 : s; buffer[j*3+2]=s; } if (fwrite(buffer,width*3,1,f) != 1) { printf("Writing of line %i to file failed!\n",i); return; } } fclose(f); } int main(int argc, char** args) { unsigned char image [256*256];//??????????????????????? unsigned char* image_GPU; int size_image = sizeof(char) * 256*256; int Xnum; int Ynum; switch (argc){ case 2: Xnum = atoi(args[1]); Ynum = 1; break; case 3: Xnum = atoi(args[1]); Ynum = atoi(args[2]); break; default: printf("ERROR!!! Correct usage: ./output <**> (<**>)\n"); } cudaMalloc((void**)&image_GPU, size_image); // NOTE: one thread to process one pixel, // say we have 16 block, then we need 256*256/16 = 4096 = 16*16*16 threads per block // dim3 threadsPerBlock(Xnum,Ynum); dim3 numBlocks(256/Xnum,256/Ynum); //assert(image != NULL);//will terminate the program if image == null //calc_fractal(256,256,0.28,0.008,image);//gonna change calc_fractal <<<numBlocks, threadsPerBlock>>> (256,256,0.28,0.008,image_GPU); cudaMemcpy(image, image_GPU, size_image, cudaMemcpyDeviceToHost); write_ppm(image,256,256,"julia.ppm");//not changed //free(image); return 0; }
5,068
// Copyright (c) 2020 Saurabh Yadav // // This software is released under the MIT License. // https://opensource.org/licenses/MIT #include <stdio.h> #include <cuda_runtime.h> #define NUM_OF_ELEMENTS 40000U #define ARRAY_A_ELEMENT ((int) 'A') #define ARRAY_B_ELEMENT ((int) 'B') //Compute vector sum C = A+B //Each thread performs one pair-wise addition __global__ void vecAddKernel(const int *arr_A, const int *arr_B, int *arr_C, int n) { int index = threadIdx.x + blockDim.x*blockIdx.x; if(index < n) { arr_C[index] = arr_A[index] + arr_B[index]; } } int main() { // Error code to check return values for CUDA calls cudaError_t err = cudaSuccess; size_t arr_size = NUM_OF_ELEMENTS * sizeof(int); // Allocate the host input and outout vectors int *host_arr_A = (int *)malloc(arr_size); int *host_arr_B = (int *)malloc(arr_size); int *host_arr_C = (int *)malloc(arr_size); // Verify that all allocations succeeded if (host_arr_A == NULL || host_arr_B == NULL || host_arr_C == NULL) { fprintf(stderr, "Failed to allocate host vectors!\n"); exit(EXIT_FAILURE); } // Initialize the host input vectors for (int i = 0; i < NUM_OF_ELEMENTS; ++i) { host_arr_A[i] = ARRAY_A_ELEMENT; host_arr_B[i] = ARRAY_B_ELEMENT; } // Allocate the device input vector A int *dev_arr_A = NULL; err = cudaMalloc((void **)&dev_arr_A, arr_size); if (err != cudaSuccess) { fprintf(stderr, "Failed to allocate device vector A (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Allocate the device input vector B int *dev_arr_B = NULL; err = cudaMalloc((void **)&dev_arr_B, arr_size); if (err != cudaSuccess) { fprintf(stderr, "Failed to allocate device vector B (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Allocate the device output vector C int *dev_arr_C = NULL; err = cudaMalloc((void **)&dev_arr_C, arr_size); if (err != cudaSuccess) { fprintf(stderr, "Failed to allocate device vector C (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Copy the host input vectors A and B in host memory to the device input vectors in // device memory printf("Copy input data from the host memory to the CUDA device\n"); err = cudaMemcpy(dev_arr_A, host_arr_A, arr_size, cudaMemcpyHostToDevice); if (err != cudaSuccess) { fprintf(stderr, "Failed to copy vector A from host to device (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } err = cudaMemcpy(dev_arr_B, host_arr_B, arr_size, cudaMemcpyHostToDevice); if (err != cudaSuccess) { fprintf(stderr, "Failed to copy vector B from host to device (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Launch the Vector Add CUDA Kernel int threadsPerBlock = 256; int blocksPerGrid =(NUM_OF_ELEMENTS + threadsPerBlock - 1) / threadsPerBlock; printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid, threadsPerBlock); vecAddKernel<<<blocksPerGrid, threadsPerBlock>>>(dev_arr_A, dev_arr_B, dev_arr_C, NUM_OF_ELEMENTS); err = cudaGetLastError(); if (err != cudaSuccess) { fprintf(stderr, "Failed to launch vectorAdd kernel (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Copy the device result vector in device memory to the host result vector // in host memory. printf("Copy output data from the CUDA device to the host memory\n"); err = cudaMemcpy(host_arr_C, dev_arr_C, arr_size, cudaMemcpyDeviceToHost); if (err != cudaSuccess) { fprintf(stderr, "Failed to copy vector C from device to host (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Verify that the result vector is correct for (int i = 0; i < NUM_OF_ELEMENTS; ++i) { if (fabs(host_arr_A[i] + host_arr_B[i] - host_arr_C[i]) > 1e-5) { fprintf(stderr, "Result verification failed at element %d!\n", i); exit(EXIT_FAILURE); } } printf("Test PASSED\n"); // Free device global memory err = cudaFree(dev_arr_A); if (err != cudaSuccess) { fprintf(stderr, "Failed to free device vector A (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } err = cudaFree(dev_arr_B); if (err != cudaSuccess) { fprintf(stderr, "Failed to free device vector B (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } err = cudaFree(dev_arr_C); if (err != cudaSuccess) { fprintf(stderr, "Failed to free device vector C (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } // Free host memory free(host_arr_A); free(host_arr_B); free(host_arr_C); printf("Done\n"); return EXIT_SUCCESS; }
5,069
#include <cuda_runtime.h> #include <device_launch_parameters.h> #include <stdio.h> #include <iostream> #include <cstring> using namespace std; __global__ void multiplyDigits(char* d_str1, char* d_str2, int* d_matrix, int str1_len, int str2_len) { int row = blockDim.y * blockIdx.x + threadIdx.y; int col = blockDim.x * blockIdx.y + threadIdx.x; int idx = row * str1_len + (col + (str2_len * row)) + 1 + (row); d_matrix[idx] = (d_str2[row] - '0') * (d_str1[col] - '0'); } __global__ void propagateCarries(int* d_matrix, int numCols) { int idx = blockDim.x * blockIdx.x + threadIdx.x * numCols; int carry = 0; for (int i = numCols - 1; i >= 0; i--) { int rowVal = (d_matrix[idx + i] + carry) % 10; carry = (d_matrix[idx + i] + carry) / 10; d_matrix[idx + i] = rowVal; } } __global__ void sumCols(int* d_matrix, int* d_result, int numRows, int numCols) { int sum = 0; int idx = blockDim.x * blockIdx.x + threadIdx.x; for (int i = 0; i < numRows; i++) { sum += d_matrix[idx + (numCols * i)]; } d_result[idx] = sum; } __host__ void propagateCarryInFinalResult(int* h_result, int numCols) { int carry = 0; for (int i = numCols - 1; i >= 0; i--) { int rowVal = (h_result[i] + carry) % 10; carry = (h_result[i] + carry) / 10; h_result[i] = rowVal; } } int main() { char* h_str1 = "111111"; char* h_str2 = "111111"; char* d_str1; char* d_str2; int* h_matrix; int* h_result; int* d_matrix; int* d_result; int row = strlen(h_str2); int col = strlen(h_str1) + row; h_matrix = new int[row * col]; h_result = new int[col]; cudaMalloc(&d_str1, sizeof(char) * strlen(h_str1)); cudaMalloc(&d_str2, sizeof(char) * strlen(h_str2)); cudaMalloc(&d_matrix, sizeof(int) * (row * col)); cudaMemset(&d_matrix, 0, sizeof(int) * (row * col)); cudaMalloc(&d_result, sizeof(int) * col); cudaMemcpy(d_str1, h_str1, sizeof(char) * strlen(h_str1), cudaMemcpyHostToDevice); cudaMemcpy(d_str2, h_str2, sizeof(char) * strlen(h_str2), cudaMemcpyHostToDevice); dim3 gridDim(strlen(h_str1) / 2, strlen(h_str2) / 2); dim3 blockDim(2, 2); //multiplyDigits<<<gridDim, dim3(strlen(h_str1) / 2, strlen(h_str2) / 2)>>>(d_str1, d_str2, d_matrix, strlen(h_str1), strlen(h_str2)); multiplyDigits<<<gridDim, blockDim>>>(d_str1, d_str2, d_matrix, strlen(h_str1), strlen(h_str2)); propagateCarries<<<row / 2, 2>>>(d_matrix, col); sumCols<<<col / 2, 2>>>(d_matrix, d_result, row, col); cudaMemcpy(h_result, d_result, sizeof(int) * col, cudaMemcpyDeviceToHost); propagateCarryInFinalResult(h_result, col); for (int i = 0; i < col; i++) { cout << h_result[i]; } cout << endl; cudaFree(d_str1); cudaFree(d_str2); cudaFree(d_result); delete[] h_matrix; delete[] h_result; cin.get(); return 0; }
5,070
#include "includes.h" __global__ void global_max(int *values, int *max, int *reg_maxes, int num_regions, int n) { int i = blockIdx.x * blockDim.x + threadIdx.x; int region = i % num_regions; if(i < n) { int val = values[i]; if(atomicMax(&reg_maxes[region], val) < val) { atomicMax(max, val); }//end of if statement }//end of if i < n }
5,071
#include <cuda_runtime.h> #include <stdio.h> #include <stdlib.h> #include <time.h> #include <device_launch_parameters.h> #define ARRAY_SIZE 1024*1024 #define NUM_THREADS 1024 // Saxpi 1 - Versin en C void saxpi_c(int n, float a, float* x, float* y) { for (int i = 0; i < n; i++) y[i] = a * x[i] + y[i]; } __global__ void saxpi_1(int n, float a, float* x, float* y) { for (int i = 0; i < n; i++) y[i] = a * x[i] + y[i]; } __global__ void saxpi_1Block(int n, float a, float* x, float* y) { int idx = threadIdx.x; int numElem = ARRAY_SIZE / NUM_THREADS; int offset = idx * numElem; if (offset + numElem < n) { for (int i = 0; i < numElem; i++) y[offset + i] = a * x[offset + i] + y[offset + i]; } } __global__ void saxpi_nBlock(int n, float a, float* x, float* y) { int idx = threadIdx.x + (blockIdx.x * blockDim.x); if (idx < n) { y[idx] = a * x[idx] + y[idx]; } } int main(void) { float* h_x, * h_y; // Apuntadores en host float* d_x, * d_y; // Apuntadores en device cudaEvent_t inicioG, finG; cudaEventCreate(&inicioG); cudaEventCreate(&finG); size_t sz = ARRAY_SIZE * sizeof(float); clock_t inicio, fin; // Asigna memoria h_x = (float*)malloc(sz); h_y = (float*)malloc(sz); cudaMalloc(&d_x, sz); cudaMalloc(&d_y, sz); // inicializa arreglos for (int i = 0; i < ARRAY_SIZE; i++) { h_x[i] = 1.0f; h_y[i] = 2.0f; } cudaMemcpy(d_x, h_x, sz, cudaMemcpyHostToDevice); cudaMemcpy(d_y, h_y, sz, cudaMemcpyHostToDevice); cudaEventRecord(inicioG); //saxpi_1Block <<<1, NUM_THREADS>>> (ARRAY_SIZE, 2.0, d_x, d_y); saxpi_nBlock <<< ARRAY_SIZE / NUM_THREADS, NUM_THREADS>>> (ARRAY_SIZE, 2.0, d_x, d_y); cudaEventRecord(finG); /* inicio = clock(); saxpi_c(ARRAY_SIZE, 2.0, h_x, h_y); fin = clock(); double t_exec = (double)(fin - inicio) / CLOCKS_PER_SEC; */ cudaMemcpy(h_y, d_y, sz, cudaMemcpyDeviceToHost); cudaEventSynchronize(finG); float t_exec = 0; cudaEventElapsedTime(&t_exec, inicioG, finG); printf("Algunos resultados: "); for (int i = 0; i < 10; i++) { printf("%3.2f, ", h_y[i]); } printf("\n Tiempo de ejecucion: %2.7f\n", t_exec); // Free memory free(h_x); free(h_y); cudaFree(d_x); cudaFree(d_y); return 0; }
5,072
/**********************************************************************\ * Author: Jose A. Iglesias-Guitian * * C/C++ code * * Introduction to CUDA * /**********************************************************************/ // Instructions: How to compile this program. // nvcc 0_hello_world.cu -L /usr/local/cuda/lib -lcudart -o 0_hello-world #include<stdio.h> int main(void) { printf("Hello World! \n"); return 0; }
5,073
#include <cassert> #include <cstdlib> #include <iostream> #define MASK_DIM 7 #define MASK_OFFSET (MASK_DIM / 2) __constant__ int mask[7 * 7]; __global__ void conv2d(int *matrix, int *result, int N) { int y = blockIdx.y * blockDim.y + threadIdx.y; int x = blockIdx.x * blockDim.x + threadIdx.x; int s_y = y - MASK_OFFSET; int s_x = x - MASK_OFFSET; int temp = 0; for (int i = 0; i < MASK_DIM; i++) { for (int j = 0; j < MASK_DIM; j++) { if((s_y + i) >= 0 && (s_y + i) < N) { if((s_x + j) >= 0 && (s_x + j) < N) { temp += matrix[N * (s_y + i) + (s_x + j)] * mask[MASK_DIM * i + j]; } } } } result[N * y + x] = temp; } void init_matrix(int *m, int n) { for (int i = 0; i < n; i++) { for (int j = 0; j < n; j++) { m[n * i + j] = rand() % 100; } } } void verify_result(int *matrix, int *result, int *mask, int N) { int temp, o_y, o_x; for(int y = 0; y < N; y++) { for(int x = 0; x < N; x++) { temp = 0; for (int i = 0; i < MASK_DIM; i++) { o_y = y - MASK_OFFSET + i; for (int j = 0; j < MASK_DIM; j++) { o_x = x - MASK_OFFSET + j; if(o_y >= 0 && o_y < N) { if(o_x >=0 && o_x < N) { temp += matrix[N * o_y + o_x] * mask[MASK_DIM * i + j]; } } } } assert(result[N * y + x] == temp); } } printf("result[%d] = %d == %d = temp \n", N * N - 1, result[N * N - 1], temp); } int main() { int N = 1 << 10; size_t bytes_n = sizeof(int) * N * N; size_t bytes_m = sizeof(int) * MASK_DIM * MASK_DIM; int *matrix = new int[N * N]; init_matrix(matrix, N); int *result = new int[N * N]; int *h_mask = new int[MASK_DIM * MASK_DIM]; init_matrix(h_mask, MASK_DIM); int *d_matrix, *d_result; cudaMalloc(&d_matrix, bytes_n); cudaMalloc(&d_result, bytes_n); cudaMemcpy(d_matrix, matrix, bytes_n, cudaMemcpyHostToDevice); cudaMemcpyToSymbol(mask, h_mask, bytes_m); int n_threads = 32; int n_blocks = (N + n_threads - 1) / n_threads; printf("%d, %d; ", n_threads, n_blocks); dim3 block_dim(n_threads, n_threads); dim3 grid_dim(n_blocks, n_blocks); conv2d<<<grid_dim, block_dim>>>(d_matrix, d_result, N); cudaMemcpy(result, d_result, bytes_n, cudaMemcpyDeviceToHost); verify_result(matrix, result, h_mask, N); std::cout << "COMPLETED SUCCESSFULLY! \n"; delete[] matrix; delete[] result; delete[] h_mask; cudaFree(d_matrix); cudaFree(d_result); return 0; }
5,074
#define NSTREAM 4 #include<stdio.h> __global__ void addVec(int* a, int* b, int* c, int const len){ int i = blockDim.x*blockIdx.x + threadIdx.x; if (i<len) c[i] = a[i] + b[i]; }; int main(){ int const totalLen = 1<<16; int const mSize = totalLen*sizeof(int); int* h_a; int* h_b; int* h_c; cudaHostAlloc((void**)&h_a, mSize, cudaHostAllocDefault); cudaHostAlloc((void**)&h_b, mSize, cudaHostAllocDefault); cudaHostAlloc((void**)&h_c, mSize, cudaHostAllocDefault); for (int i=0; i<totalLen; i++){ h_a[i] = i; h_b[i] = totalLen - i; } int* d_a; int* d_b; int* d_c; cudaMalloc((void**)&d_a, mSize); cudaMalloc((void**)&d_b, mSize); cudaMalloc((void**)&d_c, mSize); int const lenPerStream = totalLen/NSTREAM; int const mSizePerStream = mSize/NSTREAM; cudaStream_t lsStream[NSTREAM]; for (int i=0; i<NSTREAM; i++){ cudaStreamCreate(&lsStream[i]); } int const block = 256; int const grid = lenPerStream/block; for (int i=0; i<NSTREAM; i++){ int offset = i*lenPerStream; cudaMemcpyAsync(&d_a[offset], &h_a[offset], mSizePerStream, cudaMemcpyHostToDevice, lsStream[i]); cudaMemcpyAsync(&d_b[offset], &h_b[offset], mSizePerStream, cudaMemcpyHostToDevice, lsStream[i]); addVec<<<grid, block, 0, lsStream[i]>>>(&d_a[offset], &d_b[offset], &d_c[offset], lenPerStream); cudaMemcpyAsync(&h_c[offset], &d_c[offset], mSizePerStream, cudaMemcpyDeviceToHost, lsStream[i]); } for (int i=0; i<NSTREAM; i++){ cudaStreamSynchronize(lsStream[i]); } for (int i=0; i<totalLen; i++){ if (h_c[i]!=totalLen) { printf("error, %d, %d \n", h_c[i], i); break; } } for (int i=0; i<NSTREAM; i++){ cudaStreamDestroy(lsStream[i]); } return 0; }
5,075
#include <stdio.h> #include <string.h> #include <cmath> #include <iostream> #include <fstream> #include <ctime> #include <random> using namespace std; //use seed from time to generate random values std::mt19937 rng(time(0)); //used to set time to collision as a high value to indicate no collision int const NO_VALUE = std::numeric_limits<int>::max(); //type of collision enum CollType{NOCOLL, LEFTWALL, RIGHTWALL, TOPWALL, BOTTOMWALL, PARTCOLL}; //used to return what kind of collision happened with default values if //a collision did not happen struct Collision{ //colliding particle #1 (only one colliding if hitting a wall) int partId1 = NO_VALUE; //colliding particle #2 int partId2 = NO_VALUE; //tells if we collided with a particle, with which wall //or didn't collide CollType type = NOCOLL; //time it took to collide double t = NO_VALUE; }; typedef enum { MODE_PRINT, MODE_PERF } simulation_mode_t; //used to present particles in the simulation struct particle_t { int i = -1; double x = NO_VALUE; double y = NO_VALUE; double vx = NO_VALUE; double vy = NO_VALUE; int p_collisions = 0; int w_collisions = 0; bool collided = false ; bool stopPart = false; }; __constant__ int l, r, s; __constant__ int n; __managed__ particle_t* particles; __managed__ Collision wallCol; __managed__ Collision partCol; int host_n; // device method to atomically set minimum float value __device__ __forceinline__ float atomicMinD (float * addr, float value) { float old; old = (value >= 0) ? __int_as_float(atomicMin((int *)addr, __float_as_int(value))) : __uint_as_float(atomicMax((unsigned int *)addr, __float_as_uint(value))); return old; } // device method to calculate wall collision time __device__ Collision calcCollisionTime(particle_t &p1, int iL, int iR, double t){ Collision rCol; double l = iL; double r = iR; if(p1.vx < 0){ double t_help = (l - r - (l - p1.x))/(-p1.vx); if(t_help >= 0. && t_help < t){ rCol.t = t_help; rCol.type = LEFTWALL; rCol.partId1 = p1.i; } } if(p1.vx > 0){ double t_help = (l - r - p1.x)/p1.vx; if(t_help >= 0. && t_help < t){ rCol.t = t_help; rCol.type = RIGHTWALL; rCol.partId1 = p1.i; } } if(p1.vy < 0){ double t_help = (l - r - (l - p1.y))/(-p1.vy); if(t_help >= 0. && t_help < t && t_help < rCol.t){ rCol.t = t_help; rCol.type = BOTTOMWALL; rCol.partId1 = p1.i; } } if(p1.vy > 0){ double t_help = (l - r - p1.y)/p1.y; if(t_help >= 0. && t_help < t && t_help < rCol.t){ rCol.t = t_help; rCol.type = TOPWALL; rCol.partId1 = p1.i; } } return rCol; } // device method to calculate particle collision time __device__ Collision calcCollisionTime(particle_t &p1, particle_t &p2, double t){ Collision rCol; double a = (p2.vx-p1.vx)*(p2.vx-p1.vy) + (p2.vy-p1.vy)*(p2.vy-p1.vy); double b = 2.0*((p2.x-p1.x)*(p2.vx-p1.vx) + (p2.y-p1.y)*(p2.vy-p1.vy)); double c = (p2.x-p1.x)*(p2.x-p1.x) + (p2.y-p1.y)*(p2.y-p1.y) - (2*r)*(2*r); double det = b*b - 4*a*c; if (a != 0){ double t_help = (-b - sqrt(det))/(2.0*a); if (t_help >= 0. && t_help < t){ rCol.type = PARTCOLL; rCol.t = t_help; rCol.partId1 = p1.i; rCol.partId2 = p2.i; } } return rCol; } // Method updates fastest wall or particle collision depending on thread block __global__ void find_first_collisions(int num_threads, double t, int offset) { __shared__ int helpN; // collision indexes that only gets a value in block 2 int newN = NO_VALUE; Collision retCol; // fastest collision within thread __shared__ float blocktime; // fastest time in block if (threadIdx.x == 0) blocktime = NO_VALUE; helpN = NO_VALUE; //barrier to ensure blocktime is initialized to a comparable default before continuing __syncthreads(); int index = threadIdx.x; if (n > num_threads){ index += offset; } //wall collision, checked by 1st block if(blockIdx.x == 0 && index < n){ //If particle is stopped at a wall it can be ignored if(!particles[index].stopPart){ retCol = calcCollisionTime(particles[index], l, r, t); // comparing to remove collisions that are too slow to need atomic comparing if (retCol.type != NOCOLL && (float) retCol.t < blocktime) { atomicMinD(&blocktime, (float) retCol.t); } } } //////////////////////////////////////////////////////////////////////////////////////////////// //Block2 used to calculate particle-particle collisions else if(blockIdx.x == 1 && index < n){ //if hasn't collided yet if(!particles[index].collided){ //find collision time with each particle for(int j=0; j<n; j++){ if (particles[index].i == particles[j].i) { continue; } //neither particle has collided yet if(!particles[index].collided && !particles[j].collided){ retCol = calcCollisionTime(particles[index], particles[j], t); } } } if (retCol.type == PARTCOLL && (float) retCol.t < blocktime){ // atomic compare & replace fastest block time to found time if sooner // float conversion needed to further convert into int for atomicMin atomicMinD(&blocktime, (float) retCol.t); } } // barrier must be done outside of conditional __syncthreads(); // if there was a particle collision // and this thread's collision time = the block minimum if(retCol.type == PARTCOLL && (float) retCol.t == blocktime){ // calculate and compare index of collision newN = retCol.partId1 + retCol.partId2; //if new N is the same as current (reverse of existing collision or different collision with the same col index), void N if(atomicMin(&helpN, newN) == newN){ newN = NO_VALUE; } else { //printf("th%i p%i t%f N%i compared with N%i\n", threadIdx.x, index, blocktime, newN, helpN); } } // else if particle did collide with a wall (retCol has a value) // and thread's collision time = the block minimum else if (retCol.type != NOCOLL && (float) retCol.t == blocktime){ // if kernel is being run the first time (shortcircuits to save global access) // or global wall collision is later if(offset == 0 || wallCol.t > retCol.t){ // update rest of wallCol struct wallCol.t = retCol.t; wallCol.type = retCol.type; wallCol.partId1 = retCol.partId1; } } // make sure all the fastest threads in particle block have compared indexes __syncthreads(); // if this thread has a collision index and it's the chosen one // not accounting for identical indexes here, only 1 col should execute if (newN != NO_VALUE && newN == helpN) { // if kernel is being run the first time (shortcircuits to save global access) // or global particle collision is later if(offset == 0 || partCol.t > retCol.t){ // update partCol partCol.type = PARTCOLL; partCol.t = retCol.t; partCol.partId1 = retCol.partId1; partCol.partId2 = retCol.partId2; //printf("partCol updated\n"); } } } // advances particles in parallel __global__ void advanceParticlesP(int num_threads, double t, int offset){ int i = threadIdx.x + blockIdx.x*num_threads + offset; if (i < n && !particles[i].stopPart){ particles[i].y += particles[i].vy*t; particles[i].x += particles[i].vx*t; } } __host__ void checkCudaErrors() { cudaError_t rc; rc = cudaGetLastError(); if (rc != cudaSuccess) { printf("Last CUDA error %s\n", cudaGetErrorString(rc)); } } __host__ void print_particles(int step) { int i; for (i = 0; i < host_n; i++) { printf("%d %d %f %f %f %f\n", step, i, particles[i].x, particles[i].y, particles[i].vx, particles[i].vy); } } __host__ void print_statistics(int num_step) { int i; for (i = 0; i < host_n; i++) { printf("%d %d %f %f %f %f %d %d\n", num_step, i, particles[i].x, particles[i].y, particles[i].vx, particles[i].vy, particles[i].p_collisions, particles[i].w_collisions); } } // returns soonest out of fastest wall and particle collisions __host__ Collision whichCollision() { //If both collisions happen, choose the earlier one if(wallCol.type != NOCOLL && partCol.type == PARTCOLL){ //NOW WE PREFER WALLCOL IF BOTH T ARE SAME if(wallCol.t <= partCol.t){ return wallCol; } else { return partCol; } } //only one happened if(wallCol.type != NOCOLL){ return wallCol; } return partCol; //if no collisions happened the returned partCol //will have default info, this is checked later and no collision is done } __host__ void resetUnifiedCols(){ wallCol.type = NOCOLL; wallCol.t = NO_VALUE; wallCol.partId1 = NO_VALUE; partCol.type = NOCOLL; partCol.t = NO_VALUE; partCol.partId1 = NO_VALUE; partCol.partId2 = NO_VALUE; } __host__ void collideParticles(particle_t &p1, particle_t &p2) { //printf("colliding particles %i and %i, at %fx%f and %fx%f\n", p1.i, p2.i, p1.x, p1.y, p2.x, p2.y); //projecting the velocity unit normal and unit tangent vectors onto the unit //normal and unit tangent vectors, which is done by taking the dot product double vnx = (p2.x - p1.x); double vny = (p2.y - p1.y); double length = std::sqrt(vnx*vnx + vny*vny); double vunx = 0; double vuny = 0; if (length != 0){ vunx = vnx/length; vuny = vny/length; } double vutx = -vuny; double vuty = vunx; //printf("normal unit %fx%f, length %f, tangent unit %fx%f\n", vunx, vuny, length, vutx, vuty); //dot products double v1n = vunx*p1.vx + vuny*p1.vy; double v2n = vunx*p2.vx + vuny*p2.vy; double v1t = vutx*p1.vx + vuty*p1.vy; double v2t = vutx*p2.vx + vuty*p2.vy; //double v1n = vun.dot(p1.v); //double v2n = vun.dot(p2.v); //double v1t = vut.dot(p1.v); //double v2t = vut.dot(p2.v); //find the new tangential velocities (after the collision). double v1tPrime = v1t; double v2tPrime = v2t; //find the new normal velocities. //this is where we use the one-dimensional collision formulas. //for m1 = m2 we get double v1nPrime = v2n; double v2nPrime = v1n; //convert the scalar normal and tangential velocities into vectors. //multiply the unit normal vector by the scalar normal velocity double v_v1nPrimex = v1nPrime*vunx; double v_v1nPrimey = v1nPrime*vuny; double v_v2nPrimex = v2nPrime*vunx; double v_v2nPrimey = v2nPrime*vuny; double v_v1tPrimex = v1tPrime*vutx; double v_v1tPrimey = v1tPrime*vuty; double v_v2tPrimex = v2tPrime*vutx; double v_v2tPrimey = v2tPrime*vuty; //the final velocity vectors by adding the normal and tangential //components for each object p1.vx = v_v1nPrimex + v_v1tPrimex; p1.vy = v_v1nPrimey + v_v1tPrimey; p2.vx = v_v2nPrimex + v_v2tPrimex; p2.vy = v_v2nPrimey + v_v2tPrimey; //info that collided in cycle and up the collision count p1.collided = true; p1.p_collisions += 1; p2.collided = true; p2.p_collisions += 1; } //collides given particle with the given wall (changes the velociy) __host__ void collideWall(int i, CollType col) { //info that collided in cycle and up the collision count particles[i].collided = true; particles[i].w_collisions += 1; //fabs returns the absolute value //depending on which wall is hit, either vx or vy will //change sign if(col == LEFTWALL){ particles[i].vx = fabs(particles[i].vx); } else if (col == RIGHTWALL){ particles[i].vx = -fabs(particles[i].vx); } else if (col == TOPWALL){ particles[i].vy = -fabs(particles[i].vy); } else if (col == BOTTOMWALL){ particles[i].vy = fabs(particles[i].vy); } } /////////////////////////////////////////////////////////////// int main(int argc, char** argv) { int i, x, y, vx, vy; int num_blocks, num_threads; int step, offset; int host_l, host_r, host_s; simulation_mode_t mode; char mode_buf[6]; if (argc == 2){ num_threads = atoi(argv[1]); } else if (argc == 1){ num_threads = 512; } else { printf("Usage:\n%s (optional) num_threads < input\n", argv[0]); return 1; } scanf("%d", &host_n); scanf("%d", &host_l); scanf("%d", &host_r); scanf("%d", &host_s); scanf("%5s", mode_buf); // blocks to address all particles with the given number of threads = ratio rounded up num_blocks = ceil((float)num_threads/(float)host_n); printf("%d particles, %d threads per block, %d blocks.\n", host_n, num_threads, num_blocks); cudaMallocManaged(&particles, sizeof(particle_t) * host_n); for (i = 0; i < host_n; i++) { particles[i].i = -1; particles[i].p_collisions = 0; particles[i].w_collisions = 0; } i = 0; //added while (scanf("%d %d %d %d %d", &i, &x, &y, &vx, &vy) != EOF) { particles[i].i = i; particles[i].x = x; particles[i].y = y; particles[i].vx = vx; particles[i].vy = vy; } if (particles[0].i == -1) { //generate a distribution for random values in given range std::uniform_int_distribution<int> dist_pos(host_r,host_l-host_r); std::uniform_int_distribution<int> dist_v(host_l/(8*host_r), host_l/4); //Generate particles and store into data structure for(int j = 0; j < host_n; j++){ x=dist_pos(rng); y=dist_pos(rng); vx=dist_v(rng); vy=dist_v(rng); particles[j].i = j; particles[j].x = x; particles[j].y = y; particles[j].vx = vx; particles[j].vy = vy; } } mode = strcmp(mode_buf, "print") == 0 ? MODE_PRINT : MODE_PERF; /* Copy to GPU constant memory */ cudaMemcpyToSymbol(n, &host_n, sizeof(n)); cudaMemcpyToSymbol(l, &host_l, sizeof(l)); cudaMemcpyToSymbol(r, &host_r, sizeof(r)); cudaMemcpyToSymbol(s, &host_s, sizeof(s)); //printf("particles%i size%i radius%i steps%i\n", host_n, host_l, host_r, host_s); //simulation of step for (step = 0; step < host_s; step++) { if (mode == MODE_PRINT || step == 0) { print_particles(step); } for (i = 0; i < host_n; i++) { particles[i].collided = false; particles[i].stopPart = false; } double tMoved = 0.0; Collision deviceCol; //loop while within the step while(tMoved < 1.0){ // if number of particles is less than threads: // one block for each calculation will suffice // one block of N threads for wall collisions, one for particle // kernel is called with 2 blocks, numthreads in 1D each block // if block1/block2 statement seperates wall and particle collisions // syncthreads breaks up block seperation, then it returns to update wall/partCol // if particles exceeds threads: // a new grid is created with code to start at offset of particles already addressed. // each section of [numthreads] particles is calculated sequentially. for (offset = 0; offset < host_n; offset += num_threads) { /* Call the kernel */ find_first_collisions<<<2, num_threads>>>(num_threads, 1.0-tMoved, offset); //printf("Step %i tMoved %lf\n", step, tMoved); /* Barrier */ cudaDeviceSynchronize(); checkCudaErrors(); } deviceCol = whichCollision(); resetUnifiedCols(); //we had a collision if(deviceCol.type != NOCOLL && deviceCol.t >= 0){ //printf("returned collision type %i in %f\n", deviceCol.type, deviceCol.t); //move particles until collision time, using particle index offset if threads < n for (offset = 0; offset < host_n; offset += num_blocks*num_threads){ advanceParticlesP<<<num_blocks, num_threads>>>(num_threads, deviceCol.t, offset); cudaDeviceSynchronize(); checkCudaErrors(); //printf("advancing: t%d b%d time%lf, offset%d\n", num_threads, num_blocks, deviceCol.t, offset); } //advanceParticles(deviceCol.t); tMoved += deviceCol.t; //Collision between 2 particles if(deviceCol.type == PARTCOLL){ collideParticles(particles[deviceCol.partId1], particles[deviceCol.partId2]); } //wall collision else { //If particle hasn't collided yet if(!particles[deviceCol.partId1].collided){ collideWall(deviceCol.partId1, deviceCol.type); } //If it has, we need to stop it's movement at the wall else{ particles[deviceCol.partId1].stopPart = true; } } } else { //no remaining collisions so advance to end of step for (offset = 0; offset < host_n; offset += num_blocks*num_threads){ advanceParticlesP<<<num_blocks, num_threads>>>(num_threads, 1 - tMoved, offset); cudaDeviceSynchronize(); checkCudaErrors(); } tMoved = 1; } } } print_statistics(host_s); return 0; }
5,076
#include "includes.h" __global__ void ComputeBiasTermKernel( float *biasTerm, float cFactor, float *winningFraction, int activeCells, int maxCells ) { int threadId = blockDim.x*blockIdx.y*gridDim.x //rows preceeding current row in grid + blockDim.x*blockIdx.x //blocks preceeding current block + threadIdx.x; if(threadId < maxCells) { biasTerm[threadId] = cFactor * ( 1.00f / activeCells - winningFraction[threadId]); } }
5,077
#include "includes.h" __global__ void compute_l(double *dev_w, int n_patch) { int tid = threadIdx.x + blockIdx.x * blockDim.x; int N = n_patch * n_patch; while (tid < N) { dev_w[tid] = ((tid % (n_patch + 1) == 0) ? 1.0 : 0.0) - dev_w[tid]; tid += blockDim.x * gridDim.x; } }
5,078
#include <iostream> #include <vector> #include <cmath> #include <string> using namespace std::string_literals; #include <thrust/device_vector.h> #include <thrust/host_vector.h> #include <thrust/count.h> __global__ void add(unsigned int N, thrust::device_ptr<float> a, thrust::device_ptr<float> b) { auto index = blockIdx.x * blockDim.x + threadIdx.x; auto stride = blockDim.x * gridDim.x; auto a_raw_ptr = thrust::raw_pointer_cast(a); auto b_raw_ptr = thrust::raw_pointer_cast(b); for (auto i = index; i < N; i += stride) a_raw_ptr[i] += b_raw_ptr[i]; } void gpgpuWrapper() { auto constexpr N = 1'048'576u; auto constexpr kBLOCK_DIM = 256; auto constexpr kGRID_DIM = (N + kBLOCK_DIM - 1) / kBLOCK_DIM; thrust::device_vector<float> device_a(N, 1.f); thrust::device_vector<float> device_b(N, 2.f); add<<<kGRID_DIM, kBLOCK_DIM>>>(N, device_a.data(), device_b.data()); auto error_code = cudaDeviceSynchronize(); if (error_code != cudaError::cudaSuccess) std::cout << "An error is happened: "s << std::string{cudaGetErrorString(error_code)} << '\n'; auto count = thrust::count(device_a.begin(), device_a.end(), 3.f); std::cout << std::boolalpha << "Range is different: "s << (count != N) << '\n'; }
5,079
// vAdd.cu // // driver and kernel call #include <stdio.h> #define THREADS_PER_BLOCK 32 __global__ void vAdd_d (int *a_d, int *b_d, int *c_d, int n) { int x = blockIdx.x * blockDim.x + threadIdx.x; if (x < n) c_d[x] = a_d[x] + b_d[x]; } extern "C" void gpuAdd (int *a, int *b, int *c, int arraySize) { int *a_d, *b_d, *c_d; cudaMalloc ((void**) &a_d, sizeof(int) * arraySize); cudaMalloc ((void**) &b_d, sizeof(int) * arraySize); cudaMalloc ((void**) &c_d, sizeof(int) * arraySize); cudaMemcpy (a_d, a, sizeof(int) * arraySize, cudaMemcpyHostToDevice); cudaMemcpy (b_d, b, sizeof(int) * arraySize, cudaMemcpyHostToDevice); vAdd_d <<< ceil((float) arraySize/THREADS_PER_BLOCK), THREADS_PER_BLOCK >>> (a_d, b_d, c_d, arraySize); cudaError_t err = cudaGetLastError(); if (err != cudaSuccess) printf ("CUDA error: %s\n", cudaGetErrorString(err)); cudaMemcpy (c, c_d, sizeof(int) * arraySize, cudaMemcpyDeviceToHost); cudaFree (a_d); cudaFree (b_d); cudaFree (c_d); }
5,080
#include <stdio.h> #define NUM_BLOCKS 16 #define BLOCK_WIDTH 1 __global__ void hello() { printf("Hello world! I'm a thread in block %d\n", blockIdx.x); // It has 16! different ways in which the thread blocks can be run } int main(int argc,char **argv) { // launch the kernel hello<<<NUM_BLOCKS, BLOCK_WIDTH>>>(); // force the printf()s to flush cudaDeviceSynchronize(); printf("That's all!\n"); return 0; }
5,081
/* Daniel Sá Barretto Prado Garcia 10374344 Tiago Marino Silva 10734748 Felipe Guilermmo Santuche Moleiro 10724010 Laura Genari Alves de Jesus 10801180 */ #include <stdio.h> #include <stdlib.h> #include <string.h> #define THREADS 32 #define INF 0x7fffffff __global__ void prodEscalar(int* A, int* B, int* somaDosProd, int dim) { int i = blockIdx.x * blockDim.x + threadIdx.x; int j = blockIdx.y * blockDim.y + threadIdx.y; if(i < dim && j < dim ){ int produto = A[i*dim + j] * B[i*dim + j]; atomicAdd(somaDosProd, produto); } } __global__ void min_max_elementos(int* A, int* B, int* max_comp, int* min_comp, int dim) { //Calcula o índice global da thread int i = blockIdx.x * blockDim.x + threadIdx.x; int j = blockIdx.y * blockDim.y + threadIdx.y; if (i < dim && j < dim) { //Faz a soma entre elemento do vetor no índice idx e o conteúdo de soma int local_max = A[i*dim + j] > B[i*dim + j] ? A[i*dim + j] : B[i*dim + j]; int local_min = A[i*dim + j] < B[i*dim + j] ? A[i*dim + j] : B[i*dim + j]; atomicMax(max_comp, local_max); atomicMin(min_comp, local_min); } } int main(int argc, char **argv) { // Declara as matrizes int *A, *B; // Declara as variáveis de índice int i, j, dim; FILE *inputfile; // handler para o arquivo de entrada char *inputfilename; // nome do arquivo de entrada if (argc < 2) { printf("Please run with input file name, i.e., num_perf_mpi inputfile.ext\n"); exit(-1); } inputfilename = (char *)malloc(256 * sizeof(char)); strcpy(inputfilename, argv[1]); if ((inputfile = fopen(inputfilename, "r")) == 0) { printf("Error openning input file.\n"); exit(-1); } fscanf(inputfile, "%d\n", &dim); // Lê a dimensão das matrizes // Aloca as matrizes A = (int *)malloc(dim * dim * sizeof(int)); B = (int *)malloc(dim * dim * sizeof(int)); // Lê a matriz A for (i = 0; i < dim; i++) { for (j = 0; j < dim; j++) { fscanf(inputfile, "%d ", &(A[i * dim + j])); } } // Lê a matriz B for (i = 0; i < dim; i++) { for (j = 0; j < dim; j++) { fscanf(inputfile, "%d ", &(B[i * dim + j])); } } // fecha o arquivo de entrada fclose(inputfile); int *A_d, *B_d; //aloca matrizes na gpu cudaMalloc(&A_d,dim * dim * sizeof(int)); cudaMalloc(&B_d,dim * dim * sizeof(int)); cudaMemcpy(A_d, A, dim*dim*sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(B_d, B, dim*dim*sizeof(int), cudaMemcpyHostToDevice); int max = -INF, min = INF, soma = 0; int *min_D, *max_D, *soma_D; cudaMalloc(&min_D, sizeof(int)); cudaMalloc(&max_D, sizeof(int)); cudaMalloc(&soma_D, sizeof(int)); cudaMemcpy(max_D, &max, sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(min_D, &min, sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(soma_D, &soma, sizeof(int), cudaMemcpyHostToDevice); //Define a quantidade de threads por bloco dim3 threadsPerBlock(THREADS,THREADS); //Define a quantidade de blocos por grade dim3 blocksPerGrid((dim+(threadsPerBlock.x-1)) / threadsPerBlock.x, (dim+(threadsPerBlock.y-1)) / threadsPerBlock.y); min_max_elementos<<<blocksPerGrid, threadsPerBlock>>>(A_d, B_d, max_D, min_D, dim); prodEscalar<<<blocksPerGrid, threadsPerBlock>>>(A_d, B_d, soma_D, dim); //Copia o resultado da soma de volta para o host //cudaMemcpy(prod_escalar, prod_D, dim * sizeof(int), cudaMemcpyDeviceToHost); cudaMemcpy(&max, max_D,sizeof(int), cudaMemcpyDeviceToHost); cudaMemcpy(&min, min_D,sizeof(int), cudaMemcpyDeviceToHost); cudaMemcpy(&soma, soma_D,sizeof(int), cudaMemcpyDeviceToHost); //Imprime o resultado printf("%d %d %d\n", soma, min, max); // Liberação de memória alocada free(A); free(B); cudaFree(B_d); cudaFree(A_d); cudaFree(min_D); cudaFree(max_D); cudaFree(soma_D); }
5,082
#include "includes.h" __global__ void HessianPositiveDefiniteKernel( char *d_hessian_pd, float *d_Src, int imageW, int imageH, int imageD ) { __shared__ float s_Data[HES_BLOCKDIM_Z+2][HES_BLOCKDIM_Y+2][(HES_RESULT_STEPS + 2 * HES_HALO_STEPS) * HES_BLOCKDIM_X]; //Offset to the left halo edge const int baseX = (blockIdx.x * HES_RESULT_STEPS - HES_HALO_STEPS) * HES_BLOCKDIM_X + threadIdx.x; const int baseY = blockIdx.y * HES_BLOCKDIM_Y + threadIdx.y-1; const int baseZ = blockIdx.z * HES_BLOCKDIM_Z + threadIdx.z-1; const int idx = (baseZ * imageH + baseY) * imageW + baseX; d_Src += idx; d_hessian_pd += idx; if(baseZ < 0 || baseZ >= imageD || baseY < 0 || baseY >= imageH) { for (int i = 0; i < HES_HALO_STEPS + HES_RESULT_STEPS + HES_HALO_STEPS; i++) { s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] = 0; } return; } //Load main data #pragma unroll for (int i = HES_HALO_STEPS; i < HES_HALO_STEPS + HES_RESULT_STEPS; i++) { s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] = (float)d_Src[i * HES_BLOCKDIM_X]; } //Load left halo #pragma unroll for (int i = 0; i < HES_HALO_STEPS; i++) { s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] = (baseX + i * HES_BLOCKDIM_X >= 0) ? (float)d_Src[i * HES_BLOCKDIM_X] : 0; } //Load right halo #pragma unroll for (int i = HES_HALO_STEPS + HES_RESULT_STEPS; i < HES_HALO_STEPS + HES_RESULT_STEPS + HES_HALO_STEPS; i++) { s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] = (baseX + i * HES_BLOCKDIM_X < imageW) ? (float)d_Src[i * HES_BLOCKDIM_X] : 0; } // yz edge is no need to compute if (threadIdx.z == 0 || threadIdx.z == HES_BLOCKDIM_Z+1 || threadIdx.y == 0 || threadIdx.y == HES_BLOCKDIM_Y+1) return; //Compute and store results __syncthreads(); #pragma unroll for (int i = HES_HALO_STEPS; i < HES_HALO_STEPS + HES_RESULT_STEPS; i++) { float xx,xy,xz,yy,yz,zz; xx = s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X - 1] + s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X + 1] - s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] * 2; xy = s_Data[threadIdx.z][threadIdx.y + 1][threadIdx.x + i * HES_BLOCKDIM_X + 1] + s_Data[threadIdx.z][threadIdx.y - 1][threadIdx.x + i * HES_BLOCKDIM_X - 1] - s_Data[threadIdx.z][threadIdx.y + 1][threadIdx.x + i * HES_BLOCKDIM_X - 1] - s_Data[threadIdx.z][threadIdx.y - 1][threadIdx.x + i * HES_BLOCKDIM_X + 1]; xz = s_Data[threadIdx.z + 1][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X + 1] + s_Data[threadIdx.z - 1][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X - 1] - s_Data[threadIdx.z + 1][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X - 1] - s_Data[threadIdx.z - 1][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X + 1]; yy = s_Data[threadIdx.z][threadIdx.y + 1][threadIdx.x + i * HES_BLOCKDIM_X] + s_Data[threadIdx.z][threadIdx.y - 1][threadIdx.x + i * HES_BLOCKDIM_X] - s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] * 2; yz = s_Data[threadIdx.z + 1][threadIdx.y + 1][threadIdx.x + i * HES_BLOCKDIM_X] + s_Data[threadIdx.z - 1][threadIdx.y - 1][threadIdx.x + i * HES_BLOCKDIM_X] - s_Data[threadIdx.z + 1][threadIdx.y - 1][threadIdx.x + i * HES_BLOCKDIM_X] - s_Data[threadIdx.z - 1][threadIdx.y + 1][threadIdx.x + i * HES_BLOCKDIM_X]; zz = s_Data[threadIdx.z + 1][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] + s_Data[threadIdx.z - 1][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] - s_Data[threadIdx.z][threadIdx.y][threadIdx.x + i * HES_BLOCKDIM_X] * 2; xy *= 0.25; xz *= 0.25; yz *= 0.25; d_hessian_pd[i * HES_BLOCKDIM_X] = (xx < 0 && xx*yy-xy*xy < 0 && xx*yy*zz + 2*xy*yz*xz - xx*yz*yz - yy*xz*xz - zz*xy*xy < 0) ? 1 : 0; } }
5,083
#include "includes.h" __global__ void profilePhaseSolve_kernel() {}
5,084
#include <assert.h> #include <iostream> #include <cstdlib> #include<sys/time.h> #include <cmath> #include "cuda_runtime.h" const int LANGE = 16; __global__ void vecAdd(double *d_a, double *d_b, double *d_c, int N) { int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx < (N / LANGE)) { int large = N / LANGE; for (int i = 0; i < LANGE; i++) if (idx % 2 == 0){ d_c[idx + i * large] = d_a[idx + i * large] + d_b[idx + i *large]; }else{ d_c[idx + i * large] = d_a[idx + i * large] - d_b[idx + i * large]; } } } int main (void) { int n; std::cin>>n; assert(n % LANGE == 0); double *H_a, *H_b, *H_c; size_t bytes = n * sizeof(double); H_a = (double*)malloc(bytes); H_b = (double*)malloc(bytes); H_c = (double*)malloc(bytes); for (int i = 0; i < n; i++) { H_a[i] = sin(i) * sin(i); H_b[i] = cos(i) * cos(i); } //GPU parper double *d_a, *d_b, *d_c; cudaMalloc(&d_a, bytes); cudaMalloc(&d_b, bytes); cudaMalloc(&d_c, bytes); cudaMemcpy(d_a, H_a, bytes, cudaMemcpyHostToDevice); cudaMemcpy(d_b, H_b, bytes, cudaMemcpyHostToDevice); int blockSize = 1024; int gridSize = ((n-1)/LANGE)/blockSize + 1; cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); //gpu running cudaEventRecord(start); vecAdd<<<gridSize, blockSize>>>(d_a, d_b, d_c, n); cudaEventRecord(stop); cudaDeviceSynchronize(); cudaMemcpy(H_c, d_c, bytes, cudaMemcpyDeviceToHost); cudaEventSynchronize(stop); float timeGPU = 0; cudaEventElapsedTime(&timeGPU, start, stop); std::cout << "Runtime for GPU is: " << timeGPU<<" ms "<< std::endl; //CPU running..... struct timeval startCPU,endCPU; gettimeofday(&startCPU,NULL); for (int i = 0; i < n; i++) { H_c[i] = H_a[i] + H_b[i]; } gettimeofday(&endCPU,NULL); double timeCPU = endCPU.tv_sec - startCPU.tv_sec + (double)(endCPU.tv_usec - startCPU.tv_usec)/1000000; std::cout << "Runtime for CPU is: " << timeCPU <<" ms "<< std::endl; //summary if(timeGPU<timeCPU){ std::cout<<"GPU is faster than CPU for "<<timeCPU-timeGPU<<" ms "<<std::endl; }else if(timeGPU>timeCPU){ std::cout<<"CPU is faster than GPU for "<<timeGPU-timeCPU<<" ms "<<std::endl; }else{ std::cout<<"same time for GPU and CPU"<<std::endl; } free(H_a); free(H_b); free(H_c); cudaFree(d_a); cudaFree(d_b); cudaFree(d_c); return 0; }
5,085
#include <iostream> #include <cmath> using namespace std; __global__ void add(double *x, double *y, int N) { int i, ind, stride; ind = blockIdx.x*blockDim.x + threadIdx.x; stride = gridDim.x * blockDim.x; for(i=ind; i<N; i+=stride) { y[i] += x[i]; } } int main() { double *d_x, *d_y, *x, *y, err{0.}; int N = 1e6; int i, size = N*sizeof(double); cudaMalloc((void **)&d_x, size); cudaMalloc((void **)&d_y, size); x = new double[N]; y = new double[N]; for(i=0; i<N; ++i) { x[i] = 1.; y[i] = 2.; } cudaMemcpy(d_x, x, size, cudaMemcpyHostToDevice); cudaMemcpy(d_y, y, size, cudaMemcpyHostToDevice); add<<<20,512>>>(d_x, d_y, N); cudaMemcpy(y, d_y, size, cudaMemcpyDeviceToHost); for (i=0; i<N; ++i) { err += (y[i]-3.)*(y[i]-3.); } cout << "err = " << err << endl; delete [] x; delete [] y; cudaFree(d_x); cudaFree(d_y); }
5,086
#include "includes.h" #define NOMINMAX const unsigned int BLOCK_SIZE = 512; __global__ void addKernel(float *c, const float *a, const float *b) { int i = threadIdx.x; c[i] = a[i] + b[i]; }
5,087
#include "includes.h" __global__ void addWalkers ( const int dim, const int nwl, const float *xx0, const float *xxW, float *xx1 ) { int i = threadIdx.x + blockDim.x * blockIdx.x; int j = threadIdx.y + blockDim.y * blockIdx.y; int t = i + j * dim; if ( i < dim && j < nwl ) { xx1[t] = xx0[t] + xxW[t]; } }
5,088
#include <sstream> #include <iomanip> #include <cuda.h> #include <thrust/host_vector.h> #include <thrust/device_vector.h> #include <curand.h> #include <curand_kernel.h> #include <iostream> using namespace std; #define NUM_POINTS_PER_THREAD 1000 __global__ void kernel_initializeRand( curandState * randomGeneratorStateArray, unsigned long seed, int totalNumThreads) { int id = (blockIdx.x * blockDim.x) + threadIdx.x; if( id >= totalNumThreads){ return; } curand_init( seed, id, 0, &randomGeneratorStateArray[id]); } __global__ void kernel_generatePoints( curandState * globalState, int* counts, int totalNumThreads) { int index = (blockIdx.x * blockDim.x) + threadIdx.x; float x,y; if(index >= totalNumThreads){ return; } curandState localState = globalState[index]; for(int i = 0; i < NUM_POINTS_PER_THREAD; i++) { x = curand_uniform( &localState); y = curand_uniform( &localState); if(x*x+y*y <=1){ counts[index]++; } } globalState[index] = localState; } int main(int argc, char** argv) { if( argc < 2){ std::cerr << "error, incorrect param" << endl; exit(0); } int numThreads; { stringstream ss1(argv[1]); ss1 >> numThreads; } dim3 threadsPerBlock(1024,1,1); dim3 numberofBlocks((numThreads + threadsPerBlock.x-1)/threadsPerBlock.x,1,1); curandState* devRandomGeneratorStateArray; cudaMalloc (&devRandomGeneratorStateArray, numThreads*sizeof(curandState)); thrust::host_vector<int> hostCounts(numThreads,0); thrust::device_vector<int> deviceCounts(hostCounts); int* dcountsptr = thrust::raw_pointer_cast(&deviceCounts[0]); kernel_initializeRand <<< numberofBlocks, threadsPerBlock >>> ( devRandomGeneratorStateArray, time(NULL), numThreads); kernel_generatePoints <<< numberofBlocks, threadsPerBlock >>> (devRandomGeneratorStateArray, dcountsptr, numThreads); int sum = thrust::reduce(deviceCounts.begin(), deviceCounts.end(), 0, thrust::plus<int>()); std::cout << "our approx of pi = " <<std::setprecision(10) << (float(sum)/(numThreads*NUM_POINTS_PER_THREAD))*4 << std::endl; }
5,089
__global__ void add_kernel(int *x, int a, int b) { x[0] = a + b; } void add(int *x, int a, int b) { add_kernel<<<1, 1>>>(x, a, b); }
5,090
#include <stdio.h> #include <stdlib.h> #include <cuda.h> #include <math.h> // #include <stdexcept> #define CUDA_CALL(x) do { if((x)!=cudaSuccess) { \ printf("Error at %s:%d\n",__FILE__,__LINE__);\ return EXIT_FAILURE;}} while(0) __global__ void prepare_function(float * d_out, int n_points, float x_min, float x_max) { int id = threadIdx.x + blockIdx.x * blockDim.x; float x = x_min + (x_max - x_min) * id / n_points; d_out[id] = exp(-pow(x, 2)); // d_out[id] = (float) id; } // __global__ void blelloch_reduce(float * d_in, int n_points) // { // /* Assuming n_points is a power of two */ // int n_current = 2; // int id = threadIdx.x + blockIdx.x * blockDim.x; // // while(n_current <= n_points) // { // if ((id + 1) % n_current == 0) // { // d_in[id] += d_in[id - n_current/2]; // } // __syncthreads(); // n_current = n_current * 2; // } // // } __global__ void blelloch_reduce_step(float * d_in, int n_points, int stride_step) { int id = threadIdx.x + blockIdx.x * blockDim.x; if ((id + 1) % stride_step == 0) { d_in[id] += d_in[id - stride_step/2]; } } __global__ void kill_last(float * d_in, int n_points) { int id = threadIdx.x + blockIdx.x * blockDim.x; if (id == n_points - 1) { d_in[id] = 0; } } /* __global__ void blelloch_downsweep(float * d_in, int n_points) { int n_current = n_points; int id = threadIdx.x + blockIdx.x * blockDim.x; float tmp; if (id == n_points - 1) { d_in[id] = 0; } __syncthreads(); while(n_current >= 2) { if ((id + 1) % n_current == 0) { tmp = d_in[id]; d_in[id] += d_in[id - n_current/2]; d_in[id - n_current/2] = tmp; } n_current = n_current / 2; __syncthreads(); } }*/ __global__ void blelloch_downsweep_step(float * d_in, int n_points, int stride_step) { int id = threadIdx.x + blockIdx.x * blockDim.x; float tmp; if ((id + 1) % stride_step == 0) { tmp = d_in[id]; d_in[id] += d_in[id - stride_step/2]; d_in[id - stride_step/2] = tmp; } // __syncthreads(); } int main(int argc, char* argv[]) { float minus_infty = -8; float x_max = 0; int n_blocks = 1024; int n_points_per_block = 1024; int n_points = n_points_per_block * n_blocks; int stride_step; float dx; float *devFunVals; float *hostFunVals; float *hostFunVals2; if (argc > 1) { sscanf(argv[1], "%f", &x_max); // printf("%f\n", x_max); if (x_max < minus_infty) { printf("0\n"); return 0; } } else { printf("Usage: ./scan <number> \n"); return 0; } dx = (x_max - minus_infty) / (float) n_points; // printf("dx: %e\n", dx); // printf("n_points: %d\n", n_points); // if (n_points < 0 || ((n_points & (n_points - 1)) != 0)) { printf("n_points is not a power of two\n"); return 1; } hostFunVals = (float *)calloc(n_points, sizeof(float)); hostFunVals2 = (float *)calloc(n_points, sizeof(float)); CUDA_CALL(cudaMalloc((void **)&devFunVals, n_points*sizeof(float))); prepare_function<<<n_blocks, n_points_per_block>>>(devFunVals, n_points, minus_infty, x_max); // blelloch_reduce<<<n_blocks, n_points_per_block>>>(devFunVals, n_points); stride_step = 2; while(stride_step <= n_points) { blelloch_reduce_step<<<n_blocks, n_points_per_block>>>(devFunVals, n_points, stride_step); stride_step = stride_step * 2; } stride_step = n_points; CUDA_CALL(cudaMemcpy(hostFunVals, devFunVals, n_points*sizeof(float), cudaMemcpyDeviceToHost)); kill_last<<<n_blocks, n_points_per_block>>>(devFunVals, n_points); while(stride_step >= 2) { blelloch_downsweep_step<<<n_blocks, n_points_per_block>>>(devFunVals, n_points, stride_step); stride_step = stride_step / 2; } CUDA_CALL(cudaMemcpy(hostFunVals2, devFunVals, n_points*sizeof(float), cudaMemcpyDeviceToHost)); // for(int i=0; i<n_points; i++) // { // printf("%1.4f \n", hostFunVals[i]); // } // printf("=========\n"); // // for(int i=0; i<n_points; i++) // { // printf("%1.4f \n", hostFunVals2[i]); // } // printf("\n"); // printf("Func value: %1.5f\n", hostFunVals[n_points - 1]); printf("Integral value: %1.5e\n", hostFunVals2[n_points - 1] * dx); free(hostFunVals); free(hostFunVals2); CUDA_CALL(cudaFree(devFunVals)); return 0; }
5,091
#include <stdio.h> __global__ void kernel1( int *a ) { int idx = blockIdx.x*blockDim.x + threadIdx.x; a[idx] = 7; // output: 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 } __global__ void kernel2( int *a ) { int idx = blockIdx.x*blockDim.x + threadIdx.x; a[idx] = blockIdx.x; // output: 0 0 0 0 1 1 1 1 2 2 2 2 3 3 3 3 } __global__ void kernel3( int *a ) { int idx = blockIdx.x*blockDim.x + threadIdx.x; a[idx] = threadIdx.x; // output: 0 1 2 3 1 2 3 4 0 1 2 3 0 1 2 3 } int main() { int *host_array; int *dev_array; host_array = (int *) malloc(sizeof(int)*16); cudaMalloc(&dev_array, sizeof(int)*16); cudaMemset(dev_array, 0, 16); kernel1<<<4, 4>>>(dev_array); cudaMemcpy(host_array, dev_array, sizeof(int)*16, cudaMemcpyDeviceToHost); for(int i = 0; i < 16; i++) printf(" %d ", host_array[i]); printf("\n"); free(host_array); cudaFree(dev_array); cudaDeviceReset(); return 0; }
5,092
#include <cuda.h> #include <cuda_runtime_api.h> #define N_MEM_OPS_PER_KERNEL 2 //----------------------------------------------------------------------------- // Simple test kernel template for memory ops test // @param d_counters - Simple memory location to exploit for lots of memory accesses // @param n_threads - Total number of threads per block //----------------------------------------------------------------------------- __global__ void max_flops_kernel(float* d_in, float* d_out, int offset) { // Increment the counter int it = blockIdx.x * blockDim.x + threadIdx.x + offset; d_out[it] = d_in[it]; }
5,093
/* Kernel for vector squaring */ __global__ void gpusquare(float in[], float out[], int n) { int i = blockDim.x * blockIdx.x + threadIdx.x; if (i < n) { out[i] = in[i] * in[i]; } }
5,094
#include "includes.h" __device__ unsigned int getGid3d3d(){ int blockId = blockIdx.x + blockIdx.y * gridDim.x + gridDim.x * gridDim.y * blockIdx.z; int threadId = blockId * (blockDim.x * blockDim.y * blockDim.z) + (threadIdx.y * blockDim.x) + (threadIdx.z * (blockDim.x * blockDim.y)) + threadIdx.x; return threadId; } __global__ void scalarDiv_wfcNorm(double2* in, double dr, double* pSum, double2* out){ unsigned int gid = getGid3d3d(); double2 result; double norm = sqrt((pSum[0])*dr); result.x = (in[gid].x/norm); result.y = (in[gid].y/norm); out[gid] = result; }
5,095
#include "includes.h" __global__ void cuda_mul(int* A, int* B, int* C, int w) { int tid,tx,ty; //range of tx,ty 0 ~ w tx = blockDim.x * blockIdx.x + threadIdx.x; ty = blockDim.y * blockIdx.y + threadIdx.y; tid = w*ty + tx; int v = 0; int a = 0; int b = 0; /* oooo oxo xxxx X oxo oooo oxo oxo */ for(int i=0;i< w;i++) { a = A[ty * w + i]; b = B[i * w + tx]; v += a+b; } C[tid]= v; }
5,096
#include <stdio.h> // add sera ejecuta en el device // add será llamada desde el host // add corre en device asi que a,b y c deben apuntar a memoria del device __global__ void add(int *a, int *b, int *c){ *c = *a + *b; } int main(void){ int a, b, c; // Copias de a b y c en el host int *d_a, *d_b, *d_c; // Copias de a, b y c en el device int size = sizeof(int); // Obtenemos espacio para las copias de a,b y c en device cudaMalloc((void **)&d_a, size); cudaMalloc((void **)&d_b, size); cudaMalloc((void **)&d_c, size); // Valores input a = 2; b = 7; // Copiamos inputs a device cudaMemcpy(d_a, &a, size, cudaMemcpyHostToDevice); cudaMemcpy(d_b, &b, size, cudaMemcpyHostToDevice); // Lanzamos add() kernen en la GPU add<<<1,1>>>(d_a, d_b, d_c); // Copiamos el resultado al host cudaMemcpy(&c, d_c, size, cudaMemcpyDeviceToHost); // Limpiamos cudaFree(d_a); cudaFree(d_b); cudaFree(d_c); printf("%d",c); return 0; }
5,097
//nvcc -ptx EM2.cu -ccbin "F:Visual Studio\VC\Tools\MSVC\14.12.25827\bin\Hostx64\x64" __device__ void EM1( double * x, double * y, double * z, double * vx, double * vy, double * vz, double * r, double * phi, double * vr, int * parDelete, const int parNum, const double Rp, const double Lp, const double PHIp, const double dr, const double dz, const double dt ) { int globalBlockIndex = blockIdx.x + blockIdx.y * gridDim.x; int localThreadIdx = threadIdx.x + blockDim.x * threadIdx.y; int threadsPerBlock = blockDim.x*blockDim.y; int n = localThreadIdx + globalBlockIndex*threadsPerBlock; if ( n >= parNum || parDelete[n]==1 ){ return; } x[n] = x[n] + 0.5 * vx[n] * dt; y[n] = y[n] + 0.5 * vy[n] * dt; z[n] = z[n] + 0.5 * vz[n] * dt; r[n] = sqrt( x[n]*x[n] + y[n]*y[n] ); phi[n] = atan( y[n]/x[n] ); if (r[n] > Rp){ parDelete[n] = 1; } double vx1; while(phi[n]<0){ phi[n] = phi[n] + PHIp; vx1 = vx[n] * cos(PHIp) - vy[n] * sin(PHIp); vy[n] = vx[n] * sin(PHIp) + vy[n] * cos(PHIp); vx[n] = vx1; } while(phi[n]>PHIp){ phi[n] = phi[n] - PHIp; vx1 = vx[n] * cos(PHIp) + vy[n] * sin(PHIp); vy[n] = -vx[n] * sin(PHIp) + vy[n] * cos(PHIp); vx[n] = vx1; } x[n] = r[n] * cos(phi[n]); y[n] = r[n] * sin(phi[n]); if (z[n]>Lp){ z[n] = 2*Lp - z[n]; vz[n] = -vz[n]; } if (z[n]<0){ z[n] = -z[n]; vz[n] = -vz[n]; } vr[n] = vx[n]*cos(phi[n]) + vy[n]*sin(phi[n]) ; } __global__ void processMandelbrotElement( double * x, double * y, double * z, double * vx, double * vy, double * vz, double * r, double * phi, double * vr, int * parDelete, const int parNum, const double Rp, const double Lp, const double PHIp, const double dr, const double dz, const double dt ) { EM1(x, y, z, vx, vy, vz, r, phi, vr, parDelete, parNum, Rp, Lp, PHIp, dr, dz, dt); }
5,098
#include <stdio.h> #include <stdlib.h> #include <time.h> /*#define M(row, col) *(M.elements + (row) (*) M.width + col)*/ typedef struct { int width; int height; float* elements; } Matrix; //a h w B h w C void MatMul(const Matrix A, const Matrix B, Matrix C) { for (int i = 0; i < A.height; i++) { for (int j = 0; j < B.width; j++) { C.elements[i * C.width + j] = 0; for (int k = 0; k < A.width; k++) { C.elements[i * C.width + j] += A.elements[i * A.width + k] * B.elements[k * B.width + j]; } } } } void fill_Matrix(Matrix A) { for (int i = 0; i < A.height ; i++) { for (int j = 0; j < A.width; j++) { A.elements[i * A.width + j] = rand() / (float)RAND_MAX * 10; } } } void print_Matrix(Matrix A) { /*for (int i = 0; i < A.height; i++) {*/ /*for (int j = 0; j < A.width; j++) {*/ /*printf("%4.1f ", A.elements[i * A.width + j]);*/ /*}*/ /*printf("\n");*/ /*}*/ } int main(int argc, char **argv) { if (argc != 2) { printf("usage: n\n"); return -1; } int nnn = atoi(argv[1]); int n = 1 << nnn; srand(time(0)); Matrix A, B, C; A.width = A.height = n; A.elements = (float *)malloc(sizeof(float) * n * n); B.width = B.height = n; B.elements = (float *)malloc(sizeof(float) * n * n); C.width = C.height = n; C.elements = (float *)malloc(sizeof(float) * n * n); fill_Matrix(A); print_Matrix(A); printf("\n"); fill_Matrix(B); print_Matrix(B); printf("\n"); MatMul(A, B, C); print_Matrix(C); }
5,099
#include <cstdio> #include <cstdlib> #include <cmath> #define N 9999 // number of bodies #define MASS 0 // row in array for mass #define X_POS 1 // row in array for x position #define Y_POS 2 // row in array for y position #define Z_POS 3 // row in array for z position #define X_VEL 4 // row in array for x velocity #define Y_VEL 5 // row in array for y velocity #define Z_VEL 6 // row in array for z velocity #define G 10 // "gravitational constant" (not really) float dt = 0.05; // time interval // JJ: one could compute the distance matrix using tiles first in pararrel, then compute forces, and finnaly update the velocities and positions. In this naive project, I chose not to implement it. // each thread computes new position of one body __global__ void nbody(float *dev_body, float dt) { // Index for body i int i = blockDim.x*blockIdx.x + threadIdx.x; if(i<N) // if index is larger than number of bodies, no need to calculate. { float x_diff, y_diff, z_diff, r; float Fx_dir=0, Fy_dir=0, Fz_dir=0; for (int j=0; j<N && j!=i; j++) // force on body i due to body j { x_diff = dev_body[j*7+X_POS] - dev_body [i*7+X_POS]; // difference in x direction y_diff = dev_body[j*7+Y_POS] - dev_body [i*7+Y_POS]; // difference in y direction z_diff = dev_body[j*7+Z_POS] - dev_body [i*7+Z_POS]; // difference in z direction // calculate distance between i and j (r) r = sqrt(x_diff * x_diff + y_diff * y_diff + z_diff * z_diff); // force between bodies i and x float F = G * dev_body[i*7+MASS] * dev_body[j*7+MASS] / r; //JJ: It should be r^2 on my opinion. The same for the following. if (r > 10.0) { Fx_dir += F * x_diff / r; // resolve forces in x and y directions Fy_dir += F * y_diff / r; // and accumulate forces Fz_dir += F * z_diff / r; // } else { // if too close, anti-gravitational force Fx_dir -= F * x_diff / r; // resolve forces in x and y directions Fy_dir -= F * y_diff / r; // and accumulate forces Fz_dir -= F * z_diff / r; // } } // update velocities dev_body[i*7+X_VEL] += Fx_dir * dt / dev_body[i*7+MASS]; dev_body[i*7+Y_VEL] += Fy_dir * dt / dev_body[i*7+MASS]; dev_body[i*7+Z_VEL] += Fz_dir * dt / dev_body[i*7+MASS]; // update positions dev_body[i*7+X_POS] += dev_body[i*7+X_VEL] * dt; dev_body[i*7+Y_POS] += dev_body[i*7+Y_VEL] * dt; dev_body[i*7+Z_POS] += dev_body[i*7+Z_VEL] * dt; } } int main(int argc, char **argv) { float *body; // host data array of bodies float *dev_body; // device data array of bodies int tmax = 0; if (argc != 2) { fprintf(stderr, "Format: %s { number of timesteps }\n", argv[0]); exit (-1); } tmax = atoi(argv[1]); printf("Requested Timesteps: %d.\n", tmax); // allocate memory size for the body int bodysize = N * 7 * sizeof(float); body = (float *)malloc(bodysize); cudaMalloc((void**) &dev_body, bodysize); // assign each body a random position for (int i = 0; i < N; i++) { body[i * 7 + MASS] = 10; body[i * 7 + X_POS] = drand48() * 100.0; body[i * 7 + Y_POS] = drand48() * 100.0; body[i * 7 + Z_POS] = drand48() * 100.0; body[i * 7 + X_VEL] = 0.0; body[i * 7 + Y_VEL] = 0.0; body[i * 7 + Z_VEL] = 0.0; } // print out initial positions in PDB format FILE * fp; fp = fopen ("NBody.pdb", "w"); fprintf(fp, "MODEL %8d\n", 0); for (int i = 0; i < N; i++) { fprintf(fp, "%s%7d %s %s %s%4d %8.3f%8.3f%8.3f %4.2f %4.3f\n", "ATOM", i+1, "CA ", "GLY", "A", i+1, body[i * 7 + X_POS], body[i * 7 + Y_POS], body[i * 7 + Z_POS], 1.00, 0.00); fflush(fp); } fprintf(fp, "TER\nENDMDL\n"); // step through each time step for (int t = 0; t < tmax; t++) { // copy nbody info over to GPU cudaMemcpy(dev_body, body, bodysize, cudaMemcpyHostToDevice); dim3 blockDim(1024); dim3 gridDim((int)ceil(N / blockDim.x) + 1); // run nbody calculation nbody<<<gridDim, blockDim>>>(dev_body, dt); cudaThreadSynchronize(); cudaError_t err = cudaGetLastError(); if (err != cudaSuccess) { printf("Error: %s when t=%d\n", cudaGetErrorString(err),t); return -1; } // copy nbody info back to CPU cudaMemcpy(body, dev_body, bodysize, cudaMemcpyDeviceToHost); // print out positions in PDB format fprintf(fp, "MODEL %8d\n", t+1); for (int i = 0; i < N; i++) { int error = fprintf(fp, "%s%7d %s %s %s%4d %8.3f%8.3f%8.3f %4.2f %4.3f\n","ATOM", i+1, "CA ", "GLY", "A", i+1, body[i * 7 + X_POS], body[i * 7 + Y_POS], body[i * 7 + Z_POS], 1.00, 0.00); if (error ==-1) { printf("printf error when t=%d,i=%d\n", t+1,i); break; } else fflush(fp); } fprintf(fp, "TER\nENDMDL\n"); } // end of time period loop fclose(fp); free(body); cudaFree(dev_body); }
5,100
// write your code into this file #define TILE_X 16 #define TILE_Y 8 #define TILE_Z 8 #define PADDING 1 __global__ void compute_cell(int* in_array, int* out_array, int dim); void solveGPU(int **dCells, int dim, int iters) { dim3 threadsPerBlock(TILE_X, TILE_Y, TILE_Z); dim3 numBlocks((int)ceil(dim/(float)(TILE_X-2)), (int)ceil(dim/(float)(TILE_Y-2)), (int)ceil(dim/(float)(TILE_Z-2))); int* result_array; cudaMalloc((void**)&result_array, dim*dim*dim*sizeof(int)); int* tmp; int* array_in; int* array_out; array_in = *dCells; array_out = (int*)result_array; cudaFuncSetCacheConfig(compute_cell, cudaFuncCachePreferShared); for (int i = 0; i < iters; i++) { compute_cell<<<numBlocks, threadsPerBlock>>>(array_in, array_out, dim); result_array = array_out; tmp = array_in; array_in = array_out; array_out = tmp; } *dCells = result_array; // result array from loop above cudaFree(array_out); } __global__ void compute_cell(int* in_array, int* out_array, int dim) { __shared__ int tile[TILE_Z][TILE_Y][TILE_X]; int mat_idx_x = blockIdx.x*(blockDim.x-2) + threadIdx.x-1; int mat_idx_y = blockIdx.y*(blockDim.y-2) + threadIdx.y-1; int mat_idx_z = blockIdx.z*(blockDim.z-2) + threadIdx.z-1; int dim2 = dim*dim; unsigned short thread_exceeds_matrix = 1; if ((mat_idx_x < dim) && (mat_idx_y < dim) && (mat_idx_z < dim) && (mat_idx_x >= 0) && (mat_idx_y >= 0) && (mat_idx_z >= 0)) { thread_exceeds_matrix = 0; tile[threadIdx.z][threadIdx.y][threadIdx.x] = in_array[mat_idx_x+(mat_idx_y*dim)+(mat_idx_z*dim2)]; } else tile[threadIdx.z][threadIdx.y][threadIdx.x] = 0; __syncthreads(); int result = 0; int shared_value = tile[threadIdx.z][threadIdx.y][threadIdx.x]; int is_hull = 1; if (threadIdx.x > 0 && threadIdx.y > 0 && threadIdx.z > 0 && threadIdx.x < (TILE_X-1) && threadIdx.y < (TILE_Y-1) && threadIdx.z < (TILE_Z-1) ) { is_hull = 0; } if (!is_hull) { for (int i = -1; i < 2; i++) { for (int j = -1; j < 2; j++) { for (int k = -1; k < 2; k++) { result += tile[threadIdx.z+i][threadIdx.y+j][threadIdx.x+k]; } } } result -= shared_value; if ((result < 4) || (result > 5)) { result = 0; } else if (result == 5) { result = 1; } else { result = shared_value; } } __syncthreads(); // loading back to global if (!is_hull && !thread_exceeds_matrix) { out_array[mat_idx_x+(mat_idx_y*dim)+(mat_idx_z*dim2)] = result; } /* __syncthreads(); if (mat_idx_x == 0 && mat_idx_y == 0 && mat_idx_z == 0) { printf("gpu:\n"); for (int i = 0; i<TILE_X;i++) for (int j = 0; j<TILE_Y; j++) { for (int k = 0;k< TILE_Z; k++) printf("%d ", tile[i][j][k]); printf("\n"); } } */ // cell life computation }