serial_no
int64
1
24.2k
cuda_source
stringlengths
11
9.01M
12,901
//pass //--blockDim=32 --gridDim=2 #include <cuda.h> __global__ void test_Prog(int *A, int N) { int tid = threadIdx.x; int bid = blockIdx.x; int idx = blockDim.x * bid + tid; int alpha = 0; if (idx % 2 == 0) { alpha = A[tid + 2]; } if (idx % 6 == 0) { A[tid] = A[tid] + alpha; } }
12,902
#include <cstdio> #include <iostream> #include <string> #include <vector> #include <time.h> #include <cuda_runtime.h> #include <cmath> #include <cstdlib> #define BLOCK_SIZE 12288 //number of floats processed per SM #define MAX_THREADS_PER_BLOCK 1024 //Limit of GTX 1080 #define BLOCK_HEIGHT 32 #define BLOCK_WIDTH 32 #define ELM_PER_THREAD 12 // BLOCK_SIZE / MAX_THREADS using namespace std; timespec start_time; timespec stop_time; void start_clock(); void stop_clock(); double get_clock_result_seconds(); void print_time_seconds(double seconds); void PrintPartialMatrix(size_t n, float* matrix) { if(n < 5) { printf("Matrix is too small to print.\n"); return; } for(size_t i = 0; i < 5; ++i) { for(size_t j = 0; j < 5; j++) { printf("%0.2f\t", matrix[(i*n) + j]); } printf("\n"); } } __global__ void transpose_one_to_one(size_t n, float* input, float* output) { int global_col = (blockDim.x * blockIdx.x) + threadIdx.x; int global_row = (blockDim.y * blockIdx.y) + threadIdx.y; int i = (n * global_row) + global_col; //printf("block(%i, %i)\tthread(%i, %i)\ti:%i\n", blockIdx.x, blockIdx.y, threadIdx.x, threadIdx.y, i); if(i >= n*n) return; output[((i % n)*n) + (i / n)] = input[i]; } __global__ void transpose_optimized(size_t n, float* input, float* output) { __shared__ float s_data[BLOCK_SIZE]; //unsigned int global_col = ((blockDim.x * blockIdx.x) * ELM_PER_THREAD) + (threadIdx.x * ELM_PER_THREAD); //unsigned int global_row = (blockDim.y * blockIdx.y) + threadIdx.y; //unsigned int i = (n * (global_row)) + (global_col); unsigned int i = (n * ((blockDim.y * blockIdx.y) + threadIdx.y)) + (((blockDim.x * blockIdx.x) * ELM_PER_THREAD) + (threadIdx.x * ELM_PER_THREAD)); unsigned int block_level_index = ((threadIdx.y * blockDim.x) + threadIdx.x) * ELM_PER_THREAD; unsigned int start = i + block_level_index; unsigned int stop = start + ELM_PER_THREAD; int s_idx = block_level_index; for(unsigned int i = start; i < stop; i++, s_idx++) { if(i >= n*n) break; s_data[s_idx] = input[i]; } __syncthreads(); s_idx = block_level_index; for(int i = start; i < stop; i++, s_idx++) { if(i >= n*n) break; output[((i % n)*n) + (i / n)] = s_data[s_idx]; } } int main(int argc, char** argv) { if (argc < 2) { printf("Not enough arguments\n"); printf("Usage is ./a.out [matrix dim]\n"); return 1; } string dimension_arg = argv[1]; size_t N = 0; try{ N = strtoul(dimension_arg.c_str(), NULL, 10); } catch(...){ printf("Matrix dimension argument %s is not valid\n", dimension_arg.c_str()); return 2; } size_t matrix_size = N*N*sizeof(float); float* d_input_matrix; float* d_resultant_matrix; cudaMalloc((void **)&d_input_matrix, matrix_size); cudaMalloc((void **)&d_resultant_matrix, matrix_size); float* h_input_matrix = new float[N*N]; float* h_resultant_matrix_1 = new float[N*N]; float* h_resultant_matrix_2 = new float[N*N]; srand(time(NULL)); for( int i = 0; i < N*N; i++) h_input_matrix[i] = (float)rand() / (float)RAND_MAX; //h_input_matrix[i] = 0.11; cudaMemcpy(d_input_matrix, h_input_matrix, matrix_size, cudaMemcpyHostToDevice); dim3 block(BLOCK_WIDTH, BLOCK_HEIGHT); size_t grid_width = N / BLOCK_WIDTH; grid_width += N % BLOCK_WIDTH > 0 ? 1 : 0; size_t grid_height = N / (BLOCK_HEIGHT); grid_height += N % BLOCK_HEIGHT > 0 ? 1 : 0; dim3 grid(grid_width, grid_height); printf("grid(%lu, %lu)\n", grid_width, grid_height); start_clock(); transpose_one_to_one<<<grid, block>>>(N, d_input_matrix, d_resultant_matrix); cudaDeviceSynchronize(); stop_clock(); printf("naive time:\t"); print_time_seconds(get_clock_result_seconds()); printf("\n\n"); cudaMemcpy(h_resultant_matrix_1, d_resultant_matrix, matrix_size, cudaMemcpyDeviceToHost); block = dim3(BLOCK_WIDTH, BLOCK_HEIGHT); grid_width = N / (BLOCK_WIDTH * ELM_PER_THREAD); grid_width += N % BLOCK_WIDTH > 0 ? 1 : 0; grid_height = N / (BLOCK_HEIGHT); grid_height += N % BLOCK_HEIGHT > 0 ? 1 : 0; grid = dim3(grid_width, grid_height); printf("grid(%lu, %lu)\n", grid_width, grid_height); start_clock(); transpose_optimized<<<grid, block>>>(N, d_input_matrix, d_resultant_matrix); cudaDeviceSynchronize(); stop_clock(); printf("optimized time:\t"); print_time_seconds(get_clock_result_seconds()); printf("\n"); cudaMemcpy(h_resultant_matrix_2, d_resultant_matrix, matrix_size, cudaMemcpyDeviceToHost); if (memcmp(h_resultant_matrix_1, h_resultant_matrix_2, matrix_size) != 0) { printf("Results DO NOT match!\n");\ for(size_t i = 0; i < matrix_size/sizeof(float); i++) { if(h_resultant_matrix_1[i] != h_resultant_matrix_2[i]) { printf("index %lu doesn't match\n", i); printf("--1: %0.2f\t2: %0.2f--\n", h_resultant_matrix_1[i], h_resultant_matrix_2[i]); printf("Input:\n"); PrintPartialMatrix(N, h_input_matrix); printf("\nOutput 1:\n"); PrintPartialMatrix(N, h_resultant_matrix_1); printf("\nOutput 2:\n"); PrintPartialMatrix(N, h_resultant_matrix_2); break; } } } else { printf("Results match\n"); } cudaFree(d_input_matrix); cudaFree(d_resultant_matrix); delete[] h_input_matrix; delete[] h_resultant_matrix_1; delete[] h_resultant_matrix_2; return 0; } void start_clock() { clock_gettime(CLOCK_THREAD_CPUTIME_ID, &start_time); } void stop_clock() { clock_gettime(CLOCK_THREAD_CPUTIME_ID, &stop_time); } double get_clock_result_seconds() { double result = stop_time.tv_sec - start_time.tv_sec; result += (double)(stop_time.tv_nsec - start_time.tv_nsec) / 1000000000; return result; } void print_time_seconds(double seconds) { #ifdef _WIN32 printf("%0.3f seconds", seconds); #elif __linux__ printf("%0.9f seconds", seconds); #endif }
12,903
#include <stdio.h> #include <math.h> #include <stdlib.h> //Global Variable Declaration float* p; //Memory Allocation Function void memAlloc(float **data_ptr, int dim_x, int dim_y) { float *data; data = (float *) malloc(sizeof(float *) * dim_x * dim_y); *data_ptr = data; } //void cleanp() //{ //if(p) // free(p); //} /* ---------------------------------------------------- main method for Cholesky decomposition. input n size of matrix input/output a Symmetric positive def. matrix output p vector of resulting diag of a ----------------------------------------------------- */ int choldc1(int n, float* a) { int i,j,k; float sum; for (i = 0; i < n; i++) { for (j = i; j < n; j++) { sum = a[i * n + j]; for (k = i - 1; k >= 0; k--) { sum -= a[i * n + k] * a[j * n + k]; } if (i == j) { if (sum <= 0) { printf(" S is not positive definite!\n"); return 0; } p[i] = sqrt(sum); } else { a[j * n + i] = sum / p[i]; } } } return 1; } /* ----------------------------------------------------- Inverse of Cholesky decomposition. input n size of matrix input A Symmetric positive def. matrix output a inverse of lower deomposed matrix uses choldc1(int,MAT,VEC) ----------------------------------------------------- */ int choldcsl(int n, float* A, float* a) { int i,j,k; float sum; int success; for (i = 0; i < n; i++) for (j = 0; j < n; j++) a[i * n + j] = A[i * n + j]; success = choldc1(n, a); if (success == 0) return 0; for (i = 0; i < n; i++) { a[i * n + i] = 1 / p[i]; for (j = i + 1; j < n; j++) { sum = 0; for (k = i; k < j; k++) { sum -= a[j * n + k] * a[k * n + i]; } a[j * n + i] = sum / p[j]; } } return 1; } /* --------------------------------------------------- Matrix inverse using Cholesky decomposition input n size of matrix input A Symmetric positive def. matrix output a inverse of A uses choldc1(MAT, VEC) --------------------------------------------------- */ int inverse(int n, float* A, float* a) { int i,j,k,success; //temp memory allocation for p memAlloc(&p,n,n); //printf("\n memory allocation done \n"); success = choldcsl(n,A,a); if( success == 0) { //cleanp(); return 0; } for (i = 0; i < n; i++) { for (j = i + 1; j < n; j++) { a[i * n + j] = 0.0; } } for (i = 0; i < n; i++) { a[i * n + i] *= a[i * n + i]; for (k = i + 1; k < n; k++) { a[i * n + i] += a[k * n + i] * a[k * n + i]; } for (j = i + 1; j < n; j++) { for (k = j; k < n; k++) { a[i * n + j] += a[k * n + i] * a[k * n + j]; } } } for (i = 0; i < n; i++) { for (j = 0; j < i; j++) { a[i * n + j] = a[j * n + i]; } } //cleanp(); return 1; } //Inversion Complete //Other Matrix operations //Addition void add(float* C, float* A, float* B, int h, int w) { int i,j; for (i = 0; i < h; i++) { for(j = 0; j < w ;j++) { C[i * w +j] = A[i * w + j] + B[i * w + j]; } } } //subtraction void sub(float* C, float* A, float* B, int h, int w) { int i,j; for (i = 0; i < h; i++) { for(j = 0; j < w ;j++) { C[i * w + j] = A[i * w + j] - B[i * w + j]; } } } //Multiplication void mult(float* C, float* A, float* B, int ha, int wa, int hb, int wb) { int i,j,k; float sum,a = 0,b = 0; for (i = 0; i < ha; i++) { for(j = 0; j < wb ;j++) { sum = 0; for(k = 0; k < wa; k++) { a = A[i * wa + k]; b = B[k * wb + j]; sum += a * b; } C[i * wb + j] = sum; } } } //Transpose void transpose(float* B, float* A,int h, int w) { int i,j; for (i = 0; i < h; i++) { for(j = 0; j < w ;j++) { B[j * h + i] = A[i * w + j]; } } } //print the matrix void matPrint(float *A, int h, int w) { int i,j; for(i = 0;i < h;i++) { for(j = 0;j < w;j++) { printf("%f ", A[i * w + j]); } printf("\n");} } //Matrix Copy void matcopy(float *B, float *A, int h, int w) { int i; for(i = 0;i < (h*w);i++) B[i] = A[i]; } // generating L void generateL(float *L, int n) { int i,j; srand(1); for (i = 0; i < n; i++) { for(j = 0; j < n ;j++) { if(j <= i) L[i*n + j]= (rand() % 10) + 1; else L[i*n + j] = 0; } } } //Random Initialize void RandomInit(float* data, int n1, int n2) { srand(1); for (int i = 0; i < (n1*n2); ++i) data[i] = (rand() % 10) + 1; } //Ideintity Matrix Generation void Identity(float *data, int n) { for (int i = 0; i < (n*n); i=i+1) { if((i%(n+1))==0) data[i] = 1; else data[i] = 0; } } void Initialize(float *X,float *P,float *F,float *Z,float *H,float *E,float *I,float *Ht,float *Ft,float *s, int ns, int no) { RandomInit(X, ns, 1); RandomInit(Z, no, 1); RandomInit(H, no, ns); transpose(Ht,H,no,ns); //printf("\n Transpose of H successful\n"); float *P1; float *P2; memAlloc(&P1,ns,ns); memAlloc(&P2,ns,ns); generateL(P1,ns); transpose(P2,P1,ns,ns); mult(P,P1,P2,ns,ns,ns,ns); if(P1) free(P1); if(P2) free(P2); float *F1; float *F2; memAlloc(&F1,ns,ns); memAlloc(&F2,ns,ns); generateL(F1,ns); transpose(F2,F1,ns,ns); mult(F,F1,F2,ns,ns,ns,ns); if(F1) free(F1); if(F2) free(F2); transpose(Ft,F,ns,ns); //printf("\n Transpose of F successful\n"); float *E1; float *E2; memAlloc(&E1,no,no); memAlloc(&E2,no,no); generateL(E1,no); transpose(E2,E1,no,no); mult(E,E1,E2,no,no,no,no); if(E1) free(E1); if(E2) free(E2); float *s1; float *s2; memAlloc(&s1,no,no); memAlloc(&s2,no,no); generateL(s1,no); transpose(s2,s1,no,no); mult(s,s1,s2,no,no,no,no); if(s1) free(s1); if(s2) free(s2); Identity(I, ns); }
12,904
#include<cmath> #include<cstdio> #define M 2 #define N 2 #define K 2 #define n 4 __global__ void multiply(int*A,int*B,int*C) { int k,i,j,tmp; for( i=0;i<M*K;i++) { printf("%d\n",A[i]);} int I=blockIdx.x*blockDim.x+threadIdx.x; int J=blockIdx.y*blockDim.y+threadIdx.y; if( I < M || j < N) { for( k=0;k<K;k++){ C[I*N+J]+= A[I*K+k]*B[k*N+J]; } } for (i = 0; i < M*N; i++) { printf("[%d] =%d\n",i, C[i]); } } int main(){ int A[M][K]={{1,2},{3,1}}; int B[K][N]={{2,4},{5,2}}; int C[M][N]={{0,0},{0,0}}; int* d_A;int* d_B;int* d_C; cudaMalloc(&d_A,n* sizeof(int));//let memory store that m*n space for you of size ints cudaMalloc(&d_B,n* sizeof(int)); cudaMalloc(&d_C,n* sizeof(int)); //copy Aand B FROM HOST TO DEVICE cudaMemcpy(d_A, &A[0],n* sizeof(int) , cudaMemcpyHostToDevice); cudaMemcpy(d_B, &B[0],n *sizeof(int) , cudaMemcpyHostToDevice); cudaMemcpy(d_C, &C[0],n*sizeof(int) , cudaMemcpyHostToDevice); multiply<<<1,1>>>(d_A,d_B,d_C); //COPY RESULT BACK TO HOST cudaMemcpy(&C[0], d_C,n* sizeof(int), cudaMemcpyDeviceToHost); //printf("%d", C[0]); cudaFree(A);//TO FREE MEMORY cudaFree(B); cudaFree(C); }
12,905
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdlib.h> #include <stdio.h> int *IntArray(int length, int first = 0, int step = 0) { int *av = (int *)malloc(sizeof(int) * length); for (int i = 0; i < length; i++) { av[i] = first + step * i; } return av; } bool CompIntArrays(int *a, int *b, int length) { for (int i = 0; i < length; i++) { if (a[i] != b[i]) return false; } return true; } bool CompFloatArrays(float *a, float *b, int length) { for (int i = 0; i < length; i++) { if (a[i] != b[i]) return false; } return true; } void PrintFloatArray(float *aa, int width, int length) { for (int i = 0; i < length; i++) { printf("%3.3f ", aa[i]); if ((i>0) && ((i + 1) % width == 0)) printf("\n"); } printf("\n"); } void PrintIntArray(int *aa, int width, int length) { for (int i = 0; i < length; i++) { printf("%d ", aa[i]); if ((i>0) && ((i + 1) % width == 0)) printf("\n"); } printf("\n"); } void PrintUintArray(unsigned int *aa, int width, int length) { for (int i = 0; i < length; i++) { printf("%d ", aa[i]); if ((i>0) && ((i + 1) % width == 0)) printf("\n"); } printf("\n"); } float *RndFloat0to1(int arraySize) { float *temp = (float*)malloc(arraySize * sizeof(float)); for (int i = 0; i<arraySize; i++) { temp[i] = (float)rand() / (float)(RAND_MAX); } return temp; } unsigned int *RndInts(int arraySize) { unsigned int *temp = (unsigned int*)malloc(arraySize * sizeof(int)); for (int i = 0; i<arraySize; i++) { temp[i] = rand(); } return temp; } int *Rnd0or1(int arraySize, float fracOnes) { int *temp = (int*)malloc(arraySize * sizeof(float)); for (int i = 0; i<arraySize; i++) { float fv = (float)rand() / (float)(RAND_MAX); temp[i] = fv < fracOnes ? 1 : 0; } return temp; } int *Rnd_m1or1(int arraySize, float fracOnes) { int *temp = (int*)malloc(arraySize * sizeof(float)); for (int i = 0; i<arraySize; i++) { float fv = (float)rand() / (float)(RAND_MAX); temp[i] = fv < fracOnes ? 1 : -1; } return temp; } unsigned int SqrtPow2Lb(unsigned int rhs) { unsigned int fRet = 1; unsigned int nv = fRet; while (true) { nv = fRet * 2; if (nv * nv > rhs) { return fRet; } fRet = nv; } } float *LeftRightGradient(unsigned int span, float low_val, float high_val) { float delta = (high_val - low_val) / (span / 2.0f); unsigned int hs = span / 2; float *outputs = (float*)malloc(span * span * sizeof(float)); for (int i = 0; i < span; i++) { for (int j = 0; j < hs; j++) { int index = i * span + j; outputs[index] = high_val - j * delta; } for (int j = hs; j < span; j++) { int index = i * span + j; outputs[index] = low_val + (j - hs) * delta; } } return outputs; }
12,906
#include <iostream> #include <math.h> #include <chrono> #include <ctime> #include <limits.h> using namespace std; __global__ void cadd(int n, float*x, float *y){ int index = blockIdx.x * blockDim.x + threadIdx.x; int stride = blockDim.x * gridDim.x; for(int i = index; i < n; i+= stride){ y[i] = x[i] + y[i]; } } float* add(int n, float *x, float *y){ for(int i = 0; i < n; i++){ y[i] = x[i] + y[i]; } return y; } int main(void){ int N = 1<<20;//INT_MAX/10; float *x = new float[N]; float *y = new float[N]; for(int i =0; i < N; i++){ x[i] = i+1.0f; y[i] = i + 2.0f; } //CPU auto start = chrono::system_clock::now(); float *z = add(N,x,y); auto end = chrono::system_clock::now(); chrono::duration<double> timevar = end-start; std::cout << "Time to add: " << timevar.count() << std::endl; // for(int i = 0; i < sizeof(*z); i ++){ // std::cout << x[i] << " + " << y[i] << " = " << z[i] << endl; // } delete [] x; delete [] y; //GPU auto cstart = chrono::system_clock::now(); float *cx, *cy; cudaMallocManaged(&cx, N*sizeof(float)); cudaMallocManaged(&cy, N*sizeof(float)); for(int i =0; i <N;i++){ cx[i] = i + 1.0f; cy[i] = i + 2.0f; } int blocksize = 256; int numBlocks = (N + blocksize -1)/blocksize; cadd<<<numBlocks, blocksize>>>(N,x,y); cudaDeviceSynchronize(); auto cend = chrono::system_clock::now(); chrono::duration<double> ctimevar = cend-cstart; cout << "time to complete: " << ctimevar.count() << " seconds" << endl; // for(int i =0; i < sizeof(*z); i++){ // cout << cx[i] << " + " << cy[i] << " = " << cz[i] << endl; // } cudaFree(x); cudaFree(y); return 0; }
12,907
#include <stdio.h> #include <math.h> #include <stdlib.h> // drand48 #include <time.h> //#define DUMP __global__ void MoveParticles(const int nParticles,float* x, float* y, float* z, float* vx , float* vy, float* vz, const float dt) { int i = threadIdx.x + blockIdx.x * blockDim.x; // Loop over particles that experience force float Fx = 0, Fy = 0, Fz = 0; // Components of the gravity force on particle i // Loop over positions that exert force for (int j = 0; j < nParticles; j++) { // No self interaction if (i != j) { // Avoid singularity and interaction with self const float softening = 1e-20; // Newton's law of universal gravity const float dx = x[j] - x[i]; const float dy = y[j] - y[i]; const float dz = z[j] - z[i]; const float drSquared = dx*dx + dy*dy + dz*dz + softening; const float drPower32 = powf(drSquared, 3.0/2.0); // Calculate the net force Fx += dx / drPower32; Fy += dy / drPower32; Fz += dz / drPower32; } } // Accelerate particles in response to the gravitational force vx[i] += dt*Fx; vy[i] += dt*Fy; vz[i] += dt*Fz; // Move particles according to their velocities // O(N) work, so using a serial loop //#pragma acc parallel loop x[i] += vx[i]*dt; y[i] += vy[i]*dt; z[i] += vz[i]*dt; } void dump(int iter, int nParticles, float* x, float* y ,float* z) { char filename[64]; snprintf(filename, 64, "output_cuda_%d.txt", iter); FILE *f; f = fopen(filename, "w+"); int i; for (i = 0; i < nParticles; i++) { fprintf(f, "%e %e %e\n", x[i], y[i], z[i]); } fclose(f); } int main(const int argc, const char** argv) { // Problem size and other parameters const int nParticles = (argc > 1 ? atoi(argv[1]) : 16384); // Duration of test const int nSteps = (argc > 2)?atoi(argv[2]):10; // Particle propagation time step const float dt = 0.0005f; float* x = (float*)malloc(nParticles*sizeof(float)); float* y = (float*)malloc(nParticles*sizeof(float)); float* z = (float*)malloc(nParticles*sizeof(float)); float* vx = (float*)malloc(nParticles*sizeof(float)); float* vy = (float*)malloc(nParticles*sizeof(float)); float* vz = (float*)malloc(nParticles*sizeof(float)); // Initialize random number generator and particles srand48(0x2020); int i; for (i = 0; i < nParticles; i++) { x[i] = 2.0*drand48() - 1.0; y[i] = 2.0*drand48() - 1.0; z[i] = 2.0*drand48() - 1.0; vx[i] = 2.0*drand48() - 1.0; vy[i] = 2.0*drand48() - 1.0; vz[i] = 2.0*drand48() - 1.0; } // Perform benchmark printf("\nPropagating %d particles using 1 thread...\n\n", nParticles ); float rate = 0, dRate = 0; // Benchmarking data const int skipSteps = 3; // Skip first iteration (warm-up) printf("\033[1m%5s %10s %10s %8s\033[0m\n", "Step", "Time, s", "Interact/s", "GFLOP/s"); fflush(stdout); for (int step = 1; step <= nSteps; step++) { float *d_x,*d_y,*d_z,*d_vx,*d_vy,*d_vz ; size_t size = nParticles*sizeof(float); cudaMalloc(&d_x, size);cudaMalloc(&d_y, size); cudaMalloc(&d_z, size); cudaMalloc(&d_vx, size);cudaMalloc(&d_vy, size); cudaMalloc(&d_vz, size); cudaMemcpy(d_x, x, size, cudaMemcpyHostToDevice); cudaMemcpy(d_y, y, size, cudaMemcpyHostToDevice); cudaMemcpy(d_z, z, size, cudaMemcpyHostToDevice); cudaMemcpy(d_vx, vx, size, cudaMemcpyHostToDevice); cudaMemcpy(d_vy, vy, size, cudaMemcpyHostToDevice); cudaMemcpy(d_vz , vz, size, cudaMemcpyHostToDevice); int threadPerBlocs = 256; /* Ceil */ int blocksPerGrid = (nParticles + threadPerBlocs - 1) / threadPerBlocs; clock_t tStart = clock(); // Start timing MoveParticles<<< blocksPerGrid, threadPerBlocs >>>(nParticles,d_x,d_y,d_z,d_vx,d_vy,d_vz, dt); clock_t tEnd = clock(); // End timing float time_spent = (tStart - tEnd)/ CLOCKS_PER_SEC; cudaMemcpy(x, d_x, size, cudaMemcpyDeviceToHost); cudaMemcpy(y, d_y, size, cudaMemcpyDeviceToHost); cudaMemcpy(z, d_z, size, cudaMemcpyDeviceToHost); cudaMemcpy(vx, d_vx, size, cudaMemcpyDeviceToHost); cudaMemcpy(vy, d_vy, size, cudaMemcpyDeviceToHost); cudaMemcpy(vz , d_vz, size, cudaMemcpyDeviceToHost); cudaFree(d_x); cudaFree(d_y);cudaFree(d_z); cudaFree(d_vx); cudaFree(d_vy);cudaFree(d_vz); const float HztoInts = ((float)nParticles)*((float)(nParticles-1)) ; const float HztoGFLOPs = 20.0*1e-9*((float)(nParticles))*((float)(nParticles-1)); if (step > skipSteps) { // Collect statistics rate += HztoGFLOPs/(time_spent); dRate += HztoGFLOPs*HztoGFLOPs/((time_spent)*(time_spent)); } printf("%5d %10.3e %10.3e %8.1f %s\n", step, (time_spent), HztoInts/(time_spent), HztoGFLOPs/(time_spent), (step<=skipSteps?"*":"")); fflush(stdout); #ifdef DUMP dump(step, nParticles, x,y,z); #endif } rate/=(float)(nSteps-skipSteps); dRate=sqrt(dRate/(float)(nSteps-skipSteps)-rate*rate); printf("-----------------------------------------------------\n"); printf("\033[1m%s %4s \033[42m%10.1f +- %.1f GFLOP/s\033[0m\n", "Average performance:", "", rate, dRate); printf("-----------------------------------------------------\n"); printf("* - warm-up, not included in average\n\n"); free(x);free(y);free(z); free(vx);free(vz);free(vz); return 0; }
12,908
#ifdef FAST_NOISE extern "C" __global__ void kFastNoise1_kernel( int numAtoms, int paddedNumAtoms, int numModes, float kT, float4 *noiseVal, float4 *velm, float4 *modes, float *modeWeights, const float4 *__restrict__ random, unsigned int randomIndex, float maxEigenvalue, float stepSize ) { extern __shared__ float dotBuffer[]; const float val = stepSize / 0.002; const float noisescale = sqrt( 2 * kT * 1.0f / maxEigenvalue ); for( int mode = blockIdx.x; mode < numModes; mode += gridDim.x ) { float dot = 0.0f; unsigned int seed = 100; for( int atom = threadIdx.x; atom < numAtoms; atom += blockDim.x ) { const float4 n = random[randomIndex + blockIdx.x * blockDim.x + threadIdx.x]; const float4 randomNoise = make_float4( n.x * noisescale, n.y * noisescale, n.z * noisescale, n.w * noisescale ); noiseVal[atom] = randomNoise; float4 m = modes[mode * numAtoms + atom]; dot += randomNoise.x * m.x + randomNoise.y * m.y + randomNoise.z * m.z; } dotBuffer[threadIdx.x] = dot; __syncthreads(); if( threadIdx.x == 0 ) { float sum = 0; for( int i = 0; i < blockDim.x; i++ ) { sum += dotBuffer[i]; } modeWeights[mode] = sum; } } } extern "C" __global__ void kFastNoise2_kernel( int numAtoms, int numModes, float4 *posq, float4 *noiseVal, float4 *velm, float4 *modes, float *modeWeights ) { /* Load the weights into shared memory.*/ extern __shared__ float weightBuffer[]; for( int mode = threadIdx.x; mode < numModes; mode += blockDim.x ) { weightBuffer[mode] = modeWeights[mode]; } __syncthreads(); /* Compute the projected forces and update the atom positions.*/ for( int atom = threadIdx.x + blockIdx.x * blockDim.x; atom < numAtoms; atom += blockDim.x * gridDim.x ) { const float invMass = velm[atom].w, sqrtInvMass = sqrt( invMass ); float3 r = make_float3( 0.0f, 0.0f, 0.0f ); for( int mode = 0; mode < numModes; mode++ ) { float4 m = modes[mode * numAtoms + atom]; float weight = weightBuffer[mode]; r.x += m.x * weight; r.y += m.y * weight; r.z += m.z * weight; } noiseVal[atom] = make_float4( noiseVal[atom].x - r.x, noiseVal[atom].y - r.y, noiseVal[atom].z - r.z, 0.0f ); noiseVal[atom].x *= sqrtInvMass; noiseVal[atom].y *= sqrtInvMass; noiseVal[atom].z *= sqrtInvMass; float4 pos = posq[atom]; pos.x += noiseVal[atom].x; pos.y += noiseVal[atom].y; pos.z += noiseVal[atom].z; posq[atom] = pos; } } #endif
12,909
#include <iostream> #include <stdio.h> #include <time.h> #define LENGTH 1000 #define T_WIDTH 16 using namespace std; __global__ void matmul_shared(float *a, float *b, float *c, int width) { int tx = threadIdx.x; int ty = threadIdx.y; int bx = blockIdx.x; int by = blockIdx.y; __shared__ float s_a[T_WIDTH][T_WIDTH]; __shared__ float s_b[T_WIDTH][T_WIDTH]; int row = ty + by*blockDim.y; int col = tx + bx*blockDim.x; float result = 0; for(int k = 0; k < width/T_WIDTH; ++k) { s_a[ty][tx] = a[row*width + (k*T_WIDTH + tx)]; s_b[ty][tx] = b[(k*T_WIDTH + ty)*width + col]; __syncthreads(); for(int p = 0; p < T_WIDTH; ++p){ result += s_a[ty][p] * s_b[p][tx]; } __syncthreads(); } c[row*width + col] = result; } __global__ void matmul(float *a, float *b, float *c, int width) { float result = 0; int row = blockIdx.y * blockDim.y + threadIdx.y; int col = blockIdx.x * blockDim.x + threadIdx.x; for(int k=0; k<width; k++){ result += a[row*width + k]*b[k*width + col]; } c[row*width + col] = result; } int main(){ static float a_vec[LENGTH][LENGTH]; static float b_vec[LENGTH][LENGTH]; /* static float h_c[LENGTH][LENGTH]; */ float *h_a, *h_b, *h_c; float *d_a, *d_b, *d_c; h_a = (float*)malloc(LENGTH*LENGTH*sizeof(float)); h_b = (float*)malloc(LENGTH*LENGTH*sizeof(float)); h_c = (float*)malloc(LENGTH*LENGTH*sizeof(float)); for(int i=0 ; i< LENGTH; i++) { for(int j=0;j<LENGTH;j++){ a_vec[i][j] = 1; b_vec[i][j] = 1; } } int k = 0; for(int i=0;i<LENGTH;i++) { for(int j=0;j<LENGTH;j++) { h_a[k] = a_vec[i][j]; h_b[k] = b_vec[i][j]; k+=1; } } cudaMalloc((void**)&d_a, LENGTH*LENGTH*sizeof(float)); cudaMalloc((void**)&d_b, LENGTH*LENGTH*sizeof(float)); cudaMalloc((void**)&d_c, LENGTH*LENGTH*sizeof(float)); cudaMemcpy(d_a, h_a, LENGTH*LENGTH*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(d_b, h_b, LENGTH*LENGTH*sizeof(float), cudaMemcpyHostToDevice); cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start); dim3 dimBlock1(16, 16, 1); dim3 dimGrid1(LENGTH/16, LENGTH/16, 1); matmul<<<dimGrid1, dimBlock1>>>(d_a, d_b, d_c, LENGTH); /* matmul<<<LENGTH*LENGTH/256, 256>>>(d_a, d_b, d_c, LENGTH); */ cudaDeviceSynchronize(); cudaEventRecord(stop); cudaEventSynchronize(stop); // free the memory cudaFree(d_a); cudaFree(d_b); cudaFree(d_c); float milliseconds = 0; cudaEventElapsedTime(&milliseconds, start, stop); std::cout << "Time taken (normal) in ms: " << milliseconds << std::endl; cudaMalloc((void**)&d_a, LENGTH*LENGTH*sizeof(float)); cudaMalloc((void**)&d_b, LENGTH*LENGTH*sizeof(float)); cudaMalloc((void**)&d_c, LENGTH*LENGTH*sizeof(float)); cudaMemcpy(d_a, h_a, LENGTH*LENGTH*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(d_b, h_b, LENGTH*LENGTH*sizeof(float), cudaMemcpyHostToDevice); cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start); dim3 threads(16, 16, 1); dim3 blocks(LENGTH/16, LENGTH/16, 1); matmul_shared<<<blocks, threads>>>(d_a, d_b, d_c, LENGTH); /* matmul_shared<<<LENGTH*LENGTH/100, 100>>>(d_a, d_b, d_c, LENGTH); */ cudaDeviceSynchronize(); cudaEventRecord(stop); cudaEventSynchronize(stop); /* cudaMemcpy(h_c, d_c, LENGTH*LENGTH*sizeof(float), cudaMemcpyDeviceToHost); */ /* for(int i=11000;i<11100;i++){ */ /* std::cout<<h_c[i]<<std::endl; */ /* } */ // free the memory /* cudaFree(d_a); */ /* cudaFree(d_b); */ /* cudaFree(d_c); */ /* free(milliseconds); */ float millisecond = 0; cudaEventElapsedTime(&millisecond, start, stop); std::cout << "Time taken (shared memory) in ms: " <<fixed<<millisecond << std::endl; }
12,910
#include "includes.h" __global__ void d_bucketsort(unsigned int * d_in, unsigned int * d_indices, unsigned int * d_sublist, unsigned int * r_outputlist, unsigned int * d_bucketoffsets, int itemCount) { int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx < itemCount) { int newpos = d_bucketoffsets[d_sublist[idx]] + d_indices[idx]; r_outputlist[newpos] = d_in[idx]; } }
12,911
#include "includes.h" __global__ void complexmult_conj_kernal(float *afft, const float *bfft, int totaltc) { const uint ridx = 2*(threadIdx.x + (blockIdx.x + blockIdx.y*gridDim.x)*MAX_THREADS); //maybe use float2 to improve coalessing.... if (ridx < totaltc){ const uint iidx = ridx + 1; float afftr = afft[ridx]; float affti = afft[iidx]; float bfftr = bfft[ridx]; float bffti = bfft[iidx]; afft[ridx] = afftr*bfftr + affti*bffti; //real portion afft[iidx] = affti*bfftr - afftr*bffti; //imaginary portion } }
12,912
#include <stdio.h> // these are just for timing measurments #include <time.h> // error checking macro #define cudaCheckErrors(msg) \ do { \ cudaError_t __err = cudaGetLastError(); \ if (__err != cudaSuccess) { \ fprintf(stderr, "Fatal error: %s (%s at %s:%d)\n", \ msg, cudaGetErrorString(__err), \ __FILE__, __LINE__); \ fprintf(stderr, "*** FAILED - ABORTING\n"); \ exit(1); \ } \ } while (0) const int block_size = 4; // CUDA maximum is 1024 *total* threads in block //const float A_val = 3.0f; //const float B_val = 2.0f; // matrix multiply (naive) kernel: C = A * B __global__ void mmul(const float *A, const float *B, float *C, int r1,int c1, int r2, int c2) { int idx = threadIdx.x+blockDim.x*blockIdx.x; // create thread x index int idy = threadIdx.y+blockDim.y*blockIdx.y; // create thread y index if ((idx < c2) && (idy < c1)){ float temp = 0; for (int i = 0; i < c1; i++) temp += A[idy*c1+i] * B[i*c2+idx]; // dot product of row and column C[idy*c2+idx] = temp; } } int main(){ float *h_A, *h_B, *h_C, *d_A, *d_B, *d_C; // these are just for timing clock_t t0, t1, t2; double t1sum=0.0; double t2sum=0.0; // start timing t0 = clock(); //getting matrix from user int r1,c1,r2,c2; printf("Enter the row of 1st Matrix: "); scanf("%d",&r1); printf("Enter the column of 1st Matrix: "); scanf("%d",&c1); printf("Enter the row of 2st Matrix: "); scanf("%d",&r2); printf("Enter the column of 2st Matrix: "); scanf("%d",&c2); if (c1!=r2){ printf("Invalid Matrix"); } h_A = new float[r1*c1]; h_B = new float[r2*c2]; h_C = new float[r1*c2]; FILE* matrixA; matrixA = fopen("A.txt","r"); if(matrixA==NULL){ printf("Matrix A did not open \n"); return 0 ; } int i =0; while(fscanf(matrixA,"%f", &h_A[i] )!= EOF){ i++; } FILE* matrixB; matrixB = fopen("B.txt","r"); if(matrixA==NULL){ printf("Matrix B did not open \n"); return 0; } i =0; while(fscanf(matrixB,"%f",&h_B[i] )!= EOF){ i++; } //Printing values on screen for degugg /* for (int i = 0; i < r1*c1; i++){ printf("%.1f ", h_A[i]); } printf("\n"); for (int i = 0; i < r2*c2; i++){ printf("%.1f ", h_B[i]); } */ //If values was assigned in the program and not through input files /* for (int i = 0; i < r1*c1; i++){ h_A[i] = A_val; } for (int i = 0; i < r2*c2; i++){ h_B[i] = B_val; } for (int i = 0; i < r1*c2; i++){ h_C[i] = 0;} */ // Initialization timing t1 = clock(); t1sum = ((double)(t1-t0))/CLOCKS_PER_SEC; printf("Init took %f seconds. Begin compute\n", t1sum); // Allocate device memory and copy input data over to GPU cudaMalloc(&d_A, r1*c1*sizeof(float)); cudaMalloc(&d_B, r2*c2*sizeof(float)); cudaMalloc(&d_C, r1*c2*sizeof(float)); cudaCheckErrors("cudaMalloc failure"); cudaMemcpy(d_A, h_A, r1*c1*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(d_B, h_B, r2*c2*sizeof(float), cudaMemcpyHostToDevice); cudaCheckErrors("cudaMemcpy H2D failure"); // Cuda processing sequence step 1 is complete // Launch kernel dim3 block(block_size, block_size); // dim3 variable holds 3 dimensions dim3 grid(((c1+r2)/2+block.x-1)/block.x, ((c1+r2)/2+block.y-1)/block.y); mmul<<<grid, block>>>(d_A, d_B, d_C, r1,c1,r2,c2 ); cudaCheckErrors("kernel launch failure"); // Cuda processing sequence step 2 is complete // Copy results back to host cudaMemcpy(h_C, d_C, r1*c2*sizeof(float), cudaMemcpyDeviceToHost); for (int i = 0; i < r1*c1; i++){ printf("%.1f ", h_A[i]); } printf("\n"); for (int i = 0; i < r2*c2; i++){ printf("%.1f ", h_B[i]); } printf("\n"); // Verify results for (int i = 0; i < r1*c2; i++){ printf("%.1f ", h_C[i]); } FILE* matrixC; matrixC = fopen("C.txt","w"); if(matrixC==NULL){ printf("Matrix A did not open \n"); return 0 ; } for(int i =0; i < r1*c2; i++){ fprintf(matrixC, "%.1f ",h_C[i]); } fclose(matrixA); fclose(matrixB); fclose(matrixC); // GPU timing t2 = clock(); t2sum = ((double)(t2-t1))/CLOCKS_PER_SEC; printf ("Done. Compute took %f seconds\n", t2sum); // Cuda processing sequence step 3 is complete // Verify results cudaCheckErrors("kernel execution failure or cudaMemcpy H2D failure"); return 0; }
12,913
#include "includes.h" __global__ void symmetrize2D( float *h, int natoms ) { const int elementNum = blockIdx.x * blockDim.x + threadIdx.x; const int dof = 3 * natoms; if( elementNum >= dof * dof ) { return; } int r = elementNum / dof; int c = elementNum % dof; if( r > c ) { return; } else { const float avg = 0.5 * ( h[r * dof + c] + h[c * dof + r] ); h[r * dof + c] = avg; h[c * dof + r] = avg; } }
12,914
#include "includes.h" // CUDA runtime // includes extern "C" { } #define MEMSIZE 30 /* Function computing the final string to print */ __global__ void kern_compute_string(char *res, char *a, char *b, char *c, int length) { int i; i = blockIdx.x * blockDim.x + threadIdx.x; if (i < length) { res[i] = a[i] + b[i] + c[i]; } }
12,915
#include "includes.h" __global__ void gamma_norm_kernel(float* img, int image_height, int image_width, int image_step) { // The thread block has size (3,n). The first dimension of the thread block // corresponds to color channels. int channel = threadIdx.x; // The columns of the image are mapped to the first dimension of the block // grid, but to the second dimension of the thread block, as the first // already corresponds to color channels. int pixel_x = blockIdx.x * blockDim.y + threadIdx.y; // If current position is outside the image, stop here if(pixel_x >= image_width) { return; } // The columns of the image are mapped to the second dimension of the block // grid, but to the third dimension of the thread block. int pixel_y = blockIdx.y * blockDim.z + threadIdx.z; // If current position is outside the image, stop here if(pixel_y >= image_height) { return; } // Each row has image_step pixels and each pixel has three channels int in_pixel_idx = pixel_y * image_step + pixel_x * 3 + channel; // Finally perform the normalization img[in_pixel_idx] = sqrt(img[in_pixel_idx] / 256.0f); }
12,916
#include "includes.h" __device__ double2 make_complex(double in, int evolution_type){ double2 result; switch(evolution_type){ // No change case 0: result.x = in; result.y = 0; break; // Im. Time evolution case 1: result.x = exp(-in); result.y = 0; break; // Real Time evolution case 2: result.x = cos(-in); result.y = sin(-in); break; } return result; } __global__ void make_complex_kernel(double *in, int *evolution_type, double2 *out){ //int id = threadIdx.x + blockIdx.x*blockDim.x; //out[id] = make_complex(in[id], evolution_type[id]); for (int i = 0; i < 3; ++i){ out[i] = make_complex(in[i], evolution_type[i]); } }
12,917
#include "includes.h" __global__ void _reluforw(int n, float *y) { int i = threadIdx.x + blockIdx.x * blockDim.x; while (i < n) { if (y[i] < 0) y[i] = 0; i += blockDim.x * gridDim.x; } }
12,918
#include<stdio.h> #include<iostream> #include<cuda.h> __global__ void add(int* a){ // Addition Kernel int tid = blockIdx.x*blockDim.x+threadIdx.x; a[tid]+=tid; } int main(int argc, char* argv[]){ if(argc!=3){ std::cout<<"Usage: "<<argv[0]<<" Numblocks BlockDim\n"; return 0; } int nBlocks= atoi(argv[1]); int bDim = atoi(argv[2]); if(bDim>1024){ std::cout<<"BlockDim should be less than or equal to 1024\n"; return 0; } int* da; cudaMallocManaged (&da, nBlocks*bDim*sizeof(int));// Allocate CPU/GPU Memory for(int i=0;i<nBlocks*bDim;i++) // Initalize CPU array da[i]=i; add<<<nBlocks,bDim>>>(da); // Call addition kernel cudaDeviceSynchronize(); for(int i=0;i<nBlocks*bDim;i++) // Print final results std::cout<<da[i]<<"\n"; cudaFree(da); da=NULL; }
12,919
/** * cuda_delta_knn.cu * block loading delta calculation. * system('nvcc -ptx -m 64 -arch sm_35 jrc3_cuda_rho.cu') * J. James Jun, Flatiron Institute, 2018 July 5 */ #include <cuda_runtime.h> // #include "cublas_v2.h" #include <math.h> #define ABS(my_val) ((my_val) < 0) ? (-1*(my_val)) : (my_val) #define MIN(A,B) ((A)<(B)) ? (A) : (B) #define MAX(A,B) ((A)>(B)) ? (A) : (B) #define NTHREADS 128 #define NC (3*20) // number of Channels #define CHUNK 16 //previously defined as CHUNK #define SINGLE_INF (3.402E+38) __global__ void cuda_delta_knn(float *D, unsigned int *N, const float *B, const float *A, const float *R_B, const float *R_A, const int *vnConst){ int nB = vnConst[0]; int nA = vnConst[1]; int nC = vnConst[2]; int tx = threadIdx.x; int iA = (blockIdx.x + blockIdx.y * gridDim.x) * CHUNK + tx; int nThreads = blockDim.x; // must be less than NTHREADS __shared__ float sA[NC][CHUNK], sR_A[CHUNK]; // march through A __shared__ float sD[NTHREADS][CHUNK]; // count then divide later __shared__ unsigned int sN[NTHREADS][CHUNK]; // initialize if (tx < CHUNK){ if (iA < nA){ int iA_ = tx; for (int iC=0; iC<nC; ++iC) sA[iC][tx] = A[iC + iA*nC]; // copy A->sA sR_A[iA_] = R_A[iA]; // copy R_A->sR_A for (int iB_=0; iB_<nThreads; ++iB_) sD[iB_][iA_] = SINGLE_INF; // sD = inf } } __syncthreads(); // Search min distance having a greater rho for (int iB=tx; iB<nB; iB+=nThreads){ // compute distance float dist_[CHUNK]; for (int iA_=0; iA_<CHUNK; ++iA_) dist_[iA_] = 0.0f; for (int iC=0; iC<nC; ++iC){ float b_ = B[iC + iB * nC]; for (int iA_=0; iA_<CHUNK; ++iA_){ float d_ = b_ - sA[iC][iA_]; dist_[iA_] += (d_ * d_); } } // Compare the index and distance float rb_ = R_B[iB]; int iB_ = tx; for (int iA_=0; iA_<CHUNK; ++iA_){ if (rb_ > sR_A[iA_] && dist_[iA_] < sD[iB_][iA_]){ sD[iB_][iA_] = dist_[iA_]; sN[iB_][iA_] = iB; } } } // while __syncthreads(); // final count if (tx < CHUNK){ if (iA < nA){ int iA_ = tx; float dmin_ = SINGLE_INF; unsigned int imin_ = iA; // point to self initially for (int iB_=0; iB_<nThreads; ++iB_){ if (sD[iB_][iA_] < dmin_){ dmin_ = sD[iB_][iA_]; imin_ = sN[iB_][iA_]; } } D[iA] = sqrtf(ABS(dmin_)); N[iA] = imin_ + 1; //Matlab index output } } } // func
12,920
#include "includes.h" using namespace std; // https://stackoverflow.com/questions/26853363/dot-product-for-dummies-with-cuda-c __global__ void dotCuda(float* tmp, float* t1, float* t2, int size) { //unsigned int tid = threadIdx.x; unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; tmp[i] = t1[i] * t2[i]; __syncthreads(); int mididx = size / 2; while (i < mididx) { tmp[i] += tmp[i + mididx]; mididx /= 2; __syncthreads(); } //atomicAdd(tmp, p); }
12,921
#include <stdio.h> #include <stdlib.h> #define N 10000 void handleError(cudaError_t error) { if(error != cudaSuccess) { printf("Error: %s\n", cudaGetErrorString(error)); exit(EXIT_FAILURE); } } __global__ void add(int *a, int *b, int *c) { int tid = blockIdx.x; if(tid < N) { c[tid] = a[tid] + b[tid]; } } int main(int argc, char *argv[]) { int a[N], b[N], c[N]; int *dev_a, *dev_b, *dev_c; handleError(cudaMalloc((void **) &dev_a, sizeof(int) * N)); handleError(cudaMalloc((void **) &dev_b, sizeof(int) * N)); handleError(cudaMalloc((void **) &dev_c, sizeof(int) * N)); for(int i = 0; i < N; i++) { a[i] = -i; b[i] = i * i; } handleError(cudaMemcpy(dev_a, a, sizeof(int) * N, cudaMemcpyHostToDevice)); handleError(cudaMemcpy(dev_b, b, sizeof(int) * N, cudaMemcpyHostToDevice)); add<<<N,1>>>(dev_a, dev_b, dev_c); handleError(cudaMemcpy(c, dev_c, sizeof(int) * N, cudaMemcpyDeviceToHost)); for(int i = 0; i < N; i++) { printf("%d + %d = %d\n", a[i], b[i], c[i]); } cudaFree(dev_a); cudaFree(dev_b); cudaFree(dev_c); return 0; }
12,922
//--------------------------------------------// //Histogram for bag of words CPU //Authors: Hong Xu and Rajath Javali //--------------------------------------------// #include <stdio.h> #include <iterator> #include <sys/types.h> #include <dirent.h> #include <fstream> #include <sstream> #include <string> #include <vector> #include <iterator> #include <iostream> #include <algorithm> #include <typeinfo> #include <thrust/device_vector.h> #include <thrust/host_vector.h> using namespace std; #define BLOCK_SIZE 1024 #define BLOCK_SIZE2 32 #define MAX_K 20000 typedef std::vector<std::string> stringvec; //List of structs struct Sift{ float* sift; int numSift; int totalNum; int elementSize; }; struct Histogram{ float* vals; int size; string filename; }; //List of functions int countlines(char *filename); void readFromFile(int size,float *v, char *file); Histogram createHist(Sift cent, Sift im); Sift readSift(char *filename); void printSift(Sift sift, char option); void freeSift(Sift sift); void printHisto(Histogram h, char option); void freeHisto(Histogram h); void printFiles(stringvec v); void read_directory(const std::string& name, stringvec& v); __global__ void createHistCuda (float* siftCentroids, float* siftImage, int linesCent, int linesIm,float*temp); vector<Sift> readMultiSifts(stringvec v); vector<Histogram> createHistogramCudaAux(Sift& siftCentroids, vector<Sift>& imSifts, stringvec& v); vector<vector<int> > matchCudaAux(vector<Histogram>& query, vector<Histogram>& database); __global__ void matchHistCuda(float*qSet, float*dbSet, size_t qSize, size_t dbSize, size_t hSize, float*out); void histogram2array(float* array, vector<Histogram>& hists); template <class T> void arrayto2Dvec(T* array, vector<vector<T> >& vec, size_t x, size_t y); template <class T> void print2DVec(vector<vector<T> >& vec); template<class T> void print_array(T& array, int m, int n); __global__ void rowMin(float* input, int* output, size_t rowS, size_t rowNum); string extName(string s); vector<vector<string> > comFinal(vector<Histogram>& query, vector<Histogram>& database, vector<vector<int> >& values, int num); void analize(vector<vector<string> >& results); string extRight(string s); void freeHistograms(vector<Histogram> histos); //---------------------------------------------// //Main //---------------------------------------------// int main(int argc, char *argv[]) { //Header printf("\t\t###########################################\n\t\t####### Histogram Program #############\n\t\t###########################################\n\n"); //Getting query file names stringvec vquery; string name = "../input/querySift/"; read_directory(name, vquery); printFiles(vquery); //Getting database file names stringvec vdatabase; name = "../input/databaseSift/"; read_directory(name, vdatabase); printFiles(vdatabase); //Readsing Centroid sifts Sift siftCentroids = readSift((char *)"../input/querySift/1.png.txt"); printSift(siftCentroids, 'i'); //Reads query sifts from file std::vector<Sift> imSiftsQuery = readMultiSifts(vquery); //Reads database sifts from file std::vector<Sift> imSiftsDatabase = readMultiSifts(vdatabase); //Creates query histograms CUDA vector<Histogram> queryHistograms = createHistogramCudaAux(siftCentroids, imSiftsQuery,vquery); //Creates database histograms CUDA vector<Histogram> databaseHistograms = createHistogramCudaAux(siftCentroids, imSiftsDatabase,vdatabase); //Freeing sift memory freeSift(siftCentroids); for(int i = 0; i < imSiftsQuery.size(); i++){ freeSift(imSiftsQuery[i]); } for(int i = 0; i < imSiftsDatabase.size(); i++){ freeSift(imSiftsDatabase[i]); } //Matching, and sorting the histograms vector<vector<int> > values = matchCudaAux(queryHistograms, databaseHistograms); cout << "Matched" << endl; //Computing results vector<vector<string> > results = comFinal(queryHistograms, databaseHistograms, values, 5); cout << "Results Computed" << endl; //Final analysis and results analize(results); //Free histograms freeHistograms(queryHistograms); freeHistograms(databaseHistograms); printf("Done!\n\n"); return 0; } //---------------------------------------------// //Free vector of histograms //---------------------------------------------// void freeHistograms(vector<Histogram> histos){ for(int i = 0; i < histos.size(); i++){ freeHisto(histos[i]); } } //---------------------------------------------// //Analyses and displays final results //---------------------------------------------// void analize(vector<vector<string> >& results){ printf("\t\t###########################################\n\t\t######### Result Analysis #############\n\t\t###########################################\n\n"); cout << "\n-------------------\nMatches\n-------------------\n"<< endl; int corrects = 0; for(int i = 0; i < results.size(); i++){ string dbIm = results[i][0]; string rightAns = extRight(dbIm); int right = 0; cout << "Database image: " << dbIm << "\nRight Answer: "<< rightAns << endl; for(int j = 1; j < results[i].size(); j++){ if(rightAns == results[i][j]){ cout << "\t==" << results[i][j] << endl; right = 1; } else{ cout << "\t" << results[i][j] << endl; } } if(right == 1){ corrects++; } cout << endl; } float accuracy = ((float)corrects)/((float)results.size()); cout << "-------------------------------\n----Final Accuracy: " << accuracy << "-------\n-------------------------------\n" << endl; } //---------------------------------------------// //Extracts right answer //---------------------------------------------// string extRight(string s){ int lastSlash = s.length() - 1; while(lastSlash >= 0 && s[lastSlash] != '_'){ lastSlash--; } return s.substr(0, lastSlash); } //---------------------------------------------// //Computes final results //---------------------------------------------// vector<vector<string> > comFinal(vector<Histogram>& query, vector<Histogram>& database, vector<vector<int> >& values, int num){ vector<vector<string> > ans; for(int i = 0; i < values.size(); i++){ vector<string> temp; temp.push_back(extName(database[i].filename)); //cout << extName(database[i].filename) << " "; for(int j = 0; j < num; j++){ temp.push_back(extName(query[values[i][j]].filename)); //cout << extName(query[values[i][j]].filename) << " "; } ans.push_back(temp); //cout << endl; } return ans; } //---------------------------------------------// //Name extraction //---------------------------------------------// string extName(string s){ int lastSlash = s.length() - 1; while(lastSlash >= 0 && s[lastSlash] != '/'){ lastSlash--; } int count = 0; int dotAfterSlash = lastSlash; while(dotAfterSlash < s.length() && s[dotAfterSlash] != '.'){ dotAfterSlash++; count++; } return s.substr(lastSlash+1, count - 1); } //---------------------------------------------// //Cuda Creates Histogram //---------------------------------------------// __global__ void createHistCuda (float* siftCentroids, float* siftImage, int linesCent, int linesIm, float* temp) { __shared__ float cosines[BLOCK_SIZE][2]; size_t idx = blockIdx.x*blockDim.x + threadIdx.x; size_t idy = blockIdx.y; size_t tid = threadIdx.x; if(idx < linesCent){ int centin = idx * 128; int imin = idy * 128; //Cosine similarity code ------------ float sumab = 0; float suma2 = 0; float sumb2 = 0; for(int k = 0; k < 128; k++){ sumab += siftCentroids[centin + k] * siftImage[imin + k]; suma2 += siftImage[imin + k] * siftImage[imin + k]; sumb2 += siftCentroids[centin + k] * siftCentroids[centin + k]; } float cossim = sumab/(sqrtf(suma2)/sqrtf(sumb2)); //debug[idy*linesCent + idx] = cossim; cosines[threadIdx.x][0] = cossim; cosines[threadIdx.x][1] = idx; __syncthreads(); for (unsigned int s=blockDim.x/2; s>0; s>>=1) { if (tid < s){ size_t tid2 = tid + s; if(cosines[tid2][0] > cosines[tid][0]){ cosines[tid][0] = cosines[tid2][0]; cosines[tid][1] = cosines[tid2][1]; } } __syncthreads(); } if (tid == 0){ temp[(blockIdx.y*gridDim.x + blockIdx.x)*2] = cosines[0][0]; temp[(blockIdx.y*gridDim.x + blockIdx.x)*2+1] = cosines[0][1]; } } } //---------------------------------------------// //Histogram Matching CUDA //---------------------------------------------// __global__ void matchHistCuda(float*qSet, float*dbSet, size_t qSize, size_t dbSize, size_t hSize, float*out){ size_t idx = blockIdx.x*blockDim.x + threadIdx.x; size_t idy = blockIdx.y*blockDim.y + threadIdx.y; if(idx < qSize && idy < dbSize){ size_t qi = idx*hSize; size_t dbi = idy*hSize; //Cosine similarity code ------------ float sumab = 0; float suma2 = 0; float sumb2 = 0; for(int k = 0; k < hSize; k++){ sumab += qSet[qi+k] * dbSet[dbi+k]; suma2 += qSet[qi+k] * qSet[qi+k]; sumb2 += dbSet[dbi+k] * dbSet[dbi+k]; } float cossim = sumab/(sqrtf(suma2)/sqrtf(sumb2)); out[idy*qSize + idx] = cossim; } } //---------------------------------------------// //Sorting each row //---------------------------------------------// __global__ void rowMin(float* input, int* output, size_t rowS, size_t rowNum){ size_t id = blockIdx.x*blockDim.x + threadIdx.x; if(id < rowNum){ float temp[MAX_K/2][2]; size_t inId = id * rowS; for(int i = 0; i< rowS;i++){ temp[i][0] = input[inId + i]; temp[i][1] = (float)i; } for(int i = 0; i< rowS; i++){ float best = temp[i][0]; int bestInd = i; for(int j = i; j < rowS; j++){ if(temp[j][0] > best){ best = temp[j][0]; bestInd = j; } } float iVal = temp[i][0]; float iInd = temp[i][1]; temp[i][0] = temp[bestInd][0]; temp[i][1] = temp[bestInd][1]; temp[bestInd][0] = iVal; temp[bestInd][1] = iInd; } for(int i = 0; i< rowS; i++){ output[inId+i] = (int)temp[i][1]; } } } //---------------------------------------------// //Histogram Matching Auxiliary //Input: Query histograms and database histograms //---------------------------------------------// vector<vector<int> > matchCudaAux(vector<Histogram>& query, vector<Histogram>& database){ //vector<vector<float> > chart; //Get constant sizes const size_t hSize = query[0].size; const size_t querySize = query.size(); const size_t databaseSize = database.size(); //Get the histogram info into 1D arrays float* h_qSet = (float*) malloc(querySize*hSize*sizeof(float)); histogram2array(h_qSet, query); float* h_dbSet = (float*) malloc(databaseSize*hSize*sizeof(float)); histogram2array(h_dbSet, database); //Malloc GPU memory for histograms float* d_qSet; cudaMalloc((void **) &d_qSet, querySize*hSize*sizeof(float)); cudaMemcpy(d_qSet, h_qSet, querySize*hSize*sizeof(float),cudaMemcpyHostToDevice); float* d_dbSet; cudaMalloc((void **) &d_dbSet, databaseSize*hSize*sizeof(float)); cudaMemcpy(d_dbSet, h_dbSet, databaseSize*hSize*sizeof(float),cudaMemcpyHostToDevice); //Malloc GPU memory for results array float * d_results; cudaMalloc((void **) &d_results, querySize*databaseSize*sizeof(float)); dim3 grid((querySize + BLOCK_SIZE2 - 1)/BLOCK_SIZE2, (databaseSize+ BLOCK_SIZE2 - 1)/BLOCK_SIZE2); dim3 block(BLOCK_SIZE2,BLOCK_SIZE2); matchHistCuda<<<grid,block>>>(d_qSet, d_dbSet, querySize, databaseSize, hSize, d_results); //Copying results back float* out_results = (float*) malloc(querySize*databaseSize*sizeof(float)); cudaMemcpy(out_results, d_results,querySize*databaseSize*sizeof(float),cudaMemcpyDeviceToHost); //arrayto2Dvec<float>(out_results, chart, querySize, databaseSize); //print_array<float*>(out_results, querySize, databaseSize); //print2DVec<float>(chart); //Free matchHistCuda memory free(h_qSet); free(h_dbSet); cudaFree(d_qSet); cudaFree(d_dbSet); vector<vector<int> > res; //Preparing input for row min, which is the same as the results of the last kernel float * d_input; cudaMalloc((void **) &d_input, querySize*databaseSize*sizeof(float)); cudaMemcpy(d_input, out_results, databaseSize*querySize*sizeof(float),cudaMemcpyHostToDevice); //Preparing output for row min int * d_output; cudaMalloc((void **) &d_output, querySize*databaseSize*sizeof(int)); dim3 grid2((databaseSize + BLOCK_SIZE - 1)/BLOCK_SIZE); dim3 block2(BLOCK_SIZE); rowMin<<<grid2,block2>>>(d_input, d_output, querySize, databaseSize); //Copying back results int * out_output = (int*) malloc(querySize*databaseSize*sizeof(int)); cudaMemcpy(out_output, d_output,querySize*databaseSize*sizeof(int),cudaMemcpyDeviceToHost); //print_array<int*>(out_output, querySize, databaseSize); arrayto2Dvec<int>(out_output, res, querySize, databaseSize); //Freeing memory free(out_results); cudaFree(d_input); cudaFree(d_output); free(out_output); return res; } //---------------------------------------------// //Print_array //---------------------------------------------// template<class T> void print_array(T& array, int m, int n) { for(int i = 0; i < m; i++) { for(int j = 0; j < n; j++) { std::cout << array[i * n + j] << " "; } std::cout << std::endl; } } //---------------------------------------------// //1D array to 2D vector //---------------------------------------------// template <class T> void arrayto2Dvec(T* array, vector<vector<T> >& vec, size_t x, size_t y){ for(int j = 0; j < y;j++){ vector<T> vecLine; for(int i = 0; i < x; i++){ vecLine.push_back(array[j*x + i]); } vec.push_back(vecLine); } } //---------------------------------------------// //Print 2D Vector //---------------------------------------------// template <class T> void print2DVec(vector<vector<T> >& vec){ cout << "Vec x size: " << vec.size() << "\nVec y size: " << vec[0].size() << "\nType: " << typeid(vec[0][0]).name() << endl; for(size_t i = 0; i < vec.size(); i++){ for(size_t j = 0; j < vec[0].size(); j++){ cout << vec[i][j] << " "; } cout << endl; } } //---------------------------------------------// //Histograms to 1D array //---------------------------------------------// void histogram2array(float* array, vector<Histogram>& histos){ for(int i = 0; i < histos.size(); i++){ for(int j = 0; j < histos[i].size;j++){ array[i*histos[i].size + j] = histos[i].vals[j]; } } } //---------------------------------------------// //Reads in sifts //---------------------------------------------// vector<Sift> readMultiSifts(stringvec v){ std::vector<Sift> imSifts; for(int i = 0; i < v.size(); i++){ if(v[i] != "." && v[i] != ".." && v[i] != "all.txt"){ char* fn = new char[v[i].size()+1]; strcpy(fn,v[i].c_str()); Sift siftIm = readSift(fn); imSifts.push_back(siftIm); delete[] fn; } } return imSifts; } //---------------------------------------------// //Auxiliary creates histogram cuda //Inputs: siftCentroids, vector of sift images, vector of filenames //Output: vector of histograms //---------------------------------------------// vector<Histogram> createHistogramCudaAux(Sift& siftCentroids, vector<Sift>& imSifts, stringvec& v){ vector<Histogram> histograms; //Passed Vectors float* d_cent; //Memory allocation for centroids int centS = siftCentroids.numSift; int centTotal = siftCentroids.totalNum; cudaMalloc((void **) &d_cent, centTotal*sizeof(float)); cudaMemcpy(d_cent, siftCentroids.sift, centTotal*sizeof(float),cudaMemcpyHostToDevice); for(int i = 0; i < imSifts.size(); i++){ //Memory allocation for image sifts float* d_im; int imS = imSifts[i].numSift; int imTotal = imSifts[i].totalNum; cudaMalloc((void **) &d_im, imTotal*sizeof(float)); cudaMemcpy(d_im, imSifts[i].sift, imTotal*sizeof(float),cudaMemcpyHostToDevice); //Memory allocation for histogram float* hist = (float*) malloc(centS*sizeof(float)); memset(hist,0,centS*sizeof(float)); dim3 grid((centS+BLOCK_SIZE-1)/BLOCK_SIZE, imS); dim3 block(BLOCK_SIZE); //Temporary array creation float* d_temp; cudaMalloc((void **) &d_temp, imS*(centS+BLOCK_SIZE-1)/BLOCK_SIZE*2*sizeof(float)); cudaMemset(d_temp, 0, imS*(centS+BLOCK_SIZE-1)/BLOCK_SIZE*2*sizeof(float)); //Runs cuda code createHistCuda<<<grid,block>>>(d_cent,d_im,centS,imS,d_temp); //Copy temp back size_t tempS = imS*(centS+BLOCK_SIZE-1)/BLOCK_SIZE*2; float* out_temp = (float*) malloc(tempS*sizeof(float)); cudaMemcpy(out_temp, d_temp,tempS*sizeof(float),cudaMemcpyDeviceToHost); size_t tempSx = imS; size_t tempSy = (centS+BLOCK_SIZE-1)/BLOCK_SIZE; for(size_t j = 0; j < tempSx; j++){ float max = 0; int maxInd = 0; for(size_t k = 0; k < tempSy; k++){ size_t ind = (j*tempSy + k)*2; if(out_temp[ind] > max){ max = out_temp[ind]; maxInd = (int)out_temp[ind+1]; } } hist[maxInd] += 1.0f/(float)imS; } Histogram th = {.vals = hist, .size = centS, .filename = v[i]}; histograms.push_back(th); cudaFree(d_im); cudaFree(d_temp); free(out_temp); } cudaFree(d_cent); return histograms; } //---------------------------------------------// //Print directory files //---------------------------------------------// void printFiles(stringvec v){ printf("\nDirectory List\nSize: %d\n", v.size()); for(int i = 0; i < v.size(); i++){ std::cout << "\tName: " << v[i] << "\n"; } printf("\n\n"); } //---------------------------------------------// //Extracts Filenames in Directory //---------------------------------------------// void read_directory(const std::string& name, stringvec& v) { DIR* dirp = opendir(name.c_str()); struct dirent * dp; while ((dp = readdir(dirp)) != NULL) { string temp = dp->d_name; string s = name + temp; if(temp.compare(".") != 0 && temp.compare("..") != 0 && temp.compare("all.txt") != 0) v.push_back(s); } closedir(dirp); } //---------------------------------------------// //Free Histogram Memory //---------------------------------------------// void freeHisto(Histogram h){ free(h.vals); } //---------------------------------------------// //Prints Histo //Options: 'i' for info, 's' for sifts, 'b' for both //---------------------------------------------// void printHisto(Histogram h, char option){ if(option == 'i' || option == 'b'){ printf("----------------------\nSize: %d\n----------------------\n", h.size); } if(option == 's' || option == 'b'){ //Prints the histogram for(int i = 0; i < h.size; i++){ printf("%f ", h.vals[i]); } } } //---------------------------------------------// //Free Sift Memory //---------------------------------------------// void freeSift(Sift sift){ free(sift.sift); } //---------------------------------------------// //Prints Sift //Options: 'i' for info, 's' for sifts, 'b' for both //---------------------------------------------// void printSift(Sift sift, char option){ if(option == 'i' || option == 'b'){ printf("----------------------\nNum of Sifts: %d\nTotalSize: %d floats\n----------------------\n", sift.numSift, sift.totalNum); } if(option == 's' || option == 'b'){ for(int i = 0; i < sift.numSift; i++){ for(int j = 0; j < sift.elementSize; j++){ printf("%f ", sift.sift[i*sift.elementSize + j]); } } } } //---------------------------------------------// //Reads sift from file //---------------------------------------------// Sift readSift(char *filename){ int lines = countlines(filename); float * siftCentroids = (float *)malloc(lines * 128* sizeof(float)); readFromFile(lines * 128, siftCentroids, filename); printf("-----Reading-Sift-----\nFilename: %s\nSift Number: %d\n----------------------\n", filename, lines); Sift sift = {.sift = siftCentroids, .numSift = lines, .totalNum = lines*128, .elementSize = 128}; return sift; } //---------------------------------------------// //Creates histogram for single centroid-image pair //Description: Creates a histogram for an image siftImage based on centroids siftCentroids. //Input: centroids, image sifts, histogram vector, number of centroids, number of sifts for image //Output: Histogram of siftImage according to siftCentroids //---------------------------------------------// Histogram createHist(Sift cent, Sift im){ float * siftCentroids = cent.sift; float * siftImage = im.sift; int linesCent = cent.numSift; int linesIm = im.numSift; float * histo = (float*) malloc(linesCent*sizeof(float)); //Initializes histogram to 0 for(int i = 0; i < linesCent; i++){ histo[i] = 0; } //For descriptor in image for(int i = 0; i < linesIm; i++){ float bestCos = 0; int bestJ = 0; //For centroid descriptors, find the one that best matches the image descriptor for(int j = 0; j < linesCent; j++){ int centin = j*128; int imin = i*128; //Cosine similarity code ------------ float sumab = 0; float suma2 = 0; float sumb2 = 0; for(int k = 0; k < 128; k++){ sumab += siftCentroids[centin + k] * siftImage[imin + k]; suma2 += siftImage[imin + k] * siftImage[imin + k]; sumb2 += siftCentroids[centin + k] * siftCentroids[centin + k]; } float cossim = sumab/(sqrtf(suma2)/sqrtf(sumb2)); if(cossim > bestCos){ bestCos = cossim; bestJ = j; } //Cosine similarity end -------------- } histo[bestJ] += 1; } //Normalization for(int i = 0; i < linesCent; i++){ histo[i] = histo[i]/linesIm; } Histogram h = {.vals = histo, .size = linesCent, .filename = ""}; return h; } //---------------------------------------------// //Reading file into array //---------------------------------------------// void readFromFile(int size,float *v, char *file){ FILE *fp; //cout << "Openning" << endl; fp = fopen(file,"r"); int i=0; float t; while(i < size){ if(fscanf(fp,"%f",&t)==EOF){ printf("Error reading file\n"); exit(1); } //printf("%f ", t); v[i++]=t; } //cout << "Read" << endl; fclose(fp); } //---------------------------------------------// //Counting lines in file //---------------------------------------------// int countlines(char*filename) { FILE *fp; int count = 0; // Line counter (result) char c; // To store a character read from file // Open the file fp = fopen(filename, "r"); if (fp == NULL) { printf("Could not open file %s", filename); return 0; } // Extract characters from file and store in character c for (c = getc(fp); c != EOF; c = getc(fp)) if (c == '\n') // Increment count if this character is newline count = count + 1; // Close the file fclose(fp); //printf("The file %s has %d lines\n ", filename, count); return count; }
12,923
#include <stdio.h> #include <stdlib.h> #include <math.h> #define THREADS_POR_BLOCO 1024 #define DIM_BLOCO 32 #define DIM_GRID 1024 __global__ void add(int *a, int *b, int *c, int N){ //int i = threadIdx.y + blockIdx.y*blockDim.y; //int j = threadIdx.x + blockIdx.x*blockDim.x; int index = DIM_BLOCO*DIM_BLOCO*(DIM_GRID*blockIdx.x+blockIdx.y)+blockDim.y*threadIdx.x+threadIdx.y; if(index < N) c[index] = a[index] + b[index]; } int main() { int *A, *B, *C; int *d_A, *d_B, *d_C; int i, j; //Input int linhas, colunas; scanf("%d", &linhas); scanf("%d", &colunas); //Definindo tamanho dos arrays que representarão as matrizes int N = linhas*colunas; int size = sizeof(int)*N; //Alocando memória na GPU cudaMalloc((void **)&d_A,size); cudaMalloc((void **)&d_B,size); cudaMalloc((void **)&d_C,size); //Alocando memória na CPU A = (int *)malloc(size); B = (int *)malloc(size); C = (int *)malloc(size); //Inicializar for(i = 0; i < linhas; i++){ for(j = 0; j < colunas; j++){ A[i*colunas+j] = B[i*colunas+j] = i+j; //printf("%d ",A[i*colunas+j]); } //printf("\n"); } //Transferir para a memória da GPU cudaMemcpy(d_A, A, size, cudaMemcpyHostToDevice); cudaMemcpy(d_B, B, size, cudaMemcpyHostToDevice); cudaMemcpy(d_C, C, size, cudaMemcpyHostToDevice); //Computacao que deverá ser movida para a GPU // Número de blocos = Número de linhas // threads por bloco = número de colunas dim3 dimGrid(DIM_GRID,DIM_GRID); dim3 dimBlock(DIM_BLOCO,DIM_BLOCO); add<<<dimGrid,dimBlock>>>(d_A,d_B,d_C,N); //add<<<(N+THREADS_POR_BLOCO-1)/THREADS_POR_BLOCO,THREADS_POR_BLOCO>>>(d_A,d_B,d_C,N); cudaMemcpy(C,d_C,size,cudaMemcpyDeviceToHost); long long int somador=0; //Manter esta computação na CPU for(i = 0; i < linhas; i++){ for(j = 0; j < colunas; j++){ somador+=C[i*colunas+j]; //printf("%d ",C[i*colunas+j]); } //printf("\n"); } printf("%lli\n", somador); free(A); free(B); free(C); cudaFree(d_A); cudaFree(d_B); cudaFree(d_C); }
12,924
#include <iostream> #include <vector> #include <random> #include <cmath> // #include <boost/timer/timer.hpp> class Managed { public: size_t size; void *operator new(size_t len) { void *ptr; cudaMallocManaged(&ptr, len); cudaDeviceSynchronize(); return ptr; } void operator delete(void *ptr) { cudaDeviceSynchronize(); cudaFree(ptr); } void sync() { cudaDeviceSynchronize(); } }; template <class T> class CuArray : public Managed { public: int size; T * data; CuArray() : size(0), data(nullptr) { } CuArray(std::vector<T> * a) : size(a->size()) { // Allocate unified memory realloc_(a->size()); // Copy C array from vector memcpy(data, a->data(), a->size() * sizeof(T)); } CuArray(const CuArray<T> & a) : size(a.size) { realloc_(a.size); memcpy(data, a.data, a.size * sizeof(T)); } ~CuArray() { cudaFree(data); } CuArray& operator=(std::vector<T> * a) { size = a->size(); realloc_(a->size()); memcpy(data, a->data(), size * sizeof(T)); return *this; } void prefetch() { int device = -1; cudaGetDevice(&device); cudaMemPrefetchAsync(data, size * sizeof(T), device, NULL); cudaMemPrefetchAsync(&size, sizeof(int), device, NULL); } __host__ __device__ T& operator[](int pos) const { return data[pos]; } private: void realloc_(int s) { // cudaFree(data); cudaMallocManaged(&data, s * sizeof(float)); cudaDeviceSynchronize(); } }; class SplineParams : public Managed { public: CuArray<float> knotsX; CuArray<float> knotsY; CuArray<float> dydxs; CuArray<int> cells; SplineParams() {} SplineParams(CuArray<float> knotsX_, CuArray<float> knotsY_, CuArray<float> dydxs_, CuArray<int> cells_) : knotsX(knotsX_), knotsY(knotsY_), dydxs(dydxs_), cells(cells_) {} void prefetch() { knotsX.prefetch(); knotsY.prefetch(); dydxs.prefetch(); cells.prefetch(); } }; // Basically straight from the Laura++ implementation // Only a loop with trip count nKnots, probably not worth putting it on the GPU // Fit for y eventually, here just assume they are given. std::vector<float> calculateGrads(std::vector<float> &x, std::vector<float> &y) { int nKnots = x.size(); std::vector<float> a(nKnots); std::vector<float> b(nKnots); std::vector<float> c(nKnots); std::vector<float> d(nKnots); std::vector<float> grad(nKnots); a[0] = 0.; c[nKnots - 1] = 0.; // Left float xD10 = x[1] - x[0]; b[0] = 2. / xD10; c[0] = 1. / xD10; d[0] = 3. * (y[1] - y[0]) / ( xD10 * xD10 ); // Right float xk12 = x[nKnots - 1] - x[nKnots - 2]; a[nKnots - 1] = 1. / xk12; b[nKnots - 1] = 2. / xk12; d[nKnots - 1] = 3. * (y[nKnots - 1] - y[nKnots - 2]) / ( xk12 * xk12 ); // Internal for(uint i = 1; i < nKnots - 1; i++) { float xDi = x[i] - x[i - 1]; float xD1i = x[i + 1] - x[i]; a[i] = 1. / xDi; b[i] = 2. / xDi + 2. / xD1i; c[i] = 1./ xD1i; d[i] = 3. * (y[i] - y[i - 1]) / ( xDi * xDi ) + 3. * (y[i + 1] - y[i]) / ( xD1i * xD1i ); } c[0] /= b[0]; d[0] /= b[0]; for(uint i = 1; i < nKnots - 1; i++) { c[i] = c[i] / (b[i] - a[i] * c[i - 1]); d[i] = (d[i] - a[i] * d[i - 1]) / (b[i] - a[i] * c[i - 1]); } d[nKnots - 1] = (d[nKnots - 1] - a[nKnots - 1] * d[nKnots - 2]) / (b[nKnots - 1] - a[nKnots - 1] * c[nKnots - 2]); grad[nKnots - 1] = d[nKnots - 1]; for(int i = nKnots - 2; i >= 0; i--) { grad[i] = d[i] - c[i] * grad[i + 1]; } return grad; } __global__ void evalSplineKern(const int n, const CuArray<float> * xs, const SplineParams * params, CuArray<float> * splineVals) { // All threads handle blockDim.x * gridDim.x // consecutive elements (interleaved partitioning) int index = blockIdx.x * blockDim.x + threadIdx.x; int stride = blockDim.x * gridDim.x; for (int i = index; i < n; i += stride){ int cell = params->cells[i]; // // Try and avoid this on GPU if possible -> precompute and save? // while( (*xs)[i] > params->knotsX[cell+1] ) { // ++cell; // } // Might be a slow memory access.... float xLow = params->knotsX[cell]; float xHigh = params->knotsX[cell+1]; float yLow = params->knotsY[cell]; float yHigh = params->knotsY[cell+1]; float t = ((*xs)[i] - xLow) / (xHigh - xLow); float a = params->dydxs[cell] * (xHigh - xLow) - (yHigh - yLow); float b = -1. * params->dydxs[cell+1] * (xHigh - xLow) + (yHigh - yLow); (*splineVals)[i] = (1 - t) * yLow + t * yHigh + t * (1 - t) * ( a * (1 - t) + b * t ); } } void calcSplineGPU(std::vector<float> * knotsX, std::vector<float> * knotsY, std::vector<float> * dydxs, std::vector<float> * masses, std::vector<int> * cells) { int n = masses->size(); int nKnots = knotsX->size(); SplineParams * splineParams = new SplineParams(); splineParams->knotsX = knotsX; splineParams->knotsY = knotsY; splineParams->dydxs = dydxs; splineParams->cells = cells; CuArray<float> * xs = new CuArray<float>(masses); std::vector<float> * splineValsV = new std::vector<float>; splineValsV->resize(n); std::fill(splineValsV->begin(), splineValsV->end(), 0.0); CuArray<float> * splineVals = new CuArray<float>(splineValsV); splineParams->prefetch(); xs->prefetch(); splineVals->prefetch(); int blockSize = 512; int numBlocks = (n + blockSize - 1) / blockSize; { // boost::timer::auto_cpu_timer t; evalSplineKern<<<numBlocks, blockSize>>>(n, xs, splineParams, splineVals); splineParams->sync(); xs->sync(); splineVals->sync(); } // Timer std::cout << (*splineVals)[0] << std::endl; delete splineParams; } std::vector<float> calcSplineCPU(std::vector<float> & knotsX, std::vector<float> & knotsY, std::vector<float> & dydxs, std::vector<float> & masses, std::vector<int> & cells) { std::vector<float> splineVals(masses.size()); std::cout << splineVals.size() << std::endl; for (int i = 0; i < splineVals.size(); i++) { int cell = cells[i]; // while( masses[i] > knotsX[cell+1] ) { // ++cell; // } float xLow = knotsX[cell]; float xHigh = knotsX[cell+1]; float yLow = knotsY[cell]; float yHigh = knotsY[cell+1]; float t = (masses[i] - xLow) / (xHigh - xLow); float a = dydxs[cell] * (xHigh - xLow) - (yHigh - yLow); float b = -1. * dydxs[cell+1] * (xHigh - xLow) + (yHigh - yLow); splineVals[i] = (1 - t) * yLow + t * yHigh + t * (1 - t) * ( a * (1 - t) + b * t ); } return splineVals; } std::vector<int> calculateCells(std::vector<float> & masses, std::vector<float> & knotsX) { std::vector<int> cells(masses.size()); for (int i = 0; i < cells.size(); i++) { int cell = 0; while( masses[i] > knotsX[cell+1] ) { cell++; } cells[i] = cell; } return cells; } float normalDist(float mu, float sigma, float x) { float norm = 1. / (sigma * std::sqrt(2 * M_PI)); float z = (x - mu) / sigma; float arg = -0.5 * z * z; return norm * std::exp(arg); } int main(int argc, char const *argv[]) { std::default_random_engine generator; std::normal_distribution<float> normal(0.0, 1.0); int n = 100000; int nKnots = 30; std::vector<float> data(n); for (auto & d : data) d = normal(generator); std::vector<float> * knotsX = new std::vector<float>(nKnots); std::vector<float> * knotsY = new std::vector<float>(nKnots); std::vector<float> * masses = new std::vector<float>(n); float startKnot = -3.0; float endKnot = 3.0; float stepSize = (endKnot - startKnot) / nKnots; for (int i = 0; i < nKnots; i++) { (*knotsX)[i] = startKnot + i * stepSize; (*knotsY)[i] = (n / nKnots) * normalDist(0.0, 1.0, (*knotsX)[i]); } std::vector<int> cells = calculateCells(*masses, *knotsX); std::vector<float> dydxs = calculateGrads(*knotsX, *knotsY); { // boost::timer::auto_cpu_timer t; calcSplineGPU(knotsX, knotsY, &dydxs, &data, &cells); // calcSplineCPU(*knotsX, *knotsY, dydxs, data, cells); } return 0; }
12,925
#include "includes.h" __global__ void kernel_offset(int *key, int *idx, int *offset, int size) { int idxX = threadIdx.x + blockIdx.x*blockDim.x; if(idxX == 0) { offset[1] = 0; } else if(idxX < size) { int keyVal = key[idxX]; int keyValPrev = key[idxX-1]; if(keyVal != keyValPrev) { offset[keyVal+1] = idxX; } } if(idxX == size-1) { int keyVal = key[idxX]; offset[0] = keyVal+1; offset[keyVal+2] = size; } }
12,926
#include <stdio.h> #include <stdlib.h> #define CSC(call) do { \ cudaError_t e = call; \ if (e != cudaSuccess) { \ fprintf(stderr, "CUDA Error in %s:%d: %s\n", __FILE__, __LINE__, cudaGetErrorString(e)); \ exit(0); \ } \ } while(0) #define MAX_R 1024 __global__ void kernel(double* dVector, int num) { int idx = blockDim.x * blockIdx.x + threadIdx.x; int offset = gridDim.x * blockDim.x; while (idx < num) { if (dVector[idx] < 0) { dVector[idx] = -dVector[idx]; } idx += offset; } } class CUVector { public: const int n; int size; double* hVector; double* dVector; CUVector(int _n): n(_n) { size = sizeof(double) * n; hVector= (double*)malloc(size); CSC(cudaMalloc(&dVector, size)); } void init() { for (int i = 0; i < n; ++i) scanf("%lf", &hVector[i]); CSC(cudaMemcpy(dVector, hVector, size, cudaMemcpyHostToDevice)); } void run_kernel() { kernel<<<256, 256>>>(dVector, n); CSC(cudaMemcpy(hVector, dVector, size, cudaMemcpyDeviceToHost)); } void print() { for (int i = 0; i < n; ++i) printf("%.10e ", hVector[i]); printf("\n"); } ~CUVector(){ CSC(cudaFree(dVector)); free(hVector); } }; __constant__ float a[MAX_R * 2 + 1]; int main(void) { int n; scanf("%d", &n); CUVector v = CUVector(n); v.init(); v.run_kernel(); v.print(); return 0; }
12,927
#include <stdio.h> #include<stdlib.h> #define N 1000 #define MAX_ERR 1e-6 __global__ void add(int *a, int b, int c) { *a = b + c; } int main(){ int count = 0; cudaGetDevice(&count); cudaDeviceProp prop; cudaGetDeviceProperties(&prop,count); int code = cudaChooseDevice(&count,&prop); printf("%d,%d\n",code,count); int a = 0; int *dev_a; cudaMalloc((void **)&dev_a,sizeof(int)); add<<<1,1>>>(dev_a,3,5); cudaDeviceSynchronize(); cudaMemcpy(&a,dev_a, sizeof(int),cudaMemcpyDeviceToHost); printf("3 + 5 = %d\n",a); cudaFree(dev_a); return 0; }
12,928
#include <stdio.h> #include <stdlib.h> #include <cuda.h> #include <curand_kernel.h> __global__ void MyKernel(float** dev_matrix, size_t pitch, int width, int height) { int number = 0; for (int i = 0; i < height; ++i) { float* row = (float*)((char*)dev_matrix + i*pitch); for (int j = 0; j < width; ++j) { row[j] = number; number++; } } } int main (int argc , char * argv []) { int width = 4, height = 2, i, j; // float matrix[width][height]; float matrix[height][width]; float **dev_matrix; size_t pitch; printf("\nMATRIX MANIPULATION\n"); for (i = 0; i < height; i++) for (j = 0; j < width; j++) matrix[i][j] = 0.0; printf("Matrix in host memory\n"); for (i = 0; i < height; i++) { for (j = 0; j < width; j++) printf("%f ", matrix[i][j]); printf("\n"); } cudaMallocPitch(&dev_matrix, &pitch, width * sizeof(float), height); cudaMemcpy2D(dev_matrix, pitch, matrix, width * sizeof(float), width * sizeof(float), height, cudaMemcpyHostToDevice); MyKernel<<<2, 2>>>(dev_matrix, pitch, width, height); cudaMemcpy2D(matrix, width * sizeof(float), dev_matrix, pitch, width * sizeof(float), height, cudaMemcpyDeviceToHost); printf("Matrix after calculate elements in the gpu\n"); for (i = 0; i < height; i++) { for (j = 0; j < width; j++) printf("%f ", matrix[i][j]); printf("\n"); } cudaFree(dev_matrix); return 0; }
12,929
#include <iostream> using namespace std; __global__ void VecAdd(float *A, float *B, float *C){ int i = threadIdx.x; C[i] = A[i] + B[i]; } int main(){ float *A, *B, *C, *devA, *devB, *devC; int n = 100000000; A = (float*)malloc(n * sizeof(float)); B = (float*)malloc(n * sizeof(float)); C = (float*)malloc(n * sizeof(float)); cudaMalloc(&devA, n * sizeof(float)); cudaMalloc(&devB, n * sizeof(float)); cudaMalloc(&devC, n * sizeof(float)); for(int i = 0;i < n;i ++){ A[i] = float(rand()) + float(rand()%100); B[i] = float(rand()%20) + float(rand() + 32) + 93.23; } cudaMemcpy(devA, A, n * sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(devB, B, n * sizeof(float), cudaMemcpyHostToDevice); VecAdd<<<1, 1024>>>(devA, devB, devC); cudaMemcpy(C, devC, n * sizeof(float), cudaMemcpyDeviceToHost); cout << endl << C[100]; }
12,930
#include <stdio.h> #include <cuda.h> // Kernel that executes on the CUDA device __global__ void square_array(double *a, int N) { int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx<N) a[idx] = a[idx] * a[idx]; } void run_square_array(double* a_d, int N) { int block_size = 4; int n_blocks = N/block_size + (N%block_size == 0 ? 0:1); // Retrieve result from device and store it in host array square_array<<<n_blocks,block_size>>>(a_d,N); cudaError_t code = cudaGetLastError(); printf("Poseidon_kernel error: %s\n", cudaGetErrorString(code)); }
12,931
#include "includes.h" __global__ void mul_kernel(const int n, const float *a, const float *b, float *y) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < (n); i += blockDim.x * gridDim.x) { y[i] = a[i] * b[i]; } }
12,932
 #include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> int error(int* device_a, int* device_b, int* device_c); __global__ void add(int *c, const int *a, const int *b) { printf("Block: %d, Thread: %d, Block Dim: %d\n", threadIdx.x, blockIdx.x, blockDim.x); int i = threadIdx.x; c[i] = a[i] + b[i]; } int main() { cudaError_t cudaStatus; const int arraySize = 7; const int a[arraySize] = { 1, 2, 3, 5, 7, 11, 13 }; const int b[arraySize] = { 1, 2, 3, 5, 8, 13, 21 }; int c[arraySize] = { 0 }; int* device_a = 0; int* device_b = 0; int* device_c = 0; // Seleciona o dispositivo cudaStatus = cudaSetDevice(0); //Aloca memoria na GPU cudaStatus = cudaMalloc((void**)&device_a, arraySize * sizeof(int)); cudaStatus = cudaMalloc((void**)&device_b, arraySize * sizeof(int)); cudaStatus = cudaMalloc((void**)&device_c, arraySize * sizeof(int)); cudaStatus = cudaMemcpy(device_a, a, arraySize * sizeof(int), cudaMemcpyHostToDevice); cudaStatus = cudaMemcpy(device_b, b, arraySize * sizeof(int), cudaMemcpyHostToDevice); if (cudaStatus != cudaSuccess) { printf("cudaSetDevice falhou!"); return error(device_a, device_b, device_c); } add<<<1, arraySize>>>(device_c, device_a, device_b); cudaStatus = cudaGetLastError(); if (cudaStatus != cudaSuccess) { printf("add function failed: %s\n", cudaGetErrorString(cudaStatus)); return error(device_a, device_b, device_c); } cudaStatus = cudaDeviceSynchronize(); cudaStatus = cudaMemcpy(c, device_c, arraySize * sizeof(int), cudaMemcpyDeviceToHost); printf("{ 1, 2, 3, 5, 7, 11, 13 } + { 1, 2, 3, 5, 8, 13, 21 } = {%d, %d, %d, %d, %d, %d, %d}\n", c[0], c[1], c[2], c[3], c[4], c[5], c[6]); cudaStatus = cudaDeviceReset(); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaDeviceReset failed!"); return 1; } cudaFree(device_a); cudaFree(device_b); cudaFree(device_c); return 0; } int error(int* device_a, int* device_b, int* device_c) { cudaFree(device_a); cudaFree(device_b); cudaFree(device_c); return 1; }
12,933
/* * Author: scps950707 * Email: scps950707@gmail.com * Created: 2017-12-04 12:09 * Last Modified: 2017-12-04 18:42 * Filename: wave.cu * description: Serial Concurrent Wave Equation - C Version * This program implements the concurrent wave equation */ #include <stdio.h> #include <stdlib.h> #include <math.h> #include <time.h> #define MAXPOINTS 1000000 #define MAXSTEPS 1000000 #define MINPOINTS 20 #define PI 3.14159265 static void HandleError( cudaError_t err, const char *file, int line ) { if ( err != cudaSuccess ) { printf( "%s in %s at line %d\n", cudaGetErrorString( err ), file, line ); exit( EXIT_FAILURE ); } } #define HANDLE_ERROR( err ) (HandleError( err, __FILE__, __LINE__ )) int nsteps; /* number of time steps */ int tpoints; /* total points along string */ float H_currVal[MAXPOINTS + 2]; /* values at time t */ /********************************************************************** * Checks input values from parameters *********************************************************************/ void check_param( void ) { char tchar[20]; /* check number of points, number of iterations */ while ( ( tpoints < MINPOINTS ) || ( tpoints > MAXPOINTS ) ) { printf( "Enter number of points along vibrating string [%d-%d]: " , MINPOINTS, MAXPOINTS ); scanf( "%s", tchar ); tpoints = atoi( tchar ); if ( ( tpoints < MINPOINTS ) || ( tpoints > MAXPOINTS ) ) printf( "Invalid. Please enter value between %d and %d\n", MINPOINTS, MAXPOINTS ); } while ( ( nsteps < 1 ) || ( nsteps > MAXSTEPS ) ) { printf( "Enter number of time steps [1-%d]: ", MAXSTEPS ); scanf( "%s", tchar ); nsteps = atoi( tchar ); if ( ( nsteps < 1 ) || ( nsteps > MAXSTEPS ) ) { printf( "Invalid. Please enter value between 1 and %d\n", MAXSTEPS ); } } printf( "Using points = %d, steps = %d\n", tpoints, nsteps ); } /********************************************************************** * initialize points on line * Update all values along line a specified number of times *********************************************************************/ __global__ void initAndUpdate( float *D_oldVal, float *D_currVal, int tpoints, int nsteps ) { int j = blockDim.x * blockIdx.x + threadIdx.x; if ( j < tpoints ) { j += 1; /* Calculate initial values based on sine curve */ /* Initialize old values array */ float x = ( float )( j - 1 ) / ( tpoints - 1 ); D_oldVal[j] = D_currVal[j] = sin ( 6.2831853f * x ); int i; /* global endpoints */ if ( ( j == 1 ) || ( j == tpoints ) ) { D_currVal[j] = 0.0; } else { /* Update values for each time step */ for ( i = 1; i <= nsteps; i++ ) { /* Update old values with new values */ float newVal = ( 2.0 * D_currVal[j] ) - D_oldVal[j] + ( 0.09f * ( -2.0 ) * D_currVal[j] ); D_oldVal[j] = D_currVal[j]; D_currVal[j] = newVal; } } } } /********************************************************************** * Print final results *********************************************************************/ void printfinal() { int i; for ( i = 1; i <= tpoints; i++ ) { printf( "%6.4f ", H_currVal[i] ); if ( i % 10 == 0 ) { printf( "\n" ); } } } /********************************************************************** * Main program *********************************************************************/ int main( int argc, char *argv[] ) { /* Error code to check return values for CUDA calls */ sscanf( argv[1], "%d", &tpoints ); sscanf( argv[2], "%d", &nsteps ); check_param(); int threadsPerBlock = 256; int blocksPerGrid = ( tpoints + threadsPerBlock - 1 ) / threadsPerBlock; float *D_currVal, *D_oldVal; HANDLE_ERROR( cudaMalloc( ( void ** )&D_currVal, sizeof( float ) * ( tpoints + 2 ) ) ); HANDLE_ERROR( cudaMalloc( ( void ** )&D_oldVal, sizeof( float ) * ( tpoints + 2 ) ) ); printf( "Initializing points on the line...\n" ); printf( "Updating all points for all time steps...\n" ); #if __DEBUG__ clock_t t = clock(); #endif initAndUpdate <<<blocksPerGrid, threadsPerBlock>>>( D_oldVal, D_currVal, tpoints, nsteps ); HANDLE_ERROR( cudaMemcpy( H_currVal, D_currVal, sizeof( float ) * ( tpoints + 2 ), cudaMemcpyDeviceToHost ) ); #if __DEBUG__ t = clock() - t; #endif printf( "Printing final results...\n" ); printfinal(); printf( "\nDone.\n\n" ); #if __DEBUG__ printf( "time:%f\n", ( float )t / CLOCKS_PER_SEC ); #endif HANDLE_ERROR( cudaFree( D_currVal ) ); HANDLE_ERROR( cudaFree( D_oldVal ) ); return EXIT_SUCCESS; }
12,934
#include "BasicTool.cuh" //Ƕת double AngleToRad(double angle) { double temp; temp=angle*PI; return temp/180; } // float CalculateDistanceOfBipartite(Point a,Point b) { return sqrt((a.x-b.x)*(a.x-b.x)+(a.y-b.y)*(a.y-b.y)+(a.z-b.z)*(a.z-b.z)); } void UitizeVector(float *a,float *b,float *c)//һ { float length=sqrt((*a)*(*a)+(*b)*(*b)+(*c)*(*c)); *a=*a/length; *b=*b/length; *c=*c/length; } Point GetReflectedVector(Point d,Point n)//ݷĵλõ { Point reflectedPoint; float temp=2*(d.x*n.x+d.y*n.y+d.z*n.z); reflectedPoint.x=d.x-temp*n.x; reflectedPoint.y=d.y-temp*n.y; reflectedPoint.z=d.z-temp*n.z; return reflectedPoint; } Ray CalculateReflectedRayOnCPU(Ray incidentRay,Face *face,int faceCount,int *reflectedFace,int *flag)//н { Ray reflectedRay; Point defaultPoint; defaultPoint.x=0;defaultPoint.y=0;defaultPoint.z=0; reflectedRay.originalPoint=defaultPoint; reflectedRay.direction=defaultPoint; float t=50000; for (int i=0;i<faceCount;i++) { float a1=-incidentRay.direction.x,a2=-incidentRay.direction.y,a3=-incidentRay.direction.z; float b1=face[i]. B.x-face[i].A.x,b2=face[i].B.y-face[i].A.y,b3=face[i].B.z-face[i].A.z; float c1=face[i]. C.x-face[i].A.x,c2=face[i].C.y-face[i].A.y,c3=face[i].C.z-face[i].A.z; float x1=incidentRay.originalPoint.x-face[i].A.x,x2=incidentRay.originalPoint.y-face[i].A.y,x3=incidentRay.originalPoint.z-face[i].A.z; float denominator=a1*(b2*c3-b3*c2)-b1*(a2*c3-a3*c2)+c1*(a2*b3-a3*b2); float t_numerator=x1*(b2*c3-b3*c2)-b1*(x2*c3-x3*c2)+c1*(x2*b3-x3*b2); float u_numerator=a1*(x2*c3-x3*c2)-x1*(a2*c3-a3*c2)+c1*(a2*x3-a3*x2); float v_numerator=a1*(b2*x3-b3*x2)-b1*(a2*x3-a3*x2)+x1*(a2*b3-a3*b2); if (abs(denominator)>0.000001) { float u=u_numerator/denominator; float v=v_numerator/denominator; int remain=i%12; if(t_numerator/denominator<t&&t_numerator/denominator>1&&u>=0&&u<=1&&v>0&&(u+v)<1&&v<=1&&remain<8) { *flag=1; *reflectedFace=i; t=t_numerator/denominator; reflectedRay.originalPoint.x=u*b1+v*c1+face[i].A.x; reflectedRay.originalPoint.y=u*b2+v*c2+face[i].A.y; reflectedRay.originalPoint.z=u*b3+v*c3+face[i].A.z; Point n; n.x=b2*c3-b3*c2;n.y=b3*c1-b1*c3;n.z=b1*c2-c1*b2; UitizeVector(&n.x,&n.y,&n.z); reflectedRay.direction=GetReflectedVector(incidentRay.direction,n); } } } return reflectedRay; } Ray CalculateVirtualReflectedRayOnCPU(Ray incidentRay,Face face)//õijⷴ { Ray reflectedRay; Point defaultPoint; defaultPoint.x=0;defaultPoint.y=0;defaultPoint.z=0; reflectedRay.originalPoint=defaultPoint; reflectedRay.direction=defaultPoint; float t=50000; float a1=-incidentRay.direction.x,a2=-incidentRay.direction.y,a3=-incidentRay.direction.z; float b1=face.B.x-face.A.x,b2=face.B.y-face.A.y,b3=face.B.z-face.A.z; float c1=face.C.x-face.A.x,c2=face.C.y-face.A.y,c3=face.C.z-face.A.z; float x1=incidentRay.originalPoint.x-face.A.x,x2=incidentRay.originalPoint.y-face.A.y,x3=incidentRay.originalPoint.z-face.A.z; float denominator=a1*(b2*c3-b3*c2)-b1*(a2*c3-a3*c2)+c1*(a2*b3-a3*b2); float t_numerator=x1*(b2*c3-b3*c2)-b1*(x2*c3-x3*c2)+c1*(x2*b3-x3*b2); float u_numerator=a1*(x2*c3-x3*c2)-x1*(a2*c3-a3*c2)+c1*(a2*x3-a3*x2); float v_numerator=a1*(b2*x3-b3*x2)-b1*(a2*x3-a3*x2)+x1*(a2*b3-a3*b2); if (abs(denominator)>0.000001) { float u=u_numerator/denominator; float v=v_numerator/denominator; if(t_numerator/denominator<t&&t_numerator/denominator>1) { t=t_numerator/denominator; reflectedRay.originalPoint.x=u*b1+v*c1+face.A.x; reflectedRay.originalPoint.y=u*b2+v*c2+face.A.y; reflectedRay.originalPoint.z=u*b3+v*c3+face.A.z; Point n; n.x=b2*c3-b3*c2;n.y=b3*c1-b1*c3;n.z=b1*c2-c1*b2; UitizeVector(&n.x,&n.y,&n.z); reflectedRay.direction=GetReflectedVector(incidentRay.direction,n); } } return reflectedRay; } Ray CalculateReflectedRayWithTerrainOnCPU(Ray incidentRay,Face *face,int faceCount,int *reflectedFace,int *flag)//н { Ray reflectedRay; Point defaultPoint; defaultPoint.x=0;defaultPoint.y=0;defaultPoint.z=0; reflectedRay.originalPoint=defaultPoint; reflectedRay.direction=defaultPoint; float t=50000; for (int i=0;i<faceCount;i++) { float a1=-incidentRay.direction.x,a2=-incidentRay.direction.y,a3=-incidentRay.direction.z; float b1=face[i].B.x-face[i].A.x,b2=face[i].B.y-face[i].A.y,b3=face[i].B.z-face[i].A.z; float c1=face[i].C.x-face[i].A.x,c2=face[i].C.y-face[i].A.y,c3=face[i].C.z-face[i].A.z; float x1=incidentRay.originalPoint.x-face[i].A.x,x2=incidentRay.originalPoint.y-face[i].A.y,x3=incidentRay.originalPoint.z-face[i].A.z; float denominator=a1*(b2*c3-b3*c2)-b1*(a2*c3-a3*c2)+c1*(a2*b3-a3*b2); float t_numerator=x1*(b2*c3-b3*c2)-b1*(x2*c3-x3*c2)+c1*(x2*b3-x3*b2); float u_numerator=a1*(x2*c3-x3*c2)-x1*(a2*c3-a3*c2)+c1*(a2*x3-a3*x2); float v_numerator=a1*(b2*x3-b3*x2)-b1*(a2*x3-a3*x2)+x1*(a2*b3-a3*b2); if (abs(denominator)>0.000001) { float u=u_numerator/denominator; float v=v_numerator/denominator; if(t_numerator/denominator<t&&t_numerator/denominator>1&&u>=0&&u<=1&&v>0&&(u+v)<1&&v<=1) { *flag=1; *reflectedFace=i; t=t_numerator/denominator; reflectedRay.originalPoint.x=u*b1+v*c1+face[i].A.x; reflectedRay.originalPoint.y=u*b2+v*c2+face[i].A.y; reflectedRay.originalPoint.z=u*b3+v*c3+face[i].A.z; Point n; n.x=b2*c3-b3*c2;n.y=b3*c1-b1*c3;n.z=b1*c2-c1*b2; UitizeVector(&n.x,&n.y,&n.z); reflectedRay.direction=GetReflectedVector(incidentRay.direction,n); } } } return reflectedRay; } //߽ ⷢ㣩 Point GetIntersectionOfRays(Ray ray1,Ray ray2) { float a1=ray1.direction.x,a2=ray1.direction.y,a3=ray1.direction.z; float b1=-ray2.direction.x,b2=-ray2.direction.y,b3=-ray2.direction.z; float c1=ray2.originalPoint.x-ray1.originalPoint.x,c2=ray2.originalPoint.y-ray1.originalPoint.y,c3=ray2.originalPoint.z-ray1.originalPoint.z; float det=a1*(b2*c3-b3*c2)-b1*(a2*c3-a3*c2)+c1*(a2*b3-a3*b2); Point intersection; intersection.x=0;intersection.y=0;intersection.z=0; if(abs(det)<0.001) { if (abs(c1*b2-c2*b1)<0.001) { intersection.x=ray1.originalPoint.x; intersection.y=ray1.originalPoint.y; intersection.z=ray1.originalPoint.z; } else if ((abs(b2*a1-b1*a2)>0.00001)) { float u=(c1*b2-c2*b1)/(b2*a1-b1*a2); //float v=(c2*a1-c1*a2)/(b2*a1-b1*a2); intersection.x=ray1.originalPoint.x+u*a1; intersection.y=ray1.originalPoint.y+u*a2; intersection.z=ray1.originalPoint.z+u*a3; } } else { intersection.x=0;intersection.y=0;intersection.z=0; } return intersection; } bool JudgePointEqual(Point a, Point b)//жǷ { if(a.x == b.x && a.y == b.y && a.z == b.z) return true; return false; } void ExchangeTwoRay(Ray *ray1, Ray *ray2)// { Ray tempRay; tempRay = *ray1; *ray1 = *ray2; *ray2 = tempRay; }
12,935
#include <iostream> #include <string> #include <math.h> #include <stdio.h> __global__ void backWardMask(char * input, int * mask, int n) { int index = blockIdx.x * blockDim.x + threadIdx.x; int stride = blockDim.x * gridDim.x; if(index == 0) { mask[0] = 1; for(int i = index + stride; i < n; i+=stride) { mask[i] = input[i] == input[i - 1] ? 0 : 1; } return; } for(int i = index; i < n - 1; i+=stride) { mask[i] = input[i] == input[i - 1] ? 0 : 1; } } // NVIDIA's upsweep and downsweep prefix sum (prescan) // TO-DO -> eliminate bank conflicts __global__ void prescan(int *g_odata, int *g_idata, int * blockSums, int n) { extern __shared__ int temp[]; int thid = threadIdx.x; int index = 2 * blockIdx.x * blockDim.x + threadIdx.x; int offset = 1; temp[2*thid] = 0; temp[2*thid+1] = 0; if(index < n) { temp[2*thid] = g_idata[index]; temp[2*thid+1] = g_idata[index+1]; } // build sum in place up the tree for (int d = 2*blockDim.x>>1; d > 0; d >>= 1) { __syncthreads(); if (thid < d) { int ai = offset*(2*thid+1)-1; int bi = offset*(2*thid+2)-1; temp[bi] += temp[ai]; } offset *= 2; } // clear the last element if (thid == 0) { blockSums[blockIdx.x] = temp[2*blockDim.x - 1]; temp[2*blockDim.x - 1] = 0; } // traverse down tree & build scan for (int d = 1; d < 2*blockDim.x; d *= 2) { offset >>= 1; __syncthreads(); if (thid < d) { int ai = offset*(2*thid+1)-1; int bi = offset*(2*thid+2)-1; int t = temp[ai]; temp[ai] = temp[bi]; temp[bi] += t; } } __syncthreads(); temp[2*thid] = temp[2*thid + 1]; if(thid == blockDim.x - 1) { temp[2*thid + 1] = blockSums[blockIdx.x]; } else { temp[2*thid + 1] = temp[2*thid + 2]; } g_odata[index] = temp[2*thid]; g_odata[index+1] = temp[2*thid+1]; } __global__ void addOffsets(int * preScannedMask, int * blockScan) { int index = blockDim.x * blockIdx.x + threadIdx.x; if(blockIdx.x == 0) return; preScannedMask[index] += blockScan[blockIdx.x-1]; } __global__ void compactKernel(int * scannedMask, int * compactedMask, int * totalRuns, int n) { int index = blockIdx.x * blockDim.x + threadIdx.x; int stride = blockDim.x * gridDim.x; if (index == 0) { compactedMask[0] = 0; } for (int i = index; i < n; i+=stride) { if (i == (n - 1)) { compactedMask[scannedMask[i]] = i + 1; *totalRuns = scannedMask[i]; } if (scannedMask[i] != scannedMask[i - 1]) { compactedMask[scannedMask[i] - 1] = i; } } } __global__ void scatterKernel(int * compactedMask, int * totalRuns, int * in, int * symbolsOut, int * countsOut) { int n = *totalRuns; int index = blockIdx.x * blockDim.x + threadIdx.x; int stride = blockDim.x * gridDim.x; for (int i = index; i < n; i+=stride) { int a = compactedMask[i]; int b = compactedMask[i + 1]; symbolsOut[i] = in[a]; countsOut[i] = b - a; } } int main(int argc, char ** argv) { char * in = argv[1]; int input_size = (int)strlen(argv[1]); // GPU variables int * mask; int * host_mask; int * scannedMask; int * host_scannedMask; int * compactedMask; int * totalRuns; char * input; int * block_sums; int * scannedBlockSums; int * bs; int gridSize = (input_size + 1024 - 1) / 1024; //host_mask = (int *)malloc(input_size * sizeof(int)); //host_scannedMask = (int *)malloc(input_size * sizeof(int)); cudaMallocManaged(&input, input_size * sizeof(char)); cudaMallocManaged(&mask, input_size * sizeof(int)); cudaMallocManaged(&scannedMask, input_size * sizeof(int)); cudaMallocManaged(&block_sums, gridSize * sizeof(int)); cudaMallocManaged(&scannedBlockSums, gridSize * sizeof(int)); cudaMallocManaged(&bs, gridSize * sizeof(int)); cudaMemcpy(input, in, input_size * sizeof(char), cudaMemcpyHostToDevice); backWardMask<<<8, 1024>>>(in, mask, input_size); cudaDeviceSynchronize(); prescan<<<gridSize, 1024>>>(scannedMask, mask, block_sums, input_size); cudaDeviceSynchronize(); if(input_size > 2048) { // scan of block scans prescan<<<1,ceil(gridSize)>>>(scannedBlockSums, block_sums, bs, gridSize); cudaDeviceSynchronize(); // add the offset addOffsets<<<gridSize, 2048>>>(scannedMask, scannedBlockSums); } //cudaMemcpy(host_mask, mask, input_size * sizeof(int), cudaMemcpyDeviceToHost); //cudaMemcpy(host_scannedMask, scannedMask, input_size * sizeof(int), cudaMemcpyDeviceToHost); // compactKernel<<<blocks, 512>>>(); // cudaDeviceSynchronize(); // scatterKernel<<<,>>>(); // cudaDeviceSynchronize(); for(int i = 0; i < input_size; i++) { std::cout << input[i] << " "; } std::cout << std::endl; for(int i = 0; i < input_size; i++) { std::cout << mask[i] << " "; } // std::cout << std::endl; // for(int i = 0; i < input_size; i++) { // std::cout << scannedMask[i] << " "; // } // std::cout << std::endl; }
12,936
#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; } __device__ double2 pow(double2 a, int b){ double r = sqrt(a.x*a.x + a.y*a.y); double theta = atan(a.y / a.x); return{pow(r,b)*cos(b*theta),pow(r,b)*sin(b*theta)}; } __global__ void ktorus_wfc(double *x, double *y, double *z, double *items, double winding, double *phi, double2 *wfc){ int gid = getGid3d3d(); int xid = blockDim.x*blockIdx.x + threadIdx.x; int yid = blockDim.y*blockIdx.y + threadIdx.y; int zid = blockDim.z*blockIdx.z + threadIdx.z; double rad = sqrt((x[xid] - items[6]) * (x[xid] - items[6]) + (y[yid] - items[7]) * (y[yid] - items[7])) - 0.5*items[0]; wfc[gid].x = exp(-( pow((rad)/(items[14]*items[15]*0.5),2) + pow((z[zid])/(items[14]*items[17]*0.5),2) ) ); wfc[gid].y = 0.0; }
12,937
#include "includes.h" __global__ void FindPos(int *pos, bool *forest, int text_size, int order, int step) { int text_idx = blockIdx.x * blockDim.x + threadIdx.x; int offset = blockIdx.x*step; if(text_idx < text_size) { if(!forest[offset+blockDim.x+threadIdx.x]) { pos[text_idx] = 0; } else { bool isCurBlock = true; bool isLeftMost = (blockIdx.x < 1); int nodeIdx = blockDim.x+threadIdx.x; int leftBound = blockDim.x; int rightBound = 2*blockDim.x-1; int alignOrder = 0; // bottom-up while(alignOrder != order) { int leftInx; if(nodeIdx-1 < leftBound) { if(isLeftMost) break; isCurBlock = false; leftInx = offset-step+rightBound; } else { leftInx = offset+nodeIdx-1; } if(!forest[leftInx]) break; rightBound = leftBound-1; leftBound /= 2; nodeIdx /= 2; alignOrder++; } // top-down if(alignOrder == order && !isLeftMost) isCurBlock = false; nodeIdx = (!isCurBlock)? rightBound :(nodeIdx-1 < leftBound)? nodeIdx :nodeIdx-1; offset = offset - ((isCurBlock)? 0:step); while(alignOrder != 0) { if((alignOrder == order && isCurBlock) || forest[offset+2*nodeIdx+1]) { nodeIdx = 2*nodeIdx; } else { nodeIdx = 2*nodeIdx+1; } alignOrder--; } pos[text_idx] = (isCurBlock)? (threadIdx.x-(nodeIdx-blockDim.x)+(forest[offset+nodeIdx])) :(step-nodeIdx+threadIdx.x); } } }
12,938
const int threadsPerBlock = 128; unsigned int hostOffsetSize; unsigned int hostTargetSize; unsigned int hostSourceSize; static unsigned int is_set=0; static unsigned int deviceOffsetSize; static unsigned int deviceTargetSize; static unsigned int deviceSourceSize; static int *deviceOffset; static float *deviceTargetX; static float *deviceTargetY; static float *deviceTargetZ; static float *deviceTargetW; static float *deviceSourceX; static float *deviceSourceY; static float *deviceSourceZ; static float *deviceSourceG; __global__ void kernel(int* deviceOffset, float* deviceTargetX, float* deviceTargetY, float* deviceTargetZ, float* deviceTargetW, float sigma, float* deviceSourceX, float* deviceSourceY, float* deviceSourceZ, float* deviceSourceG) { int i = blockIdx.x * threadsPerBlock + threadIdx.x; int jbase,jsize,jblok,j,jb,jj; float targetX,targetY,targetZ,targetW,dx,dy,dz,coef; __shared__ float sharedSourceX[threadsPerBlock]; __shared__ float sharedSourceY[threadsPerBlock]; __shared__ float sharedSourceZ[threadsPerBlock]; __shared__ float sharedSourceG[threadsPerBlock]; targetX = deviceTargetX[i]; targetY = deviceTargetY[i]; targetZ = deviceTargetZ[i]; targetW = 0; coef = 0.5f/(sigma*sigma); jbase = deviceOffset[blockIdx.x]; jsize = deviceOffset[blockIdx.x+1]-deviceOffset[blockIdx.x]; jblok = (jsize + threadsPerBlock - 1) / threadsPerBlock; for (j = 0; j < jblok-1; j++) { jb = jbase + j * threadsPerBlock + threadIdx.x; __syncthreads(); sharedSourceX[threadIdx.x] = deviceSourceX[jb]; sharedSourceY[threadIdx.x] = deviceSourceY[jb]; sharedSourceZ[threadIdx.x] = deviceSourceZ[jb]; sharedSourceG[threadIdx.x] = deviceSourceG[jb]; __syncthreads(); #pragma unroll 32 for(jj = 0; jj < threadsPerBlock; jj++){ dx = targetX-sharedSourceX[jj]; dy = targetY-sharedSourceY[jj]; dz = targetZ-sharedSourceZ[jj]; targetW += sharedSourceG[jj]*exp(-(dx*dx+dy*dy+dz*dz)*coef); } } jb = jbase + j * threadsPerBlock + threadIdx.x; __syncthreads(); sharedSourceX[threadIdx.x] = deviceSourceX[jb]; sharedSourceY[threadIdx.x] = deviceSourceY[jb]; sharedSourceZ[threadIdx.x] = deviceSourceZ[jb]; sharedSourceG[threadIdx.x] = deviceSourceG[jb]; __syncthreads(); for(jj = 0; jj < jsize - (j * threadsPerBlock); jj++){ dx = targetX-sharedSourceX[jj]; dy = targetY-sharedSourceY[jj]; dz = targetZ-sharedSourceZ[jj]; targetW += sharedSourceG[jj]*exp(-(dx*dx+dy*dy+dz*dz)*coef); } deviceTargetW[i] = targetW/M_PI*coef; } void gpumatmult(float *hostTargetX, float *hostTargetY, float *hostTargetZ, float *hostTargetW, float *hostSourceX, float *hostSourceY, float *hostSourceZ, float *hostSourceG, int *hostOffset, int iblok, float sigma, int numCluster, int numTrunc) { hostOffsetSize = sizeof(int) * (numCluster+1); hostTargetSize = sizeof(float) * numCluster * threadsPerBlock; hostSourceSize = sizeof(float) * numCluster * numTrunc; if (is_set==0) { cudaSetDevice(0); is_set=1; } if (hostOffsetSize>deviceOffsetSize) { if(deviceOffsetSize!=0) cudaFree(deviceOffset); cudaMalloc((void**)&deviceOffset,hostOffsetSize); deviceOffsetSize=hostOffsetSize; } if (hostTargetSize>deviceTargetSize) { if(deviceTargetSize!=0) { cudaFree(deviceTargetX); cudaFree(deviceTargetY); cudaFree(deviceTargetZ); cudaFree(deviceTargetW); } cudaMalloc((void**)&deviceTargetX,hostTargetSize); cudaMalloc((void**)&deviceTargetY,hostTargetSize); cudaMalloc((void**)&deviceTargetZ,hostTargetSize); cudaMalloc((void**)&deviceTargetW,hostTargetSize); deviceTargetSize=hostTargetSize; } if (hostSourceSize>deviceSourceSize) { if(deviceSourceSize!=0) { cudaFree(deviceSourceX); cudaFree(deviceSourceY); cudaFree(deviceSourceZ); cudaFree(deviceSourceG); } cudaMalloc((void**)&deviceSourceX,hostSourceSize); cudaMalloc((void**)&deviceSourceY,hostSourceSize); cudaMalloc((void**)&deviceSourceZ,hostSourceSize); cudaMalloc((void**)&deviceSourceG,hostSourceSize); deviceSourceSize=hostSourceSize; } cudaMemcpy(deviceOffset,hostOffset,hostOffsetSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceTargetX,hostTargetX,hostTargetSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceTargetY,hostTargetY,hostTargetSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceTargetZ,hostTargetZ,hostTargetSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceSourceX,hostSourceX,hostSourceSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceSourceY,hostSourceY,hostSourceSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceSourceZ,hostSourceZ,hostSourceSize,cudaMemcpyHostToDevice); cudaMemcpy(deviceSourceG,hostSourceG,hostSourceSize,cudaMemcpyHostToDevice); dim3 block(threadsPerBlock); dim3 grid(iblok); kernel<<< grid, block >>>(deviceOffset,deviceTargetX,deviceTargetY,deviceTargetZ,deviceTargetW, sigma,deviceSourceX,deviceSourceY,deviceSourceZ,deviceSourceG); cudaMemcpy(hostTargetW,deviceTargetW,hostTargetSize,cudaMemcpyDeviceToHost); }
12,939
extern "C" __global__ void crossEntropyCostDerivative(float *desiredOutput, unsigned int length, float *networkOutput, float* result) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < length; i += blockDim.x * gridDim.x) { result[i]=-desiredOutput[i]/(0.00001f+networkOutput[i])+(1.0f-desiredOutput[i])/(1.00001f-networkOutput[i]); } }
12,940
#include <stdio.h> #include <cuda_runtime_api.h> __global__ void cuda_add(int a, int b, int *result) { *result = a + b; } int main() { int a, b; int h_result, *d_result; printf("Pleas, enter two ints\n"); scanf("%d%d", &a, &b); cudaMalloc((void **)&d_result, sizeof(int)); cuda_add<<<1, 1>>>(a, b, d_result); cudaMemcpy(&h_result, d_result, sizeof(int), cudaMemcpyDeviceToHost); cudaFree(d_result); printf("a + b == %d\n", h_result); return 0; }
12,941
/********************************************************************* * Copyright © 2011-2012, * Marwan Abdellah: <abdellah.marwan@gmail.com> * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation. * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, * MA 02110-1301, USA. ********************************************************************/ /*! * CUDA : This kernel adds a constant value to the input vector into the * output vector with length N. * * @param devArrayInput * Input vector. * * @param constVal * Constant value to be added to the input device vector. * * @param devArrayOutput * Sum vector. * * @param N * Vector length. * * @author * Marwan Abdellah <abdellah.marwan@gmail.com> * * @date * Created: August, 2012. * @date * Last Update: September, 2012. * * @note * Minimum CUDA version 3.2. * @note * Minimum Device Compute Capability 1.0. */ template <typename T> __global__ void Constant_Add_1D_Array_Kernel(T* devArrayInput, T constVal, T* devArrayOutput, int N) { int xThreadIdx = threadIdx.x; int blockWidth = blockDim.x; int index = blockIdx.x * blockWidth + xThreadIdx; #ifdef VEC_CHECK if (index < N) devArrayOutput[index] = (T) ((T) devArrayInput[index] + (T) constVal); #else devArrayOutput[index] = (T) ((T) devArrayInput[index] + (T) constVal); #endif }
12,942
#include "includes.h" __global__ void reposition (double4 *ac, double4 *ac1, double4 *ac2, double4 *af, unsigned long nextsize) { unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; if(i < nextsize){ af[i] = ac[i]; af[i + nextsize] = ac1[i]; af[i + 2*nextsize] = ac2[i]; } }
12,943
/* Author : Kim, KyoungHo (rain_woo@korea.ac.kr) Ki-Hwan Kim (wbkifun@korea.ac.kr) Written date : 2009. 6. 11 last update : Copyright : GNU GPL */ __global__ void initmem( int Ntot, int idx0, float *a ) { int idx = blockIdx.x*blockDim.x + threadIdx.x + idx0; if ( idx < Ntot ) a[idx] = 0; }
12,944
#include<iostream> #include<ctime> using namespace std; template<unsigned int DIMX, unsigned int DIMY, typename T> __global__ void transpose(T *in_data, T *out_data, unsigned int nx, unsigned int ny) { //padding = 2 __shared__ T tile[DIMY][DIMX*2 + 2]; unsigned int idx = threadIdx.x + blockDim.x * blockIdx.x * 2; unsigned int idy = threadIdx.y + blockDim.y * blockIdx.y; if(idx + blockDim.x < nx && idy < ny) { tile[threadIdx.y][threadIdx.x] = in_data[idy * nx + idx]; tile[threadIdx.y][threadIdx.x + blockDim.x] = in_data[idy * nx + idx + blockDim.x]; __syncthreads(); unsigned int posB = threadIdx.y * blockDim.x + threadIdx.x; unsigned int column = posB / blockDim.y; unsigned int row = posB % blockDim.y; idx = column + blockDim.x * blockIdx.x * 2; idy = row + blockDim.y * blockIdx.y; out_data[idx * ny + idy] = tile[row][column]; out_data[(idx + blockDim.x) * ny + idy] = tile[row][column + blockDim.x]; } } template<typename T> void transposeHost(T *in, T* out, unsigned int nx, unsigned int ny) { for(int i = 0;i < nx;++i) { for(int j = 0;j < ny;++j) { out[i * ny + j] = in[j * nx + i]; } } } int main(int argc, char *argv[]) { unsigned int nx = 1 << 9; unsigned int ny = 1 << 9; constexpr unsigned int blockx = 32; constexpr unsigned int blocky = 32; clock_t start, end; int in[nx * ny], out[nx * ny], *in_dev, *out_dev; auto init = [](auto*in ,unsigned int size)->void { for(int i = 0;i < size;++i) { in[i] = random()%1000; } }; init(in, nx * ny); cudaMalloc((void**)&in_dev, sizeof(in)); cudaMalloc((void**)&out_dev, sizeof(in)); cudaMemcpy(in_dev, in ,sizeof(in), cudaMemcpyHostToDevice); cudaDeviceSynchronize(); transposeHost(in, out, nx, ny); dim3 block(blockx, blocky); dim3 grid((nx + blockx - 1) / blockx / 2, (ny + blocky - 1) / blocky); start = clock(); transpose<blockx, blocky><<<grid, block>>>(in_dev, out_dev, nx,ny); cudaDeviceSynchronize(); end = clock(); cout <<" gpu time: " << end - start<<endl; cudaMemcpy(in, out_dev,sizeof(in), cudaMemcpyDeviceToHost); cudaDeviceSynchronize(); cudaFree(in_dev); cudaFree(out_dev); int n = 0; for (int i = 0;i < nx * ny;++i) { if(out[i] != in[i]) { n++; } } cout << n << endl; return 0; }
12,945
#include<stdio.h> __global__ void sum(int *a_d, int n,int* maxsum){ int strid=n/2; //printf("fas"); int t=threadIdx.x; while(strid>=1){ //printf("af"); __syncthreads(); if(t<strid){ a_d[t]=a_d[t]+a_d[strid+t]; // printf("threadid=%d val=%d\n",t,a_d[t]); } strid/=2; } maxsum[0]=a_d[0]; } int main(){ int n=512; int a[n]; for(int i=0; i<n; i++){ a[i]=i; } int *a_d,*maxsum; cudaMalloc((void**)&a_d,n*sizeof(int)); cudaMalloc((void**)&maxsum,sizeof(int)); //cudaMalloc((void**)&n_d,sizeof(int)); cudaMemcpy(a_d,a,n*sizeof(int),cudaMemcpyHostToDevice); //for(int i=0; i<n; i++) //printf("%d ",a[i]); //printf("\n"); sum<<<1,n>>>(a_d,n,maxsum); int maxi[n]; int max_val[1]; cudaMemcpy(maxi,a_d,n*sizeof(int),cudaMemcpyDeviceToHost); cudaMemcpy(max_val,maxsum,sizeof(int),cudaMemcpyDeviceToHost); // for(int i=0; i<n; i++) printf("%d ",max_val); printf("%d ",maxi[0]); return 0; }
12,946
#include "CosmicConstants.cuh" #include <stdio.h> __global__ void mykernel() { printf("value is %f\n", K0); } void setDT(float newDT) { cudaMemcpyToSymbol(dt, &newDT, sizeof(newDT)); } void setK0(float newK0) { mykernel << <1, 1 >> >(); cudaMemcpyToSymbol(K0, &newK0, sizeof(newK0), 0, cudaMemcpyHostToDevice); mykernel << <1, 1 >> >(); }
12,947
#include "includes.h" #define SIZ 20 #define num_inp 4 using namespace std; typedef struct edge { int first, second; } edges; __global__ void w1_kernel(double * grads_W1, double * W1, double learning_rate, int size) { int i = blockIdx.x; int j = threadIdx.x; W1[i*size + j] += (-learning_rate * grads_W1[i*size + j]); }
12,948
#include "includes.h" __global__ void Cos( float * x, size_t idx, size_t N, float W0) { for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < N; i += blockDim.x * gridDim.x) { x[(idx-1)*N+i] = cos ( W0*x[(idx-1)*N+i] ); } return; }
12,949
//xfail:BOOGIE_ERROR //--blockDim=512 --gridDim=64 --loop-unwind=2 --no-inline //kernel.cu: error: possible write-write race on B #include <cuda.h> #include <cuda_runtime_api.h> #include <stdio.h> #include <assert.h> #define N 2//512 extern "C" { __global__ void helloCUDA(float *A) { __shared__ float B[256]; for(int i = 0; i < N*2; i ++) { B[i] = A[i]; } } } int main() { float *A; float *dev_A; float size= N*sizeof(float); A=(float*)malloc(size); cudaMalloc((void**)&dev_A, size); for (int i = 0; i < N; i++) A[i] = 5; cudaMemcpy(dev_A, A, size, cudaMemcpyHostToDevice); helloCUDA<<<64,N>>>(dev_A); //ESBMC_verify_kernel(helloCUDA, 1, N, dev_A); cudaMemcpy(A, dev_A, size, cudaMemcpyDeviceToHost); cudaFree(dev_A); free(A); }
12,950
#include <stdio.h> #include <assert.h> #include <cuda_runtime.h> #include <stdlib.h> #include <time.h> #include <math.h> #include <string.h> #define BLOCK_SIZE 16 void printDeviceProp(const cudaDeviceProp &prop) { printf("Device Name : %s.\n", prop.name); printf("totalGlobalMem : %lu.\n", prop.totalGlobalMem); printf("sharedMemPerBlock : %lu.\n", prop.sharedMemPerBlock); printf("regsPerBlock : %d.\n", prop.regsPerBlock); printf("warpSize : %d.\n", prop.warpSize); printf("memPitch : %lu.\n", prop.memPitch); printf("maxThreadsPerBlock : %d.\n", prop.maxThreadsPerBlock); printf("maxThreadsDim[0 - 2] : %d %d %d.\n", prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2]); printf("maxGridSize[0 - 2] : %d %d %d.\n", prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2]); printf("totalConstMem : %lu.\n", prop.totalConstMem); printf("major.minor : %d.%d.\n", prop.major, prop.minor); printf("clockRate : %d.\n", prop.clockRate); printf("textureAlignment : %lu.\n", prop.textureAlignment); printf("deviceOverlap : %d.\n", prop.deviceOverlap); printf("multiProcessorCount : %d.\n", prop.multiProcessorCount); } bool InitCUDA() { int count; cudaGetDeviceCount(&count); if (count == 0) { fprintf(stderr, "There is no device.\n"); return false; } int i; for (i = 0; i < count; i++) { cudaDeviceProp prop; cudaGetDeviceProperties(&prop, i); printDeviceProp(prop); if (cudaGetDeviceProperties(&prop, i) == cudaSuccess) { if (prop.major >= 1) { break; } } } if (i == count) { fprintf(stderr, "There is no device supporting CUDA 1.x.\n"); return false; } cudaSetDevice(i); return true; } //CPU void matrixMulCPU(float* res,const float *matrixA,const float *matrixB,int colsA,int rowsA,int rowsB) { float sum = 0; for (int i = 0; i < rowsA; ++i) { for (int j = 0; j < rowsB; ++j) { sum = 0; for (int k = 0; k < colsA; ++k) { sum += (float)matrixA[i*colsA+k]*(float)matrixB[k*rowsB+ j]; } res[i*rowsB+j] = (float)sum; } } } // GPU // C(i,j) = sum{A(i, k)* B(k ,j)} // each thread cal C(i, j) __global__ void matrixMulGPUKernal0(float* matrixC,const float* matrixA,const float *matrixB,int colsA,int rowsB) { float sum = 0; int row = blockIdx.y*blockDim.y + threadIdx.y; int col = blockIdx.x*blockDim.x + threadIdx.x; for (int i = 0; i < colsA; ++i) { sum += matrixA[row*colsA + i] * matrixB[i*rowsB + col]; } matrixC[row*rowsB + col] = sum; } // Csub(i,j) = sum{A(i,ksub+offsetA)*B(ksub+offsetB,j)} 0 <= ksub < blockSize // C(i,j) = sum{Csub(i,j)} // each thread cal each block __global__ void matrixMulGPUKernal1(float* matrixC,const float* matrixA,const float *matrixB,int colsA,int rowsB) { int bx = blockIdx.x; int by = blockIdx.y; int tx = threadIdx.x; int ty = threadIdx.y; int aBegin = colsA*(by*BLOCK_SIZE);//A(0,by) int aEnd = aBegin + colsA - 1; int aStep = BLOCK_SIZE;//offsetA int bBegin = BLOCK_SIZE*bx;//B(bx,0) int bStep = BLOCK_SIZE*rowsB;//offsetB float cSub = 0; for (int a = aBegin,b = bBegin; a <= aEnd; a += aStep,b += bStep) { __shared__ float As[BLOCK_SIZE][BLOCK_SIZE]; __shared__ float Bs[BLOCK_SIZE][BLOCK_SIZE]; As[ty][tx] = matrixA[a + colsA*ty + tx]; Bs[ty][tx] = matrixB[b + rowsB*ty + tx]; __syncthreads(); //i * j for each thread for (int k = 0; k < BLOCK_SIZE; ++k) { cSub += As[ty][k]*Bs[k][tx]; } __syncthreads(); } int cIndex = (by*BLOCK_SIZE + ty)*rowsB + (bx*BLOCK_SIZE + tx); matrixC[cIndex] = cSub; } void copyFromCPUToGPU(const float *matrixA, float *d_a, int n) { cudaMemcpy(d_a, matrixA, sizeof(float) * n, cudaMemcpyHostToDevice); } void copyFromGPUToCPU(const float *d_c, float *matrixC, int n) { cudaMemcpy(matrixC, d_c, sizeof(float) * n, cudaMemcpyDeviceToHost); } void matrixMulGPU(float* matrixC,const float *matrixA,const float *matrixB,int colsA,int rowsA,int rowsB) { float *d_a, *d_b, *d_c; cudaMalloc((void**) &d_a, sizeof(float) * colsA*rowsA); cudaMalloc((void**) &d_b, sizeof(float) * rowsB*colsA); cudaMalloc((void**) &d_c, sizeof(float) * rowsB*rowsA); copyFromCPUToGPU(matrixA,d_a,colsA*rowsA); copyFromCPUToGPU(matrixB,d_b,rowsB*colsA); dim3 blocks(rowsB/BLOCK_SIZE, rowsA/BLOCK_SIZE); dim3 threads(BLOCK_SIZE, BLOCK_SIZE); float time_elapsed = 0; cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start, 0); matrixMulGPUKernal0<<<blocks,threads>>>(d_c,d_a,d_b,colsA,rowsA); cudaEventRecord(stop, 0); cudaEventSynchronize(start); cudaEventSynchronize(stop); cudaEventElapsedTime(&time_elapsed, start, stop); cudaEventDestroy(start); cudaEventDestroy(stop); printf(" - Running time: %f ms\n", time_elapsed); double gflop = (2.0 * (double)colsA * colsA * colsA) * 0.000001; printf(" - GFlop: %.5f GFlop/sec\n\n", gflop/time_elapsed); cudaThreadSynchronize(); copyFromGPUToCPU(d_c,matrixC,rowsB*rowsA); cudaFree(d_a); cudaFree(d_b); cudaFree(d_c); } void randomInit(float* _data,int size) { for (int i = 0; i < size; ++i) { _data[i] = rand()/(float)RAND_MAX; } } bool checkError(const float* matrixA, const float* matrixB, int size) { for (int i = 0 ; i < size; ++i) { if (fabs(matrixA[i] - matrixB[i]) > 1.0e-3) { printf(" ! Wrong index: %d\n", i); printf("%f \t %f\n",matrixA[i],matrixB[i]); return false; } } return true; } int main(int argc, char* argv[]) { if (!InitCUDA()) return 0; srand(63); printf("\n - BLOCK_SIZE: %d\n", BLOCK_SIZE); int N = (1 << 11); int colsA, colsB, colsC, rowsA, rowsB, rowsC; colsA = colsB = colsC = rowsA = rowsB = rowsC = N; printf(" - Matrix size: %d * %d\n", rowsC, rowsC); float* A , *B, *C, *C2; A = (float*) malloc(sizeof(float) * colsA * rowsA); B = (float*) malloc(sizeof(float) * colsB * rowsB); randomInit(A,colsA*rowsA); randomInit(B,colsB*rowsB); C = (float*) malloc(sizeof(float) * colsC * rowsC); memset(C,0,sizeof(float)*colsC*rowsC); C2 = (float*) malloc(sizeof(float) * colsC * rowsC); memset(C2,0,sizeof(float)*colsC*rowsC); clock_t tick1 = clock(); matrixMulCPU(C2,A,B,colsA,rowsA,colsB); printf(" - CPU use Time : %f ms\n",(double)(clock() - tick1)/CLOCKS_PER_SEC); // unsigned int timer = 0; // cutilCheckError(cutCreateTimer(&timer)); // cutilCheckError(cutStartTimer(timer)); matrixMulGPU(C,A,B,colsA,rowsA,colsB); // cutilCheckError(cutStopTimer(timer)); // printf("GPU use time: %f (ms) \n", cutGetTimerValue(timer)); // cutilCheckError(cutDeleteTimer(timer)); if (checkError(C,C2,colsC*rowsC)) { printf("Right Answer!\n"); }else{ printf("Worng Answer!\n"); } free(A); free(B); free(C); free(C2); return 0; }
12,951
extern "C" __global__ void sumSquareError (int nBatch, int rbs, int rScale, int nCoeff, float *DA, float *CA, float *EA, float *SA) { int i = blockIdx.x * blockDim.x + threadIdx.x; if (i < nBatch) { const int daOffset = i * rbs * rScale * nCoeff; const int caOffset = i * nCoeff; const int eaOffset = i * rbs * rScale; SA[i] = 0; for(int j = 0; j < rbs * rScale ; j++){ float fx = 0.0f; for(int k = 0 ; k < nCoeff ; k++){ fx += DA[daOffset + rbs * rScale * k + j] * CA[caOffset + k]; } float error = EA[eaOffset + j] - fx; SA[i] += error*error; // sum square error } } }
12,952
#include<cuda.h> #include<iostream> __global__ void divergence(/* volatile */ int* data) { int i = threadIdx.x + blockDim.x * blockIdx.x; if ((i & 0x01) == 0) { data[i+1] = data[i+1] + i; // if even, come here //__threadfence_block(); } else { data[i] = data[i] + 2*i; // if odd, come here } } int main() { const int numElems = 4; int hostArray[numElems], *devArray; //allocate memory on the device (GPU); zero out all entries in this device array cudaMalloc((void**)&devArray, sizeof(int) * numElems); cudaMemset(devArray, 0, numElems * sizeof(int)); //invoke GPU kernel, with one block that has four threads divergence <<<1, numElems >>>(devArray); //bring the result back from the GPU into the hostArray cudaMemcpy(&hostArray, devArray, sizeof(int) * numElems, cudaMemcpyDeviceToHost); //print out the result to confirm that things are looking good std::cout << "Values stored in hostArray: " << std::endl; for (int i = 0; i < numElems; i++) std::cout << hostArray[i] << std::endl; //release the memory allocated on the GPU cudaFree(devArray); return 0; }
12,953
#include <stdio.h> #include <stdlib.h> #include <math.h> #include <time.h> #include <cuda.h> #include <cuda_runtime.h> #include <curand_kernel.h> __global__ void mult(int *a, int *b, int *c, int n) { int row = blockIdx.y * blockDim.y + threadIdx.y; int col = blockIdx.x * blockDim.x + threadIdx.x; int sum = 0; if (col < n && row < n) { for (int i = 0; i < n; i++) sum += a[row * n + i] * b[i * n + col]; c[row * n + col] = sum; } } int main() { int n; int i, j, k; int *a, *b, *c; int *dev_a, *dev_b, *dev_c; printf("Please enter the size of matrix: \n"); scanf("%d", &n); cudaMalloc((void**)&dev_a, sizeof(int) * n * n); cudaMalloc((void**)&dev_b, sizeof(int) * n * n); cudaMalloc((void**)&dev_c, sizeof(int) * n * n); cudaMallocHost((void**)&a, sizeof(int) * n * n); cudaMallocHost((void**)&b, sizeof(int) * n * n); cudaMallocHost((void**)&c, sizeof(int) * n * n); for (i = 0; i < n; i++){ for (j = 0; j < n; j++){ a[i * n + j] = round(rand() % 2); b[i * n + j] = round(rand() % 2); } } printf("Start calculating...\n"); clock_t start_time = clock(); cudaMemcpy(dev_a, a, sizeof(int) * n * n, cudaMemcpyHostToDevice); cudaMemcpy(dev_b, b, sizeof(int) * n * n, cudaMemcpyHostToDevice); mult<<<n, n>>>(dev_a, dev_b, dev_c, n); clock_t end_time = clock(); printf("Time consuming of calculating %dx%d matrix using GPU is %f ms.\n", n, n, static_cast<double>(end_time - start_time)/CLOCKS_PER_SEC*1000); cudaFree(dev_a); cudaFree(dev_b); cudaFree(dev_c); cudaFreeHost(a); cudaFreeHost(b); cudaFreeHost(c); return 0; }
12,954
//#include "Field.hpp" //#include "cuda_runtime.h" //#include "device_launch_parameters.h" //we are computing FFT of the size 2**N #define N 1024 #define LOG_N 10 /*using namespace arithmetic; std::initializer_list<uint_default_t> p = { 0x2370fb049d410fbe, 0x4e761a9886e50241, 0x7d023f4018000001, 0x7e80600000000001 }; static const bignum<MAX_BITSIZE> modulus(p); using field = Field<MAX_BITSIZE, modulus>; __constant__ field dev_roots_of_unity[LOG_N]; __global__ void FFTKernel(field* input_buf, field* output_buf); int main() { //first find suitable Cuda device //TBD: or split between several CUDA devices if possible int device_count; cudaError_t cudaStatus = cudaGetDeviceCount(&device_count); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaGetDeviceCount failed!"); return 1; } if (device_count == 0) { fprintf(stderr, "No suitable CUDA devices were found!"); return 1; } cudaDeviceProp prop; cudaStatus = cudaGetDeviceProperties(&prop, 0); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaGetDeviceCount failed!"); return 1; } //TODO: check if there are enough constant memory and other additional checks //set appropriate device // Choose which GPU to run on, change this on a multi-GPU system. cudaStatus = cudaSetDevice(0); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaSetDevice failed! Do you have a CUDA-capable GPU installed?"); goto Error; } field input[N]; field output[N]; //we are goint to calculate roots of unity in corresponding powers field roots_of_unity[LOG_N]; //TODO: compute them in advance before the kernel starts for (size_t i = 0; i < N; i++) input[i] = field::random(); field* dev_input = nullptr; field* dev_output = nullptr; // Allocate GPU buffers for three vectors (one input, one output) . cudaStatus = cudaMalloc((void**)&dev_input, N * sizeof(field)); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMalloc failed!"); goto Error; } cudaStatus = cudaMalloc((void**)&dev_output, N * sizeof(field)); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMalloc failed!"); goto Error; } // Copy input vector from host memory to GPU buffers. cudaStatus = cudaMemcpy(dev_input, input, N * sizeof(field), cudaMemcpyHostToDevice); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMemcpy failed!"); goto Error; } //copy precomputed roots of unity to constant memory cudaMemcpyToSymbol(dev_roots_of_unity, roots_of_unity, LOG_N * sizeof(field)); // Launch a kernel on the GPU with one thread for each element. FFTKernel << <1, size >> > (dev_input, dev_output); // Check for any errors launching the kernel cudaStatus = cudaGetLastError(); if (cudaStatus != cudaSuccess) { fprintf(stderr, "FFTKernel launch failed: %s\n", cudaGetErrorString(cudaStatus)); goto Error; } // cudaDeviceSynchronize waits for the kernel to finish, and returns // any errors encountered during the launch. cudaStatus = cudaDeviceSynchronize(); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching addKernel!\n", cudaStatus); goto Error; } // Copy output vector from GPU buffer to host memory. cudaStatus = cudaMemcpy(output, dev_output, N * sizeof(field), cudaMemcpyDeviceToHost); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMemcpy failed!"); goto Error; } Error: cudaFree(dev_input); cudaFree(dev_output); return 0; } //NB: we have precomputed the powers of roots of unity and put them into constant memory __global__ void FFTKernel(const field* input_buf, field* output_buf) { field temp_buf[N]; const field* in_buf = input_buf; field* out_buf = (LOG_N % 2 ? output_buf : temp_buf); int thread_idx = threadIdx.x + blockIdx.x * blockDim.x; for (size_t i = 0; i < LOG_N; i++) { int idx = thread_idx; while (idx < N) { idx += blockDim.x * gridDim.x; } } }*/
12,955
#include "includes.h" __global__ void fupdate_dummy(float *f, float *z1, float *z2, float *g, float invlambda, int nx, int ny) { int px = blockIdx.x * blockDim.x + threadIdx.x; int py = blockIdx.y * blockDim.y + threadIdx.y; int idx = px + py*nx; float DIVZ; if (px<nx && py<ny) { // compute the divergence DIVZ = 0; float Z1c = z1[(idx)]; float Z2c = z2[(idx)]; // float Z1l=z1[(idx-1 )]; // float Z2d=z2[(idx-nx)]; if (!(px == (nx - 1))) DIVZ += Z1c; // if (!(px==0)) DIVZ -= Z1l; if (!(py == (ny - 1))) DIVZ += Z2c; // if (!(py==0)) DIVZ -= Z2d; // update f f[idx] = DIVZ - g[idx] * invlambda; } }
12,956
/* * a simple test */ __shared__ float data1[32][32]; __shared__ float data2[32][32]; __shared__ float data3[32][32]; __device__ void mult(float d1[32][32], float d2[32][32], float d3[32][32], int idx0, int idx1, int idx2) { int i; int iv0 = 0; int iv1 = 0; int iv2 = 0; for (i = 0; i < 32; i++) { d1[iv0][idx0] = d2[iv1][idx1] + d3[iv2][idx2]; iv0 += 1; iv1 += 1; iv2 += 1; } } __global__ void doit(int start, int end) { int i; int id0 = start; int id1 = start; int id2 = start; for (i = start; i < end; i++) { mult(data1, data1, data1, id0, id1, id2); id0 += 1; id1 += 1; id2 += 1; } } __device__ void mult1(float d1[32][32], float d2[32][32], float d3[32][32], int idx0, int idx1, int idx2) { int i; int iv0 = 0; int iv1 = 0; int iv2 = 0; for (i = 0; i < 32; i++) { d1[iv0][idx0] = d2[iv1][idx1] + d3[iv2][idx2]; iv0 += 1; iv1 += 1; iv2 += 1; } } __global__ void doit1(int start, int end) { int i; int id0 = start; int id1 = start; int id2 = start; for (i = start; i < end; i++) { mult(data1, data1, data1, id0, id1, id2); id0 += 1; id1 += 1; id2 += 1; } }
12,957
#include<stdio.h> #include<cuda_runtime.h> #include<device_launch_parameters.h> __device__ int intToBin(int x){ return 5; } __global__ void add(int *A, int *B, int m, int n) { int row = threadIdx.y; int col = threadIdx.x; if ((row%(m-1) == 0) || (col%(n-1) == 0)){ B[row*n + col] = A[row*n + col]; }else{ B[row*n + col] = intToBin(A[row*n + col]); } } int main(){ int a[100], b[100], n, m; printf("Enter m: "); scanf("%d",&m); printf("Enter n: "); scanf("%d",&n); printf("Enter Matrix:\n"); for(int i=0;i<n*m;i++) scanf("%d",&a[i]); int *d_a,*d_b; int size = sizeof(int)*m*n; cudaMalloc((void**)&d_a,size); cudaMalloc((void**)&d_b,size); cudaMemcpy(d_a,&a,size,cudaMemcpyHostToDevice); cudaMemcpy(d_b,&b,size,cudaMemcpyHostToDevice); dim3 block(m, n, 1); add<<<1, block>>>(d_a, d_b, m, n); cudaMemcpy(&b,d_b,size,cudaMemcpyDeviceToHost); for(int i=0;i<n*m;i++){ if (i % n == 0) printf("\n"); printf("%d ",b[i]); } printf("\n"); cudaFree(d_a); cudaFree(d_b); }
12,958
#include<iostream> #include<string> #include<cstring> #include<ctime> #include<cstdlib> #include<sys/time.h> #include<stdio.h> #include<iomanip> /* we need these includes for CUDA's random number stuff */ #include<curand.h> #include<curand_kernel.h> using namespace std; #define MAX 26 //int a[1000]; //array of all possible password characters int b[1000]; //array of attempted password cracks unsigned long long tries = 0; char alphabet[] = { 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z' }; size_t result = 1000 * sizeof(float); int *a = (int *) malloc(result); void serial_passwordCrack(int length){ bool cracked = false; do{ b[0]++; for(int i =0; i<length; i++){ if (b[i] >= 26 + alphabet[i]){ b[i] -= 26; b[i+1]++; }else break; } cracked=true; for(int k=0; k<length; k++) if(b[k]!=a[k]){ cracked=false; break; } if( (tries & 0x7ffffff) == 0 ) cout << "\r \r "; else if( (tries & 0x1ffffff) == 0 ) cout << "."; tries++; }while(cracked==false); } __global__ void parallel_passwordCrack(int length,int*d_output,int *a) { int idx = blockIdx.x * blockDim.x + threadIdx.x; bool cracked = false; char alphabetTable[] = { 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z' }; int newB[1000]; __shared__ int nIter; __shared__ int idT; __shared__ long totalAttempt; do{ if(idx == 0){ nIter = 0; totalAttempt = 0; } newB[0]++; for(int i =0; i<length; i++){ if (newB[i] >= 26 + alphabetTable[i]){ newB[i] -= 26; newB[i+1]++; }else break; } cracked=true; for(int k=0; k<length; k++) { if(newB[k]!=a[k]){ cracked=false; break; }else { cracked = true; } } if(cracked && nIter == 0){ idT = idx; break; } else if(nIter){ break; } totalAttempt++; }while(!cracked || !nIter); if(idx == idT){ for(int i = 0; i< length; i++){ d_output[i] = newB[i]; } } } long long start_timer() { struct timeval tv; gettimeofday(&tv, NULL); return tv.tv_sec * 1000000 + tv.tv_usec; } // Prints the time elapsed since the specified time long long stop_timer(long long start_time, std::string name) { struct timeval tv; gettimeofday(&tv, NULL); long long end_time = tv.tv_sec * 1000000 + tv.tv_usec; std::cout << std::setprecision(5); std::cout << name << ": " << ((float) (end_time - start_time)) / (1000 * 1000) << " sec\n"; return end_time - start_time; } int main() { int length; //length of password int random; //random password to be generated int *d_input = (int *) malloc(result); cout << "Enter a password length: "; cin >> length; int *h_gpu_result = (int*)malloc(1000*sizeof(int)); srand(time(NULL)); //generating random password cout << "Random generated password: " << endl; for (int i =0; i<length; i++){ random = alphabet[(rand()%26)]; a[i] = random; //adding random password to array cout << char(a[i]); }cout << "\n" << endl; long long serial_start_time = start_timer(); cout << "Serial Password Cracked: " << endl; serial_passwordCrack(length); cout << "\n"; long long serial_end_time = stop_timer(serial_start_time, "\nSerial Run Time"); for(int i=0; i<length; i++){ cout << char(b[i]); }cout << "\nNumber of tries: " << tries << endl; //long long serial_end_time = stop_timer(serial_start_time, "\nSerial Run Time"); //declare GPU memory pointers int *d_output; //allocate GPU memory cudaMalloc((void **) &d_output,1000*sizeof(int)); cudaMalloc((void **) &d_input, result); cudaError_t err = cudaSuccess; //transfer the array to the GPU err = cudaMemcpy(d_input, a, result,cudaMemcpyHostToDevice); if(err != cudaSuccess) { fprintf(stderr, "Failed to copy d_S from host to device (error code %s)!\n", cudaGetErrorString(err)); exit(EXIT_FAILURE); } //launch the kernel int threads =length; long long parallel_start_time = start_timer(); parallel_passwordCrack<<<1,threads>>>(length,d_output,d_input); long long parallel_end_time = stop_timer(parallel_start_time, "\nParallel Run Time"); //copy back the result array to the CPU cudaMemcpy(h_gpu_result,d_output,1000*sizeof(int),cudaMemcpyDeviceToHost); cout << "\nParallel Password Cracked: " << endl; for(int i=0; i<length; i++){ printf("%c\n", char(h_gpu_result[i])); } printf("\n"); cudaFree(d_output); cudaFree(d_input); free(h_gpu_result); return 0; }
12,959
#include "includes.h" __global__ void reverseArray(int *A, int *B) { int threadID = threadIdx.x; int start = (threadID * ArraySize) / 256; int end = ( ( (threadID + 1 ) * ArraySize) / 256) - 1; while(end > 0) { B[end] = A[start]; end--; start++; } }
12,960
/* Justinas Linkevicius * IFF-3/2 * L1c * 1. Kokia tvarka startuoja procesai? **** tokia, kokia užrašyti 2. Kokia tvarka vykdomi procesai? **** tokia, kokia startuoja 3. Kiek iteracijų iš eilės padaro vienas procesas? **** vieną pilna 4. Kokia tvarka atspausdinami to paties masyvo duomenys? **** tokia, kokia surašyti. */ #include "cuda_runtime.h" #include "device_launch_parameters.h" #include <string> #include <iostream> #include <fstream> #include <vector> #include <stdio.h> using namespace std; #define INPUTFILE "LinkeviciusJ.txt" #define CUDA_THREADS 10 #define MAX_DATA_PER_THREADS 10 /* Struktura saugoti vienai duomenu eilutei */ struct FileData { char stringField[255]; int intField; double doubleField; FileData() { strncpy(stringField, "", 255); intField = 0; doubleField = 0.0; } FileData(string a, int b, double c) { strncpy(stringField, a.c_str(), 255); intField = b; doubleField = c; } }; /* Struktura saugoti visiems duomenims */ struct ThreadData { FileData array[ MAX_DATA_PER_THREADS ]; }; // lauku pavadinimai string stringFieldName, intFieldName, doubleFieldName; // nuskaito pradinius duomenis void readData(ThreadData* threadDataArrays, int & threadDataSize, int & dataElementsCount) { ifstream input(INPUTFILE); input >> stringFieldName; input >> intFieldName; input >> doubleFieldName; input >> dataElementsCount; threadDataSize = ceil((double)dataElementsCount / CUDA_THREADS); int line = 0; for (int i = 0; i < CUDA_THREADS; i++) { for (int j = 0; j < threadDataSize; j++) { string stringField; int intField; double doubleField; input >> stringField >> intField >> doubleField; // jei masyvui nebera duomenu, uzpildome tusciais elementais if (line < dataElementsCount) threadDataArrays[i].array[j] = FileData(stringField, intField, doubleField); else threadDataArrays[i].array[j] = FileData(); line++; } } input.close(); } // spausdina pradinius duomenis void writeData(ThreadData* threadDataArrays, int & threadDataSize, int & dataElementsCount) { int line = 0; cout << stringFieldName << "\t" << intFieldName << "\t" << doubleFieldName << "\r\n"; for (int i = 0; i < CUDA_THREADS; i++) { cout << endl << "**** Array" << i << " ****" << endl; for (int j = 0; j < threadDataSize; j++) { line++; if (threadDataArrays[i].array[j].stringField != "") { cout.precision(2); cout << j << ") " << threadDataArrays[i].array[j].stringField << "\t" << threadDataArrays[i].array[j].intField << "\t" << fixed << threadDataArrays[i].array[j].doubleField << "\r\n"; } if (line == dataElementsCount) break; } } cout << endl; } // lygiagrecioji programos dalis // perduodamas duomenu masyvas ir kiekvieno proceso apdorojamu elementu kiekis __global__ void printKernel(ThreadData *threadData, int* size) { // get threadId int i = threadIdx.x; // spausdina proceso elementus for (int j = 0; j < *size; j++) { printf("process_%d: %d\t%s\t%d\t%.2f\n", i, j, threadData[i].array[j].stringField, threadData[i].array[j].intField, threadData[i].array[j].doubleField); } } // Helper function for using CUDA to add vectors in parallel. cudaError_t printWithCuda(ThreadData* hostData, int threadsCount, int threadDataSize) { cudaError_t cudaStatus; ThreadData* devData = 0; int* devSize = 0; // choose gpu cudaStatus = cudaSetDevice(0); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaSetDevice failed! Do you have a CUDA-capable GPU installed?"); goto Error; } // Allocate GPU buffers cudaStatus = cudaMalloc((void**)&devData, threadsCount * sizeof(ThreadData)); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMalloc failed!"); goto Error; } cudaStatus = cudaMalloc(&devSize, sizeof(int)); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMalloc failed!"); goto Error; } // Copy input vectors from host memory to GPU buffers. cudaStatus = cudaMemcpy(devData, hostData, threadsCount * sizeof(ThreadData), cudaMemcpyHostToDevice); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMemcpy failed!"); goto Error; } cudaStatus = cudaMemcpy(devSize, &threadDataSize, sizeof(int), cudaMemcpyHostToDevice); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaMemcpy failed!"); goto Error; } // Launch a kernel on the GPU with one thread for each element. printf("Starting CUDA threads!\n"); printKernel <<< 1, threadsCount >>> (devData, devSize); // Check for any errors launching the kernel cudaStatus = cudaGetLastError(); if (cudaStatus != cudaSuccess) { fprintf(stderr, "addKernel launch failed: %s\n", cudaGetErrorString(cudaStatus)); goto Error; } // cudaDeviceSynchronize waits for the kernel to finish, and returns // any errors encountered during the launch. cudaStatus = cudaDeviceSynchronize(); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaDeviceSynchronize returned error code %d after launching addKernel!\n", cudaStatus); goto Error; } printf("End of CUDA threads!\n"); Error: cudaFree(devData); cudaFree(devSize); std::cin.get(); return cudaStatus; } int main() { ThreadData threadDataArrays[CUDA_THREADS]; int dataElementsCount; int threadDataSize; readData(threadDataArrays, threadDataSize, dataElementsCount); writeData(threadDataArrays, threadDataSize, dataElementsCount); // Start CUDA cudaError_t cudaStatus = printWithCuda(threadDataArrays, CUDA_THREADS, threadDataSize); if (cudaStatus != cudaSuccess) { fprintf(stderr, "addWithCuda failed!"); return 1; } // cudaDeviceReset must be called before exiting in order for profiling and // tracing tools such as Nsight and Visual Profiler to show complete traces. cudaStatus = cudaDeviceReset(); if (cudaStatus != cudaSuccess) { fprintf(stderr, "cudaDeviceReset failed!"); return 1; } std::cin.get(); return 0; }
12,961
#include <stdio.h> #include <stdlib.h> #include <cuda_runtime.h> #include <iostream> #include <fstream> /* -------- KERNEL -------- */ __global__ void reduce_kernel(unsigned int * d_out, unsigned int * d_in, unsigned int SIZE) { // position and threadId unsigned int pos = threadIdx.x + blockIdx.x * blockDim.x; unsigned int tid = threadIdx.x; // do reduction in global memory for (unsigned int s = blockDim.x / 2; s>0; s>>=1) { if (tid < s) { if (pos+s < SIZE) // Handling out of bounds { d_in[pos] = d_in[pos] + d_in[pos+s]; } } __syncthreads(); } // only thread 0 writes result, as thread if ((tid==0) && (pos < SIZE)) { d_out[blockIdx.x] = d_in[pos]; } } /* -------- KERNEL -------- */ __global__ void reduce_kernel_volatile(unsigned int * d_out, unsigned int * d_in, unsigned int SIZE) { // position and threadId unsigned int pos = threadIdx.x + blockIdx.x * blockDim.x; unsigned int tid = threadIdx.x; volatile unsigned int a = blockDim.x / 2; // do reduction in global memory for (unsigned int s = a; s>0; s>>=1) { if (tid < s) { if (pos+s < SIZE) // Handling out of bounds { d_in[pos] = d_in[pos] + d_in[pos+s]; } } __syncthreads(); } // only thread 0 writes result, as thread if ((tid==0) && (pos < SIZE)) { d_out[blockIdx.x] = d_in[pos]; } } /* -------- KERNEL WRAPPER -------- */ void reduce(unsigned int * d_out, unsigned int * d_in, unsigned int SIZE, unsigned int NUM_THREADS) { // Setting up blocks and intermediate result holder unsigned int NUM_BLOCKS = SIZE/NUM_THREADS + ((SIZE % NUM_THREADS)?1:0); unsigned int * d_intermediate_in; unsigned int * d_intermediate_out; cudaMalloc(&d_intermediate_in, sizeof(unsigned int)*SIZE); cudaMalloc(&d_intermediate_out, sizeof(unsigned int)*NUM_BLOCKS); cudaMemcpy(d_intermediate_in, d_in, sizeof(unsigned int)*SIZE, cudaMemcpyDeviceToDevice); // Recursively solving, will run approximately log base NUM_THREADS times. do { reduce_kernel<<<NUM_BLOCKS, NUM_THREADS>>>(d_intermediate_out, d_intermediate_in, SIZE); // Updating SIZE SIZE = NUM_BLOCKS;//SIZE / NUM_THREADS + SIZE_REST; // Updating input to intermediate cudaMemcpy(d_intermediate_in, d_intermediate_out, sizeof(unsigned int)*NUM_BLOCKS, cudaMemcpyDeviceToDevice); // Updating NUM_BLOCKS to reflect how many blocks we now want to compute on NUM_BLOCKS = SIZE/NUM_THREADS + ((SIZE % NUM_THREADS)?1:0); } while(SIZE > NUM_THREADS); // if it is too small, compute rest. // Computing rest reduce_kernel<<<1, SIZE>>>(d_out, d_intermediate_out, SIZE); cudaFree(d_intermediate_in); cudaFree(d_intermediate_out); } /* -------- MAIN -------- */ int main(int argc, char **argv) { std::ofstream myfile; myfile.open ("par_reduce.csv"); // Setting NUM_THREADS const unsigned int times = 10; for (unsigned int rounds = 0; rounds<30; rounds++) { // printf("Round: %d\n", rounds); unsigned int NUM_THREADS = 1<<10; // Making non-bogus data and setting it on the GPU unsigned int SIZE = 1<<rounds; unsigned int * d_in; unsigned int * d_out; cudaMalloc(&d_in, sizeof(unsigned int)*SIZE); cudaMalloc(&d_out, sizeof(unsigned int)*SIZE); unsigned int * h_in = (unsigned int *)malloc(SIZE*sizeof(int)); for (unsigned int i = 0; i < SIZE; i++) h_in[i] = 1; cudaMemcpy(d_in, h_in, sizeof(unsigned int)*SIZE, cudaMemcpyHostToDevice); // Running kernel wrapper // setting up time cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); // kernel time!!! cudaEventRecord(start, 0); for (unsigned int i = 0; i < times; i++) { reduce(d_out, d_in, SIZE, NUM_THREADS); } cudaEventRecord(stop, 0); cudaEventSynchronize(stop); // calculating time float elapsedTime; cudaEventElapsedTime(&elapsedTime, start, stop); elapsedTime = elapsedTime / ((float) times); // printf("time!: %.5f\n", elapsedTime); unsigned int h_out; cudaMemcpy(&h_out, d_out, sizeof(int), cudaMemcpyDeviceToHost); printf("%d \n", h_out); myfile << elapsedTime << ","; } myfile.close(); return 0; }
12,962
#include <iostream> #include <random> #include <cuda_runtime_api.h> #include <sys/time.h> #include <vector> struct RGBPoint { float x; float y; float z; float r; float g; float b; float a; RGBPoint() {} RGBPoint(float x, float y, float z, float r, float g, float b, float a) : x(x), y(y), z(z), r(r), g(g), b(b), a(a) {} }; __global__ void TestKernel(RGBPoint *d_img_RGB, const int row, const int col) { int h = threadIdx.x + blockIdx.x * blockDim.x; int w = threadIdx.y + blockIdx.y * blockDim.y; if ((h >= row) || (w >= col)) { return; } int index = h * col + w; if (index % 120 == 0) { d_img_RGB[index].a = 3.0f; d_img_RGB[index].r = 4.0f; d_img_RGB[index].g = 5.0f; d_img_RGB[index].b = 6.0f; d_img_RGB[index].x = 7.0f; d_img_RGB[index].y = 8.0f; d_img_RGB[index].z = 9.0f; } } int main() { struct timeval start, end; float t1, t2; const int row = 640; const int col = 480; const size_t size_RGB = row * col * sizeof(RGBPoint); RGBPoint *img_RGB = (RGBPoint*)malloc(size_RGB); RGBPoint *d_img_RGB; cudaMalloc(&d_img_RGB, size_RGB); dim3 block_size(4, 32); dim3 grid_size((row - 1) / block_size.x + 1, (col - 1) / block_size.y + 1); gettimeofday(&start, nullptr); TestKernel<<<grid_size, block_size>>>(d_img_RGB, row, col); cudaMemcpy(img_RGB, d_img_RGB, size_RGB, cudaMemcpyDeviceToHost); gettimeofday(&end, nullptr); t1 = ((end.tv_sec - start.tv_sec) * 1000000.0f + (end.tv_usec - start.tv_usec)) / 1000.0f; std::vector<RGBPoint> result_RGB; gettimeofday(&start, nullptr); for (int h = 0; h < row; h++) { for (int w = 0; w < col; w++) { int index = h * col + w; if (img_RGB[index].a > 0) { result_RGB.push_back(img_RGB[index]); } } } gettimeofday(&end, nullptr); t2 = ((end.tv_sec - start.tv_sec) * 1000000.0f + (end.tv_usec - start.tv_usec)) / 1000.0f; free(img_RGB); cudaFree(d_img_RGB); std::cout << "kernel and data transfer time: " << t1 << " ms" << std::endl; std::cout << "postprocess time: " << t2 << " ms" << std::endl; return 0; }
12,963
#include <math.h> #include <stdio.h> #define N 2048*2048 // Number of elements in each vector #define rowcol2idx(num, r, c) ((r)*(num)+(c)) /* * Optimize this already-accelerated codebase. Work iteratively, * and use nvprof to support your work. * * Aim to profile `saxpy` (without modifying `N`) running under * 20us. * * Some bugs have been placed in this codebase for your edification. */ __global__ void saxpy(int *a, int *b, int *c, int base) { /* int tid = threadIdx.x + blockIdx.x * blockDim.x; for (; tid < N; tid += stride) c[tid] = (a[tid]<<1) + b[tid]; */ int row = blockIdx.y*blockDim.y + threadIdx.y; int col = blockIdx.x*blockDim.x + threadIdx.x; int tid = rowcol2idx(gridDim.x*blockDim.x, row, col)+base; if (tid < N) c[tid] = (a[tid]<<1) + b[tid]; } /* __global__ void init_array(int *a, int *b, int *c) { int tid = threadIdx.x + blockIdx.x * blockDim.x; int stride = blockDim.x * gridDim.x; for (; tid < N; tid += stride) { a[tid] = 2; b[tid] = 1; c[tid] = 0; } } */ __global__ void init_array(int *a, int target, int stride) { for (int tid = threadIdx.x + blockIdx.x * blockDim.x; tid < N; tid += stride) a[tid] = target; } int main() { int *a, *b, *c; int size = N*sizeof(int); // The total number of bytes per vector cudaMallocManaged(&a, size); cudaMallocManaged(&b, size); cudaMallocManaged(&c, size); int deviceId; cudaDeviceProp props; cudaGetDevice(&deviceId); cudaGetDeviceProperties(&props, deviceId); int multiProcessorCount = props.multiProcessorCount; size_t threadsPerBlock = 1024; size_t numberOfBlocks = ((N>>10)/multiProcessorCount+1)*multiProcessorCount; // i first prefetch pages... cudaMemPrefetchAsync(a, size, deviceId); cudaMemPrefetchAsync(b, size, deviceId); init_array<<<numberOfBlocks, threadsPerBlock>>>(a,2,threadsPerBlock*numberOfBlocks); init_array<<<numberOfBlocks, threadsPerBlock>>>(b,1,threadsPerBlock*numberOfBlocks); // we have no need to initialize array c, because of default value is 0. //init_array<<<numberOfBlocks, threadsPerBlock>>>(c,0,threadsPerBlock*numberOfBlocks); cudaDeviceSynchronize(); //printf("sm numer = %d\n", multiProcessorCount); cudaMemPrefetchAsync(c, size, deviceId); //saxpy <<<numberOfBlocks, threadsPerBlock>>>(a,b,c,threadsPerBlock*numberOfBlocks); int len = 40; dim3 threads_per_block(32, 32, 1); dim3 number_of_blocks(len, len, 1); saxpy<<<number_of_blocks, threads_per_block>>>(a, b, c, 0); saxpy<<<number_of_blocks, threads_per_block>>>(a, b, c, 1600*1024); saxpy<<<number_of_blocks, threads_per_block>>>(a, b, c, 3200*1024); cudaDeviceSynchronize(); cudaMemPrefetchAsync(c, size, cudaCpuDeviceId); // Print out the first and last 5 values of c for a quality check for( int i = 0; i < 5; ++i ) printf("c[%d] = %d, ", i, c[i]); printf ("\n"); for( int i = N-5; i < N; ++i ) printf("c[%d] = %d, ", i, c[i]); printf ("\n"); cudaFree(a); cudaFree(b); cudaFree(c); }
12,964
#include <stdio.h> #include <cuda_runtime.h> __global__ void kernel() { double a = 2.71828; //register variables, automatic double c[100]; //local variable, automatic __shared__ double b; //shared variable int tx = threadIdx.x; //register variable if (tx == 0) { b = 3.1415926f; } //__syncthreads(); // run with/without this line printf("id = %d, a=%7.5f, b=%9.7f\n", tx, a, b); } int main() { kernel<<<1,8>>>(); cudaDeviceReset(); return 0; }
12,965
template<typename T> __device__ void vectorAddVector(const T* A, const T* B, T* C, const int length) { int bx = blockIdx.x; int tx = threadIdx.x; int index = bx * blockDim.x + tx; if (index < length) { C[index] = A[index] + B[index]; } } template<typename T> __device__ void vectorSubVector(const T* A, const T* B, T* C, const int length) { int bx = blockIdx.x; int tx = threadIdx.x; int index = bx * blockDim.x + tx; if (index < length) { C[index] = A[index] - B[index]; } } template<typename T> __device__ void vectorTimesVector(const T* A, const T* B, T* C, const int length) { int bx = blockIdx.x; int tx = threadIdx.x; int index = bx * blockDim.x + tx; if (index < length) { C[index] = A[index] * B[index]; } } template<typename T> __device__ void vectorDivVector(const T* A, const T* B, T* C, const int length) { int bx = blockIdx.x; int tx = threadIdx.x; int index = bx * blockDim.x + tx; if (index < length) { C[index] = A[index] / B[index]; } }
12,966
// From CUDA for Engineering // dist_v2/kernel.cu #include <stdio.h> #include <cuda_runtime.h> #include <iostream> #include <iomanip> #define TPB 32 #define N 256000 #define M 5 // number of times to do cudaMemcpy #define DEBUG 0 __device__ float distance(float x1, float x2) { return sqrt((x2 - x1) * (x2 - x1)); } __global__ void distanceKernel(float *d_out, float *d_in, float ref, int len) { const int i = blockIdx.x * blockDim.x + threadIdx.x; if (i >= len) { return; } const float x = d_in[i]; d_out[i] = distance(x, ref); } void distanceArray(float *out, float *in, float ref, int len) { // alloc cuda memory float *d_in = 0; float *d_out = 0; cudaMalloc(&d_in, len * sizeof(float)); cudaMalloc(&d_out, len * sizeof(float)); // memcpy to device struct timespec t0 = {0,0}; struct timespec t1 = {0,0}; clock_gettime(CLOCK_REALTIME, &t0); for (int i = 0; i < M; i++) { cudaMemcpy(d_in, in, len * sizeof(float), cudaMemcpyHostToDevice); } clock_gettime(CLOCK_REALTIME, &t1); std::cout << "Data transfer time (ms) = " << (t1.tv_sec-t0.tv_sec)*1e3 + (t1.tv_nsec-t0.tv_nsec)/1e6 << "\n"; // call wrapper clock_gettime(CLOCK_REALTIME, &t0); distanceKernel<<<(len+TPB-1)/TPB, TPB>>>(d_out, d_in, ref, len); cudaDeviceSynchronize(); clock_gettime(CLOCK_REALTIME, &t1); std::cout << "Kernel time (ms) = " << (t1.tv_sec-t0.tv_sec)*1e3 + (t1.tv_nsec-t0.tv_nsec)/1e6 << "\n"; // memcpy from device cudaMemcpy(out, d_out, len * sizeof(float), cudaMemcpyDeviceToHost); // free cuda memory cudaFree(d_in); cudaFree(d_out); } float scale(int i, int n) { return ((float)i) / (n - 1); } int main() { std::cout << "dist_v2_cuda\n"; const float ref = 0.5f; float *in = (float*)calloc(N, sizeof(float)); float *out = (float*)calloc(N, sizeof(float)); for (int i = 0; i < N; i++) { in[i] = scale(i, N); } distanceArray(out, in, ref, N); #if DEBUG std::cout << std::fixed << std::setprecision(4); for (int i = 0; i < N; i++) { std::cout << "i = " << i << "\tin: " << in[i] << "\tout: " << out[i] << "\n"; } #endif free(in); free(out); return 0; }
12,967
#include<stdio.h> #include<cuda_runtime.h> #include<math.h> #define SIZE 1024 #define TILE_WIDTH 16 float h_M[SIZE*SIZE],h_N[SIZE*SIZE],h_P[SIZE*SIZE]; __global__ void multiplication_kernel(float *d_M,float *d_N,float *d_P) { __shared__ float ds_M[TILE_WIDTH][TILE_WIDTH]; __shared__ float ds_N[TILE_WIDTH][TILE_WIDTH]; int tx = threadIdx.x; int ty = threadIdx.y; int bx = blockIdx.x; int by = blockIdx.y; int col = TILE_WIDTH * bx + tx; int row = TILE_WIDTH * by + ty; float prod_value = 0; int m,k; for(m=0;m<SIZE/TILE_WIDTH;m++) { ds_M[ty][tx] = d_M[row*SIZE+(m*TILE_WIDTH+tx)]; ds_N[ty][tx] = d_N[(m*TILE_WIDTH+ty)*SIZE+col]; __syncthreads(); for(k=0;k<TILE_WIDTH;k++) prod_value+=ds_M[ty][k]*ds_N[k][tx]; __syncthreads(); } d_P[row*SIZE+col] = prod_value; } void matrix_multiplication(float *d_M,float *d_N,float *d_P) { dim3 dimBlock(TILE_WIDTH,TILE_WIDTH,1); dim3 dimGrid(SIZE/TILE_WIDTH,SIZE/TILE_WIDTH,1); multiplication_kernel<<<dimGrid,dimBlock>>>(d_M,d_N,d_P); } void display_matrix(float mat[]) { int i,j; for(i=0;i<SIZE;i++) { for(j=0;j<SIZE;j++) printf("%f ",mat[i*SIZE+j]); printf("\n"); } } int main() { int deviceCount; cudaGetDeviceCount(&deviceCount); if(!deviceCount){ fprintf(stderr,"No devices supporting cuda\n"); exit(EXIT_FAILURE); } int deviceId = 0; cudaSetDevice(deviceId); const int ARRAY_BYTES = SIZE*SIZE*sizeof(float); float *d_M,*d_N,*d_P; cudaMalloc((void**)&d_M,ARRAY_BYTES); cudaMalloc((void**)&d_N,ARRAY_BYTES); cudaMalloc((void**)&d_P,ARRAY_BYTES); int i,j; for(i=0;i<SIZE;i++) { for(j=0;j<SIZE;j++) { h_M[i*SIZE+j] = rand()%101; h_N[i*SIZE+j] = rand()%101; } } cudaMemcpy(d_M,h_M,ARRAY_BYTES,cudaMemcpyHostToDevice); cudaMemcpy(d_N,h_N,ARRAY_BYTES,cudaMemcpyHostToDevice); cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start,0); matrix_multiplication(d_M,d_N,d_P); cudaEventRecord(stop,0); cudaEventSynchronize(stop); float elapsedTime; cudaEventElapsedTime(&elapsedTime,start,stop); cudaMemcpy(h_P,d_P,ARRAY_BYTES,cudaMemcpyDeviceToHost); /* printf("M is \n"); display_matrix(h_M); printf("N is \n"); display_matrix(h_N); printf("Product of M and N is \n"); display_matrix(h_P); */ printf("Elapsed time is %f\n",elapsedTime); cudaFree(d_M); cudaFree(d_N); cudaFree(d_P); return 0; }
12,968
#include <cuda.h> #include <stdio.h> const int N = 1 << 20; __global__ void kernel(float *x, int n){ int tid = threadIdx.x + blockIdx.x * blockDim.x; for(int i = tid; i < n; i += blockDim.x * gridDim.x) { x[i] = sqrt(pow(3.14159,i)); } } int main(int argc, char **argv){ // seteo de numero de streams if(argc != 2){ fprintf(stderr, "run as ./prog numstreams\n"); exit(EXIT_FAILURE); } const int num_streams = atoi(argv[1]); cudaStream_t streams[num_streams]; float *data[num_streams]; // creacion de streams y de datos for(int i = 0; i < num_streams; i++){ printf("creando stream %i\n", i); cudaStreamCreate(&streams[i]); cudaMalloc(&data[i], N * sizeof(float)); } // ejecucion de cada kernel cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); printf("ejecutando con %i streams....", num_streams); fflush(stdout); cudaEventRecord(start); for (int i = 0; i < num_streams; i++) { // launch one worker kernel per stream kernel<<<1, 64, 0, streams[i]>>>(data[i], N); kernel<<<1, 1>>>(0, 0); } cudaDeviceSynchronize(); cudaEventRecord(stop); printf("ok\n"); fflush(stdout); cudaEventSynchronize(stop); float milliseconds = 0.0; cudaEventElapsedTime(&milliseconds, start, stop); printf("Time GPU: %f\n", milliseconds ); cudaDeviceReset(); return 0; }
12,969
//#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> /* ------------------------102- Introducction to parallel programing ------------------------------------------------------------------------------------------------------------------------------------------------------- Context (instruccion por turnos) - Collection of data about process which allows processor to suspend or hold the execution of a process and restart the execution later. - Memory addresses - Program counter states Thread (secuendia mas pequea de instruccion programada - Process - threads (subprocess) Parallel Process - Tipos - Paralelismo a nivel de tarea - los nucleos realizan tareas distintas con datos distintos o los mismos - Paralelismo a nivel de datos - los nucleos realizan la misma tarea con diferentes datos Paralelismo vs concurrencia. - Concurrencia = realizacion de procesos en distintos tiempos(secuenciales) de milesimas de segundo que aparentan simultaneidad o paralelismo - Paralelismo = distintos nucleos realizan tareas al mismo tiempo ------------------------104- Install ------------------------------------------------------------------------------------------------------------------------------------------------------- Revisar compatibilidad en wikipedia en ingles. GPGPU- windows + r (dxdiag) // visualiza las caracteristicas del PC windows + r (cmd) //escribir (nvcc --version) para saber la version de CUDA instalado del PC */ //------------------------105 - Basic steps of a CUDA program---------------------------------------------------- /*RESUMEN * - initization of data from CPU * - transfer data from CPU context to GPU context * - Kernel launc with needed grid/block size * - Transfer results back to CPU context from CPU context * - Reclaim the memory from both CPU and GPU * * - IMPORTANTE * -Grid - Grid is a collection of all the threads launch for a kernel ( coleccion de todos los hilos lanzados para un kernel) * - En el ejemplo hello CUDA world se tienen 20 subprocesos, los hilos en una cuadricula estan organizados * en un grupo llamado bloques de hilos * -Block - subconjunto de hilos dentro de un GRID que se pueden representar como un cubo(3d) mas pequeo que a su vez esta subdividido en pequeos cubos que representan a los hilos o threads -GRID (cubo general en x, y z) -BLOCK (subcubo dentro de GRID que forma un subconjunto de hilos) -THREADS kernel_ name <<< number_of_blocks, // especifica cuantos bloques de hilos en la cuadracula en cada dimension thread_per_block // especifica cuantos hilos en un bloque en cada dimension >>> (arguments) // TODO ESTO ES EN UNA DIMENSION * - Para especificar cuadriculas y bloques multidimensionales * -dim3 variable_name (x,y,z) // se inicializa por defecto en 1 * - dim3 variable_name(x,y,z) // puede acceder a cada valor de dimension * - variable_name.x * * - variable_name.y * * - variable_name.z *EJEMPLO UNIDIMENSIONAL * - 8 bloques de hilos, donde cada bloque tiene 4 hilos en la dimension x -la dimension de nuestro bloque es de cuatro hilos en la dimension x y 1 hilo en las dimensiones Y y Z. -dim3 grid(8,1,1) // nos referimos a todos los hilos lanzados para un kernel como grid. -dim3 block(4,1,1) _________________________ _________________________ _________________________ _________________________ _________________________ _________________________ _________________________ _________________________ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |1| |2| |3| |4| | | |1| |2| |3| |4| | | |1| |2| |3| |4| | | |1| |2| |3| |4| | | |1| |2| |3| |4| | | |1| |2| |3| |4| | | |1| |2| |3| |4| | | |1| |2| |3| |4| | |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| |__|_|___|_|___|_|___|_|__| 8 bloques unidimensionales con 8 hilos unidimensionales - si no se especifican las dimensiones se inicializaran como 1 -LIMITES DE TAMAO DE BLOQUE - 1024 HILOS PARA DIMENSION X - 1024 HILOS PARA DIMENSION Y - 64 HILOS PARA DIMENSION Z - x* y* x <= 1024 la multiplicacion del numero de subprocesos en cada dimension deber ser menor o igual a 1024 -LIMITES DE TAMAO DE CUADRICULA - 65536 (1<<32-1) BLOQUES PARA DIMENSION X - 65536 (2^32-1) BLOQUES PARA DIMENSION Y,Z */ //// EJEMPLOS //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // ////kernel ///* //* funcion asincrona ( el host puede continuar con las siguientes instrucciones) a //* menos que se especifique que debe esperar (cudaDeviceSynchronize()) //* //* cudaDeviceReset(); reestablece el dispositivo //*/ //__global__ void hello_cuda() //{ // printf("Hello CUDA world \n"); // //cout << "hello CPU world" << endl; // esta instruccion no funciona dentro del kernel //} // // //int main() //{ // //************************// EJEMPLO 1 *** // // //hello_cuda <<<1,1 >>>(); // kernel con parametros de lanzamiento // /* // * - El segundo parametro hace referencia al numero de subprocesos que se ejecutanran en el DEVICE // * // */ // //hello_cuda << <1, 10 >> > (); // imprime 10 veces hello CUDA world // // //cudaDeviceSynchronize(); // hace que el CPU o host espere en este punto, hasta que termine el proceso de DEVICE // // //cudaDeviceReset(); //reestablece dispositivo // // //cout << "hello CPU world" << endl; // // //return 0; // // //************************// EJEMPLO 2 "8 BLOQUES(8X1) CON 4 HILOS(4X1)" imprime 32 veces hello CUDA world*** // // //dim3 grid(8); // conjunto de 8 blocks en X y 1 en las dimensiones Y,Z. // //dim3 block(4); // bloque con tamao 4 en X y 1 en Y,Z. // // // //// el primer parametro(grid) es el numero de bloques de hilos en cada dimension // //// el segundo parametro(block) es el numero de hilos en cada dimension del bloque // // //hello_cuda << <grid, block >> > (); // // //cudaDeviceSynchronize(); // hace que el CPU o host espere en este punto, hasta que termine el proceso de DEVICE // // //cudaDeviceReset(); //reestablece dispositivo // // //cout << "hello CPU world" << endl; // // //return 0; // // //************************// EJEMPLO 3 "4 BLOQUES (2X2) CON 16 HILOS (8X2) " imprime 32 veces hello CUDA world*** // // // int nx; // variables dinamicas para ir modificando en tiempo de ejecucion // int ny; // nx = 16; // ny = 4; // // // dim3 block(8,2); // 16 hilos en cada bloque // dim3 grid(nx/block.x, ny/block.y); // 16/8=2 , 4/2=2 4 bloques en total //// _____________________________________________ //// | | | | | | | | | | | | | | | | | //// | |1| |2| |3| |4| |5| |6| |7| |8| == 4 BLOQUES (GRID) IGUALES A ESTE. //// | |1| |2| |3| |4| |5| |6| |7| |8| //// |__|_|___|_|___|_|___|_|__|_|__|_|___|_|___|_| // // // el primer parametro(grid) es el numero de bloques de hilos en cada dimension // // el segundo parametro(block) es el numero de hilos en cada dimension del bloque // // hello_cuda << <grid, block >> > (); // // cudaDeviceSynchronize(); // hace que el CPU o host espere en este punto, hasta que termine el proceso de DEVICE // // cudaDeviceReset(); //reestablece dispositivo // // cout << "hello CPU world" << endl; // // return 0; // //} //------------------------106 - Organization of threads in a CUDA program 1---------------------------------------------------- //1D // ______________________ ____________________ // | |A| |B| |C| |D| |E| |F| |G| |H| //Threadlx.X| |0| |1| |2| |3| |0| |1| |2| |3| //Threadlx.Y| |0| |0| |0| |0| |0| |0| |0| |0| 2 bloques : ejemplo de identificacion de hilo //Threadlx.Z| |0| |0| |0| |0| |0| |0| |0| |0| C = 2,0,0 // |__|_|___|_|___|_|___|_| |_|__|_|_____|_|___|_| //2D // __0_____1____2____3___ _0___1_______2___3__ // | || |X| || || |P| || || || // | || |Y| || || || || |Q| || // ______________________ ____________________ // X Y P Q R S T U // ______________________ ____________________ Threadlcx.X 1 1 0 2 0 3 1 0 // | |R| || || || || |T| || || Threadlcx.Y 0 1 0 1 0 1 0 1 // | || || || |S| |U| || || || // ______________________ ____________________ // //************************// EJEMPLO 1 -> GRID 2X2 CON 8 HILOS CADA BLOQUE *** // //************************// EJEMPLO 1 -> IDENTIFICACION DE HILOS *** //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // //__global__ void print_threadIds() //{ // printf("threadIdx.x : %d, threadIdx.y : %d, threadIdx.z : %d \n", threadIdx.x, threadIdx.y, threadIdx.z); //} // //int main() //{ // int nx, ny; // nx = 2; // ny = 2; // // dim3 block(2, 2); // 8 subprocesos en la dimension X y 8 subprocesos en la dimension Y. // dim3 grid(nx / block.x, ny / block.y); // grid de 2x2 // // print_threadIds << <grid, block >> > (); // cudaDeviceSynchronize(); // da la orden para que el host o int main espere a que termine el kernel o __global__. // cudaDeviceReset(); // return 0; // // /* // ORDEN GRID 2X2 = 4 BLOCKS , 4 HILOS POR BLOCK = 16 HILOS // // PRIMER BLOCK 1 // _____ // | | // | A B | // | C D | // |_____| // // *BLOCKS 1 | 2 | 3 | 4 // A B C D E F G H I J K L M N O P // Threadidx.X 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 // Threadidx.Y 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 // Threadidx.Z 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 // // */ //} //------------------------107 - Organization of threads in a CUDA program 2---------------------------------------------------- /* * En tiempo de ejecucion CUDA la variable blckldx inicializada de forma unica para cada hilo dependiendo de las coordenadas de la pertenencia blockldx.X = coordenadas de cada hilo tomando como base cada block //1D 0 1 // ____________________ __________________ // |P| || |Q| || || |R| || |S| // 0 ____________________ __________________ P Q R S T U V X // blockldx.X 0 0 1 1 0 0 1 1 // ____________________ __________________ blockldx.Y 0 0 0 0 1 1 1 1 // |T| || |U| || |V| || || |X| // 1 ____________________ __________________ //-------------------------------------------------------------------------- // __________________X_______________ //2D 0 1 // ____________________ ___________________ // || |X| || || |P| || || || // |0 // | || |Y| || || || || |Q| || // | ______________________ ____________________ // Y| X Y P Q R S T U // | ______________________ ____________________ blockldx.X 0 0 1 1 0 0 1 1 // | |R| || || || || |T| || || blockldx.Y 0 0 0 0 1 1 1 1 // |1 blockDim.X = 4 // || || || |S| |U| || || || blockDim.Y = 2 // ______________________ ____________________ GridDim.X = 2 GridDim.Y = 2 blockDim = es la dimension del bloque ej. blockDim.x=4 y blockDim.y = 2 da como resultado un bloque de 8 hilos GridDim = es la dimension de la rejilla ej. gridDim.x = 2 y gridDim.y = 2 da como resultado 4 bloques de hilos */ //EJEMPLO //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // //__global__ void print_details() //{ // printf("blockIdx.x : %d, blockIdx.y : %d, blockIdx.z : %d, blockDim.x : %d, blockDim.y : %d, gridDim.x : %d, gridDim.y : %d \n", blockIdx.x, blockIdx.y, blockIdx.z, blockDim.x, blockDim.y, gridDim.x, gridDim.y); //} // //int main() //{ // int nx, ny; // nx = 4; // ny = 1; // // dim3 block(1, 1); // 8 subprocesos en la dimension X y 8 subprocesos en la dimension Y. // dim3 grid(nx / block.x, ny / block.y); // grid de 2x2 // // print_details << <grid, block >> > (); // cudaDeviceSynchronize(); // da la orden para que el host o int main espere a que termine el kernel o __global__. // cudaDeviceReset(); // return 0; //} //------------------------108 - Ejercicio grid 3d y block 3d---------------------------------------------------- //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // //__global__ void print_details() //{ // printf("threadIdx.x : %d, threadIdx.y : %d, threadIdx.z : %d \n", threadIdx.x, threadIdx.y, threadIdx.z); // printf("blockIdx.x : %d, blockIdx.y : %d, blockIdx.z : %d, blockDim.x : %d, blockDim.y : %d, blockDim.z : %d, gridDim.x : %d, gridDim.y : %d, gridDim.z : %d \n", blockIdx.x, blockIdx.y, blockIdx.z, blockDim.x, blockDim.y, blockDim.z, gridDim.x, gridDim.y, gridDim.z); //} // //int main() //{ // int nx, ny, nz; // nx = 4; // ny = 4; // nz = 4; // // dim3 block(2, 2, 2); // 8 subprocesos en la dimension X y 8 subprocesos en la dimension Y. // dim3 grid(nx / block.x, ny / block.y, nz / block.z); // grid de 2x2 // // print_details << <grid, block >> > (); // cudaDeviceSynchronize(); // da la orden para que el host o int main espere a que termine el kernel o __global__. // cudaDeviceReset(); // return 0; //} //------------------------109 Unique index calculation using threadIdx blockId and blockDim-------------------- //************************************Ejemplo 1 ////asignar valores de un array a cada hilo // //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // //__global__ void unique_idx_calc_threadIdx(int* input) //{ // int tid = threadIdx.x; // printf("threadIdx : %d, value : %d \n", tid, input[tid]); //} // //int main() //{ // int array_size = 8; // int array_byte_size = sizeof(int) * array_size; // int h_data[] = { 23,9,4,53,65,12,1,33 }; // // for (int i = 0; i < array_size; i++) // { // cout << h_data[i] << " "; // } // // cout << endl; // cout << endl; // // int* d_data; // cudaMalloc((void**)&d_data, array_byte_size); // cudaMemcpy(d_data, h_data, array_byte_size, cudaMemcpyHostToDevice); // // //dim3 block(8); //8 threads en un bloque // //dim3 grid(1); // // dim3 block(4); // 8 threads en 2 bloques de 4 cada uno // dim3 grid(2); // // unique_idx_calc_threadIdx << <grid, block >> > (d_data); // cudaDeviceReset(); // return 0; // // //} //************************************//Ejemplo 2 //asignar valores de un array continuos a un grupo de blocks (grid 1D con 16 hilos en 4 bloques) // gid = tid + offset // gid = tid + blackldx.x * blockDim.x //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // // // //__global__ void unique_gid_calculation(int * input) //{ // int tid = threadIdx.x; // int offset = blockIdx.x * blockDim.x; // numero de hilos que componen un bloque // int gid = tid + offset; //indice en el que empezara a asignar valores a cada bloque de hilos // // //ejemplo de 3 blocks ///* // ////1D 0 1 3 //// _______________________ ________________________ ________________________ //// |23| |9| |4| |53| |65| |12| |1| |33| |65| |12| |1| |33| //tid(threadIdx) = 0 1 2 3 0 1 2 3 0 1 2 3 //blockIdx.x = 0 0 0 0 1 1 1 1 2 2 2 2 //blockDim.x = 4 4 4 4 4 4 4 4 4 4 4 4 //offset = 0 0 0 0 4 4 4 4 8 8 8 8 //gid = 0 1 2 3 4 5 6 7 8 9 10 11 // //*/ // // printf("blockIdx.x : %d, threadIdx.x : %d, gid: %d, value : %d \n", // blockIdx.x, tid, gid, input[gid]); // //} // //int main() //{ // int array_size = 16; // int array_byte_size = sizeof(int) * array_size; // int h_data[] = { 23,9,4,53,65,12,1,33,22,1,1,3,5,2,1,3 }; // // for (int i = 0; i < array_size; i++) // { // cout << h_data[i] << " "; // } // // cout << endl; // cout << endl; // // int* d_data; // cudaMalloc((void**)&d_data, array_byte_size); // cudaMemcpy(d_data, h_data, array_byte_size, cudaMemcpyHostToDevice); // // //dim3 block(8); //8 threads en un bloque // //dim3 grid(1); // // dim3 block(4); // 4 threads en 4 bloques // dim3 grid(4); // // unique_gid_calculation << <grid, block >> > (d_data); // cudaDeviceReset(); // return 0; // // //} //------------------------110 Unique index calculation for 2D grid 1-------------------- //------------------------110 calculo del indice global para cuadricula 2D 1 (GRID DE 2X2 CON 4x1 hilos) ------ /* * * Formula para calcular el indice unico para identificar los hilos que estan en una segunda fila * * Index = row offset + block offset + tid * row offset = number of threads in one thread block row (blockldx.y) * block offset = number of threads in thread block(blockldx.x) * tid = threadldx.x * * gid = gridDim.x * blockDim.x * blockldx.y + blockldx.x * blockDim.x + threadldx.x */ //asignar valores de un array continuos a un grupo de blocks (grid 2D con 16 hilos en 4 bloques de 4x1 hilos) //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // // // //__global__ void unique_gid_calculation_2d(int * input) //{ // int tid = threadIdx.x; // int offset = blockIdx.x * blockDim.x; // numero de hilos que componen un bloque // // int row_offset = blockDim.x * gridDim.x * blockIdx.y; // // int gid = tid + offset + row_offset; //indice en el que empezara a asignar valores a cada bloque de hilos //ejemplo de 3 blocks /* //2D (4 BLOQUES EN UN GRID DE 2X2 // 0 1 // _______________________ ________________________ //fila 1 de bloques |23| |9| |4| |53| |22| |1| |1| |3| // _______________________ ________________________ //fila 2 de bloques |65| |12| |1| |33| |5| |2| |1| |3| //fila 1 de bloques |23| |9| |4| |53| |65| |12| |1| |33| tid(threadIdx.X)= 0 1 2 3 0 1 2 3 blockIdx.x = 0 0 0 0 1 1 1 1 blockDim.x = 4 4 4 4 4 4 4 4 offset = 0 0 0 0 4 4 4 4 blockIdx.y = 0 0 0 0 0 0 0 0 gridDim.x = 2 2 2 2 2 2 2 2 rowOffset = 0 0 0 0 0 0 0 0 gid = 0 1 2 3 4 5 6 7 //fila 2 de bloques |65| |12| |1| |33| |5| |2| |1| |3| tid(threadIdx.X)= 0 1 2 3 0 1 2 3 blockIdx.x = 0 0 0 0 1 1 1 1 blockDim.x = 4 4 4 4 4 4 4 4 offset = 0 0 0 0 4 4 4 4 blockIdx.y = 1 1 1 1 1 1 1 1 gridDim.x = 2 2 2 2 2 2 2 2 rowOffset = 8 8 8 8 8 8 8 8 gid = 8 9 10 11 12 13 14 15 rowOffset = blockDim.x * gridDim.x * blockIdx.y; gid = tid + offset + row_offset; //indice en el que empezara a asignar valores a cada bloque de hilos */ // printf("blockIdx.x : %d, blockIdx.y: %d, threadIdx.x: %d, gid: %d - input: %d \n", // blockIdx.x, blockIdx.y, tid, gid, input[gid]); // //} // //int main() //{ // int array_size = 16; // int array_byte_size = sizeof(int) * array_size; // int h_data[] = { 23,9,4,53,65,12,1,33,22,1,1,3,5,2,1,3 }; // // for (int i = 0; i < array_size; i++) // { // cout << h_data[i] << " "; // } // // cout << endl; // cout << endl; // // int* d_data; // cudaMalloc((void**)&d_data, array_byte_size); // cudaMemcpy(d_data, h_data, array_byte_size, cudaMemcpyHostToDevice); // // //dim3 block(8); //8 threads en un bloque // //dim3 grid(1); // // dim3 block(4); // 4 threads en 4 bloques // dim3 grid(2,2); // // unique_gid_calculation_2d << <grid, block >> > (d_data); // cudaDeviceReset(); // return 0; // // //} //-----------------111 Unique index calculation for 2D grid 2-------------------- //-----------------111 calculo del indice global para cuadricula 2D (GRID DE 2X2 CON 2x2 hilos) ----- /* * * Formula para calcular el indice unico para identificar los hilos que estan en una segunda fila * * Index = row offset + block offset + tid * row offset = number of threads in one thread block row (blockldx.y) * block offset = number of threads in thread block(blockldx.x) * tid = threadldx.x * * gid = gridDim.x * blockDim.x * blockldx.y + blockldx.x * blockDim.x + threadldx.x */ //asignar valores de un array continuos a un grupo de blocks (grid 2D con 16 hilos en 4 bloques de 2x2 hilos) //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //using namespace std; // // // //__global__ void unique_gid_calculation_2d_2d(int * input) //{ // int tid = blockDim.x * threadIdx.y + threadIdx.x; // // int num_threads_in_a_block = blockDim.x * blockDim.y; // int block_offset = blockIdx.x * num_threads_in_a_block; // // int num_threads_in_a_row = num_threads_in_a_block * gridDim.x; // int row_offset = num_threads_in_a_row * blockIdx.y; // // int gid = tid + block_offset + row_offset; //indice en el que empezara a asignar valores a cada bloque de hilos //ejemplo de 3 blocks /* //2D (4 BLOQUES EN UN GRID DE 2X2 // 0 1 //bloques 1 y 2 ___________ __________ //fila 1 |23| |9| |22| |1| //fila 2 |4| |53| |1| |3| //bloques 3 y 4 ____________ __________ //fila 1 |65| |12| |5| |2| //fila 2 |1| |33| |1| |3| tid = blockDim.x * threadIdx.y + threadIdx.x; num_threads_in_a_block = blockDim.x * blockDim.y; block_offset = blockIdx.x * num_threads_in_a_block; num_threads_in_a_row = num_threads_in_a_block * gridDim.x; row_offset = num_threads_in_a_row * blockIdx.y; gid = tid + block_offset + row_offset; //fila 1 de bloques |23| |9| |4| |53| |22| |1| |1| |3| blockDim.x = 2 2 2 2 2 2 2 2 treadsIdx.x = 0 1 0 1 0 1 0 1 treadsIdx.y = 0 0 1 1 0 0 1 1 tid = 0 1 2 3 0 1 2 3 blockDim.y = 2 2 2 2 2 2 2 2 num_threads_in_a_block = 4 4 4 4 4 4 4 4 blockIdx.x = 0 0 0 0 1 1 1 1 block_offset = 0 0 0 0 4 4 4 4 gridDim.x = 2 2 2 2 2 2 2 2 num_threads_in_a_row = 8 8 8 8 8 8 8 8 blockIdx.y = 0 0 0 0 0 0 0 0 rowOffset = 0 0 0 0 0 0 0 0 gid = 0 1 2 3 4 5 6 7 tid = blockDim.x * threadIdx.y + threadIdx.x; num_threads_in_a_block = blockDim.x * blockDim.y; block_offset = blockIdx.x * num_threads_in_a_block; num_threads_in_a_row = num_threads_in_a_block * gridDim.x; row_offset = num_threads_in_a_row * blockIdx.y; gid = tid + block_offset + row_offset; //fila 2 de bloques |65| |12| |1| |33| |5| |2| |1| |3| blockDim.x = 2 2 2 2 2 2 2 2 treadsIdx.x = 0 1 0 1 0 1 0 1 treadsIdx.y = 0 0 1 1 0 0 1 1 tid = 0 1 2 3 0 1 2 3 blockDim.y = 2 2 2 2 2 2 2 2 num_threads_in_a_block = 4 4 4 4 4 4 4 4 blockIdx.x = 0 0 0 0 1 1 1 1 block_offset = 0 0 0 0 4 4 4 4 gridDim.x = 2 2 2 2 2 2 2 2 num_threads_in_a_row = 8 8 8 8 8 8 8 8 blockIdx.y = 1 1 1 1 1 1 1 1 rowOffset = 8 8 8 8 8 8 8 8 gid = 8 9 10 11 12 13 14 15 */ // printf("blockIdx.x : %d, blockIdx.y: %d, threadIdx.x: %d, gid: %d - input: %d \n", // blockIdx.x, blockIdx.y, tid, gid, input[gid]); // //} // //int main() //{ // int array_size = 16; // int array_byte_size = sizeof(int) * array_size; // int h_data[] = { 23,9,4,53,65,12,1,33,22,1,1,3,5,2,1,3 }; // // for (int i = 0; i < array_size; i++) // { // cout << h_data[i] << " "; // } // // cout << endl; // cout << endl; // // int* d_data; // cudaMalloc((void**)&d_data, array_byte_size); // cudaMemcpy(d_data, h_data, array_byte_size, cudaMemcpyHostToDevice); // // //dim3 block(8); //8 threads en un bloque // //dim3 grid(1); // // dim3 block(2,2); // 4 threads en cada block (2x2) // dim3 grid(2,2); // 4 blocks (2x2) // // unique_gid_calculation_2d_2d << <grid, block >> > (d_data); // cudaDeviceReset(); // return 0; // // //} //-----------------112 CUDA MEMORY TRANSFER -------------------------------------------------- /* -Two devices -HOST ( cpu- memory) - CPU - CACHES AND DRAM -DEVICE ( gpu - internal gpu memory)- SM (stream multiprocess) - CACHES AND DRAM - Para transferir memoria entre el host y el dispositivo cudaMemCpy( destination ptr, source ptr, size in byte, direction) * ptr = puntero *destination ptr = hostToDevice o DeviceToHost o HostToHost (cudamemcpyhtod, cudamemcpydtoh, cudamemcpydtod) */ //Ejemplo 1. pasar datos a memoria del device en un solo bloque de hilos //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //#include <stdlib.h> //#include <time.h> //using namespace std; // //__global__ void mem_trs_test(int* input) // kernel que toma como un puntero a una matriz de enteros //{ // //cuadricula 1D con 2 bloques de hilos // int gid = blockIdx.x * blockDim.x + threadIdx.x; //indice global para acceder a elementos de la matriz // printf("tid: %d, gid: %d, value: %d \n", threadIdx.x, gid, input[gid]); //} // // // //int main() //{ // int size = 128; // tamao de la matriz // int byte_size = size * sizeof(int);// cantidad de bytes que necesitamos para asignar a esta matriz // int* h_input; //asignar memoria del Host (la h_ es para indicar que es una variable del lenguaje principal) // // //asignacion de memoria usando funcion malloc. // h_input = (int*)malloc(byte_size); // asinacion de bytes necesarios // // //inicializacion aleatoria de la matriz con secuencia aleatoria de numeros // time_t t; // srand((unsigned)time(&t)); // for (int i = 0; i < size; i++) // { // h_input[i] = (int)(rand() & 0xff);//valor aleatoria entre 0 y 255 // } // // int* d_input; // se utiliza d_ para indicar que es una variable de dispositivo // // //asignacion de memoria en el dispositivo(gpu) // /* // C CUDA // malloc cudaMalloc -- asignar memoria // memset cudaMemset -- establece valores para una ubicacion de memoria dada // free cudaFree -- recupera la ubicacion de memoria especificada // // */ // // // ** = puntero doble o puntero a un puntero // // &d_input = especifica tamao de la memoria // cudaMalloc((void**)&d_input,byte_size); // // cudaMemcpy(d_input,h_input,byte_size,cudaMemcpyHostToDevice);// tranferir la matriz inicializada en el host al dispositivo // // h_input = puntero de origen // // d_input = puntero de destino en el device // // //parametros de lanzamiento // dim3 block(64); // TODO: POR LO GENERAL SE MANTIENE EL TAMAO EN MULTIPLOS DE 32 // dim3 grid(2); // // mem_trs_test << <grid, block >> > (d_input); // cudaDeviceSynchronize();// hace que la ejecucion espere en este punto // // cudaFree(d_input); // recuperar memoria // free(h_input); // recuperar memoria // // cudaDeviceReset(); // return 0; // //} //*********************************************** //Ejemplo 2. pasar datos a memoria del device en varios bloques de hilos //#include "cuda_runtime.h" //#include "device_launch_parameters.h" // //#include <stdio.h> //#include <iostream> //#include <stdlib.h> //#include <time.h> //using namespace std; // //__global__ void mem_trs_test2(int* input, int size) // kernel que toma como un puntero a una matriz de enteros //{ //int size = tamao matriz // // //cuadricula 1D con 2 bloques de hilos // int gid = blockIdx.x * blockDim.x + threadIdx.x; //indice global para acceder a elementos de la matriz // // // // CON ESTA VERIFICACION SOLO SE UTILIZAN LOS HILOS QUE MANEJARAN DATOS DADO EL INPUT // /*if (gid < size) // { // printf("tid: %d, gid: %d, value: %d \n", threadIdx.x, gid, input[gid]); // }*/ // // // SIN LA VERIFICACION SE ACCEDE A LOS HILOS DE TODO EL GRID AUN CUANDO NO MANEJEN DATOS // printf("tid: %d, gid: %d, value: %d \n", threadIdx.x, gid, input[gid]); //} // // // //int main() //{ // int size = 150; // tamao de la matriz // int byte_size = size * sizeof(int);// cantidad de bytes que necesitamos para asignar a esta matriz // int* h_input; //asignar memoria del Host (la h_ es para indicar que es una variable del lenguaje principal) // // //asignacion de memoria usando funcion malloc. // h_input = (int*)malloc(byte_size); // asinacion de bytes necesarios // // //inicializacion aleatoria de la matriz con secuencia aleatoria de numeros // time_t t; // srand((unsigned)time(&t)); // for (int i = 0; i < size; i++) // { // h_input[i] = (int)(rand() & 0xff);//valor aleatoria entre 0 y 255 // } // // int* d_input; // se utiliza d_ para indicar que es una variable de dispositivo //asignacion de memoria en el dispositivo(gpu) /* C CUDA malloc cudaMalloc -- asignar memoria memset cudaMemset -- establece valores para una ubicacion de memoria dada free cudaFree -- recupera la ubicacion de memoria especificada */ // ** = puntero doble o puntero a un puntero // &d_input = especifica tamao de la memoria // cudaMalloc((void**)&d_input, byte_size); // // cudaMemcpy(d_input, h_input, byte_size, cudaMemcpyHostToDevice);// tranferir la matriz inicializada en el host al dispositivo // // h_input = puntero de origen // // d_input = puntero de destino en el device // // //parametros de lanzamiento // dim3 block(32); // TODO: POR LO GENERAL SE MANTIENE EL TAMAO EN MULTIPLOS DE 32 // dim3 grid(5); // // mem_trs_test2 << <grid, block >> > (d_input,size); // cudaDeviceSynchronize();// hace que la ejecucion espere en este punto // // cudaFree(d_input); // recuperar memoria // free(h_input); // recuperar memoria // // cudaDeviceReset(); // return 0; // //} //-----------------112 exercise GRID 3D -------------------------------------------------- //-----------------114 Sum array example with validity check -------------------------------------------------- //#include "cuda_runtime.h" //#include "device_launch_parameters.h" ////#include "cuda_common.cuh" // //#include <stdio.h> //#include "common.h" // incluye metodo para comparar matrices // //// for random initialize //#include <stdlib.h> //#include <time.h> // //// for memset //#include <cstring> //using namespace std; // //__global__ void sum_array_gpu(int* a, int* b, int* c, int size) //{ // int gid = blockIdx.x * blockDim.x + threadIdx.x; // // if (gid < size) // verificar si el indice global esta dentro del tamao de nuestra matriz // { // c[gid] = a[gid] + b[gid]; // } //} // //// funcion para verificar resultado de gpu //void sum_array_cpu(int* a, int* b, int* c, int size) //{ // for (int i = 0; i < size; i++) // { // c[i] = a[i] + b[i]; // } //} // //int main() //{ // int size = 10000; // tamao de la matriz // int block_size = 128; // tamao del bloque en 128 // int num_bytes = size * sizeof(int); // tamao necesario en bytes // // // punteros host // int* h_a, * h_b, * gpu_results; // // int* h_c; // para verificacion en cpu // // //asignacion de memoria para cada puntero // h_a = (int*)malloc(num_bytes); // h_b = (int*)malloc(num_bytes); // gpu_results = (int*)malloc(num_bytes); // // h_c = (int*)malloc(num_bytes);// para verificacion en cpu // // //inicializacion aleatoria de cada matriz // time_t t; // srand((unsigned)time(&t)); // for (int i = 0; i < size; i++) // { // h_a[i] = (int)(rand() & 0xFF); // valor generado entre 0 y 255 // } // for (int i = 0; i < size; i++) // { // h_b[i] = (int)(rand() & 0xFF); // } // // sum_array_cpu(h_a, h_b, h_c, size); // // memset(gpu_results, 0, num_bytes); // // // punteros device // int* d_a, * d_b, * d_c; // cudaMalloc((int**)&d_a, num_bytes); // cudaMalloc((int**)&d_b, num_bytes); // cudaMalloc((int**)&d_c, num_bytes); // // //tranferencia de matriz h_a y h_b // cudaMemcpy(d_a, h_a, num_bytes, cudaMemcpyHostToDevice); // cudaMemcpy(d_b, h_b, num_bytes, cudaMemcpyHostToDevice); // // //launching the grid // dim3 block(block_size); //tamao de bloque 128 en la dimension X // dim3 grid((size / block.x) + 1); // (10000 / 128) + 128 = GRID 1D de 79 block de 128 hilos cada uno // // sum_array_gpu << <grid, block >> > (d_a, d_b, d_c, size); // cudaDeviceSynchronize(); // // cudaMemcpy(gpu_results, d_c, num_bytes, cudaMemcpyDeviceToHost); // puntero de origen d_c, puntero de destino gpu_results // // // COMPARACION DE RESULTADOS CPU Y GPU // compare_arrays(gpu_results, h_c, size); // // cudaFree(d_c); // cudaFree(d_b); // cudaFree(d_a); // // free(gpu_results); // free(h_b); // free(h_a); // // cudaDeviceReset(); // return 0; // // //} //-----------------115 Error handling -------------------------------------------------- /*Types error * -Compile time errors * -Errors language syntax. * * -Run time errors * -Errors happens while program is running * */ //ejemplo con la suma anterior //#include "cuda_runtime.h" //#include "device_launch_parameters.h" ////#include "cuda_common.cuh" // //#include <stdio.h> //#include "common.h" // incluye metodo para comparar matrices // //// for random initialize //#include <stdlib.h> //#include <time.h> // //// for memset //#include <cstring> // //#include "cuda_common.cuh" // //using namespace std; // //__global__ void sum_array_gpu(int* a, int* b, int* c, int size) //{ // int gid = blockIdx.x * blockDim.x + threadIdx.x; // // if (gid < size) // verificar si el indice global esta dentro del tamao de nuestra matriz // { // c[gid] = a[gid] + b[gid]; // } //} // //// funcion para verificar resultado de gpu //void sum_array_cpu(int* a, int* b, int* c, int size) //{ // for (int i = 0; i < size; i++) // { // c[i] = a[i] + b[i]; // } //} // //int main() //{ // int size = 10000; // tamao de la matriz // int block_size = 128; // tamao del bloque en 128 // int num_bytes = size * sizeof(int); // tamao necesario en bytes // // //ERROR (comprobacion) // cudaError error; // // // punteros host // int* h_a, * h_b, * gpu_results; // // int* h_c; // para verificacion en cpu // // //asignacion de memoria para cada puntero // h_a = (int*)malloc(num_bytes); // h_b = (int*)malloc(num_bytes); // gpu_results = (int*)malloc(num_bytes); // // h_c = (int*)malloc(num_bytes);// para verificacion en cpu // // //inicializacion aleatoria de cada matriz // time_t t; // srand((unsigned)time(&t)); // for (int i = 0; i < size; i++) // { // h_a[i] = (int)(rand() & 0xFF); // valor generado entre 0 y 255 // } // for (int i = 0; i < size; i++) // { // h_b[i] = (int)(rand() & 0xFF); // } // // sum_array_cpu(h_a, h_b, h_c, size); // // memset(gpu_results, 0, num_bytes); // // // punteros device // int* d_a, * d_b, * d_c; // // //--------------------------- // //ERROR FORMA MANUAL // /*error = cudaMalloc((int**)&d_a, num_bytes); // if (error != cudaSuccess) // { // fprintf(stderr, "Error : %s \n", cudaGetErrorString(error)); // }*/ // // //ERROR UTILIZANDO cuda_common.cuh // // gpuErrchk(cudaMalloc((int**)&d_a, num_bytes)); // gpuErrchk(cudaMalloc((int**)&d_b, num_bytes)); // gpuErrchk(cudaMalloc((int**)&d_c, num_bytes)); // // //------------------------------- // //cudaMalloc((int**)&d_a, num_bytes); // //cudaMalloc((int**)&d_b, num_bytes); // //cudaMalloc((int**)&d_c, num_bytes); // // //tranferencia de matriz h_a y h_b // cudaMemcpy(d_a, h_a, num_bytes, cudaMemcpyHostToDevice); // cudaMemcpy(d_b, h_b, num_bytes, cudaMemcpyHostToDevice); // // //launching the grid // dim3 block(block_size); //tamao de bloque 128 en la dimension X // dim3 grid((size / block.x) + 1); // (10000 / 128) + 128 = GRID 1D de 79 block de 128 hilos cada uno // // sum_array_gpu << <grid, block >> > (d_a, d_b, d_c, size); // cudaDeviceSynchronize(); // // cudaMemcpy(gpu_results, d_c, num_bytes, cudaMemcpyDeviceToHost); // puntero de origen d_c, puntero de destino gpu_results // // // COMPARACION DE RESULTADOS CPU Y GPU // compare_arrays(gpu_results, h_c, size); // // cudaFree(d_c); // cudaFree(d_b); // cudaFree(d_a); // // free(gpu_results); // free(h_b); // free(h_a); // // cudaDeviceReset(); // return 0; // // //} //-----------------116 Sum array example with timing -------------------------------------------------- // tiempo de ejecucion en cpu y gpu //#include "cuda_runtime.h" //#include "device_launch_parameters.h" ////#include "cuda_common.cuh" // //#include <stdio.h> //#include "common.h" // incluye metodo para comparar matrices // //// for random initialize //#include <stdlib.h> //#include <time.h> // //// for memset //#include <cstring> // //#include "cuda_common.cuh" // //using namespace std; // //__global__ void sum_array_gpu(int* a, int* b, int* c, int size) //{ // int gid = blockIdx.x * blockDim.x + threadIdx.x; // // if (gid < size) // verificar si el indice global esta dentro del tamao de nuestra matriz // { // c[gid] = a[gid] + b[gid]; // } //} // //// funcion para verificar resultado de gpu //void sum_array_cpu(int* a, int* b, int* c, int size) //{ // for (int i = 0; i < size; i++) // { // c[i] = a[i] + b[i]; // } //} // //int main() //{ // int size = 10000; // tamao de la matriz // int block_size = 128; // tamao del bloque en 128 // int num_bytes = size * sizeof(int); // tamao necesario en bytes // // //ERROR (comprobacion) // cudaError error; // // // punteros host // int* h_a, * h_b, * gpu_results; // // int* h_c; // para verificacion en cpu // // //asignacion de memoria para cada puntero // h_a = (int*)malloc(num_bytes); // h_b = (int*)malloc(num_bytes); // gpu_results = (int*)malloc(num_bytes); // // h_c = (int*)malloc(num_bytes);// para verificacion en cpu // // //inicializacion aleatoria de cada matriz // time_t t; // srand((unsigned)time(&t)); // for (int i = 0; i < size; i++) // { // h_a[i] = (int)(rand() & 0xFF); // valor generado entre 0 y 255 // } // for (int i = 0; i < size; i++) // { // h_b[i] = (int)(rand() & 0xFF); // } // // //******************* // // PARA SUMA DE TIEMPo en la funcion sum_array_cpu // clock_t cpu_start, cpu_end; // cpu_start = (double)clock(); // printf("cpu_start: %0.5f \n", (double)cpu_start); // sum_array_cpu(h_a, h_b, h_c, size); // cpu_end = (double)clock(); // printf("cpu_end: %0.5f \n", (double)cpu_end); // //******************* // // // memset(gpu_results, 0, num_bytes); // memset(gpu_results, 0, num_bytes); // // // // // punteros device // int* d_a, * d_b, * d_c; // // //--------------------------- // //ERROR FORMA MANUAL // /*error = cudaMalloc((int**)&d_a, num_bytes); // if (error != cudaSuccess) // { // fprintf(stderr, "Error : %s \n", cudaGetErrorString(error)); // }*/ // // //ERROR UTILIZANDO cuda_common.cuh // // gpuErrchk(cudaMalloc((int**)&d_a, num_bytes)); // gpuErrchk(cudaMalloc((int**)&d_b, num_bytes)); // gpuErrchk(cudaMalloc((int**)&d_c, num_bytes)); // // //------------------------------- // //cudaMalloc((int**)&d_a, num_bytes); // //cudaMalloc((int**)&d_b, num_bytes); // //cudaMalloc((int**)&d_c, num_bytes); // // //******************* // // TIEMPO QUE UTILIZA GPU PARA tranferir datos a memoria // clock_t htod_start, htod_end; // htod_start = clock(); // //tranferencia de matriz h_a y h_b // cudaMemcpy(d_a, h_a, num_bytes, cudaMemcpyHostToDevice); // cudaMemcpy(d_b, h_b, num_bytes, cudaMemcpyHostToDevice); // htod_end = clock(); // //******************* // // //launching the grid // dim3 block(block_size); //tamao de bloque 128 en la dimension X // dim3 grid((size / block.x) + 1); // (10000 / 128) + 128 = GRID 1D de 79 block de 128 hilos cada uno // // //******************* // // TIEMPO QUE UTILIZA GPU PARA ejecutar el kernel // clock_t gpu_start, gpu_end; // gpu_start = clock(); // sum_array_gpu << <grid, block >> > (d_a, d_b, d_c, size); // gpu_end = clock(); // //******************* // // cudaDeviceSynchronize(); // // // //******************* // // TIEMPO QUE UTILIZA GPU PARA tranferir datos a host // clock_t dtoh_start, dtoh_end; // dtoh_start = clock(); // cudaMemcpy(gpu_results, d_c, num_bytes, cudaMemcpyDeviceToHost); // puntero de origen d_c, puntero de destino gpu_results // dtoh_end = clock(); // //******************* // // // COMPARACION DE RESULTADOS CPU Y GPU // compare_arrays(gpu_results, h_c, size); // // cudaFree(d_c); // cudaFree(d_b); // cudaFree(d_a); // // // VELOCIDAD DE RELOJ DE CPU // printf("sum array CPU execution time: %4.6f \n", (double)((double)(cpu_end - cpu_start) / CLOCKS_PER_SEC)); // // // VELOCIDAD DE RELOJ DE CPU // printf("sum array GPU execution time: %4.6f \n", (double)((double)(gpu_end - gpu_start) / CLOCKS_PER_SEC)); // printf("htod mem transfer time: %4.6f \n", (double)((double)(htod_end - htod_start) / CLOCKS_PER_SEC)); // printf("dtod mem transfer time: %4.6f \n", (double)((double)(dtoh_end - dtoh_start) / CLOCKS_PER_SEC)); // // printf("Sum array GPU total execution time: %4.6f \n", (double)((double)(dtoh_end - htod_start) / CLOCKS_PER_SEC)); // // // free(gpu_results); // free(h_b); // free(h_a); // // cudaDeviceReset(); // return 0; // // //} //----------------------------------117 Device properties -------------------------------------------------- #include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> void query_device() { int deviceCount = 0; cudaGetDeviceCount(&deviceCount); if (deviceCount == 0) { printf("No CUDA support device found"); } int devNo = 0; cudaDeviceProp iProp; cudaGetDeviceProperties(&iProp, devNo); printf("Device %d: %s\n", devNo, iProp.name); printf(" Number of multiprocessors: %d\n", iProp.multiProcessorCount); printf(" clock rate : %d\n", iProp.clockRate); printf(" Compute capability : %d.%d\n", iProp.major, iProp.minor); printf(" Total amount of global memory: %4.2f KB\n", iProp.totalGlobalMem / 1024.0); printf(" Total amount of constant memory: %4.2f KB\n", iProp.totalConstMem / 1024.0); printf(" Total amount of shared memory per block: %4.2f KB\n", iProp.sharedMemPerBlock / 1024.0); printf(" Total amount of shared memory per MP: %4.2f KB\n", iProp.sharedMemPerMultiprocessor / 1024.0); printf(" Total number of registers available per block: %d\n", iProp.regsPerBlock); printf(" Warp size: %d\n", iProp.warpSize); printf(" Maximum number of threads per block: %d\n", iProp.maxThreadsPerBlock); printf(" Maximum number of threads per multiprocessor: %d\n", iProp.maxThreadsPerMultiProcessor); printf(" Maximum number of warps per multiprocessor: %d\n", iProp.maxThreadsPerMultiProcessor / 32); printf(" Maximum Grid size : (%d,%d,%d)\n", iProp.maxGridSize[0], iProp.maxGridSize[1], iProp.maxGridSize[2]); printf(" Maximum block dimension : (%d,%d,%d)\n", iProp.maxThreadsDim[0], iProp.maxThreadsDim[1], iProp.maxThreadsDim[2]); } int main() { query_device(); }
12,970
#include "includes.h" __global__ void init(int *arr, int sqroot, int limit) { int c; for(c = 2; c <= sqroot; c++) { if(arr[c] == 0) { /* #pragma omp parallel for shared(arr, limit, c) private(m) for(m = c+1; m < limit; m++) { if(m%c == 0) { arr[m] = 1; } } */ int tid = c+1+ threadIdx.x + (blockIdx.x * blockDim.x); if (tid<limit){ if (tid % c ==0) { arr[tid] = 1; } } } } }
12,971
#include <stdio.h> __global__ void kernel( void ) { /* Do something fun! */ } int main(void) { kernel<<<1,1>>>(); printf( "Hello, World!\n" ); return 0; }
12,972
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> #include <stdlib.h> #include <fstream> #include <iostream> #include <chrono> __global__ void convoluteGPU(int* pixeliIntrare, int* pixeliIesire, int linii, int coloane, int canaleCuloare) { int kernel[5][5] = { { 0, 0, -1, 0, 0}, { 0, -1, -2, -1, 0}, {-1, -2, 16, -2, -1}, { 0, -1, -2, -1, 0}, { 0, 0, -1, 0, 0} }; int id = blockDim.x * blockIdx.x + threadIdx.x; if (id < linii * coloane * canaleCuloare) { //out[id] = in[id]+20; //apply kernel int linie = id / (coloane * canaleCuloare); int coloana = (id % (coloane * canaleCuloare)) / canaleCuloare; int canalCuloare = id % canaleCuloare; int pixel = 0; if (linie > 1 && linie < (linii - 2) && coloana > 1 && coloana < (coloane - 2)) { //mijloc pixel += kernel[2][2] * pixeliIntrare[id]; //N pixel += kernel[1][2] * pixeliIntrare[id - coloane * canaleCuloare]; //NE pixel += kernel[1][3] * pixeliIntrare[id - coloane * canaleCuloare + canaleCuloare]; //E pixel += kernel[2][3] * pixeliIntrare[id + canaleCuloare]; //SE pixel += kernel[3][3] * pixeliIntrare[id + coloane * canaleCuloare + canaleCuloare]; //S pixel += kernel[3][2] * pixeliIntrare[id + coloane * canaleCuloare]; //SV pixel += kernel[3][1] * pixeliIntrare[id + coloane * canaleCuloare - canaleCuloare]; //V pixel += kernel[2][1] * pixeliIntrare[id - canaleCuloare]; //NV pixel += kernel[1][1] * pixeliIntrare[id - coloane * canaleCuloare - canaleCuloare]; //conturul kernelului //N pixel += kernel[0][0] * pixeliIntrare[id - 2 * coloane * canaleCuloare - 2 * canaleCuloare]; pixel += kernel[0][1] * pixeliIntrare[id - 2 * coloane * canaleCuloare - canaleCuloare]; pixel += kernel[0][2] * pixeliIntrare[id - 2 * coloane * canaleCuloare]; pixel += kernel[0][3] * pixeliIntrare[id - 2 * coloane * canaleCuloare + canaleCuloare]; pixel += kernel[0][4] * pixeliIntrare[id - 2 * coloane * canaleCuloare + 2 * canaleCuloare]; //E pixel += kernel[1][4] * pixeliIntrare[id - 1 * coloane * canaleCuloare + 2 * canaleCuloare]; pixel += kernel[2][4] * pixeliIntrare[id - 0 * coloane * canaleCuloare + 2 * canaleCuloare]; pixel += kernel[3][4] * pixeliIntrare[id + 1 * coloane * canaleCuloare + 2 * canaleCuloare]; //S pixel += kernel[4][0] * pixeliIntrare[id + 2 * coloane * canaleCuloare - 2 * canaleCuloare]; pixel += kernel[4][1] * pixeliIntrare[id + 2 * coloane * canaleCuloare - canaleCuloare]; pixel += kernel[4][2] * pixeliIntrare[id + 2 * coloane * canaleCuloare]; pixel += kernel[4][3] * pixeliIntrare[id + 2 * coloane * canaleCuloare + canaleCuloare]; pixel += kernel[4][4] * pixeliIntrare[id + 2 * coloane * canaleCuloare + 2 * canaleCuloare]; //V pixel += kernel[1][0] * pixeliIntrare[id - 1 * coloane * canaleCuloare - 2 * canaleCuloare]; pixel += kernel[2][0] * pixeliIntrare[id - 0 * coloane * canaleCuloare - 2 * canaleCuloare]; pixel += kernel[3][0] * pixeliIntrare[id + 1 * coloane * canaleCuloare - 2 * canaleCuloare]; pixel = pixel / 1; } else { pixel = 0; } pixeliIesire[id] = pixel; } } int* mapareMatricePixeliRGBLaVector(int*** imagine, int linii, int coloane, int canaleCuloare) { int* vector = (int*)malloc(linii * coloane * canaleCuloare * sizeof(int)); int id = 0; for (int i = 0; i < linii; i++) { for (int j = 0; j < coloane; j++) { for (int c = 0; c < canaleCuloare; c++) { vector[id] = imagine[i][j][c]; id++; } } } return vector; } int*** mapareVectorLaMatricePixeliRGB(int* vector, int linii, int coloane, int canaleCuloare) { int*** imagine = (int***)malloc(linii * sizeof(int**)); int id = 0; for (int i = 0; i < linii; i++) { imagine[i] = (int**)malloc(coloane * sizeof(int*)); for (int j = 0; j < coloane; j++) { imagine[i][j] = (int*)malloc(canaleCuloare * sizeof(int)); for (int c = 0; c < canaleCuloare; c++) { imagine[i][j][c] = vector[id]; id++; } } } return imagine; } void aplicareFiltru() { //citim matricea de pixeli RGB std::ifstream in("pixels.txt"); int linii, coloane, canaleCuloare; in >> linii >> coloane >> canaleCuloare; int BLOCK_SIZE = 1000; int blockCount = ((linii * coloane * canaleCuloare) / BLOCK_SIZE) + 1; //citire in memorie int*** matrix = (int***)malloc(linii * sizeof(int**)); for (int i = 0; i < linii; i++) { matrix[i] = (int**)malloc(coloane * sizeof(int*)); for (int j = 0; j < coloane; j++) { int* line = (int*)malloc(canaleCuloare * sizeof(int)); in >> line[0] >> line[1] >> line[2]; matrix[i][j] = line; } } int dimensiune = linii * coloane * canaleCuloare; //maparea matricei la vector int* vector = mapareMatricePixeliRGBLaVector(matrix, linii, coloane, canaleCuloare); int* rezultat = (int*)malloc(dimensiune * sizeof(int)); //copiem vectorul de pixeli in vectorDevice int* vectorDevice; int* rezultatDevice; cudaMalloc(&vectorDevice, dimensiune * sizeof(int)); cudaMalloc(&rezultatDevice, dimensiune * sizeof(int)); cudaMemcpy( vectorDevice, vector, dimensiune * sizeof(int), cudaMemcpyHostToDevice ); //apelam filtrul convolutional (better: multiplu de 2 ca numar de thread-uri) (test: different block sizes) convoluteGPU <<< blockCount, BLOCK_SIZE >>> (vectorDevice, rezultatDevice, linii, coloane, canaleCuloare); //copiem rezultatDevice in rezultat cudaMemcpy( rezultat, rezultatDevice, dimensiune * sizeof(int), cudaMemcpyDeviceToHost ); int*** imagine = mapareVectorLaMatricePixeliRGB(rezultat, linii, coloane, canaleCuloare); std::ofstream out("pixels.txt"); out << linii << " " << coloane << " " << canaleCuloare << "\n"; for (int i = 0; i < linii; i++) { for (int j = 0; j < coloane; j++) { for (int k = 0; k < canaleCuloare; k++) { out << imagine[i][j][k] << " "; } out << "\n"; } } out.close(); } int main() { char* pathFisierIntrare = "python in.py C:/Users/George/source/repos/P2/P2/landscape.png"; char* pathFisierIesire = "python out.py C:/Users/George/source/repos/P2/P2/landscape1.png"; system(pathFisierIntrare); //citim si scriem valoarea pixelilor in pixels.txt auto start = std::chrono::steady_clock::now(); aplicareFiltru(); auto stop = std::chrono::steady_clock::now(); system(pathFisierIesire); //scriem pixelii in imaginea filtrata // we cuda've done that auto diff = stop - start; std::cout << std::chrono::duration <double, std::milli>(diff).count() << " ms" << std::endl; return 0; }
12,973
/*----------- * * atomics.cu * * This is the source file of antomic operations. * * This kernel is based on CUDA samples. simpleAtomicIntrinsics.cuh * * streamsOptBenchmark/atomics.cu * * By Hao Li * *------------ */ #include <time.h> #include <cuda_runtime.h> // #include "functions.cuh" __global__ void atomicFunc(float *g_idata, float *g_odata) { for(int l = 0; l < 100000; l++) { // access thread id const unsigned int tid = blockDim.x * blockIdx.x + threadIdx.x; // Test various atomic instructions // Arithmetic atomic instructions int i = 0; while(g_odata[i] != NULL) { g_odata[i] = g_idata[i]; // Atomic addition atomicAdd(&g_odata[i], 10.0); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic subtraction (final should be 0) atomicSub((int *)&g_odata[i], 10); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic exchange atomicExch(&g_odata[i], (float)tid); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic maximum atomicMax((int *)&g_odata[i], tid); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic minimum atomicMin((int *)&g_odata[i], tid); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic increment (modulo 17+1) atomicInc((unsigned int *)&g_odata[i], 17); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic decrement atomicDec((unsigned int *)&g_odata[i], 137); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic compare-and-swap atomicCAS((int *)&g_odata[i], tid-1, tid); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Bitwise atomic instructions // Atomic AND atomicAnd((int *)&g_odata[i], 2*tid+7); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic OR atomicOr((int *)&g_odata[i], 1 << tid); if(g_odata[i++] != NULL) break; g_odata[i] = g_idata[i]; // Atomic XOR atomicXor((int *)&g_odata[i], tid); i++; } } } // int main(int argc, char **argv) // { // unsigned int numThreads = 256; // unsigned int numBlocks = 64; // unsigned int numData = 1000000; // unsigned int memSize = sizeof(int) * numData; // //allocate mem for the result on host side // int *hOData = (int *) malloc(memSize); // //initalize the memory // for (unsigned int i = 0; i < numData; i++) // hOData[i] = 0; // //To make the AND and XOR tests generate something other than 0... // hOData[8] = hOData[10] = 0xff; // // allocate device memory for result // float *dOData; // cudaMalloc((void **) &dOData, sizeof(float) * memSize); // // copy host memory to device to initialize to zers // cudaMemcpy(dOData, hOData, sizeof(float) * memSize, cudaMemcpyHostToDevice); // cudaEvent_t start; // error = cudaEventCreate(&start); // if (error != cudaSuccess) // { // fprintf(stderr, "Failed to create start event (error code %s)!\n", cudaGetErrorString(error)); // exit(EXIT_FAILURE); // } // cudaEvent_t stop; // error = cudaEventCreate(&stop); // if (error != cudaSuccess) // { // fprintf(stderr, "Failed to create stop event (error code %s)!\n", cudaGetErrorString(error)); // exit(EXIT_FAILURE); // } // // Record the start event // error = cudaEventRecord(start, NULL); // if (error != cudaSuccess) // { // fprintf(stderr, "Failed to record start event (error code %s)!\n", cudaGetErrorString(error)); // exit(EXIT_FAILURE); // } // // execute the kernel // atomicFunc<<<numBlocks, numThreads>>>(dOData); // // Record the stop event // error = cudaEventRecord(stop, NULL); // if (error != cudaSuccess) // { // fprintf(stderr, "Failed to record stop event (error code %s)!\n", cudaGetErrorString(error)); // exit(EXIT_FAILURE); // } // // Wait for the stop event to complete // error = cudaEventSynchronize(stop); // if (error != cudaSuccess) // { // fprintf(stderr, "Failed to synchronize on the stop event (error code %s)!\n", cudaGetErrorString(error)); // exit(EXIT_FAILURE); // } // float msecTotal = 0.0f; // error = cudaEventElapsedTime(&msecTotal, start, stop); // printf("Running Time: %f ms\n", msecTotal); // cudaMemcpy(hOData, dOData, memSize, cudaMemcpyDeviceToHost); // free(hOData); // cudaFree(dOData); // return 0; // }
12,974
#include <stdio.h> #include <string> #include <iostream> using namespace std; //no longer require std:: prefix for string functions int getSPcores(cudaDeviceProp devProp, std::string& arch) { int cores = 0; int mp = devProp.multiProcessorCount; arch="unknown"; switch (devProp.major){ case 2: // Fermi arch="Fermi"; if (devProp.minor == 1) cores = mp * 48; else cores = mp * 32; break; case 3: // Kepler arch="Kepler"; cores = mp * 192; break; case 5: // Maxwell arch="Maxwell"; cores = mp * 128; break; case 6: // Pascal arch="Pascal"; if (devProp.minor == 1) cores = mp * 128; else if (devProp.minor == 0) cores = mp * 64; else printf("Unknown device type\n"); break; case 7: // Volta and Turing arch="Volta|Turing"; if ((devProp.minor == 0) || (devProp.minor == 5)) cores = mp * 64; else printf("Unknown device type\n"); break; default: printf("Unknown device type\n"); break; } return cores; } int main() { string arch("undefined"); cudaDeviceProp prop; int nDevices; cudaGetDeviceCount(&nDevices); for (int i = 0; i < nDevices; i++) { cudaGetDeviceProperties(&prop, i); printf("Device Number: %d\n", i); printf(" Device name: %s\n", prop.name); printf(" CUDA Cores: %d\n", getSPcores(prop, arch)); cout<< " Architecture: "+arch+"\n"; printf(" Memory Clock Rate (KHz): %d\n", prop.memoryClockRate); printf(" Memory Bus Width (bits): %d\n", prop.memoryBusWidth); printf(" Peak Memory Bandwidth (GB/s): %f\n\n", 2.0*prop.memoryClockRate*(prop.memoryBusWidth/8)/1.0e6); } }
12,975
#include <stdio.h> #include <cuda.h> #include <stdlib.h> #include <time.h> __global__ void matrixMult(int *a, int *b, int *c, int N) { int row = blockIdx.y * blockDim.y + threadIdx.y; int col = blockIdx.x * blockDim.x + threadIdx.x; int sum = 0; int index, i; if (row < N && col < N) { index = col + (row * N); for (i = 0; i < N; i++) { sum += a[i + (row * N)] * b[col + (i * N)]; } c[index] = sum; } } int main(void) { int N, T, B, repeat; repeat = 1; while (repeat == 1) { printf("Enter size of matrices: "); scanf("%d", &N); while (N <= 0) { printf( "Size of matrices must be greater than 0. Enter a valid size of matrices: "); scanf("%d", &N); } printf("Enter number of threads in a block: "); scanf("%d", &T); while (T <= 0) { printf( "Number of threads must be greater than 0. Enter number of threads in a block: "); scanf("%d", &T); } while (T > 1024) { printf( "Number of threads must not exceed the device bandwidth. Enter number of threads in a block: "); scanf("%d", &T); } printf("Enter number of blocks in a grid: "); scanf("%d", &B); while (B <= 0) { printf( "Number of blocks must be greater than 0. Enter number of blocks in a grid: "); scanf("%d", &B); } while (B > 65535) { printf( "Number of blocks must not exceed the device bandwidth. Enter number of blocks in a grid: "); scanf("%d", &B); } int *a, *b, *c, *deviceC; int *dev_a, *dev_b, *dev_c; int i, j, k; int ssd = 0; cudaEvent_t start, stop; float elapsedTime; cudaEventCreate(&start); cudaEventCreate(&stop); cudaMalloc((void**) &dev_a, (N * N) * sizeof(int)); cudaMalloc((void**) &dev_b, (N * N) * sizeof(int)); cudaMalloc((void**) &dev_c, (N * N) * sizeof(int)); a = (int *) malloc((N * N) * sizeof(int)); b = (int *) malloc((N * N) * sizeof(int)); c = (int *) malloc((N * N) * sizeof(int)); deviceC = (int *) malloc((N * N) * sizeof(int)); srand(time(NULL)); //loop will generate the matrix with random integers from 0 to 9 for (i = 0; i < N; i++) { for (j = 0; j < N; j++) { a[j + (i * N)] = (int) rand() % 10; b[j + (i * N)] = (int) rand() % 10; } } /*******************begin host code*****************************/ clock_t begin, end; begin = clock(); for (i = 0; i < N; i++) { for (j = 0; j < N; j++) { c[j + i * N] = 0; for (k = 0; k < N; k++) { c[j + i * N] = c[j + i * N] + a[k + i * N] * b[j + k * N]; } } } end = clock(); printf("It took %f seconds for the host to do the matrix operation.\n", (float) (end - begin) / (CLOCKS_PER_SEC)); /*******************end host code*****************************/ /*******************begin device code*****************************/ cudaMemcpy(dev_a, a, (N * N) * sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(dev_b, b, (N * N) * sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(dev_c, deviceC, (N * N) * sizeof(int), cudaMemcpyHostToDevice); dim3 grid(B, B); dim3 block(T, T); cudaEventRecord(start, 0); matrixMult<<<grid, block>>>(dev_a, dev_b, dev_c, N); cudaThreadSynchronize(); cudaEventRecord(stop, 0); cudaEventSynchronize(stop); cudaMemcpy(deviceC, dev_c, (N * N) * sizeof(int), cudaMemcpyDeviceToHost); cudaEventElapsedTime(&elapsedTime, start, stop); printf( "It took %f seconds for the device to do the matrix operation.\n", (elapsedTime / 1000)); for (i = 0; i < N; i++) { for (j = 0; j < N; j++) { ssd += deviceC[j + (i * N)] - c[j + (i * N)]; } } printf("The sum of square difference is %d.\n", ssd); printf("The speedup factor is %f.\n", ((float) (end - begin) / (CLOCKS_PER_SEC)) / (elapsedTime / 1000)); cudaEventDestroy(start); cudaEventDestroy(stop); free(a); free(b); free(c); free(deviceC); cudaFree(dev_a); cudaFree(dev_b); cudaFree(dev_c); /*******************end device code*****************************/ printf("Enter 1 to continue: "); scanf("%d", &repeat); } return 0; }
12,976
#include <chrono> #include <iostream> __global__ void _cuda_parallel_multiplication(int count, int* test_data, int magnitude); int main() { int count = 60000000; // 60 million elements int* test_data = new int[count]; for(int i = 0; i < count; i++) test_data[i] = i; // Perform calculation on host CPU auto t1 = std::chrono::high_resolution_clock::now(); for(int i = 0; i < count; i++) test_data[i] = test_data[i] * 5; auto t2 = std::chrono::high_resolution_clock::now(); // Copy data to device int* d_test_data; cudaMalloc(&d_test_data, count * sizeof(int)); cudaMemcpy(d_test_data, test_data, count * sizeof(int), cudaMemcpyHostToDevice); // Launch kernel int block_count = ceil((double)count / 1024); _cuda_parallel_multiplication<<<10, 1024>>>(count, d_test_data, 5); cudaDeviceSynchronize(); cudaMemcpy(test_data, d_test_data, count * sizeof(int), cudaMemcpyDeviceToHost); cudaFree(d_test_data); for(int i = 0; i < 10; i++) std::cout << i << ": " << test_data[i] << std::endl; // Copy results back to device std::cout << "CPU time: " << std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count() << "ms" << std::endl; } __global__ void _cuda_parallel_multiplication(int count, int* test_data, int magnitude) { int globalIdx = blockIdx.x * blockDim.x + threadIdx.x; while (globalIdx < count) { test_data[globalIdx] = test_data[globalIdx] * magnitude; globalIdx += blockDim.x * gridDim.x; __syncthreads(); } }
12,977
#include <stdio.h> #include <time.h> int generateInitialPrimes(int *intialTempArray, int **PL, int initialPrimesRange); __global__ void calcPrimes(int *d_IL, int *d_PL, int numOfPrimes, int lenInputList); #define LEN_IL 1000000 #define LEN_INITIAL_PRIMES 1000 #define THREADS_PER_BLOCK 32 int main() { int *IL = NULL, *PL = NULL, *tempPL = NULL; int *d_IL = NULL, *d_PL = NULL; cudaEvent_t start, stop; float time; cudaEventCreate(&start); cudaEventCreate(&stop); clock_t t; //int count = 0; t = clock(); int numOfInitialPrimes = generateInitialPrimes(tempPL, &PL, LEN_INITIAL_PRIMES); t = clock() - t; double time_taken = ((double)t)/CLOCKS_PER_SEC; // in seconds // Print the initial range of primes calculated in the CPU, which will be passed to the GPU: printf("\nThe initial primes calculated are:\n"); for(int i=0; i < numOfInitialPrimes; i++) { printf("%d ", PL[i]); } printf("\nNumber of initial primes = %d\n\n", numOfInitialPrimes); // Space for host copies: IL = (int*) malloc(LEN_IL * sizeof(int)); //PL = (int*) malloc(LEN_INITIAL_PRIMES * sizeof(int)); // Allocated in the generate function instead int size_IL = LEN_IL * sizeof(int); int size_PL = numOfInitialPrimes * sizeof(int); // Initialize Input list: 0 -> Not prime: for(int i=0; i<LEN_IL; i++) { IL[i] = 1; } // Space for device copies: cudaMalloc((void **) &d_IL, size_IL); cudaMalloc((void **) &d_PL, size_PL); // Copying the data to the device (GPU): cudaMemcpy(d_IL, IL, size_IL, cudaMemcpyHostToDevice); cudaMemcpy(d_PL, PL, size_PL, cudaMemcpyHostToDevice); // Launching the kernel and measuring the time taken: cudaEventRecord(start, 0); calcPrimes<<<(numOfInitialPrimes/THREADS_PER_BLOCK) + 1, THREADS_PER_BLOCK>>> (d_IL, d_PL, numOfInitialPrimes, LEN_IL); cudaEventRecord(stop, 0); cudaEventSynchronize(stop); cudaEventElapsedTime(&time, start, stop); // Space allocated to store the modified form of input array, with marking for prime and non-prime: int *result = (int*) malloc(size_IL); // Copy the result back to the host: cudaMemcpy(result, d_IL, size_IL, cudaMemcpyDeviceToHost); // Extract indexes of primes in 'result' to get the actual new prime numbers: printf("********* New Primes List **********\n"); int *newPrimes = (int*)malloc(LEN_IL / 4 * sizeof(int)); // Arbitrary size; which is '1/4th' of numbers list size int newPrimesCount = 0; for(int i=LEN_INITIAL_PRIMES; i<LEN_IL; i++) { int num = result[i]; if(num == 1) { newPrimes[newPrimesCount] = num; newPrimesCount++; printf("%d ", i); } } printf("\n\nNumber of new primes found = %d\n\n", newPrimesCount); printf("Time taken to find initial primes on CPU = %f ms\n", time_taken * 1000); printf("Parallel Job time for current iteration = %f ms\n\n", time); // Free memory: cudaFree(d_IL); cudaFree(d_PL); free(IL); free(PL); free(result); free(newPrimes); return 0; } // Generate initial prime numbers in the CPU: // Returns: Number of primes found from 1 to 'LEN_INITIAL_PRIMES' int generateInitialPrimes(int *intialTempArray, int **PL, int initialPrimesRange) { int primesCount = 0; //int intialTempArray[initialPrimesRange]; intialTempArray = (int*) malloc(LEN_INITIAL_PRIMES * sizeof(int)); *PL = (int*) malloc(LEN_INITIAL_PRIMES / 2 * sizeof(int)); // Taking half size of initial (full) primes array // Initialize array with all 1's: for(int i=0; i < initialPrimesRange; i++) { intialTempArray[i] = 1; } // Make non-primes as '0': for(int i=2; i*i <= initialPrimesRange; i++) { for(int j=2*i; j <= initialPrimesRange; j=j+i) { intialTempArray[j] = 0; } } // Store the actual primes in a new array which will be copied later to the device (converting 'prime num indexes' to 'prime numbers') : for(int i=2; i<=initialPrimesRange; i++) { if(intialTempArray[i] == 1) { (*PL)[primesCount] = i; primesCount++; } } free(intialTempArray); return primesCount; } // GPU Kernel (Parallel Processing): __global__ void calcPrimes(int *d_IL, int *d_PL, int numOfPrimes, int lenInputList) { int index = threadIdx.x + blockIdx.x * blockDim.x; if(index < numOfPrimes) { for(int i = d_PL[numOfPrimes-1]+1; i < lenInputList; i++) { if(i % d_PL[index] == 0) { d_IL[i] = 0; } } } }
12,978
// // Created by Peter Rigole on 2019-04-26. // #ifndef AXONBITS_CYCLEPARITY_H #define AXONBITS_CYCLEPARITY_H /** * All threads working on a cycle have the same cycle parity. This parity is used to identify the activity variable * in the neuron that is to be updated (the next activity) versus the one that must be used as the neuron's current * activity. */ enum CycleParity { EvenCycle, OddCycle }; #endif //AXONBITS_CYCLEPARITY_H
12,979
// 1D version of the code #include<stdio.h> #include<stdlib.h> #include<cuda.h> #define N 5 #define M 6 __global__ void dkernel(unsigned *mat) { unsigned xId = blockIdx.x, yId = threadIdx.x, col = blockDim.x; mat[xId * col + yId] = xId * col + yId; } int main() { //dim3 block(N, M, 1); unsigned *matrix, *hmatrix, i = 0, j = 0; cudaMalloc(&matrix, N * M * sizeof(unsigned)); hmatrix = (unsigned *)malloc(N * M * sizeof(unsigned)); dkernel<<<N, M>>>(matrix); cudaMemcpy(hmatrix, matrix, N * M * sizeof(unsigned), cudaMemcpyDeviceToHost); for(i = 0; i < N; i++) { for(j = 0; j < M; j++) { printf("%d\t", hmatrix[i * M + j]); } printf("\n"); } return 0; }
12,980
#include <iostream> #include <cuda.h> using namespace std; __global__ void dkernel(int *arr, int N){ unsigned id = blockIdx.x * blockDim.x + threadIdx.x; while(id < N) { arr[id] = 0; id += blockDim.x * gridDim.x ; } } __global__ void dkernel_add(int *arr, int N) { unsigned id = blockIdx.x * blockDim.x + threadIdx.x; while (id < N) { arr[id] += id; id += blockDim.x * gridDim.x; } } int main() { int *gpuArray, *cpuArray; //int *cpuArray = new int[32]; cudaMallocManaged(&cpuArray, 32*sizeof(int)); cudaMallocManaged(&gpuArray, 32*sizeof(int)); dkernel<<<1, 32>>>(gpuArray,32); cudaMemcpy(cpuArray, gpuArray, 32*sizeof(int), cudaMemcpyDeviceToHost); for (int i = 0; i < 32 ; i++) { cout <<"cpuArray["<<i<<"]"<<cpuArray[i]<< endl; } cudaMallocManaged(&gpuArray, 1024*sizeof(int)); dkernel<<<1, 1024>>>(gpuArray, 1024); dkernel_add<<<1, 1024>>>(gpuArray, 1024); //cpuArray = new int[1024]; cudaMallocManaged(&cpuArray, 1024*sizeof(int)); cudaMemcpy(cpuArray, gpuArray, 1024*sizeof(int), cudaMemcpyDeviceToHost); for (int i = 0 ; i < 1024 ; i++) { cout << "cpuArray["<<i<<"]"<< cpuArray[i] << endl; } cudaMallocManaged(&gpuArray, 8000*sizeof(int)); dkernel<<<8000/128, 128>>>(gpuArray, 8000); dkernel_add<<<8000,128>>>(gpuArray, 8000); cpuArray = new int[8000]; cudaMemcpy(cpuArray, gpuArray, 8000*sizeof(int), cudaMemcpyDeviceToHost); for(int i = 0; i < 8000 ; i++) { cout << "cpuArray["<<i<<"]" << cpuArray[i] << endl; } }
12,981
/*Created by Alessandro Bigiotti*/ // Kernel Function to calculate the Unit Clause of the formula __global__ void UnitClause(int NValid, int b, int t, int *nextpos, int *num_args, int *lett, int *clause, int *matrixelem, int *col, int *row, int *poslet, int *tipo, int n, int m, int nk) { int idx = threadIdx.x; int idblock = blockIdx.x; int indexpos = 0; int indexclause = 0; if (idblock*t + idx < m){ if (tipo[idblock*t + idx] == 1){ indexpos = poslet[idblock*t + idx]; indexclause = idblock*t + idx; nextpos[indexclause] = 1; nextpos[m] = 1; lett[indexpos] = 1; } } } // kernel_function to calculate the Propagation of the assignments __global__ void Propagate(int NValid, int b, int t, int *nextpos, int *num_args, int *lett, int *clause, int *matrixelem, int *col, int *row, int *poslet, int *tipo, int n, int m, int nk) { int idx = threadIdx.x; int idblock = blockIdx.x; int indexpos = 0; int indexnextpos = 0; if (idblock*t + idx < m){ if (nextpos[idblock*t + idx] == 1){ indexpos = poslet[idblock*t + idx]; nextpos[idblock*t + idx] = 0; for (int i = row[indexpos]; i < row[indexpos + 1]; i++){ if (matrixelem[i] == 2){ int old = atomicSub(num_args + col[i], 1); if (old == 1){ indexnextpos = poslet[col[i]]; if (indexnextpos != NValid){ if (lett[indexnextpos] == 0){ lett[indexnextpos] = 1; nextpos[col[i]] = 1; nextpos[m] = 1; } } else{ if (tipo[col[i]] % 2 == 0){ nextpos[m + 1] = NValid; break; } } } } } } } }
12,982
#include "includes.h" //FILE IO RELATED //max number of lines in the training dataset #define MAX_ROWS_TRAINING 16896 // max number of columns/features in the training dataset #define MAX_COLUMNS_TRAINING 26 // max number of rows in the testing dataset #define MAX_ROWS_TESTING 4096 // max number of columns in the testing data #define MAX_COLUMNS_TESTING 26 //max number of characters/line #define MAX_CHAR 300 __constant__ int features = 26; __constant__ int num_rows = 16896; long mem_cpy_time = 0; long beta_cpy_time = 0; // parallelized across the rows // parallelized across the features __global__ void log_gradient(float* log_func_v, float* gradient, float* betas, float* data, int* yvec) { // the logistic function itself has been pulled out int feature_index = blockIdx.x * blockDim.x + threadIdx.x; float temp = 0.0f; for(int i = 0; i < num_rows; i++) { float sub = log_func_v[i] - yvec[i]; float accessed_data = data[(i * features) + feature_index]; temp += sub * accessed_data; } gradient[feature_index] = temp; }
12,983
#include <dirent.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <malloc.h> #define MAP_COUNT __device__ void mapCount(char*key,char*value,size_t key_size, size_t value_size,int*key_im_size,int*value_im_size,int*map_im_num,int threadID) #define EMIT_IM_COUNT(im_key_size,im_value_size) emitMapCount(im_key_size,im_value_size,word_num,key_im_size,value_im_size,map_im_num,threadID) /*extern */MAP_COUNT; typedef struct MapFileList { char* filename; struct MapFileList* next; }MapFileList; typedef enum InputFormat{TextInputFormat,KeyValueInputFormat,SequenceFileInputFormat} input_format; typedef struct Index{ size_t key_offset; size_t key_size; size_t value_offset; size_t value_size; }Index; typedef struct MapReduceSpec{ MapFileList* map_file_list; char* map_input_keys; char* map_input_values; Index* map_input_index; int* map_im_key_size; int* map_im_value_size; int* map_im_num; int map_input_num; int map_block_num; int map_thread_num; input_format map_input_format; }MapReduceSpec; void init_map_file_list(MapFileList* list){ list->filename=NULL; list->next=NULL; } void free_map_file_list(MapFileList* list){ MapFileList* del; MapFileList* tmp; del=list; tmp=list->next; while(tmp){ if(del->filename!=NULL) free(del->filename); free(del); del=tmp; tmp=tmp->next; } if(del->filename!=NULL) free(del->filename); free(del); } void init_mapreduce_spec(MapReduceSpec* spec){ spec->map_file_list=NULL; spec->map_input_keys=NULL; spec->map_input_values=NULL; spec->map_input_index=NULL; spec->map_im_key_size=NULL; spec->map_im_value_size=NULL; spec->map_im_num=NULL; spec->map_input_num=0; spec->map_block_num=0; spec->map_thread_num=512; spec->map_input_format=TextInputFormat; } void free_spec(MapReduceSpec* spec){ free_map_file_list(spec->map_file_list); free(spec->map_input_keys); free(spec->map_input_values); free(spec->map_input_index); free(spec->map_im_key_size); free(spec->map_im_value_size); free(spec->map_im_num); free(spec); } char *my_strncpy(char *dest, const char *src, size_t n) { size_t i; for (i = 0; i < n && src[i] != '\0'; i++) dest[i] = src[i]; for ( ; i < n; i++) dest[i] = '\0'; return dest; } void map_input_split(MapReduceSpec* spec){ MapFileList* file_list_entry; size_t buffer_size=(size_t)256*1024*1024; size_t buffer_used=0; FILE* pFile; file_list_entry=spec->map_file_list; size_t file_size; size_t key_array_size; size_t value_array_size; size_t index_array_size; if(spec->map_input_format==TextInputFormat){ file_size=key_array_size=value_array_size=index_array_size=0; while(file_list_entry->filename!=NULL){ pFile=fopen(file_list_entry->filename,"rb"); if (pFile==NULL) {fputs ("File error\n",stderr); exit (1);} fseek (pFile , 0 , SEEK_END); file_size = ftell (pFile); rewind (pFile); if(buffer_used+file_size<=buffer_size){ ssize_t result=0; while (result!= -1) { size_t value_size = 0; size_t key_size=0; char* temp_key=NULL; char* temp_value=NULL; temp_key=(char*)malloc(10); sprintf(temp_key,"%d",(int)ftell(pFile)); key_size=strlen(temp_key)+1; //get the new key's size spec->map_input_keys=(char*)realloc(spec->map_input_keys,key_array_size+key_size); //reallocate key_array, so that it can contain new keys my_strncpy((spec->map_input_keys)+key_array_size,temp_key,key_size); result=getline(&(temp_value), &value_size, pFile); value_size=strlen(temp_value)+1; spec->map_input_values=(char*)realloc(spec->map_input_values,value_array_size+value_size); //reallocate value_size, so that it can contain new values strcpy((char*)(spec->map_input_values+value_array_size),temp_value); spec->map_input_index=(Index*)realloc(spec->map_input_index,(index_array_size+1)*sizeof(Index)); //reallocate index array, so that it can contain new <key,value> information spec->map_input_index[index_array_size].key_offset=key_array_size; spec->map_input_index[index_array_size].key_size=key_size; spec->map_input_index[index_array_size].value_offset=value_array_size; spec->map_input_index[index_array_size].value_size=value_size; key_array_size+=key_size; value_array_size+=value_size; index_array_size++; free(temp_key); free(temp_value); } buffer_used=buffer_used+file_size; } else printf("Buffer full!!\n"); file_list_entry=file_list_entry->next; fclose(pFile); } spec->map_input_num=index_array_size; printf("Map Input entry number: %i, %u, %u, %u\n",spec->map_input_num,key_array_size,value_array_size,index_array_size*sizeof(Index)); } } __device__ bool isChar(char c){ if(((c<='z')&&(c>='a'))||((c<='Z')&&(c>='A'))) return true; else return false; } __device__ void emitMapCount(int key_size, int value_size,int word_num,int*key_im_size_array,int*value_im_size_array,int*map_im_num,int threadID){ *(key_im_size_array+threadID)=key_size; *(value_im_size_array+threadID)=value_size; *(map_im_num+threadID)=word_num; } __global__ void map_count_warp(char*keys,char*values,Index*index,int*map_im_key_size,int*map_im_value_size,int*map_im_num,int input_num){ int i=blockDim.x*blockIdx.x+threadIdx.x; if(i<input_num){ mapCount((keys+((index+i)->key_offset)),(values+((index+i)->value_offset)),(index+i)->key_size,(index+i)->value_size,map_im_key_size,map_im_value_size,map_im_num,i); } } void map_count_phase(MapReduceSpec* spec){ char* d_map_input_keys; char* d_map_input_values; Index* d_map_input_index; int* d_map_im_key_size; int* d_map_im_value_size; int* d_map_im_num; size_t map_im_size=(spec->map_input_num)*sizeof(int); spec->map_im_key_size=(int*)malloc(map_im_size); spec->map_im_value_size=(int*)malloc(map_im_size); spec->map_im_num=(int*)malloc(map_im_size); size_t keys_size=malloc_usable_size(spec->map_input_keys); size_t values_size=malloc_usable_size(spec->map_input_values); size_t index_size=malloc_usable_size(spec->map_input_index); //printf("%u,%u,%u\n",malloc_usable_size(spec->map_input_keys),malloc_usable_size(spec->map_input_values),malloc_usable_size(spec->map_input_index)); cudaMalloc(&d_map_input_keys,keys_size); cudaMalloc(&d_map_input_values,values_size); cudaMalloc(&d_map_input_index,index_size); cudaMalloc(&d_map_im_key_size,map_im_size); cudaMalloc(&d_map_im_value_size,map_im_size); cudaMalloc(&d_map_im_num,map_im_size); cudaMemcpy(d_map_input_keys,spec->map_input_keys,keys_size,cudaMemcpyHostToDevice); cudaMemcpy(d_map_input_values,spec->map_input_values,values_size,cudaMemcpyHostToDevice); cudaMemcpy(d_map_input_index,spec->map_input_index,index_size,cudaMemcpyHostToDevice); spec->map_block_num=((spec->map_input_num)+(spec->map_thread_num)-1)/(spec->map_thread_num); // printf("%d\n",spec->map_block_num); map_count_warp<<<spec->map_block_num,spec->map_thread_num>>>(d_map_input_keys,d_map_input_values,d_map_input_index,d_map_im_key_size,d_map_im_value_size,d_map_im_num,spec->map_input_num); cudaMemcpy(spec->map_im_key_size,d_map_im_key_size,map_im_size,cudaMemcpyDeviceToHost); cudaMemcpy(spec->map_im_value_size,d_map_im_value_size,map_im_size,cudaMemcpyDeviceToHost); cudaMemcpy(spec->map_im_num,d_map_im_num,map_im_size,cudaMemcpyDeviceToHost); printf("%s\n",(spec->map_input_values+((spec->map_input_index+1)->value_offset))); printf("%d %d %d\n",*(spec->map_im_key_size+1),*(spec->map_im_value_size+1),*(spec->map_im_num+1)); cudaFree(d_map_input_keys); cudaFree(d_map_input_values); cudaFree(d_map_input_index); cudaFree(d_map_im_key_size); cudaFree(d_map_im_value_size); cudaFree(d_map_im_num); } void add_input_path(char *path,MapReduceSpec* spec){ MapFileList* plist; plist=(MapFileList*)malloc(sizeof(MapFileList)); spec->map_file_list=plist; struct dirent* entry = NULL; DIR *pDir; pDir=opendir(path); while((entry=readdir(pDir))!=NULL){ if(entry->d_type==DT_REG){ plist->filename=(char*)malloc(strlen(path)+strlen(entry->d_name)+1); strcpy(plist->filename,path); strcat(plist->filename,entry->d_name); plist->next=(MapFileList*)malloc(sizeof(MapFileList)); plist=plist->next; } } map_input_split(spec); map_count_phase(spec); } MAP_COUNT{ unsigned int i; unsigned int im_key_size=0; unsigned int im_value_size=0; int word_num=0; for(i=0;i<value_size;){ while((i<value_size)&&!isChar(*(value+i))) i++; int start = i; while((i<value_size)&&isChar(*(value+i))) i++; if(start<i){ im_key_size+=(i-start); im_value_size+=sizeof(int); word_num++; } } EMIT_IM_COUNT(im_key_size,im_value_size); //emitMapCount(im_key_size,im_value_size,word_num,key_im_size,value_im_size,map_im_num,threadID); } int main(int argc, char **argv){ MapReduceSpec* spec=(MapReduceSpec*)malloc(sizeof(MapReduceSpec)); init_mapreduce_spec(spec); add_input_path(argv[1],spec); free(spec); }
12,984
#include <assert.h> #include <stdio.h> #include <cuda.h> #define ARRAY_SIZE 10 const int ARRAY_BYTES = ARRAY_SIZE * sizeof(int); //kernel __global__ void addKernel( int *d_a, int *d_b, int *d_result){ int idx = threadIdx.x; d_result[idx] = d_a[idx] + d_b[idx]; } void onDevice( int *h_a, int *h_b, int *h_result ){ int *d_a, *d_b, *d_result; //allocate memory on the device cudaMalloc( (void**)&d_a, ARRAY_BYTES ); cudaMalloc( (void**)&d_b, ARRAY_BYTES ); cudaMalloc( (void**)&d_result, ARRAY_BYTES ); //copythe arrays 'a' and 'b' to the device cudaMemcpy( d_a, h_a, ARRAY_BYTES, cudaMemcpyHostToDevice ); cudaMemcpy( d_b, h_b, ARRAY_BYTES, cudaMemcpyHostToDevice ); //run the kernel addKernel<<<1,ARRAY_SIZE>>>( d_a, d_b, d_result); // copy the array 'result' back from the device to the CPU cudaMemcpy( h_result, d_result, ARRAY_BYTES, cudaMemcpyDeviceToHost ); // free device memory cudaFree( d_a ); cudaFree( d_b ); cudaFree( d_result ); } void onHost(){ int *h_a, *h_b, *h_result; //allocate memory on the host h_a = (int*)malloc(ARRAY_BYTES); h_b = (int*)malloc(ARRAY_BYTES); h_result = (int*)malloc(ARRAY_BYTES); // filling the arrays 'a' and 'b' on the CPU for (int i=0; i<ARRAY_SIZE; i++) { h_a[i] = i; h_b[i] = i*i; h_result[i]=0; } onDevice(h_a, h_b, h_result); // check the results for (int i=0; i<ARRAY_SIZE; i++) { assert( h_a[i] + h_b[i] == h_result[i] ); } printf("-: successful execution :-\n"); // free host memory free(h_a); free(h_b); free(h_result); } int main(){ onHost(); return 0; }
12,985
/*** substitui os valores aleatórios de determinado vetor de tamanho N por valores ordenados de 0 a N ***/ #include <stdio.h> #include <assert.h> #include <cuda.h> #define N 2//64 __global__ void foo(int* glob) { int a; int* p; a = 0; p = &a; *p = threadIdx.x; glob[*p] = threadIdx.x; }
12,986
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> extern "C" __global__ void matrixMulkernelShared(float * A, float *B, int wA, int wB, float *C) { //BLOCK index int bx = blockIdx.x; int by = blockIdx.y; //Thread index int tx = threadIdx.x; int ty = threadIdx.y; //Index of the first sub-matrix of A processed by the block int aBegin = wA * 16 * by; //Index of the last sub-matrix of A processed by the block int aEnd = aBegin + wA - 1; //Step size used to iterate thorugh the sub-matrices of A int aStep = 16; //Index of the first sub-matrix of B processed by the block int bBegin = 16 * bx; //Step size used to iterate through the sub-matrices of B int bStep = 16 * wB; // The element of the block sub-matrix that is computed // by the thread float Csub = 0; // Loop over all the sub-matrices of A and B required to // compute the block sub-matrix for (int a = aBegin, b = bBegin; a <= aEnd; a += aStep, b += bStep) { // Shared memory for the sub-matrix of A __shared__ float As[16][16]; // Shared memory for the sub-matrix of B __shared__ float Bs[16][16]; // Load the matrices from global memory to shared memory; // each thread loads one element of each ma trix As[ty][tx] = A[a + wA * ty + tx]; Bs[ty][tx] = B[b + wB * ty + tx]; // Synchronize to make sure the matrices are loaded __syncthreads(); // Multiply the two matrices together; // each thread computes one element // of the block sub-matrix for (int k = 0; k < 16; ++k) { Csub += As[ty][k] * Bs[k][tx]; // Synchronize to make sure that the preceding // computation is done before loading two new // sub-matrices of A and B in the next iteration __syncthreads(); } } // Write the block sub-matrix to global memory; // each thread writes one element int c = wB * 16 * by + 16 * bx; C[c + wB * ty + tx] = Csub; }
12,987
#include<bits/stdc++.h> using namespace std; void vector_sum(vector<float> a,float &cpu_res,int n) { for (int i = 0; i < n; i++) { cpu_res += a[i]; } } void vector_min(vector<float> a,float &cpu_res,int n){ cpu_res = INT_MAX; for (int i = 0; i < n; i++) { cpu_res = min(cpu_res,a[i]); } } void vector_max(vector<float> a,float &cpu_res,int n){ cpu_res = INT_MIN; for (int i = 0; i < n; i++) { cpu_res = max(cpu_res,a[i]); } } void vector_sd(vector<float> a,float sum,double &cpu_res_sd,int n){ double mean = (double)sum/(double)n; double s = 0; for(int i=0;i<n;i++){ s += ((a[i]-mean)*(a[i]-mean)); } cpu_res_sd = (double)s/(double)n; } __global__ void cuda_vector_sum(float* a,int n) { const int tid=threadIdx.x; int no_of_threads=blockDim.x; for(int step=1;step < n; step *= 2,no_of_threads /= 2){ if (tid <= no_of_threads){ int ind=2*step*tid; if((ind+step) >= n){ a[ind] = a[ind] + 0; }else{ a[ind] = a[ind] + a[ind+step]; } } } } __global__ void cuda_vector_min(float* a,int n) { const int tid=threadIdx.x; int no_of_threads=blockDim.x; for(int step=1;step < n; step *= 2,no_of_threads /= 2){ if (tid <= no_of_threads){ int ind=2*step*tid; if((ind+step) >= n){ a[ind] = min(a[ind],FLT_MAX); }else{ a[ind] = min(a[ind],a[ind+step]); } } } } __global__ void cuda_vector_max(float* a,int n) { const int tid=threadIdx.x; int no_of_threads=blockDim.x; for(int step=1;step < n; step *= 2,no_of_threads /= 2){ if (tid <= no_of_threads){ int ind=2*step*tid; if((ind+step) >= n){ a[ind] = max(a[ind],FLT_MIN); }else{ a[ind] = max(a[ind],a[ind+step]); } } } } __global__ void cuda_update_arr(float *a,double mean){ const int tid=threadIdx.x; a[tid] = (a[tid]-mean)*(a[tid]-mean); } int main() { int N = 2048; vector<float> a(N); srand(time(0)); generate(begin(a), end(a), []() { return (float(rand())/float((RAND_MAX)) * 100.0); }); for(auto item:a) cout<<item<<" "; cout<<'\n'; float cpu_res=0,gpu_res=0; double cpu_res_sd = 0,gpu_res_sd = 0; cout<<"CPU: "<<'\n'; //------------------------------------------------------------------- // Sum calculation vector_sum(a,cpu_res,N); cout << "Vector Sum using CPU :"<<cpu_res<<" \n"; // Average calculation cout << "Vector Average using CPU :"<<(double)cpu_res/(double)N<<" \n"; vector_sd(a,cpu_res,cpu_res_sd,N); cout << "Vector Standard Deviation using CPU :"<<fixed<<setprecision(2)<<sqrt(cpu_res_sd)<<" \n"; vector_min(a,cpu_res,N); cout << "Vector Min using CPU :"<<cpu_res<<" \n"; vector_max(a,cpu_res,N); cout << "Vector Max using CPU :"<<cpu_res<<" \n"; cout<<"GPU: "<<'\n'; //------------------------------------------------------------------- // Allocate memory on the device size_t bytes = sizeof(float) * N; float* d_a; cudaMalloc(&d_a, bytes); //------------------------------------------------------------------- // Copy data from the host to the device (CPU to GPU) cudaMemcpy(d_a, a.data(), bytes, cudaMemcpyHostToDevice); cuda_vector_sum <<<1,N/2>>> (d_a,N); cudaMemcpy(&gpu_res, d_a, sizeof(float), cudaMemcpyDeviceToHost); cout << "Vector Sum using GPU :"<<gpu_res<<" \n"; //------------------------------------------------------------------- cout << "Vector Average using GPU :"<<(double)gpu_res/(double)N<<" \n"; //------------------------------------------------------------------- double mean = (double)gpu_res/(double)N; cudaMemcpy(d_a, a.data(), bytes, cudaMemcpyHostToDevice); cuda_update_arr<<<1,N>>>(d_a,mean); cuda_vector_sum<<<1,N/2>>>(d_a,N); cudaMemcpy(&gpu_res, d_a, sizeof(float), cudaMemcpyDeviceToHost); gpu_res = (double)gpu_res/(double)N; cout << "Vector Standard Deviation using GPU :"<<fixed<<setprecision(2)<<sqrt(gpu_res)<<" \n"; //------------------------------------------------------------------- cudaMemcpy(d_a, a.data(), bytes, cudaMemcpyHostToDevice); gpu_res = INT_MAX; cuda_vector_min <<<1,N/2>>> (d_a,N); cudaMemcpy(&gpu_res, d_a, sizeof(float), cudaMemcpyDeviceToHost); cout << "Vector Min using GPU :"<<gpu_res<<" \n"; //------------------------------------------------------------------- cudaMemcpy(d_a, a.data(), bytes, cudaMemcpyHostToDevice); gpu_res = INT_MIN; cuda_vector_max <<<1,N/2>>> (d_a,N); cudaMemcpy(&gpu_res, d_a, sizeof(float), cudaMemcpyDeviceToHost); cout << "Vector Max using GPU :"<<gpu_res<<" \n"; //------------------------------------------------------------------- // Free memory on device cudaFree(d_a); }
12,988
#include <cuda.h> #include <stdio.h> __global__ void sum(float* a, float* b, float* c) { int index = threadIdx.x + blockDim.x * blockIdx.x; //int index = blockIdx.x; c[index] = a[index] + b[index]; } #define CUDA_CHECK_RETURN(value) ((cudaError_t)value != cudaSuccess) ? printf("Error %s at line %d in the file %s\n", cudaGetErrorString((cudaError_t)value), __LINE__, __FILE__) : printf("") int main() { int n, k; scanf("%d%d", &n, &k); float* a = new float[n * k]; float* b = new float[n * k]; float* c = new float[n * k]; for(int i = 0; i < n * k; i++) { a[i] = i; b[i] = i; } float* dev1; float* dev2; float* dev3; float elapsedTime; cudaEvent_t start, stop; CUDA_CHECK_RETURN(cudaEventCreate(&start)); CUDA_CHECK_RETURN(cudaEventCreate(&stop)); CUDA_CHECK_RETURN(cudaMalloc((void**)&dev1, n * k * sizeof(float))); CUDA_CHECK_RETURN(cudaMalloc((void**)&dev2, n * k * sizeof(float))); CUDA_CHECK_RETURN(cudaMalloc((void**)&dev3, n * k * sizeof(float))); CUDA_CHECK_RETURN(cudaMemcpy(dev1, a, n * k * sizeof(float), cudaMemcpyHostToDevice)); CUDA_CHECK_RETURN(cudaMemcpy(dev2, b, n * k * sizeof(float), cudaMemcpyHostToDevice)); CUDA_CHECK_RETURN(cudaEventRecord(start, 0)); sum <<< n, k >>> (dev1, dev2, dev3); CUDA_CHECK_RETURN(cudaEventRecord(stop, 0)); CUDA_CHECK_RETURN(cudaEventSynchronize(stop)); CUDA_CHECK_RETURN(cudaGetLastError()); CUDA_CHECK_RETURN(cudaEventElapsedTime(&elapsedTime, start, stop)); fprintf(stderr, "gTest took %g\n", elapsedTime); CUDA_CHECK_RETURN(cudaEventDestroy(start)); CUDA_CHECK_RETURN(cudaEventDestroy(stop)); CUDA_CHECK_RETURN(cudaMemcpy(c, dev3, n * k * sizeof(float), cudaMemcpyDeviceToHost)); for(int i = (n * k) - 5; i < n * k; i++) { printf("Element #%i: %f\n", i, c[i]); } free(a); free(b); free(c); CUDA_CHECK_RETURN(cudaFree(dev1)); CUDA_CHECK_RETURN(cudaFree(dev2)); CUDA_CHECK_RETURN(cudaFree(dev3)); return 0; }
12,989
// // Created by root on 2020/11/30. // #include "kernel.cuh" #include "stdio.h" #define TPB 64 #define ATOMIC 1 // 0 for non-atomic add __global__ void dotKernel(int *d_res, int *d_a, int *d_b, int n) { int idx = threadIdx.x + blockDim.x * blockIdx.x; if (idx >= n) { return; } int s_dix = threadIdx.x; __shared__ int s_prod[TPB]; s_prod[s_dix] = d_a[idx] * d_b[idx]; __syncthreads(); if (s_dix == 0) { int blockSum = 0; for (int i = 0; i < blockDim.x; i++) { blockSum += s_prod[i]; } printf("Block %d, block sum = %d\n", blockIdx.x, blockSum); if (ATOMIC) { atomicAdd(d_res, blockSum); } else { *d_res += blockSum; } } } void dotLauncher(int *res, int *a, int *b, int n) { int *d_res; int *d_a = 0, *d_b = 0; cudaMalloc(&d_res, sizeof(int )); cudaMalloc(&d_a, n * sizeof(int )); cudaMalloc(&d_b, n * sizeof(int )); cudaMemset(d_res, 0, sizeof(int )); cudaMemcpy(d_a, a, n * sizeof(int ), cudaMemcpyHostToDevice); cudaMemcpy(d_b, b, n * sizeof(int ), cudaMemcpyHostToDevice); dotKernel<<<(n + TPB - 1) / TPB, TPB>>>(d_res, d_a, d_b, n); cudaMemcpy(res, d_res, sizeof(int ), cudaMemcpyDeviceToHost); cudaFree(d_res); cudaFree(d_a); cudaFree(d_b); }
12,990
#include <cassert> struct type { /* a regular type */ }; __host__ __device__ void ordinary_function(type* tc_operand, int block, int thread) { assert(block == 0 && thread == 0); // use tc_operand } __global__ void entry_point_function(type* tc_operand) { ordinary_function(tc_operand, blockIdx.x, threadIdx.x); } #include <memory> template <class T> struct managed { typedef T value_type; managed () = default; template <class U> constexpr managed (const managed<U>&) noexcept {} T* allocate(std::size_t n) { void* out = nullptr; cudaMallocManaged(&out, n*sizeof(T)); return static_cast<T*>(out); } void deallocate(T* p, std::size_t) noexcept { cudaFree(p); } }; int main() { auto managed_object = managed<type>().allocate(1); entry_point_function<<<1,1>>>(managed_object); cudaDeviceSynchronize(); ordinary_function(managed_object, 0, 0); managed<type>().deallocate(managed_object, 1); return 0; }
12,991
// nvcc ani2PCF.cu -o par.out && ./par.out data.dat rand0.dat 32768 30 180 #include <iostream> #include <fstream> //manejo de archivos #include <string.h> #include <time.h> #include <math.h> using namespace std; struct PointW3D{ float x; float y; float z; float w; }; struct Node{ int len; // Cantidad de elementos en el nodo. PointW3D *elements; // Elementos del nodo. }; void open_files(string name_file, int pts, PointW3D *datos){ /* Función para abrir nuestros archivos de datos */ ifstream file; string mypathto_files = "../../../fake_DATA/DATOS/"; //This creates the full path to where I have my data files name_file.insert(0,mypathto_files); file.open(name_file.c_str(), ios::in | ios::binary); //le indico al programa que se trata de un archivo binario con ios::binary if (file.fail()){ cout << "Error al cargar el archivo " << endl; exit(1); } for ( int c = 0; c < pts; c++) { file >> datos[c].x >> datos[c].y >> datos[c].z >> datos[c].w; } file.close(); } //==================================================================== /* void save_histogram(string name, int bns, double *histo){ //Función para guardar nuestros archivos de histogramas int i, j; ofstream file; file.open(name.c_str(),ios::out | ios::binary); if (file.fail()){ cout << "Error al guardar el archivo " << endl; exit(1); } for (i=0; i<bns; ++i){ for (j=0; j<bns; ++j) { file << histo[i][j] << " "; } file << "\n"; } file.close(); } */ //=================================================================== void add(PointW3D *&array, int &lon, float _x, float _y, float _z, float _w){ lon++; PointW3D *array_aux; cudaMallocManaged(&array_aux, lon*sizeof(PointW3D)); for (int i=0; i<lon-1; i++){ array_aux[i].x = array[i].x; array_aux[i].y = array[i].y; array_aux[i].z = array[i].z; array_aux[i].w = array[i].w; } cudaFree(array); array = array_aux; array[lon-1].x = _x; array[lon-1].y = _y; array[lon-1].z = _z; array[lon-1].z = _w; } void make_nodos(Node ***nod, PointW3D *dat, unsigned int partitions, float size_node, unsigned int np){ /* Función para crear los nodos con los datos y puntos random Argumentos nod: arreglo donde se crean los nodos. dat: datos a dividir en nodos. */ int row, col, mom; // Inicializamos los nodos vacíos: for (row=0; row<partitions; row++){ for (col=0; col<partitions; col++){ for (mom=0; mom<partitions; mom++){ nod[row][col][mom].len = 0; cudaMallocManaged(&nod[row][col][mom].elements, sizeof(PointW3D)); } } } // Llenamos los nodos con los puntos de dat: for (int i=0; i<np; i++){ row = (int)(dat[i].x/size_node); col = (int)(dat[i].y/size_node); mom = (int)(dat[i].z/size_node); add(nod[row][col][mom].elements, nod[row][col][mom].len, dat[i].x, dat[i].y, dat[i].z, dat[i].w); } } //==================================================================== //============ Sección de Kernels ================================== //=================================================================== __device__ void count_distances11(float **XX, PointW3D *elements, int len, float ds, float dd_max){ /* Funcion para contar las distancias entre puntos en un mismo Nodo. */ int bi, bj; float v; float x1,y1,z1,w1,x2,y2,z2,w2; float ddz, dd_ort; for (int i=0; i<len-1; ++i){ x1 = elements[i].x; y1 = elements[i].y; z1 = elements[i].z; w1 = elements[i].w; for (int j=i+1; j<len; ++j){ x2 = elements[j].x; y2 = elements[j].y; z2 = elements[j].z; w2 = elements[j].w; dd_ort = (x2-x1)*(x2-x1)+(y2-y1)*(y2-y1); ddz = (z2-z1)*(z2-z1); if (ddz < dd_max && dd_ort < dd_max){ bi = int(sqrt(ddz)*ds); bj = int(sqrt(dd_ort)*ds); v = 2*w1*w2; atomicAdd(&XX[bi][bj],2); } } } } __device__ void count_distances12(float **XX, PointW3D *elements1, int len1, PointW3D *elements2, int len2, float ds, float dd_max){ /* Funcion para contar las distancias entre puntos en un mismo Nodo. */ int bi, bj; float d, v; float x1,y1,z1,w1,x2,y2,z2,w2; float ddz, dd_ort; for (int i=0; i<len1; ++i){ x1 = elements1[i].x; y1 = elements1[i].y; z1 = elements1[i].z; w1 = elements1[i].w; for (int j=0; j<len2; ++j){ x2 = elements2[j].x; y2 = elements2[j].y; z2 = elements2[j].z; w2 = elements2[j].w; dd_ort = (x2-x1)*(x2-x1)+(y2-y1)*(y2-y1); ddz = (z2-z1)*(z2-z1); if (ddz < dd_max && dd_ort < dd_max){ bi = int(sqrt(ddz)*ds); bj = int(sqrt(dd_ort)*ds); v = 2*w1*w2; atomicAdd(&XX[bi][bj],2); } } } } __global__ void make_histoXX(float **XX_A, float **XX_B, Node ***nodeD, int partitions, float ds, float dd_max, int did_max, int did_max2){ int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx<(partitions*partitions*partitions)){ //Get the node positon in this thread int mom = (int) (idx/(partitions*partitions)); int col = (int) ((idx%(partitions*partitions))/partitions); int row = idx%partitions; if (nodeD[row][col][mom].len > 0){ if (idx%2==0){ count_distances11(XX_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, ds, dd_max); } else { count_distances11(XX_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, ds, dd_max); } int u,v,w; //Posicion del nodo 2 unsigned int dx_nod12, dy_nod12, dz_nod12, dd_nod12; //Nodo2 solo movil en z for(w = mom+1; w<partitions && w-row<=did_max; w++){ if (idx%2==0){ //Si es par lo guarda en histograma A, si no en el B count_distances12(XX_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeD[row][col][w].elements, nodeD[row][col][w].len, ds, dd_max); } else { count_distances12(XX_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeD[row][col][w].elements, nodeD[row][col][w].len, ds, dd_max); } } //Nodo2 movil en ZY for(v=col+1; v<partitions && v-col<=did_max; v++){ dy_nod12 = v-col; for(w=(mom-did_max)*(mom>did_max); w<partitions && w-mom<=did_max; w++){ dz_nod12 = w-mom; dd_nod12 = dz_nod12*dz_nod12 + dy_nod12*dy_nod12; if (dd_nod12<=did_max2){ if (idx%2==0){ //Si es par lo guarda en histograma A, si no en el B count_distances12(XX_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeD[row][v][w].elements, nodeD[row][v][w].len, ds, dd_max); } else { count_distances12(XX_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeD[row][v][w].elements, nodeD[row][v][w].len, ds, dd_max); } } } } //Nodo movil en XYZ for(u = row+1; u < partitions && u-row< did_max; u++){ dx_nod12 = u-row; for(v = (col-did_max)*(col>did_max); v < partitions && v-col< did_max; v++){ dy_nod12 = v-col; for(w = (mom-did_max)*(mom>did_max); w < partitions && w-mom< did_max; w++){ dz_nod12 = w-mom; dd_nod12 = dz_nod12*dz_nod12 + dy_nod12*dy_nod12 + dx_nod12*dx_nod12; if (dd_nod12<=did_max2){ if (idx%2==0){ //Si es par lo guarda en histograma A, si no en el B count_distances12(XX_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeD[u][v][w].elements, nodeD[u][v][w].len, ds, dd_max); } else { count_distances12(XX_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeD[u][v][w].elements, nodeD[u][v][w].len, ds, dd_max); } } } } } } } } __global__ void make_histoXY(float **XY_A, float **XY_B, Node ***nodeD, Node ***nodeR, int partitions, float ds, float dd_max, int did_max, int did_max2){ int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx<(partitions*partitions*partitions)){ //Get the node positon in this thread int mom = (int) (idx/(partitions*partitions)); int col = (int) ((idx%(partitions*partitions))/partitions); int row = idx%partitions; if (nodeD[row][col][mom].len > 0){ int u,v,w; //Posicion del nodo 2 unsigned int dx_nod12, dy_nod12, dz_nod12, dd_nod12; //Nodo2 solo movil en z w = 0;//(mom-did_max)*(mom>did_max); for(w = (mom-did_max)*(mom>did_max); w<partitions && w-row<=did_max; w++){ if (idx%2==0){ //Si es par lo guarda en histograma A, si no en el B count_distances12(XY_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeR[row][col][w].elements, nodeR[row][col][w].len, ds, dd_max); } else { count_distances12(XY_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeR[row][col][w].elements, nodeR[row][col][w].len, ds, dd_max); } } //Nodo2 movil en ZY for(v = (col-did_max)*(col>did_max); v<partitions && v-col<=did_max; v++){ dy_nod12 = v-col; for(w = (mom-did_max)*(mom>did_max); w<partitions && w-mom<=did_max; w++){ dz_nod12 = w-mom; dd_nod12 = dz_nod12*dz_nod12 + dy_nod12*dy_nod12; if (dd_nod12<=did_max2){ if (idx%2==0){ //Si es par lo guarda en histograma A, si no en el B count_distances12(XY_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeR[row][v][w].elements, nodeR[row][v][w].len, ds, dd_max); } else { count_distances12(XY_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeR[row][v][w].elements, nodeR[row][v][w].len, ds, dd_max); } } } } //Nodo movil en XYZ for(u = (row-did_max)*(row>did_max); u < partitions && u-row< did_max; u++){ dx_nod12 = u-row; for(v = (col-did_max)*(col>did_max); v < partitions && v-col< did_max; v++){ dy_nod12 = v-col; for(w = (mom-did_max)*(mom>did_max); w < partitions && w-mom< did_max; w++){ dz_nod12 = w-mom; dd_nod12 = dz_nod12*dz_nod12 + dy_nod12*dy_nod12 + dx_nod12*dx_nod12; if (dd_nod12<=did_max2){ if (idx%2==0){ //Si es par lo guarda en histograma A, si no en el B count_distances12(XY_A, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeR[u][v][w].elements, nodeR[u][v][w].len, ds, dd_max); } else { count_distances12(XY_B, nodeD[row][col][mom].elements, nodeD[row][col][mom].len, nodeR[u][v][w].elements, nodeR[u][v][w].len, ds, dd_max); } } } } } } } } int main(int argc, char **argv){ unsigned int np = stoi(argv[3]), bn = stoi(argv[4]); float dmax = stof(argv[5]); float ds = (float)(bn)/dmax, dd_max=dmax*dmax, size_box = 250.0, alpha = 2.176; float size_node = alpha*(size_box/pow((float)(np),1/3.)); int did_max = (int)(ceil(dmax/size_node)); int did_max2 = (int)(ceil(dd_max/(size_node*size_node))); unsigned int partitions = (int)(ceil(size_box/size_node)); //int np = 32768, bn = 10; //float dmax = 180.0; float **DD_A, **RR_A, **DR_A, **DD_B, **RR_B, **DR_B; double **DD, **RR, **DR; PointW3D *dataD; PointW3D *dataR; cudaMallocManaged(&dataD, np*sizeof(PointW3D));// Asignamos meoria a esta variable cudaMallocManaged(&dataR, np*sizeof(PointW3D)); // Nombre de los archivos string nameDD = "DDiso.dat", nameRR = "RRiso.dat", nameDR = "DRiso.dat"; // Asignamos memoria para los histogramas DD = new double*[bn]; RR = new double*[bn]; DR = new double*[bn]; cudaMallocManaged(&DD_A, bn*sizeof(float*)); cudaMallocManaged(&RR_A, bn*sizeof(float*)); cudaMallocManaged(&DR_A, bn*sizeof(float*)); cudaMallocManaged(&DD_B, bn*sizeof(float*)); cudaMallocManaged(&RR_B, bn*sizeof(float*)); cudaMallocManaged(&DR_B, bn*sizeof(float*)); for (int i=0; i<bn; ++i){ *(DD+i) = new double[bn]; *(RR+i) = new double[bn]; *(DR+i) = new double[bn]; cudaMallocManaged(&*(DD_A+i), bn*sizeof(float)); cudaMallocManaged(&*(RR_A+i), bn*sizeof(float)); cudaMallocManaged(&*(DR_A+i), bn*sizeof(float)); cudaMallocManaged(&*(DD_B+i), bn*sizeof(float)); cudaMallocManaged(&*(RR_B+i), bn*sizeof(float)); cudaMallocManaged(&*(DR_B+i), bn*sizeof(float)); } //Inicializar en 0 los histogramas for (int i = 0; i<bn; i++){ for (int j=0; j<bn; j++){ DD[i][j] = 0.0; RR[i][j] = 0.0; DR[i][j] = 0.0; DD_A[i][j] = 0.0; RR_A[i][j] = 0.0; DR_A[i][j] = 0.0; DD_B[i][j] = 0.0; RR_B[i][j] = 0.0; DR_B[i][j] = 0.0; } } // Abrimos y guardamos los datos en los en los arrays correspondientes open_files(argv[1], np, dataD); open_files(argv[2], np, dataR); //Iniciar los nodos. Node ***nodeD; Node ***nodeR; cudaMallocManaged(&nodeR, partitions*sizeof(Node**)); cudaMallocManaged(&nodeD, partitions*sizeof(Node**)); for (int i=0; i<partitions; i++){ cudaMallocManaged(&*(nodeR+i), partitions*sizeof(Node*)); cudaMallocManaged(&*(nodeD+i), partitions*sizeof(Node*)); for (int j=0; j<partitions; j++){ cudaMallocManaged(&*(*(nodeR+i)+j), partitions*sizeof(Node)); cudaMallocManaged(&*(*(nodeD+i)+j), partitions*sizeof(Node)); } } //Clasificar los puntos en los nodos make_nodos(nodeD, dataD, partitions, size_node, np); make_nodos(nodeR, dataR, partitions, size_node, np); int blocks = (int)(ceil((float)((partitions*partitions*partitions)/(float)(1024)))); dim3 grid(blocks,1,1); dim3 block(1024,1,1); clock_t begin = clock(); make_histoXX<<<grid,block>>>(DD_A, DD_B, nodeD, partitions, ds, dd_max, did_max, did_max2); make_histoXX<<<grid,block>>>(RR_A, RR_B, nodeR, partitions, ds, dd_max, did_max, did_max2); make_histoXY<<<grid,block>>>(DR_A, DR_B, nodeD, nodeR, partitions, ds, dd_max, did_max, did_max2); //Waits for the GPU to finish cudaDeviceSynchronize(); //Check here for errors cudaError_t error = cudaGetLastError(); cout << "The error code is " << error << endl; if(error != 0) { // print the CUDA error message and exit printf("CUDA error: %s\n", cudaGetErrorString(error)); exit(-1); } clock_t end = clock(); double time_spent = (double)(end - begin) / CLOCKS_PER_SEC; printf("\nTiempo en CPU usado = %.4f seg.\n", time_spent ); for (int i = 0; i < bn; i++){ for (int j = 0; j < bn; j++){ DD[i][j] = (double)(DD_A[i][j]+ DD_B[i][j]); RR[i][j] = (double)(RR_A[i][j]+ RR_B[i][j]); DR[i][j] = (double)(DR_A[i][j]+ DR_B[i][j]); } } cout << "Termine de hacer todos los histogramas" << endl; // Mostramos los histogramas cout << "\nHistograma DD:" << endl; int sum = 0; for (int i = 0; i<bn; i++){ for (int k = 0; k<bn; k++){ cout << DD[i][k] << "\t"; sum += DD[i][k]; } cout << "\n"; } cout << "Total: " << sum << endl; cout << "\nHistograma RR:" << endl; for (int i = 0; i<bn; i++){ for (int k = 0; k<bn; k++){ cout << RR[i][k] << "\t"; sum += RR[i][k]; } } cout << "\nHistograma DR:" << endl; for (int i = 0; i<bn; i++){ for (int k = 0; k<bn; k++){ cout << DR[i][k] << "\t"; sum += DR[i][k]; } cout << "\n"; } // Guardamos los histogramas //save_histogram(nameDD, bn, DD); cout << "Guarde histograma DD..." << endl; //save_histogram(nameRR, bn, RR); cout << "Guarde histograma RR..." << endl; //save_histogram(nameDR, bn, DR); cout << "Guarde histograma DR..." << endl; cudaFree(&dataD); cudaFree(&dataR); for (int i=0; i<bn; ++i){ delete[] *(DD+i); delete[] *(RR+i); delete[] *(DR+i); } delete[] DD; delete[] DR; delete[] RR; for (int i=0; i<bn; ++i){ cudaFree(&*(DD_A+i)); cudaFree(&*(RR_A+i)); cudaFree(&*(DR_A+i)); cudaFree(&*(DD_B+i)); cudaFree(&*(RR_B+i)); cudaFree(&*(DR_B+i)); } cudaFree(&DD_A); cudaFree(&RR_A); cudaFree(&DR_A); cudaFree(&DD_B); cudaFree(&RR_B); cudaFree(&DR_B); for (int i=0; i<partitions; i++){ for (int j=0; j<partitions; j++){ cudaFree(&*(*(nodeR+i)+j)); cudaFree(&*(*(nodeD+i)+j)); } cudaFree(&*(nodeR+i)); cudaFree(&*(nodeD+i)); } cudaFree(&nodeR); cudaFree(&nodeD); cout << "Programa Terminado..." << endl; return 0; }
12,992
#include <cuda.h> #include <cuda_runtime_api.h> #include <stdio.h> __global__ void kernel( int *a, int dimx, int dimy ) { int ix = blockIdx.x*blockDim.x + threadIdx.x; int iy = blockIdx.y*blockDim.y + threadIdx.y; int idx = iy*dimx + ix; a[idx] = a[idx]+1; } int main() { int dimx = 16, dimy = 16; int num_bytes = dimx*dimy*sizeof(int); int *d_a=0, *h_a=0; // device and host pointers h_a = (int*)malloc(num_bytes); cudaMalloc( (void**)&d_a, num_bytes ); if( 0==h_a || 0==d_a ) { printf("couldn't allocate memory\n"); return 1; } cudaMemset( d_a, 0, num_bytes ); dim3 grid, block; block.x = 4; block.y = 4; grid.x = dimx / block.x; grid.y = dimy / block.y; kernel<<<grid, block>>>( d_a, dimx, dimy ); cudaMemcpy(h_a, d_a, num_bytes, cudaMemcpyDeviceToHost ); free( h_a ); cudaFree( d_a ); return 0; }
12,993
#include <stdio.h> #include <stdlib.h> #include <curand_kernel.h> #include <time.h> #define X 1 #define O -1 #define BLANK 0 #define NUM_THREAD 500 #define RUNS_PER_THREAD 20 #define NUM_BLOCKS 20 #define NUM_SEQ_LOOPS 200000 //When a subBoard is won, it should be filled with that mark. (All X or all O) This is needed as a speed optimization __device__ __host__ void printSquare(int x){ switch(x) { case X: printf("X"); break; case O: printf("O"); break; case BLANK: printf("_"); break; } } bool printSquareWithNumber(int x, int num){ switch(x) { case X: printf("XX"); return false; case O: printf("OO"); return false; case BLANK: printf("%d", num); if(num<=9) { printf(" ");} return true; } return false; } void PrintBoardWithNumbers(int* board){ int x,y,j,i, num; num = 0; for(x = 0; x <3; x++) { printf("_____________\n"); for(i = 0; i < 3; i++) { printf("|"); for(y = 0; y < 3; y++) { for(j = 0; j< 3; j++) { if(printSquareWithNumber(board[(3*x+y)*9+(3*i+j)],num)) { num++; } } printf("|"); } printf("\n"); } } printf("_____________\n"); } void PrintBoard(int* board){ int x,y,j,i; for(x = 0; x <3; x++) { printf("_____________\n"); for(i = 0; i < 3; i++) { printf("|"); for(y = 0; y < 3; y++) { for(j = 0; j< 3; j++) { printSquare(board[(3*x+y)*9+(3*i+j)]); } printf("|"); } printf("\n"); } } printf("_____________\n"); } void PrintSubBoardWithNumbers(int* subBoard){ int x,y, num; num = 0; for(x = 0; x < 3; x++) { printf("|"); for(y = 0; y < 3; y++) { if(printSquareWithNumber(subBoard[3*x+y],num)) { num++; } } printf("|\n"); } } void PrintSubBoard(int* subBoard){ int x,y; for(x = 0; x < 3; x++) { printf("|"); for(y = 0; y < 3; y++) { printSquare(subBoard[3*x+y]); } printf("|\n"); } } __device__ __host__ int SubBoardWinner(int* subBoard){ int i, total; //left to right wins for(i = 0; i < 3; i++) { total = subBoard[3*i] + subBoard[3*i +1 ] + subBoard[3*i+2]; // printf("total: %d\n",total); if(abs(total) == 3) { return (total/3); } } //up to down for(i = 0; i < 3; i++) { total = subBoard[i] + subBoard[3+i] + subBoard[6+i]; // printf("total: %d\n",total); if(abs(total) == 3) { return (total/3); } } //Diagonals total = subBoard[0] + subBoard[4] + subBoard[8]; //printf("total: %d\n",total); if(abs(total) == 3) { return (total/3); } total = subBoard[2] + subBoard[4] + subBoard[6]; // printf("total: %d\n",total); if(abs(total) == 3) { return (total/3); } return 0; } int SubBoardWinner(double* subBoard){ int i, total; //left to right wins for(i = 0; i < 3; i++) { total = subBoard[3*i] + subBoard[3*i +1 ] + subBoard[3*i+2]; if(abs(total) == 3) { return (total/3); } } //up to down for(i = 0; i < 3; i++) { total = subBoard[i] + subBoard[ 3+i ] + subBoard[6+i]; if(abs(total) == 3) { return (total/3); } } //Diagonals total = subBoard[0] + subBoard[4] + subBoard[8]; if(abs(total) == 3) { return (total/3); } total = subBoard[2] + subBoard[4] + subBoard[6]; if(abs(total) == 3) { return (total/3); } return 0; } __device__ __host__ int BoardWinner(int* board){ int i,metaBoard[9]; for(i = 0; i < 9; i++) { metaBoard[i] = SubBoardWinner(board+(i*9)); } return SubBoardWinner(metaBoard); } __device__ __host__ bool IsSubBoardFull(int* subBoard){ int i; for(i = 0; i < 9; i++) { if(subBoard[i] == 0) { return false; } } return true; } __device__ __host__ bool IsSubBoardFull(double* subBoard){ int i; for(i = 0; i < 9; i++) { if(subBoard[i] != 1 || subBoard[i] != -1) { return false; } } return true; } __device__ __host__ bool IsBoardFull(int* board){ for(int i = 0; i < 81; i++) { if(board[i] == BLANK) { return false; } } return true; } __device__ __host__ int NumberOfFreeSquaresInFullBoard(int* board){ int i, count = 0; for(i = 0; i < 81; i++) { if(board[i] == 0) { count++; } } return count; } __device__ __host__ int NumberOfFreeSquaresInSubBoard(int* subBoard){ int i, count = 0; for(i = 0; i < 9; i++) { if(subBoard[i] == 0) { count++; } } return count; } __device__ __host__ int NumberOfPossibleMoves(int* board, int lastMove, bool fullBoard){ int subBoard = lastMove % 9; if(fullBoard) { return NumberOfFreeSquaresInFullBoard(board); } else { return NumberOfFreeSquaresInSubBoard(board + 9*subBoard); } } int DoEvalRow(int a, int b, int c){ int count = 0; int sum = a + b + c; if(a != 0) {count++;} if(b != 0) {count++;} if(c != 0) {count++;} return sum * count; } double DoEvalRow(double a, double b, double c){ int count = 0; double sum = a + b + c; if(sum > 0) { if(a > 0) {count++;} if(b > 0) {count++;} if(c > 0) {count++;} } else { if(a < 0) {count++;} if(b < 0) {count++;} if(c < 0) {count++;} } return sum * (double)count; } int EvalRow(int a, int b, int c){ if( a >= 0 && b >= 0 && c >= 0) { return DoEvalRow(a,b,c); } else if( a <= 0 && b <= 0 && c <= 0) { return DoEvalRow(a,b,c); } else { return 0; } } double EvalSubBoard(int* subBoard){ double sum = 0; int winner = SubBoardWinner(subBoard); switch (winner) { case BLANK: if(IsSubBoardFull(subBoard)) { return 0; } sum += EvalRow(subBoard[0],subBoard[1],subBoard[2]); sum += EvalRow(subBoard[3],subBoard[4],subBoard[5]); sum += EvalRow(subBoard[6],subBoard[7],subBoard[8]); sum += EvalRow(subBoard[0],subBoard[3],subBoard[6]); sum += EvalRow(subBoard[1],subBoard[4],subBoard[7]); sum += EvalRow(subBoard[2],subBoard[5],subBoard[8]); sum += EvalRow(subBoard[0],subBoard[4],subBoard[8]); sum += EvalRow(subBoard[2],subBoard[4],subBoard[6]); sum /= 21; break; case X: sum = 1; break; case O: sum = -1; break; } return sum; } double EvalMetaRow(double a, double b, double c){ if( a > -1 && b > -1 && c > -1) { return DoEvalRow(a,b,c); } else if( a < 1 && b < 1 && c < 1) { return DoEvalRow(a,b,c); } else { return 0; } } double EvalMetaBoard(double* subBoard){ double sum = 0; int winner = SubBoardWinner(subBoard); switch (winner) { case BLANK: if(IsSubBoardFull(subBoard)) { return 0; } sum += EvalMetaRow(subBoard[0],subBoard[1],subBoard[2]); sum += EvalMetaRow(subBoard[3],subBoard[4],subBoard[5]); sum += EvalMetaRow(subBoard[6],subBoard[7],subBoard[8]); sum += EvalMetaRow(subBoard[0],subBoard[3],subBoard[6]); sum += EvalMetaRow(subBoard[1],subBoard[4],subBoard[7]); sum += EvalMetaRow(subBoard[2],subBoard[5],subBoard[8]); sum += EvalMetaRow(subBoard[0],subBoard[4],subBoard[8]); sum += EvalMetaRow(subBoard[2],subBoard[4],subBoard[6]); break; case X: sum = 21; break; case O: sum = -21; break; } return sum; } double EvalFullBoard(int* board){ int i; double metaBoard[9]; int winner = BoardWinner(board); switch(winner) { case BLANK: if(IsBoardFull(board)) { return 0; } for(i = 0; i < 9; i++) { metaBoard[i] = EvalSubBoard(board + 9*i); } return EvalMetaBoard(metaBoard); case X: return (double)21; case O: return (double)-21; } return 98; } __device__ int EvalFullBoardKenel(int* board){ switch(BoardWinner(board)) { case X: return 1; case O: return -1; case BLANK: return 0; } return 0; } __device__ __host__ int PlaceMoveinSubBoard(int* board, int lastMove, int placement, int mark){ int subBoard, freeSquares, i; subBoard = lastMove % 9; freeSquares = 0; for(i = 0; i < 9; i++) { if(board[subBoard* 9 + i] == 0) { if(freeSquares == placement) { board[subBoard* 9 + i] = mark; freeSquares = i; break; } freeSquares++; } } if( SubBoardWinner(board + subBoard * 9 ) != 0 ) { for(i = 0; i < 9; i++) { board[subBoard* 9 + i] = mark; } } return subBoard * 9 + freeSquares; } __device__ __host__ int PlaceMarkinNthFree(int* board, int lastMove, int placement, int mark){ int subBoard, freeSquares, i; freeSquares = 0; for(i = 0; i < 81; i++) { if(board[i] == 0) { if(freeSquares == placement) { board[i] = mark; freeSquares = i; break; } freeSquares++; } } subBoard = i / 9; if( SubBoardWinner(board + subBoard * 9 ) != 0 ) { for(i = 0; i < 9; i++) { board[subBoard* 9 + i] = mark; } } return subBoard * 9 + freeSquares; } int playRandomMove(int* board, int lastMove, int mark){ int subBoard = lastMove%9; bool fullBoard = SubBoardWinner(board+9*subBoard) != 0 || IsSubBoardFull(board+subBoard*9) ; int numOfMoves = NumberOfPossibleMoves(board, lastMove, fullBoard); int index = rand() % (numOfMoves); if(fullBoard) { return PlaceMarkinNthFree(board, lastMove, index, mark); } else { return PlaceMoveinSubBoard(board, lastMove, index, mark); } } __device__ int playRandomMove(int* board, int lastMove, int mark, curandState_t state){ int subBoard = lastMove%9; bool fullBoard = SubBoardWinner(board+9*subBoard) != 0 || IsSubBoardFull(board+subBoard*9) ; int numOfMoves = NumberOfPossibleMoves(board, lastMove, fullBoard); int index = curand(&state) % (numOfMoves); if(fullBoard) { return PlaceMarkinNthFree(board, lastMove, index, mark); } else { return PlaceMoveinSubBoard(board, lastMove, index, mark); } } int MonteCarlo(int* board, int lastMove, int mark, int numRuns){ int fakeBoard[81]; int fakeLastMove; int fakeMark; int subBoard = lastMove%9; bool fullBoard = SubBoardWinner(board+9*subBoard) != 0 || IsSubBoardFull(board+subBoard*9) ; int numOfMoves = NumberOfPossibleMoves(board, lastMove, fullBoard); double score [70]; for(int i = 0; i < 70; i++) { score[i] = 0; } for(int i = 0; i < numRuns; i++) { for(int j = 0; j < 81; j++) { fakeBoard[j] = board[j]; fakeLastMove = lastMove; } int index = i % (numOfMoves); fakeMark = mark; if(BoardWinner(fakeBoard) == 0 && !IsBoardFull(fakeBoard)){ if(fullBoard) { fakeLastMove = PlaceMarkinNthFree(fakeBoard, fakeLastMove, index, fakeMark); } else { fakeLastMove = PlaceMoveinSubBoard(fakeBoard, fakeLastMove, index, fakeMark); } fakeMark = fakeMark * -1; while(BoardWinner(fakeBoard) == 0 && !IsBoardFull(fakeBoard)) { fakeLastMove = playRandomMove(fakeBoard, fakeLastMove, fakeMark); fakeMark = -1 * fakeMark; } } score[i % numOfMoves] = EvalFullBoard(fakeBoard) + score[i % numOfMoves]; } int winningIndex = 0; if(mark == X) { double max = score[0]; for(int i = 0; i < numOfMoves; i++) { if(score[i] > max) { winningIndex = i; max = score[i]; } } } else { double min = score[0]; for(int i = 0; i < numOfMoves; i++) { if(score[i] < min) { winningIndex = i; min = score[i]; } } } if(fullBoard) { return PlaceMarkinNthFree(board, lastMove, winningIndex, mark); } else { return PlaceMoveinSubBoard(board, lastMove, winningIndex, mark); } } __global__ void MonteCarloKernel(int* board, int* lastMove, int* mark, bool* fullBoard, int* numOfMoves, int* score, int Runs){ extern __shared__ int shared[]; int tId = threadIdx.x + (blockIdx.x * blockDim.x); int thread = threadIdx.x; curandState_t state; curand_init((unsigned long long)clock() + tId, 0, 0, &state); int o_board[81]; int fakeBoard[81]; int fakeLastMove; int fakeMark; if(thread < *numOfMoves) { shared[thread] = 0; } for(int j = 0; j < 81; j++) { o_board[j] = board[j]; } __syncthreads(); //offset by tID to reduce collisions on the scores for(int i = 0+tId; i < Runs +tId; i++) { //reset the board in local mem for(int j = 0; j < 81; j++) { fakeBoard[j] = o_board[j]; } int index = i % (*numOfMoves); fakeMark = *mark; fakeLastMove = *lastMove; if(BoardWinner(fakeBoard) == 0 && !IsBoardFull(fakeBoard)){ if(*fullBoard) { fakeLastMove = PlaceMarkinNthFree(fakeBoard, fakeLastMove, index, fakeMark); } else { fakeLastMove = PlaceMoveinSubBoard(fakeBoard, fakeLastMove, index, fakeMark); } fakeMark = fakeMark * -1; while(BoardWinner(fakeBoard) == 0 && !IsBoardFull(fakeBoard)) { fakeLastMove = playRandomMove(fakeBoard, fakeLastMove, fakeMark, state); fakeMark = -1 * fakeMark; } } atomicAdd(&shared[i%(*numOfMoves)], EvalFullBoardKenel(fakeBoard)); } __syncthreads(); if(thread < *numOfMoves) { atomicAdd(&score[thread], shared[thread]); } } int ParMonteCarlo(int* board, int lastMove, int mark, int Runs) { int *d_board, *d_score ,*d_numOfMoves, *d_mark, *d_lastMove; bool *d_fullBoard; int score[70]; memset(score, 0, sizeof(int) * 70); int subBoard = lastMove%9; bool fullBoard = SubBoardWinner(board+9*subBoard) != 0 || IsSubBoardFull(board+subBoard*9); int numOfMoves = NumberOfPossibleMoves(board, lastMove, fullBoard); cudaMalloc(&d_board, sizeof(int) * 81); cudaMalloc(&d_score, sizeof(int) * 70); cudaMalloc(&d_mark ,sizeof(int)); cudaMalloc(&d_lastMove ,sizeof(int)); cudaMalloc(&d_numOfMoves ,sizeof(int)); cudaMalloc(&d_fullBoard ,sizeof(bool)); cudaMemset(d_score, 0, sizeof(int) * 70); cudaMemcpy(d_board, board, sizeof(int) *81, cudaMemcpyHostToDevice); cudaMemcpy(d_mark,&mark,sizeof(int),cudaMemcpyHostToDevice); cudaMemcpy(d_numOfMoves,&numOfMoves,sizeof(int),cudaMemcpyHostToDevice); cudaMemcpy(d_lastMove,&lastMove,sizeof(int),cudaMemcpyHostToDevice); cudaMemcpy(d_fullBoard,&fullBoard,sizeof(bool),cudaMemcpyHostToDevice); MonteCarloKernel<<<NUM_BLOCKS,NUM_THREAD, sizeof(int) * 70>>>(d_board, d_lastMove, d_mark, d_fullBoard, d_numOfMoves, d_score, Runs); cudaDeviceSynchronize(); cudaMemcpy(score, d_score, sizeof(int)*70, cudaMemcpyDeviceToHost); int winningIndex = 0; if(mark == X) { double max = score[0]; for(int i = 0; i < 70; i++) { if(score[i] > max) { winningIndex = i; max = score[i]; } } } else { double min = score[0]; for(int i = 0; i < 70; i++) { if(score[i] < min) { winningIndex = i; min = score[i]; } } } cudaFree(d_board); cudaFree(d_score); cudaFree(d_mark); cudaFree(d_numOfMoves); cudaFree(d_lastMove); cudaFree(d_fullBoard); if(fullBoard) { return PlaceMarkinNthFree(board, lastMove, winningIndex, mark); } else { return PlaceMoveinSubBoard(board, lastMove, winningIndex, mark); } } int main() { clock_t start; clock_t diff; clock_t end; clock_t ParTime = 0; clock_t SeqTime = 0; int Xwin = 0; int Ywin = 0; srand(time(NULL)); for(int i = 0; i < 10; i++) { ParTime = 0; SeqTime = 0; int board[81]; memset(board, BLANK, sizeof(int)*81); int lastMove = 0; int mark = 1; bool test= true; while(BoardWinner(board) == 0 && !IsBoardFull(board) ) { if(test) { printf("Monte Carlo Turn in Parallel\n"); start = clock(); lastMove = ParMonteCarlo(board, lastMove, mark, RUNS_PER_THREAD); end = clock(); diff =end -start; ParTime += diff; printf("Par Time: %d\n", diff); } else { printf("Monte Carlo Turn in Sequence\n"); start = clock(); lastMove = MonteCarlo(board, lastMove, mark, NUM_SEQ_LOOPS); end = clock(); diff = end-start; SeqTime += diff; printf("Seq Time: %d\n", diff); } mark = mark * -1; test = !test; PrintBoard(board); } if(BoardWinner(board) == X) { Xwin++; } else if (BoardWinner(board)== O) { Ywin++; } printf("Parallel Time Total %d, Seq Time Total: %d\n",ParTime, SeqTime ); printf("BoardWinner: "); printSquare(BoardWinner(board)); printf("\n"); } printf("X won %d times\n out of 10\n",Xwin); printf("O won %d times\n out of 10\n",Ywin); return 0; }
12,994
#include "includes.h" __global__ void BFS_kernel_one_block_spill( volatile unsigned int *frontier, unsigned int frontier_len, volatile unsigned int *cost, volatile int *visited, unsigned int *edgeArray, unsigned int *edgeArrayAux, unsigned int numVertices, unsigned int numEdges, volatile unsigned int *frontier_length, const unsigned int max_local_mem) { extern volatile __shared__ unsigned int s_mem[]; //block queues unsigned int *b_q=(unsigned int *)&s_mem[0]; unsigned int *b_q2=(unsigned int *)&s_mem[max_local_mem]; volatile __shared__ unsigned int b_offset[1]; volatile __shared__ unsigned int b_q_length[1]; //get the threadId unsigned int tid=threadIdx.x; //copy frontier queue from global queue to local block queue if(tid<frontier_len) { b_q[tid]=frontier[tid]; } unsigned int f_len=frontier_len; while(1) { //Initialize the block queue size to 0 if(tid==0) { b_q_length[0]=0; b_offset[0]=0; } __syncthreads(); if(tid<f_len) { //get the nodes to traverse from block queue unsigned int node_to_process=*(volatile unsigned int *)&b_q[tid]; //remove from frontier visited[node_to_process]=0; //get the offsets of the vertex in the edge list unsigned int offset = edgeArray[node_to_process]; unsigned int next = edgeArray[node_to_process+1]; //Iterate through the neighbors of the vertex while(offset<next) { //get neighbor unsigned int nid=edgeArrayAux[offset]; //get its cost unsigned int v=atomicMin((unsigned int *)&cost[nid], cost[node_to_process]+1); //if cost is less than previously set add to frontier if(v>cost[node_to_process]+1) { int is_in_frontier=atomicExch((int *)&visited[nid],1); //if node already in frontier do nothing if(is_in_frontier==0) { //increment the warp queue size unsigned int t= atomicAdd((unsigned int *)&b_q_length[0],1); if(t< max_local_mem) { b_q2[t]=nid; } //write to global memory if shared memory full else { int off=atomicAdd((unsigned int *)&b_offset[0], 1); frontier[off]=nid; } } } offset++; } } __syncthreads(); if(tid<max_local_mem) b_q[tid]=*(volatile unsigned int *)&b_q2[tid]; __syncthreads(); //Traversal complete exit if(b_q_length[0]==0) { if(tid==0) frontier_length[0]=0; return; } // If frontier exceeds one block in size copy warp queues to //global frontier queue and exit else if( b_q_length[0] > blockDim.x || b_q_length[0] > max_local_mem) { if(tid<(b_q_length[0]-b_offset[0])) frontier[b_offset[0]+tid]= *(volatile unsigned int *)&b_q[tid]; if(tid==0) { frontier_length[0] = b_q_length[0]; } return; } f_len=b_q_length[0]; __syncthreads(); } }
12,995
#include <sys/time.h> #include <cuda.h> #include <stdio.h> #include <cuda_runtime_api.h> // For the CUDA runtime routines (prefixed with "cuda_") #include <cuda_runtime.h> //#include <cuda.h> //#include <helper_cuda.h> // time stamp function in seconds double getTimeStamp() { struct timeval tv ; gettimeofday( &tv, NULL ) ; return (double) tv.tv_usec/1000000 + tv.tv_sec ; } // host side matrix addition void h_addmat(float *A, float *B, float *C, int nx, int ny){ for (int i =0;i<nx;i++){ for(int j=0;j<ny;j++){ C[i*ny+j] = A[i*ny+j]+B[i*ny+j]; } } return; } // device-side matrix addition __global__ void f_addmat( float *A, float *B, float *C, int nx, int ny ){ // kernel code might look something like this // but you may want to pad the matrices and index into them accordingly int ix = threadIdx.x + blockIdx.x*(blockDim.x) ; int iy = threadIdx.y + blockIdx.y*(blockDim.y) ; int idx = ix*ny + iy ; //iy*ny + ix previously with <= instead of = //printf("Thread %d %d\n",ix,iy); if( (ix<nx) && (iy<ny) ){ C[idx] = A[idx] + B[idx] ; //printf("Thread %d %d\n",ix,iy); } } void initData(float *M, int x, int y, int width, int flag ){ //remove and put it in main assigining values in a single lool if(flag) { printf("A\n"); for (int i=0;i<x;i++){ for (int j=0;j<y;j++){ M[i*width+j] = (float)(i+j)/3.0; //printf("%f ",M[i*y+j]); } //printf("\n"); } } else { printf("B\n"); for (int i=0;i<x;i++){ for (int j=0;j<y;j++){ M[i*width+j] = (float)3.14*(i+j) ; //printf("%f ",M[i*y+j]); } //printf("\n"); } } } int main( int argc, char *argv[] ) { // get program arguments if (argc!=3){ printf("Fail"); exit(1); //printf("Fail"); } int nx = atoi( argv[1] ) ; // should check validity int ny = atoi( argv[2] ) ; // should check validity int mx=0,my=0; //if((nx%16) != 0){ // mx = 0; // //mx = 16 - (nx%16); //} if((ny%1024) != 0){ my = 1024 - (ny%1024); } int noElems = (nx+mx)*(ny+my) ; printf ("%d %d %d %d \n",(nx*ny),(noElems),mx,my); int bytes = noElems * sizeof(float) ; // but you may want to pad the matrices… // alloc memory host-side //float *h_A = (float *) malloc( bytes ) ; //float *h_B = (float *) malloc( bytes ) ; float *h_hC = (float *) malloc( bytes ) ; // host result //float *h_dC = (float *) malloc( bytes ) ; // gpu result // init matrices with random data //initData(h_A,nx,ny,1); initData(h_B,nx,ny,0); // alloc memory dev-side float *d_A, *d_B, *d_C ; cudaMalloc( (void **) &d_A, bytes ) ; cudaMalloc( (void **) &d_B, bytes ) ; cudaMalloc( (void **) &d_C, bytes ) ; float *h_Ap, *h_Bp, *h_dCp; cudaMallocHost( (float **) &h_Ap, bytes ) ; cudaMallocHost( (float **) &h_Bp, bytes ) ; cudaMemset(h_Ap,0,bytes); cudaMemset(h_Bp,0,bytes); initData(h_Ap,nx,ny,ny+my,1); initData(h_Bp,nx,ny,ny+my,0); for(int i=0;i<(nx+mx);i++){ for(int j=0;j<(ny+my);j++){ //if(h_hC[i*(ny+my)+j] != h_dCp[i*(ny+my)+j]) //printf("%d ",j); //printf("%f %f %d %d\n",h_Ap[i*(ny+my)+j],h_Bp[i*(ny+my)+j],i,j); //if(h_hC[i*(ny+my)+j] != h_dCp[i*(ny+my)+j]) // flag=1; } //printf("\n"); } cudaMallocHost( (float **) &h_dCp, bytes ) ; cudaMemset(h_dCp,0,bytes); double timeStampA = getTimeStamp() ; //transfer data to dev cudaMemcpy( d_A, h_Ap, bytes, cudaMemcpyHostToDevice ) ; cudaMemcpy( d_B, h_Bp, bytes, cudaMemcpyHostToDevice ) ; // note that the transfers would be twice as fast if h_A and h_B // matrices are pinned double timeStampB = getTimeStamp() ; // invoke Kernel dim3 block( 1, 1024) ; // you will want to configure this dim3 grid( (nx+block.x-1)/block.x, (ny+my)/block.y) ; //(ny+block.y-1)/block.y) printf("Grid %d %d \n",(nx+mx)/block.x,(ny+my)/block.y); f_addmat<<<grid, block>>>( d_A, d_B, d_C, nx, ny+my ) ; cudaDeviceSynchronize() ; double timeStampC = getTimeStamp() ; //copy data back cudaMemcpyAsync(h_dCp, d_C, bytes, cudaMemcpyDeviceToHost); cudaDeviceSynchronize() ; //learn how to comment and uncomment in one go /* printf("C\n"); for(int i=0;i<nx;i++){ for(int j=0;j<ny;j++){ //printf("%f ",h_dC[i*ny+j]); } //printf("\n"); } */ double timeStampD = getTimeStamp() ; //for(int i=0; i<nx; i++){ // for(int j=0; j<ny; j++){ // printf("%f ",h_dC[i*ny+j]); // } // printf("\n"); //} // free GPU resources cudaFree( d_A ) ; cudaFree( d_B ) ; cudaFree( d_C ) ; //cudaFreeHost(h_Ap); cudaFreeHost(h_Bp); //cudaFreeHost(h_dCp); //cudaDeviceReset() ; // check result printf("%f %f %f %f\n",(timeStampD-timeStampA),(timeStampB-timeStampA),(timeStampC-timeStampB),(timeStampD-timeStampC)); h_addmat( h_Ap, h_Bp, h_hC, nx+mx, ny+my ) ; int flag = 0; for(int i=0;i<(nx+mx);i++){ for(int j=0;j<(ny+my);j++){ //if(h_hC[i*(ny+my)+j] != h_dCp[i*(ny+my)+j]) //printf("%d ",j); //printf("%f %f %d %d\n",h_hC[i*(ny+my)+j],h_dCp[i*(ny+my)+j],i,j); if(h_hC[i*(ny+my)+j] != h_dCp[i*(ny+my)+j]) flag=1; } //printf("\n"); } cudaFreeHost(h_Ap); cudaFreeHost(h_Bp); cudaFreeHost(h_dCp); free(h_hC); cudaDeviceReset() ; printf("\n %d \n",flag); }
12,996
#include <iostream> #include <random> #include <functional> #include <chrono> #include <unistd.h> #ifndef BUF_KIND #define BUF_KIND 0 #endif #if(BUF_KIND == 0) #define BUF_TYPE float #else #define BUF_TYPE double #endif #define CudaWrap(EXP) \ { \ auto ret = EXP; \ if (ret != cudaSuccess) { \ std::cerr << "Error! " << cudaGetErrorString(ret) << " (" << ret << ")" << std::endl; \ return 1; \ }\ } __global__ void addKernel(const BUF_TYPE* A, BUF_TYPE* B, size_t size) { size_t i = blockIdx.x*blockDim.x+threadIdx.x; if (i < size) { B[i] = A[i] + B[i]; } } int main() { // Initialize random number generator std::mt19937_64 generator; generator.seed(42); std::uniform_real_distribution<BUF_TYPE> distribution(-1., 1.); auto gen = std::bind(distribution, generator); // Initialize arrays size_t num_gen = (1<<(30))/sizeof(BUF_TYPE); // Allocate arrays. BUF_TYPE* array1 = new BUF_TYPE[num_gen]; BUF_TYPE* array2 = new BUF_TYPE[num_gen]; // Allocate Device arrays. BUF_TYPE* dev_1 = nullptr; BUF_TYPE* dev_2 = nullptr; CudaWrap(cudaMalloc(&dev_1, num_gen*sizeof(BUF_TYPE))); CudaWrap(cudaMalloc(&dev_2, num_gen*sizeof(BUF_TYPE))); // Fill arrays with values for(size_t i=0; i < num_gen; ++i) { array1[i] = gen(); array2[i] = gen(); } // Copy data to Device cudaMemcpy(dev_1, array1, num_gen*sizeof(BUF_TYPE), cudaMemcpyHostToDevice); cudaMemcpy(dev_2, array2, num_gen*sizeof(BUF_TYPE), cudaMemcpyHostToDevice); cudaDeviceSynchronize(); auto start = std::chrono::high_resolution_clock::now(); // Compute on Device int blockSize = 256; int numBlocks = (num_gen+blockSize-1)/(blockSize); addKernel<<<numBlocks,blockSize>>>(array1, array2, num_gen); // Copy data out of Device cudaMemcpy(array2, dev_2, num_gen*sizeof(BUF_TYPE), cudaMemcpyDeviceToHost); // Wait for operations to finish. cudaDeviceSynchronize(); auto stop = std::chrono::high_resolution_clock::now(); // Do something with the arrays so the addition isn't optimized out. BUF_TYPE sum = 0.; for(size_t i=0; i < num_gen; ++i) { sum += array2[i]; } std::cout << sum << std::endl; std::cout << std::hexfloat; std::cout << sum << std::endl; auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(stop-start); std::cout << "Took " << duration.count() << " milliseconds" << std::endl; cudaFree(dev_1); cudaFree(dev_2); delete [] array1; delete [] array2; return 0; }
12,997
#include <cstdio> int main(void) { cudaDeviceProp prop; cudaGetDeviceProperties(&prop, 0); printf("-gencode arch=compute_%d%d,code=sm_%d%d\n", prop.major, prop.minor, prop.major, prop.minor); return 0; }
12,998
#include "includes.h" #define max(a, b) ((a > b)?a:b) #define THREADSPERDIM 16 #define FALSE 0 #define TRUE !FALSE // mX has order rows x cols // vectY has length rows // mX has order rows x cols // vectY has length rows __global__ void getRestricted(int countx, int county, int rows, int cols, float * mX, int mXdim, float * vY, int vYdim, float * mQ, int mQdim, float * mR, int mRdim, float * vectB, int vectBdim) { int m = blockIdx.x * THREADSPERDIM + threadIdx.x, n, i, j, k; float sum, invnorm, * X, * Y, * Q, * R, * B, * coli, * colj, * colQ, * colX; if(m >= county) return; if(m == 1) n = 0; else n = 1; X = mX + (m * mXdim); // initialize the intercepts for(i = 0; i < rows; i++) X[i] = 1.f; Y = vY + (m * countx + n) * vYdim; B = vectB + m * vectBdim; Q = mQ + m * mQdim; R = mR + m * mRdim; // initialize Q with X ... for(i = 0; i < rows; i++) { for(j = 0; j < cols; j++) Q[i+j*rows] = X[i+j*rows]; } // gramm-schmidt process to find Q for(j = 0; j < cols; j++) { colj = Q+rows*j; for(i = 0; i < j; i++) { coli = Q+rows*i; sum = 0.f; for(k = 0; k < rows; k++) sum += coli[k] * colj[k]; for(k = 0; k < rows; k++) colj[k] -= sum * coli[k]; } sum = 0.f; for(i = 0; i < rows; i++) sum += colj[i] * colj[i]; invnorm = 1.f / sqrtf(sum); for(i = 0; i < rows; i++) colj[i] *= invnorm; } for(i = cols-1; i > -1; i--) { colQ = Q+i*rows; // matmult Q * X -> R for(j = 0; j < cols; j++) { colX = X+j*rows; sum = 0.f; for(k = 0; k < rows; k++) sum += colQ[k] * colX[k]; R[i+j*cols] = sum; } sum = 0.f; // compute the vector Q^t * Y -> B for(j = 0; j < rows; j++) sum += colQ[j] * Y[j]; // back substitution to find the x for Rx = B for(j = cols-1; j > i; j--) sum -= R[i+j*cols] * B[j]; B[i] = sum / R[i+i*cols]; } }
12,999
// NV GPU compute capability detection. // build: nvcc arch-test.cu && ./a.out // // Refs: <https://github.com/BVLC/caffe/blob/master/cmake/Cuda.cmake#L18-L31> #include <cstdio> int main() { int count = 0; if (cudaSuccess != cudaGetDeviceCount(&count)) return -1; if (count == 0) return -1; for (int device = 0; device < count; ++device) { cudaDeviceProp prop; if (cudaSuccess == cudaGetDeviceProperties(&prop, device)) std::printf("%d.%d \n", prop.major, prop.minor); } return 0; }
13,000
// // Average extreme spread of five-shot group assuming impact coordinates follow standard normal distribution // // Building: // nvcc -std=c++11 es_cuda.cu -o es_cuda -lcurand // // Running: // for run in {1..10}; do ./es_cuda 15 | tee -a es_cuda.csv; done // #include <string> #include <vector> #include <numeric> #include <stdexcept> #include <typeinfo> #include <cooperative_groups.h> #include <iostream> #include <iomanip> #include <stdexcept> #include <cuda_runtime.h> #include <math.h> #include <chrono> #include <curand.h> namespace cg = cooperative_groups; using std::string; using std::vector; // First level of reduction __device__ double reduce_sum(double in, cg::thread_block cta) { extern __shared__ double sdata[]; // Write to shared memory unsigned ltid = threadIdx.x; sdata[ltid] = in; cg::sync(cta); // Do reduction in shared memory for (unsigned s = blockDim.x / 2 ; s > 0 ; s >>= 1) { if (ltid < s) { sdata[ltid] += sdata[ltid + s]; } cg::sync(cta); } return sdata[0]; } // Estimator kernel __global__ void computeValue(double* const results, const double* const points, const unsigned numGroups) { // Handle to thread block group cg::thread_block cta = cg::this_thread_block(); // Determine thread ID unsigned bid = blockIdx.x; unsigned tid = blockIdx.x * blockDim.x + threadIdx.x; unsigned step = gridDim.x * blockDim.x; // Shift the input/output pointers const double* pointx = points + tid; const double* pointy = pointx + 5 * numGroups; // Sum up extreme spread of all groups double sum = 0; for (unsigned i = tid ; i < numGroups; i += step, pointx += step * 5, pointy += step * 5) { // Pairwise distances double dx[10], dy[10]; // Unroll nested comparison loops dx[0] = pointx[0] - pointx[1]; dy[0] = pointy[0] - pointy[1]; dx[1] = pointx[0] - pointx[2]; dy[1] = pointy[0] - pointy[2]; dx[2] = pointx[0] - pointx[3]; dy[2] = pointy[0] - pointy[3]; dx[3] = pointx[0] - pointx[4]; dy[3] = pointy[0] - pointy[4]; dx[4] = pointx[1] - pointx[2]; dy[4] = pointy[1] - pointy[2]; dx[5] = pointx[1] - pointx[3]; dy[5] = pointy[1] - pointy[3]; dx[6] = pointx[1] - pointx[4]; dy[6] = pointy[1] - pointy[4]; dx[7] = pointx[2] - pointx[3]; dy[7] = pointy[2] - pointy[3]; dx[8] = pointx[2] - pointx[4]; dy[8] = pointy[2] - pointy[4]; dx[9] = pointx[3] - pointx[4]; dy[9] = pointy[3] - pointy[4]; double max_d2 = 0; for (unsigned j = 0; j < 10; j++) { auto candidate_d2 = dx[j] * dx[j] + dy[j] * dy[j]; max_d2 = max(max_d2, candidate_d2); } double es = sqrt(max_d2); sum += es; } // Reduce within the block sum = reduce_sum(sum, cta); // Store the result if (threadIdx.x == 0) { results[bid] = sum; } } double es_cuda(unsigned power_of_4, unsigned seed) { // Get device properties struct cudaDeviceProp deviceProperties; cudaError_t cudaResult = cudaGetDeviceProperties(&deviceProperties, 0); if (cudaResult != cudaSuccess) { string msg("Could not get device properties: "); msg += cudaGetErrorString(cudaResult); throw std::runtime_error(msg); } // Check precision is valid if (deviceProperties.major < 1 || (deviceProperties.major == 1 && deviceProperties.minor < 3)) { throw std::runtime_error("Device does not have double precision support"); } // Check requested size is valid const unsigned threadBlockSize = 128; if (threadBlockSize > (deviceProperties.maxThreadsPerBlock)) { throw std::runtime_error("Thread block size is greater than maxThreadsPerBlock"); } dim3 block; block.x = threadBlockSize; // Attach to GPU cudaResult = cudaSetDevice(0); if (cudaResult != cudaSuccess) { string msg("Could not set CUDA device: "); msg += cudaGetErrorString(cudaResult); throw std::runtime_error(msg); } // Aim to launch around ten or more times as many blocks as there // are multiprocessors on the target device. dim3 grid; const unsigned numGroups = 1 << (2 * power_of_4); grid.x = numGroups / threadBlockSize; while (grid.x > 20 * deviceProperties.multiProcessorCount) { grid.x >>= 1; } // Get computeValue function properties and check the maximum block size struct cudaFuncAttributes funcAttributes; cudaResult = cudaFuncGetAttributes(&funcAttributes, computeValue); if (cudaResult != cudaSuccess) { string msg("Could not get function attributes: "); msg += cudaGetErrorString(cudaResult); throw std::runtime_error(msg); } if (block.x > (unsigned)funcAttributes.maxThreadsPerBlock) { throw std::runtime_error("Block X dimension is too large for computeValue kernel"); } // Check the dimensions are valid if (block.x > (unsigned)deviceProperties.maxThreadsDim[0]) { throw std::runtime_error("Block X dimension is too large for device"); } if (grid.x > (unsigned)deviceProperties.maxGridSize[0]) { throw std::runtime_error("Grid X dimension is too large for device"); } // Allocate memory for points // Each simulation has ten random numbers to give five pairs of X and Y coordinates double* d_points = 0; cudaResult = cudaMalloc((void **)&d_points, 10 * numGroups * sizeof(double)); if (cudaResult != cudaSuccess) { string msg("Could not allocate memory on device for random numbers: "); msg += cudaGetErrorString(cudaResult); throw std::runtime_error(msg); } // Allocate memory for result // Each thread block will produce one result double* d_results = 0; cudaResult = cudaMalloc((void**)&d_results, grid.x * sizeof(double)); if (cudaResult != cudaSuccess) { string msg("Could not allocate memory on device for partial results: "); msg += cudaGetErrorString(cudaResult); throw std::runtime_error(msg); } // Generate random points curandStatus_t curandResult; curandGenerator_t prng; curandResult = curandCreateGenerator(&prng, CURAND_RNG_PSEUDO_DEFAULT); if (curandResult != CURAND_STATUS_SUCCESS) { string msg("Could not create pseudo-random number generator: "); msg += curandResult; throw std::runtime_error(msg); } curandResult = curandSetPseudoRandomGeneratorSeed(prng, seed); if (curandResult != CURAND_STATUS_SUCCESS) { string msg("Could not set seed for pseudo-random number generator: "); msg += curandResult; throw std::runtime_error(msg); } curandResult = curandGenerateNormalDouble(prng, (double*)d_points, 10 * numGroups, 0, 1); if (curandResult != CURAND_STATUS_SUCCESS) { string msg("Could not generate pseudo-random numbers: "); msg += curandResult; throw std::runtime_error(msg); } curandResult = curandDestroyGenerator(prng); if (curandResult != CURAND_STATUS_SUCCESS) { string msg("Could not destroy pseudo-random number generator: "); msg += curandResult; throw std::runtime_error(msg); } // Calculate and average group size computeValue<<<grid, block, block.x * sizeof(double)>>>(d_results, d_points, numGroups); // Copy the results back to host vector<double> results(grid.x); cudaResult = cudaMemcpy(&results[0], d_results, grid.x * sizeof(double), cudaMemcpyDeviceToHost); if (cudaResult != cudaSuccess) { string msg("Could not copy results to host: "); msg += cudaGetErrorString(cudaResult); throw std::runtime_error(msg); } // Complete sum-reduction double sum = std::accumulate(results.begin(), results.end(), double(0)); // Cleanup if (d_points) { cudaFree(d_points); } if (d_results) { cudaFree(d_results); } // Divide sum by count to get the average return sum / numGroups; } int main(int argc, char **argv) { unsigned power_of_4 = 12; if (argc == 2) { power_of_4 = atoi(argv[1]); } unsigned nt = 12; if (power_of_4 > 12) { nt <<= 2 * (power_of_4 - 12); power_of_4 = 12; } try { auto start_time = std::chrono::system_clock::now(); double avg = 0, min = 100, max = 0; __uint128_t mcg128_state = time(NULL) | 1; // can be seeded to any odd number for (unsigned j = 0; j < nt; j++) { double r = es_cuda(power_of_4, (unsigned)(mcg128_state >> 64)); avg += r; min = fmin(r, min); max = fmax(r, max); mcg128_state *= 0xda942042e4dd58b5ULL; } avg /= nt; auto end_time = std::chrono::system_clock::now(); std::chrono::duration<double> seconds = end_time - start_time; std::cout.precision(14); std::cout << "code,threads,power_of_4,min,avg,max,time\n"; std::cout << "CUDA," << nt << "," << power_of_4 << "," << min << "," << avg << "," << max << "," << seconds.count() << "\n"; } catch (std::runtime_error &e) { // es_cuda() can throw runtime exceptions fprintf(stderr, "runtime error (%s)\n", e.what()); return(EXIT_FAILURE); } return(EXIT_SUCCESS); }