serial_no
int64
1
24.2k
cuda_source
stringlengths
11
9.01M
10,801
#include <cstdio> #include <cuda.h> static const int ThreadsPerBlock = 512; static int* d_maxlen; static __global__ void collatz(const long start, const long stop, int* const maxlen) { // todo: process odd values from start (assume start to be odd) to stop (inclusively if stop is odd) with one thread per value (based on code from previous project) const long i = threadIdx.x + blockIdx.x * (long)blockDim.x; if(i+start < stop ) // Each thread does work if and only if less than the stop value { long val = 2*(i +((start-1)/2))+1; int len = 1; while(val != 1){ len++; if((val % 2) == 0)//even {val = val / 2;} else //Odd {val = 3 * val +1;} } if(len > *maxlen){ atomicMax(maxlen, len);} // If greater than greatest length, becomes new max len; } } void GPU_Init() { int maxlen = 0; if (cudaSuccess != cudaMalloc((void **)&d_maxlen, sizeof(int))) {fprintf(stderr, "ERROR: could not allocate memory\n"); exit(-1);} if (cudaSuccess != cudaMemcpy(d_maxlen, &maxlen, sizeof(int), cudaMemcpyHostToDevice)) {fprintf(stderr, "ERROR: copying to device failed\n"); exit(-1);} } void GPU_Exec(const long start, const long stop) { if (start <= stop) { collatz<<<((stop - start + 2) / 2 + ThreadsPerBlock - 1) / ThreadsPerBlock, ThreadsPerBlock>>>(start, stop, d_maxlen); } } int GPU_Fini() { int maxlen; // todo: copy the result from the device to the host and free the device memory if(cudaSuccess != cudaMemcpy(&maxlen, d_maxlen, sizeof(int), cudaMemcpyDeviceToHost)){fprintf(stderr, "Error: copying to host failed\n"); exit(-1);} cudaFree(d_maxlen); return maxlen; }
10,802
#include <stdlib.h> #include <stdio.h> #define N 10 __global__ void VecAdd(float* A, float* B, float* C) { int i = threadIdx.x; printf("tid: x=%d\n", i); C[i] = A[i] + B[i]; } int main() { float A[N] = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0}; float B[N] = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0}; float *C = (float*)malloc(sizeof(float) * N); float *dA, *dB, *dC; cudaMalloc((void**)&dA, N * sizeof(float)); cudaMalloc((void**)&dB, N * sizeof(float)); cudaMalloc((void**)&dC, N * sizeof(float)); cudaMemcpy(dA, A, N * sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(dB, B, N * sizeof(float), cudaMemcpyHostToDevice); // Kernel invocation with N threads VecAdd<<<1, N>>>(dA, dB, dC); cudaMemcpy(C, dC, N * sizeof(float), cudaMemcpyDeviceToHost); cudaFree(dA); cudaFree(dB); cudaFree(dC); for(int i=0; i<N; i++) { printf("%d: %f\n", i, *(C+i)); } free(C); getchar(); return 0; }
10,803
#include "cuda.h" __global__ void return_double_(int n, double *b, const double*a){ int i = blockIdx.x * blockDim.x + threadIdx.x; if (i<n) b[i] = 2*a[i]; } void return_double(int n, double *b, const double*a){ return_double_<<<(n+255)/256, 256>>>(n, b, a); }
10,804
/*** Original Question : https://stackoverflow.com/questions/13215614/ I am new to CUDA. I am trying to parallelize the following code. Right now it's sitting on kernel but is not using threads at all, thus slow. I tried to use this answer but to no avail so far. The kernel is supposed to generate first n prime numbers, put them into device_primes array and this array is later accessed from host. The code is correct and works fine in serial version but I need to speed it up, perhaps with use of shared memory. //CUDA kernel code __global__ void generatePrimes(int* device_primes, int n) { //int i = blockIdx.x * blockDim.x + threadIdx.x; //int j = blockIdx.y * blockDim.y + threadIdx.y; int counter = 0; int c = 0; for (int num = 2; counter < n; num++) { for (c = 2; c <= num - 1; c++) { if (num % c == 0) //not prime { break; } } if (c == num) //prime { device_primes[counter] = num; counter++; } } } My current, preliminary, and definitely wrong attempt to parallelize this looks like the following: //CUDA kernel code __global__ void generatePrimes(int* device_primes, int n) { int i = blockIdx.x * blockDim.x + threadIdx.x; int j = blockIdx.y * blockDim.y + threadIdx.y; int num = i + 2; int c = j + 2; int counter = 0; if ((counter >= n) || (c > num - 1)) { return; } if (num % c == 0) //not prime { } if (c == num) //prime { device_primes[counter] = num; counter++; } num++; c++; } But this code populates the array with data that does not make sense. In addition, many values are zeroes. Thanks in advance for any help, it's appreciated. ***/ __global__ void getPrimes(int *device_primes,int n) { int c = 0; int thread_id = blockIdx.x * blockDim.x + threadIdx.x; int num = thread_id; if (thread_id == 0) device_primes[0] = 1; __syncthreads(); while(device_primes[0] < n) { for (c = 2; c <= num - 1; c++) { if (num % c == 0) //not prime { break; } } if (c == num) //prime { int pos = atomicAdd(&device_primes[0],1); device_primes[pos] = num; } num += blockDim.x * gridDim.x; // Next number for this thread } }
10,805
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <math.h> #include<curand_kernel.h> // __device__ float calcDistance(int target_row ,int target_col,int source_row, int source_col , double * target_block ,double * source_block,int target_rows,int target_cols ,int source_rows,int source_cols) { int r = 3; double dif=0; double dif0=1,dif1=1,dif2=1; if( target_row-r<0 ) target_row = r; if( target_col-r<0 ) target_col =r; if( source_row-r<0 ) source_row = r; if( source_col-r<0 ) source_col = r; if( target_row+r>=target_rows ) target_row = target_cols-1-r; if( target_col+r>= target_cols ) target_col = target_rows-1-r; if( source_row+r>=source_rows ) source_row = source_rows-1-r; if( source_col+r>= source_cols ) source_col = source_cols-1-r; for(int i=-r ;i<=r;i++){ for(int j=-r ;j<=r;j++){ int temp = 3*((source_row+i)*source_cols+source_col+j) ; int temp2 = 3*((target_row+i)*target_cols+target_col+j) ; dif0 = source_block[ temp+ 0] - target_block[temp2+ 0] ; dif1 = source_block[ temp+ 1] - target_block[temp2+ 1] ; dif2 = source_block[ temp+ 2] - target_block[temp2+ 2] ; dif += sqrt(dif0*dif0 +dif1*dif1 +dif2*dif2); } } return dif; } __device__ int calcDistance(int target_row ,int target_col,int source_row1, int source_col1,int source_row2, int source_col2 ,int source_row3, int source_col3,double * target_block , double *source_block,int target_rows,int target_cols ,int source_rows,int source_cols) { float first2Second = calcDistance(target_row, target_col , source_row1, source_col1,target_block ,source_block,target_rows,target_cols,source_rows, source_cols); float first2Third = calcDistance( target_row, target_col , source_row2, source_col2,target_block ,source_block,target_rows,target_cols,source_rows, source_cols); float first2Fourth = calcDistance( target_row,target_col , source_row3, source_col3,target_block ,source_block,target_rows,target_cols,source_rows, source_cols); if (first2Second<=first2Third) { if (first2Second<=first2Fourth) return 1; else return 3; } else if (first2Third<= first2Fourth) return 2; else return 3; } __device__ int calcDistance(int target_row,int target_col ,int source_row1, int source_col1,int source_row2, int source_col2 ,double * target_block ,double * source_block,int target_rows,int target_cols ,int source_rows,int source_cols) { float first2Second = calcDistance(target_row, target_col , source_row1, source_col1,target_block ,source_block,target_rows,target_cols,source_rows, source_cols); float first2Third = calcDistance( target_row, target_col , source_row2, source_col2,target_block ,source_block,target_rows,target_cols,source_rows, source_cols); if (first2Second <= first2Third) return 1; return 2; } __global__ void extern PropagationGPU(double * target_block ,double * source_block , int * relation_block , int target_rows , int target_cols ,int source_rows , int source_cols) { //ݹ̣ int y = blockIdx.x; int x = threadIdx.x; //뾶 int c_r0 = relation_block[ 2*(y*target_cols+x) + 0 ]; int c_c0 = relation_block[ 2*(y*target_cols+x) + 1]; int c_r1 = relation_block[ 2*((y+1)*target_cols+x) + 0 ]-1; int c_c1 = relation_block[ 2*((y+1)*target_cols+x) + 1 ]; int c_r2 = relation_block[ 2*(y*target_cols+x+1) + 0]; int c_c2 = relation_block[ 2*(y*target_cols+x+1) + 1]-1; int patchNumber = calcDistance(y , x , c_r0 , c_c0 , c_r1, c_c1 , c_r2 , c_c2 , target_block ,source_block,target_rows,target_cols,source_rows, source_cols); switch(patchNumber) { case 2: relation_block[ 2*(y*target_cols+x) + 0 ]= c_r1; relation_block[ 2*(y*target_cols+x) + 1 ]= c_c1; break; case 3: relation_block[ 2*(y*target_cols+x) + 0 ] = c_r2; relation_block[ 2*(y*target_cols+x) + 1 ] = c_c2; break; } } __global__ void extern RandomSearchGPU(double * target_block ,double * source_block ,int * relation_block,int target_rows,int target_cols ,int source_rows,int source_cols){ //ݹ̣ int y = blockIdx.x; int x = threadIdx.x; //뾶 int c_r0 = relation_block[ 2*(y*target_cols+x) + 0 ]; int c_c0 = relation_block[ 2*(y*target_cols+x) + 1]; int c_r1 = relation_block[ 2*((y-2)*target_cols+x) + 0 ]+2; int c_c1 = relation_block[ 2*((y-2)*target_cols+x) + 1 ]; int c_r2 = relation_block[ 2*(y*target_cols+x-2) + 0]; int c_c2 = relation_block[ 2*(y*target_cols+x-2) + 1]+2; int patchNumber = calcDistance(y , x , c_r0 , c_c0 , c_r1, c_c1 , c_r2 , c_c2 , target_block ,source_block,target_rows,target_cols,source_rows, source_cols); switch(patchNumber) { case 2: relation_block[ 2*(y*target_cols+x) + 0 ]= c_r1; relation_block[ 2*(y*target_cols+x) + 1 ]= c_c1; break; case 3: relation_block[ 2*(y*target_cols+x) + 0 ] = c_r2; relation_block[ 2*(y*target_cols+x) + 1 ] = c_c2; break; } } __global__ void extern baoli(double * target_block ,double * source_block ,int * relation_block,int target_rows,int target_cols ,int source_rows,int source_cols,double *distance){ int y = threadIdx.y; int x = threadIdx.x; for(int i = 0 ; i<12;i++){ for(int j = 0 ;j< 12 ; j++){ double c = calcDistance(y , x , i , j , target_block ,source_block,target_rows,target_cols,source_rows, source_cols); if( c < distance[ y*target_cols+x ]){ relation_block[ 2*(y*target_cols+x) + 0 ]= 1; relation_block[ 2*(y*target_cols+x) + 1 ]= 1; distance[ y*target_cols+x ] = c; } } } } void extern bridge(double * target_block ,double * source_block ,int * relation_block,int target_rows,int target_cols ,int source_rows,int source_cols, double * distance){ //̴߳Сƣ /**/ for(int i = 0;i<130 ;i++){ PropagationGPU<<<target_rows ,target_cols>>>(target_block, source_block, relation_block , target_rows , target_cols , source_rows , source_cols); cudaThreadSynchronize(); RandomSearchGPU<<<target_rows ,target_cols>>>(target_block, source_block, relation_block,target_rows,target_cols,source_rows, source_cols); cudaThreadSynchronize(); } //baoli<<<target_rows ,target_cols>>>(target_block, source_block, relation_block , target_rows , target_cols , source_rows , source_cols ,distance); }
10,806
#include<stdio.h> #include<stdlib.h> #include<cuda.h> #define N 9 __global__ void sum(int *a,int *o) { int of; int id=threadIdx.x; for(of=N/2 ; of > 0 ;of=of/2) { if(id<of) { a[id]+=a[id+of]; } } if(N%2==1) { a[0]=a[0]+a[N-1]; } o[0]=a[0]; } int main() { int *h_a,*d_a,*oh_a,*od_a; int size= N * sizeof(int); h_a=(int*)malloc(size); oh_a=(int*)malloc(size); cudaMalloc(&d_a,size); cudaMalloc(&od_a,size); int i; for(i=0 ;i<N ;i++) { h_a[i] = random() % N; } printf("\n\nNumbers =>"); for(i=0 ;i<N ;i++) { printf("%d ",h_a[i]); } cudaMemcpy(d_a, h_a,size,cudaMemcpyHostToDevice); sum<<<1, N/2>>>(d_a,od_a); cudaMemcpy(oh_a, od_a,size,cudaMemcpyDeviceToHost); printf("\n\nSum => %d",oh_a[0]); float arithmeticMean=(float)oh_a[0]/N; printf("\n\nArithmetic Mean => %f",arithmeticMean); cudaFree(d_a); cudaFree(od_a); free(h_a); free(oh_a); return 0; }
10,807
#include "matmul.hh" #include "../runtime/node.hh" namespace gpu { namespace { __global__ void matmul(const dbl_t* a, const dbl_t* b, dbl_t* out, std::size_t arows, std::size_t acols, std::size_t bcols) { std::size_t row = blockIdx.x * blockDim.x + threadIdx.x; std::size_t col = blockIdx.y * blockDim.y + threadIdx.y; if (row >= arows || col >= bcols) return; dbl_t x = 0; for (std::size_t i = 0; i < acols; ++i) x += a[row * acols + i] * b[i * bcols + col]; out[row * bcols + col] = x; } __global__ void mvrow_add(const dbl_t* a, const dbl_t* b, dbl_t* out, std::size_t rows, std::size_t cols) { std::size_t row = blockIdx.x * blockDim.x + threadIdx.x; std::size_t col = blockIdx.y * blockDim.y + threadIdx.y; if (row >= rows || col >= cols) return; out[row * cols + col] = a[row * cols + col] + b[col]; } __global__ void matmul_add(const dbl_t* a, const dbl_t* b, const dbl_t* c, dbl_t* out, std::size_t arows, std::size_t acols, std::size_t bcols) { std::size_t row = blockIdx.x * blockDim.x + threadIdx.x; std::size_t col = blockIdx.y * blockDim.y + threadIdx.y; if (row >= arows || col >= bcols) return; dbl_t x = 0; for (std::size_t i = 0; i < acols; ++i) x += a[row * acols + i] * b[i * bcols + col]; out[row * bcols + col] = x + c[col]; } __global__ void tmat_mat_mul(const dbl_t* a, const dbl_t* b, dbl_t* out, std::size_t acols, std::size_t arows, std::size_t bcols) { std::size_t row = blockIdx.x * blockDim.x + threadIdx.x; std::size_t col = blockIdx.y * blockDim.y + threadIdx.y; if (row >= acols || col >= bcols) return; dbl_t x = 0; for (std::size_t i = 0; i < arows; ++i) x += a[i * acols + row] * b[i * bcols + col]; out[row * bcols + col] = x; } __global__ void mat_tmat_mul(const dbl_t* a, const dbl_t* b, dbl_t* out, std::size_t arows, std::size_t acols, std::size_t brows) { std::size_t row = blockIdx.x * blockDim.x + threadIdx.x; std::size_t col = blockIdx.y * blockDim.y + threadIdx.y; if (row >= arows || col >= brows) return; dbl_t x = 0; for (std::size_t i = 0; i < acols; ++i) x += a[row * acols + i] * b[col * acols + i]; out[row * brows + col] = x; } } void kernel_mat_mat_mul(rt::Node* node) { std::size_t arows = node->len1; std::size_t acols = node->len2; std::size_t bcols = node->len3; std::size_t block_size = 32; dim3 threads_per_block (block_size, block_size); std::size_t nb_blocks_x = (arows + block_size - 1) / block_size; std::size_t nb_blocks_y = (bcols + block_size - 1) / block_size; dim3 blocks_per_grid (nb_blocks_x, nb_blocks_y); matmul<<<blocks_per_grid, threads_per_block>>>(node->in1, node->in2, node->out1, arows, acols, bcols); } void kernel_mat_rvect_add(rt::Node* node) { std::size_t rows = node->len1; std::size_t cols = node->len2; std::size_t block_size = 32; dim3 threads_per_block (block_size, block_size); std::size_t nb_blocks_x = (rows + block_size - 1) / block_size; std::size_t nb_blocks_y = (cols + block_size - 1) / block_size; dim3 blocks_per_grid (nb_blocks_x, nb_blocks_y); mvrow_add<<<blocks_per_grid, threads_per_block>>>(node->in1, node->in2, node->out1, rows, cols); } void kernel_mat_mul_add(rt::Node* node) { std::size_t arows = node->len1; std::size_t acols = node->len2; std::size_t bcols = node->len3; std::size_t block_size = 32; dim3 threads_per_block (block_size, block_size); std::size_t nb_blocks_x = (arows + block_size - 1) / block_size; std::size_t nb_blocks_y = (bcols + block_size - 1) / block_size; dim3 blocks_per_grid (nb_blocks_x, nb_blocks_y); matmul_add<<<blocks_per_grid, threads_per_block>>>(node->in1, node->in2, node->in3, node->out1, arows, acols, bcols); } void kernel_tmat_mat_mul(rt::Node* node) { std::size_t acols = node->len1; std::size_t arows = node->len2; std::size_t bcols = node->len3; std::size_t block_size = 32; dim3 threads_per_block (block_size, block_size); std::size_t nb_blocks_x = (acols + block_size - 1) / block_size; std::size_t nb_blocks_y = (bcols + block_size - 1) / block_size; dim3 blocks_per_grid (nb_blocks_x, nb_blocks_y); tmat_mat_mul<<<blocks_per_grid, threads_per_block>>>(node->in1, node->in2, node->out1, acols, arows, bcols); } void kernel_mat_tmat_mul(rt::Node* node) { std::size_t arows = node->len1; std::size_t acols = node->len2; std::size_t brows = node->len3; std::size_t block_size = 32; dim3 threads_per_block (block_size, block_size); std::size_t nb_blocks_x = (arows + block_size - 1) / block_size; std::size_t nb_blocks_y = (brows + block_size - 1) / block_size; dim3 blocks_per_grid (nb_blocks_x, nb_blocks_y); mat_tmat_mul<<<blocks_per_grid, threads_per_block>>>(node->in1, node->in2, node->out1, arows, acols, brows); } }
10,808
#include "includes.h" __global__ void kernel_cudaPrepareProjectionIndexes(char *d_v_is_projection, int *d_nearest_neighbour_indexes, int number_of_points) { int ind=blockIdx.x*blockDim.x+threadIdx.x; if(ind<number_of_points) { if(d_v_is_projection[ind] == 0) { d_nearest_neighbour_indexes[ind] = -1; }else { d_nearest_neighbour_indexes[ind] = ind; } } }
10,809
/* @Author: 3sne ( Mukur Panchani ) @FileName: q2MatrixSummer.cu @Task: CUDA program compute sums of two matrices, using different parallelism techniques. */ #include <stdio.h> #include <stdlib.h> #include <cuda_runtime.h> __global__ void addMatRowThreads(int *a, int *b, int *c, int m, int n) { int id = threadIdx.x; for ( int i = 0; i < n; i++ ) { int ind = id * n + i; c[ind] = a[ind] + b[ind]; } } __global__ void addMatColThreads(int *a, int *b, int *c, int m, int n) { int id = threadIdx.x; for ( int i = 0; i < m; i++ ) { int ind = i * n + id; c[ind] = a[ind] + b[ind]; } } __global__ void addMatElementThread(int *a, int *b, int *c, int m, int n) { int ci = threadIdx.x; int ri = threadIdx.y; int id = ri * m + ci; c[id] = a[id] + b[id]; } int main() { int *matA, *matB, *matC; int *da, *db, *dc; int m, n; printf("== Enter Dimension of Matrix A and B (m x n) ==\n"); printf("m >> "); scanf("%d", &m); printf("n >> "); scanf("%d", &n); matA = (int*)malloc(sizeof(int) * m * n); matB = (int*)malloc(sizeof(int) * m * n); matC = (int*)malloc(sizeof(int) * m * n); printf("== Matrix A Elements ==\n"); for(int i = 0; i < m * n; i++) { scanf("%d", &matA[i]); } printf("== Matrix B Elements ==\n"); for(int i = 0; i < m * n; i++) { scanf("%d", &matB[i]); } cudaMalloc((void **) &da, sizeof(int) * m * n); cudaMalloc((void **) &db, sizeof(int) * m * n); cudaMalloc((void **) &dc, sizeof(int) * m * n); cudaMemcpy(da, matA, sizeof(int) * m * n, cudaMemcpyHostToDevice); cudaMemcpy(db, matB, sizeof(int) * m * n, cudaMemcpyHostToDevice); printf("\nChoose a degree of parallelism >> \n"); printf("1. Thread handles row\n"); printf("2. Thread handles column\n"); printf("3. Thread handles element\nChoice >> "); int choice = 0; scanf("%d", &choice); dim3 block_conf (n, m); switch(choice) { case 1://Part A: 1 Thread handles 1 row >> printf("Chose: Thread handles row\n"); addMatRowThreads<<<1,m>>>(da, db, dc, m, n); break; case 2://Part B: 1 Thread handles 1 column >> printf("Chose: Thread handles column\n"); addMatColThreads<<<1,n>>>(da, db, dc, m, n); break; case 3://Part C: 1 Thread handles 1 element >> printf("Chose: Thread handles element\n"); addMatElementThread<<<1, block_conf>>>(da, db, dc, m, n); break; default: printf("Bad Option, exiting ...\n"); exit(EXIT_FAILURE); break; } cudaMemcpy(matC, dc, sizeof(int) * m * n, cudaMemcpyDeviceToHost); printf("== Matrix C Elements (computed by choice %d)==\n", choice); for ( int i = 0; i < m; i++ ) { for ( int j = 0; j < n; j++ ) { printf("%d ", matC[i * n + j]); } printf("\n"); } cudaFree(da); cudaFree(db); cudaFree(dc); free(matA); free(matB); free(matC); return 0; }
10,810
#include<stdlib.h> #include<stdio.h> const int N = 32; __global__ void mul(int* A, int* B, int* C){ int col = blockIdx.x * blockDim.x + threadIdx.x; int lig = blockIdx.y * blockDim.y + threadIdx.y; int index = lig * N + col; if (col < N && lig < N){ int inter = 0; for (int i = 0; i<N; ++i){ inter += A[lig*N + i] * B[i*N + col]; } C[index] = inter; } } __host__ void affiche(int *A, int z){ for( int i=0;i<z;i++){ for (int j=0; j<z;j++){ printf(" %d ",A[i*z+j]); } printf("\n"); } } int main(void){ int *A, *B, *C, *da, *db, *dc; int size = N * N * sizeof(int); cudaMalloc((void **) & da, size); cudaMalloc((void **) & db, size); cudaMalloc((void **) & dc, size); A = (int *)malloc(size); B = (int *)malloc(size); C = (int *)malloc(size); for (int i=0; i<N * N; ++i){ A[i]=1; B[i]=1; } cudaMemcpy(da, A, size, cudaMemcpyHostToDevice); cudaMemcpy(db, B, size, cudaMemcpyHostToDevice); //dim3 dimBlock(N, N); dim3 dimGrid(N, N); mul<<<dimGrid, dimGrid>>>(da, db, dc); cudaMemcpy(C, dc, size, cudaMemcpyDeviceToHost); affiche(C, N); free(A); free(B); free(C); cudaFree(da); cudaFree(db); cudaFree(dc); return 0; }
10,811
#include <stdio.h> __global__ void array_sum(double *x, double *y) { int i=0; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; x[i] -= y[i]; x[i] += y[i]; }
10,812
#include "includes.h" __global__ void reduce_v0(float* in,float* out, int n){ int tx = threadIdx.x; int bx = blockIdx.x; int BX = blockDim.x; //same as THEAD_MAX int i = bx*BX+tx; __shared__ float S[THEAD_MAX]; S[tx] = i < n ? in[i] : 0; __syncthreads(); for(int s=1; s<BX ;s*=2){ if(tx%(2*s)==0) S[tx] += S[tx+s]; __syncthreads(); } if(tx==0) out[bx] = S[0]; }
10,813
/* * * Carlos Roman Rivera - A01700820 * * Programming Languages - Cuda Quiz * */ #include <stdio.h> #define N 9 #define K N/3 #define ThreadsPerBlock K #define NumBlocks K __global__ void compress(float *mat, int n, float *comp, int k){ int row = threadIdx.y + blockIdx.y * blockDim.y; int col = threadIdx.x + blockIdx.x * blockDim.x; if (row < k && col < k) { comp[col + row * k] = 0; for (int i_row = 0 ; i_row < k ; i_row++) { for (int j_col = 0 ; j_col < k ; j_col++) { comp[col + row * k] += mat[(col + j_col) + (row + i_row) * n]; } } } } void print_mat(float *mat, int n){ for (int i = 0; i < n; i++){ for (int j = 0; j < n; j++){ printf("%.1f\t", mat[i*n+j]); } printf("\n"); } printf("\n"); } void fill_mat(float *mat, int n){ int c = 0; for (int i = 0; i < n; i++){ for (int j = 0; j < n; j++){ mat[i*n+j] = c++; } } } int main(){ float *h_compress, *h_matrix; float *d_compress, *d_matrix; h_compress = (float *)malloc(sizeof(float) * K * K); h_matrix = (float *)malloc(sizeof(float) * N * N); fill_mat(h_matrix, N); fill_mat(h_compress, K); printf("Input matrix:\n"); print_mat(h_matrix, N); cudaMemcpy(d_matrix, h_matrix, sizeof(float) * N * N, cudaMemcpyHostToDevice); cudaMemcpy(d_compress, h_compress, sizeof(float) * K * K, cudaMemcpyHostToDevice); dim3 Blocks(K,K); dim3 Threads(K,K); compress<<<Blocks, Threads>>>(d_matrix, N, d_compress, K); cudaMemcpy(h_compress, d_compress, sizeof(float) * K * K, cudaMemcpyDeviceToHost); printf("Compressed matrix:\n"); print_mat(h_compress, K); free(h_matrix); free(h_compress); cudaFree(d_matrix); cudaFree(d_compress); }
10,814
#include "includes.h" __device__ double min2(double a, double b) { if (b < a) return b; return a; } __device__ double max2(double a, double b) { if (b > a) return b; return a; } __global__ void ConditionCFLKernel2D1 (double *Rsup, double *Rinf, double *Rmed, int nsec, int nrad, double *Vresidual, double *Vtheta, double *Vmoy, int FastTransport, double *SoundSpeed, double *Vrad, double *DT2D) { int j = threadIdx.x + blockDim.x*blockIdx.x; int i = threadIdx.y + blockDim.y*blockIdx.y; double dxrad, dxtheta, invdt1, invdt2, invdt3, invdt4, dvr, dvt, dt; if (i > 0 && i<nrad && j<nsec){ dxrad = Rsup[i]-Rinf[i]; dxtheta = Rmed[i]*2.0*PI/(double)nsec; if (FastTransport) Vresidual[i*nsec + j] = Vtheta[i*nsec + j]-Vmoy[i]; /* Fargo algorithm */ else Vresidual[i*nsec + j] = Vtheta[i*nsec + j]; /* Standard algorithm */ //Vresidual[i*nsec + nsec] = Vresidual[i*nsec]; invdt1 = SoundSpeed[i*nsec + j]/(min2(dxrad,dxtheta)); invdt2 = fabs(Vrad[i*nsec + j])/dxrad; invdt3 = fabs(Vresidual[i*nsec + j])/dxtheta; dvr = Vrad[(i+1)*nsec + j]-Vrad[i*nsec + j]; dvt = Vtheta[i*nsec + (j+1)%nsec]-Vtheta[i*nsec + j]; if (dvr >= 0.0) dvr = 1e-10; else dvr = -dvr; if (dvt >= 0.0) dvt = 1e-10; else dvt = -dvt; invdt4 = max2(dvr/dxrad, dvt/dxtheta); invdt4*= 4.0*CVNR*CVNR; dt = CFLSECURITY/sqrt(invdt1*invdt1+invdt2*invdt2+invdt3*invdt3+invdt4*invdt4); DT2D[i*nsec + j] = dt; // array nrad*nsec size dt } }
10,815
#include "scan.h" __global__ void scan_v1_kernel(float *d_output, float *d_input, int length) { int idx = blockDim.x * blockIdx.x + threadIdx.x; float element = 0.f; for (int offset = 0; offset < length; offset++) { if (idx - offset >= 0) element += d_input[idx - offset]; } d_output[idx] = element; } void scan_v1(float *d_output, float *d_input, int length) { dim3 dimBlock(BLOCK_DIM); dim3 dimGrid((length + BLOCK_DIM - 1) / BLOCK_DIM); scan_v1_kernel<<<dimGrid, dimBlock>>>(d_output, d_input, length); }
10,816
#include <stdio.h> struct City { int x, y; char name; }; inline void GPUassert(cudaError_t code, char * file, int line, bool Abort=true) { if (code != 0) { fprintf(stderr, "GPUassert: %s %s %d\n", cudaGetErrorString(code),file,line); if (Abort) exit(code); } } #define GPUerrchk(ans) { GPUassert((ans), __FILE__, __LINE__); } #define NUMCITIES 9 #define CITYSIZE sizeof(struct City) // #define CITYSIZE 1 __host__ __device__ void printCity(struct City c){ printf("[x=%i, y=%i]\n", c.x, c.y); } __host__ __device__ void swap(struct City *a, int x, int y){ struct City temp; temp = a[x]; a[x] = a[y]; a[y] = temp; } __device__ double get_distance(struct City c1, struct City c2){ double x = double(c1.x - c2.x); double y = double(c1.y - c2.y); return sqrt((x*x) + (y*y)); } __device__ long total_distance(struct City *path){ long distance = 0; for(int i = 0; i < NUMCITIES - 1; i++){ distance += get_distance(path[i], path[i+1]); } return (long)distance; } __device__ void shortest_path(struct City *path){ int best_path_idx = 0; for(int i = 0; i < NUMCITIES - 1; i++){ printCity(path[i]); } } __device__ void print_path(struct City *path){ for(int i = 0; i < NUMCITIES; i++){ printf("%c>", path[i].name); } printf("\n"); } __device__ void format_path(struct City *path, char *str){ for(int i = 0; i < NUMCITIES; i++){ *str = path[i].name; str++; *str = '>'; str++; } str--; *str = 0; } __device__ void permutations_kernel(struct City *a, char **paths, double *distances, int i, int length, int tid, int *count) { if (length == i){ long distance = total_distance(a); //format_path(a, paths[count[0]]); count[0] = count[0] + 1; } else { for (int j = i; j < length; j++) { swap(a, i, j); // CUDA // permutations(a, i+1, length, tid, count); permutations_kernel(a, paths, distances, i+1, length, tid, count); swap(a, i, j); } } } __global__ void permute_kernel(struct City *dev_cities, char **paths, double *distances, int size) { int tid = threadIdx.x + blockIdx.x * blockDim.x; int count[1]; count[0] = 0; struct City local_array[NUMCITIES]; for (int i=0; i<size; i++){ local_array[i] = dev_cities[i]; } //swap(local_array + threadIdx.x, local_array); //swap(local_array, threadIdx.x, 0); permutations_kernel(local_array, paths, distances, 0, NUMCITIES, tid, count); } long factorial(int i) { long result = 1; while(i > 0) { result *= i; i--; } return result; } int main(){ struct City host_cities[NUMCITIES]; for(int c = 0; c < NUMCITIES; c++){ host_cities[c].name = 'A' + c; host_cities[c].x = rand() % 20 + 5; host_cities[c].y = rand() % 20 + 5; } //char host_paths [ factorial(NUMCITIES) ][ NUMCITIES*NUMCITIES ]; char host_paths [0][0]; char **device_paths; //double host_distances[ factorial(NUMCITIES) ]; double host_distances[0]; double *device_distances; float time; cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); struct City *device_cities; cudaMalloc((void**) &device_cities, sizeof(host_cities)); //cudaMalloc((void**) &device_paths, sizeof(host_paths)); cudaMalloc((void**) &device_paths, sizeof(char) * NUMCITIES * NUMCITIES * factorial(NUMCITIES)); cudaMalloc((void**) &device_distances, sizeof(host_distances)); GPUerrchk(cudaMemcpy(device_distances, host_distances, sizeof(host_distances), cudaMemcpyHostToDevice)); GPUerrchk(cudaMemcpy(device_cities, host_cities, sizeof(host_cities), cudaMemcpyHostToDevice)); //GPUerrchk(cudaMemcpy(device_paths, host_paths, sizeof(host_paths), cudaMemcpyHostToDevice)); GPUerrchk(cudaMemcpy(device_paths, host_paths, sizeof(char) * NUMCITIES * NUMCITIES * factorial(NUMCITIES), cudaMemcpyHostToDevice)); cudaEventRecord(start,0); permute_kernel<<<1, NUMCITIES>>>(device_cities, device_paths, device_distances, NUMCITIES); cudaEventRecord(stop,0); GPUerrchk(cudaPeekAtLastError()); GPUerrchk(cudaDeviceSynchronize()); cudaEventElapsedTime( &time, start, stop ); printf("\nTiempo de Ejecucion: %f mSeg\n\n", time); cudaEventDestroy( start ); cudaEventDestroy( stop ); cudaFree(device_cities); return 0; }
10,817
#include<stdio.h> #include<stdlib.h> #include<cuda.h> #include<unistd.h> #include<time.h> __global__ void multiply(int *scval, int *sccol, int *vec, int *result, int *cols, int *rowptr) { int tid=threadIdx.x+blockIdx.x*blockDim.x; int sum=0; int i; int colidx=tid/2; // printf("\n tid=%d", tid); // printf("\ncols[%d]=%d",colidx,cols[colidx]); for(i=0;i<cols[colidx];i++) { sum += vec[sccol[rowptr[tid]+i]]*scval[rowptr[tid]+i]; if(tid==1) { printf("\n HAHAHAHA"); printf("\nrowptr[%d]=%d",tid, rowptr[tid]); // printf("\ntid=%d, %d*%d=%d",tid, scval[rowptr[tid]+i],vec[sccol[rowptr[tid]+i]],vec[sccol[rowptr[tid]+i]]*scval[rowptr[tid]+i]); // printf("\nsccol[%d]=%d",rowptr[tid]+i, sccol[rowptr[tid]+i]); // printf("\nvec[%d]=%d",sccol[rowptr[tid]+i], vec[sccol[rowptr[tid]+i]]); printf("\nSum=%d", sum); } printf("\n"); } // __syncthreads(); result[tid]=sum; } __global__ void printmatscreen(int* mat, int N) { int i; for (i=0;i<N;i++) { printf("%d ",mat[i]); } printf("\n"); } int** Make2DIntArray(int arraySizeX, int arraySizeY) { int** theArray; theArray = (int**) malloc(arraySizeX*sizeof(int*)); int i; for (i = 0; i < arraySizeX; i++) theArray[i] = (int*) malloc(arraySizeY*sizeof(int)); int j; for (i=0;i<arraySizeX;i++) { for (j=0;j<arraySizeY;j++) { theArray[i][j]=0; } } return theArray; } int** Make2DVariableIntArray(int rows, int blocks, int blocksize, int* columns) { int** theArray; theArray = (int**) malloc(rows*sizeof(int*)); int i, j, k; for (i = 0; i < blocks; i++) { k=columns[i]; for (j=0; j < blocksize; j++) { theArray[i*blocksize+j] = (int*) malloc(k*sizeof(int)); } } //int j; for (i=0;i<blocks;i++) { for (j=0;j<blocksize;j++) { for (k=0;k<columns[i];k++) { theArray[i*blocksize+j][k]=0; } } } return theArray; } int** Changeto2DVariableIntArray(int** theArray,int rows, int blocks, int blocksize, int* columns) { int** NewArray=Make2DVariableIntArray(rows,blocks,blocksize,columns); int i, j, k; for (i=0;i<blocks;i++) { for (j=0;j<blocksize;j++) { for (k=0;k<columns[i];k++) { NewArray[i*blocksize+j][k]=theArray[i*blocksize+j][k]; } } } printf("changed to multiple matrixes"); return NewArray; } void init_zeros(int** matrix, int N) { int i,j; for (i=0;i<N;i++) { for (j=0;j<N;j++) { matrix[i][j]=0; } } } void printmat(int** matrix, int N, int Nj) { int i,j; for (i=0;i<N;i++) { printf("\n"); for (j=0;j<N;j++) { printf("%d ",matrix[i][j]); } } printf("\n"); } void printtofile(int** matrix, int K, char* filename) { /* Prints original 2D matrices to file */ FILE *fp; fp=fopen(filename,"wt"); int i,j; for (i=0;i<K;i++) { fprintf(fp, "\n"); for (j=0;j<K;j++) { fprintf(fp, "%d\t", matrix[i][j]); } } } void printtofile1D(int* matrix, int K, char* filename) { /* Prints resultant matrix to a file */ FILE *fp; fp=fopen(filename,"wt"); int i,j; int counters=0; for (i=0;i<K;i++) { fprintf(fp, "\n"); for (j=0;j<K;j++) { fprintf(fp, "%d \t", matrix[counters]); counters++; } } } int* Make1DIntArray(int arraySizeX) { int* theArray; theArray = (int*)malloc(arraySizeX*sizeof(int)); int i; for (i=0;i<arraySizeX;i++) { theArray[i]=0; } return theArray; } void freese(int sizeX, int sizeY, double** ptr) { int i; for (i=0;i<sizeX;i++) free(ptr[i]); free(ptr); } int main() { int N=6; // const int Dsize=1000; FILE *arr, *vec; int i,j,maxrowwidth=0,tint=0; int** a=Make2DIntArray(N,N); // int* val=Make1DIntArray(Dsize); // int* col=Make1DIntArray(Dsize); // int* row=Make1DIntArray(Dsize); int* result=Make1DIntArray(N); int* vecX=Make1DIntArray(N); int** scval=Make2DIntArray(N,N); //sell c value int** sccol=Make2DIntArray(N,N); //sell c col int* rowwidth=Make1DIntArray(N); //number of elements in each row int* temp=Make1DIntArray(N); int *dev_vec, *dev_scval, *dev_result, *dev_sccol, *dev_cols, *dev_rowptr; int* rows=Make1DIntArray(N); int* resultsordered=Make1DIntArray(N); //int val[10],col[10],row[10]; arr=fopen("mat.txt","r"); int k=0; // struct timeval start, end; // gettimeofday(&start, NULL); //Reading the vector vec=fopen("vec.txt","r"); for (i=0;i<N;i++) { fscanf(vec,"%d",&vecX[i]); rows[i]=i; } printf("\n Vector is:\n"); for (i=0;i<N;i++) { printf("%d\n",vecX[i]); } //Reading the matrix for(i=0;i<N;i++) { printf("\n"); for(j=0;j<N;j++) { fscanf(arr,"%d",&a[i][j]); printf("%d ",a[i][j]); } } printf("\n"); //row[i]=k; //printf("\n k = %d\n ", k); //sleep(10); //gettimeofday(&end, NULL); //double delta = ((end.tv_sec - start.tv_sec) * 1000000u + // end.tv_usec - start.tv_usec) / 1.e6; // printf("\nTime spent=%f\n", delta); for(i=0;i<N;i++) { for(j=0;j<N;j++) { if(a[i][j]!=0) { scval[i][k]=a[i][j]; //printf("\n scval[%d][%d]=%d",i,k,scval[i][k]); //sleep(1); sccol[i][k]=j; //printf("\n sccol[%d][%d]=%d",i,k,sccol[i][k]); rowwidth[i]=k+1; if(rowwidth[i]>maxrowwidth) { //printf("\nrow[%d] width=%d\n",i,maxrowwidth); maxrowwidth=rowwidth[i]; }k++; } } //printf("\nRow width %d = %d", i, rowwidth[i]); k=0; } for(i=0;i<N-1;i++) { for(j=0;j<N-1;j++) { if(rowwidth[j]<rowwidth[j+1]) { /*printf("\nrow %d width=%d",j,rowwidth[j]); printf("\nscval[%d]=",j); for(k=0;k<rowwidth[j];k++) { printf("%d ", scval[j][k]); } printf("\nscval[%d]=",j+1); for(k=0;k<rowwidth[j+1];k++) { printf("%d ", scval[j+1][k]); } */ temp=scval[j]; scval[j]=scval[j+1]; scval[j+1]=temp; temp=sccol[j]; sccol[j]=sccol[j+1]; sccol[j+1]=temp; tint=rowwidth[j]; rowwidth[j]=rowwidth[j+1]; rowwidth[j+1]=tint; tint=rows[j]; rows[j]=rows[j+1]; rows[j+1]=tint; } } } for(i=0;i<N;i++) { if(scval[i][0]==0) { break; } } if(i%2==1) N=i+1; else N=i; printf("\nmaxrowwidth=%d\n",maxrowwidth); printmat(scval,N,N); printtofile(scval,N,"scval.txt"); printtofile(sccol,N,"sccol.txt"); printf("\n Vector is:\n"); for (i=0;i<N;i++) { printf("%d\n",vecX[i]); } //printmatscreen<<<1,1>>>(dev_b,N); // NEED TO FIGURE OUT A WAY TO POPULATE cols SO AS TO HAVE varmat CREATED PROPERLY. SYSTEM CRASHES OTHERWISE int c=2; int* cols=Make1DIntArray(N/c); j=0; int colsum=0; for(i=0;i<N;i=i+c) { cols[j]=rowwidth[i]; printf("\n cols[%d]=%d",j,cols[j]); colsum+=cols[j]; j++; } int** varscval=Changeto2DVariableIntArray(scval,N,N/c,c,cols); int** varsccol=Changeto2DVariableIntArray(sccol,N,N/c,c,cols); for (i=0;i<N/c;i++) { for(j=0;j<c;j++) { printf("\n"); for (k=0;k<cols[i];k++) { printf("%d ",varscval[i*c+j][k]); printf("%d \t",varsccol[i*c+j][k]); } } } int varsize=colsum*c; //flattening scval and sccol int counters=0; int* scval_flat=Make1DIntArray(varsize); int* sccol_flat=Make1DIntArray(varsize); int* rowptr=Make1DIntArray(N+1); rowptr[0]=0; int countcols=0; int z=0; for (i=0;i<N/c;i++) { for(j=0;j<c;j++) { printf("\n"); countcols=0; for (k=0;k<cols[i];k++) { scval_flat[counters]=varscval[i*c+j][k]; if (scval_flat[counters]!=0) { sccol_flat[counters]=varsccol[i*c+j][k]; } counters=counters+1; countcols=countcols+1; } rowptr[z+1]=rowptr[z]+countcols; z=z+1; } } printf("\n rowptrs:\n"); for(i=0;i<N;i++) printf("%d ",rowptr[i]); printf("\n"); cudaMalloc((void**)&dev_vec, sizeof(int)*N); cudaMalloc((void**)&dev_scval, sizeof(int)*varsize); cudaMalloc((void**)&dev_result, sizeof(int)*N); cudaMalloc((void**)&dev_sccol, sizeof(int)*varsize); cudaMalloc((void**)&dev_cols, sizeof(int)*(N/c)); cudaMalloc((void**)&dev_rowptr, sizeof(int)*N); cudaMemcpy(dev_vec, vecX, sizeof(int)*N, cudaMemcpyHostToDevice); cudaMemcpy(dev_scval, scval_flat, sizeof(int)*varsize, cudaMemcpyHostToDevice); cudaMemcpy(dev_result, result, sizeof(int)*N, cudaMemcpyHostToDevice); cudaMemcpy(dev_sccol, sccol_flat, sizeof(int)*varsize, cudaMemcpyHostToDevice); cudaMemcpy(dev_cols, cols, sizeof(int)*(N/c), cudaMemcpyHostToDevice); cudaMemcpy(dev_rowptr, rowptr, sizeof(int)*N, cudaMemcpyHostToDevice); printf("\nrowwidth:\n"); for (i=0;i<N;i++) printf("%d ",cols[i]); printf("\n"); printmatscreen<<<1,1>>>(dev_scval,varsize); printmatscreen<<<1,1>>>(dev_sccol,varsize); printmatscreen<<<1,1>>>(dev_cols,(N/c)); printmatscreen<<<1,1>>>(dev_rowptr,N); //sleep(5); multiply<<<N/c,c>>>(dev_scval, dev_sccol, dev_vec, dev_result, dev_cols, dev_rowptr); //__syncthreads(); cudaMemcpy(result, dev_result, sizeof(int)*N, cudaMemcpyDeviceToHost); for (i=0;i<N;i++) { printf("\nrow[%d]=%d",i,rows[i]); resultsordered[rows[i]]=result[i]; } for (i=0;i<N;i++) { printf("\n%d",resultsordered[i]); } cudaFree(dev_vec); cudaFree(dev_scval); cudaFree(dev_result); cudaFree(dev_sccol); cudaFree(dev_cols); return 0; }
10,818
#include "includes.h" /******************************************************************************* * *******************************************************************************/ /************************************************************************* /*************************************************************************/ /*************************************************************************/ __global__ void drawGray(unsigned char* optr, const float* outSrc) { // map from threadIdx/BlockIdx to pixel position int x = threadIdx.x + blockIdx.x * blockDim.x; int y = threadIdx.y + blockIdx.y * blockDim.y; int offset = x + y * blockDim.x * gridDim.x; float val = outSrc[offset]; val = (val / 50.0) + 0.5; //get {-25 to 25} range into {0 to 1} range if (val < 0) val = 0; if (val > 1) val = 1; optr[offset * 4 + 0] = 255 * val; // red optr[offset * 4 + 1] = 255 * val; // green optr[offset * 4 + 2] = 255 * val; // blue optr[offset * 4 + 3] = 255; // alpha (opacity) }
10,819
#include "includes.h" __global__ void cumulativeOffspringToAncestorKernel(const int* cumulativeOffspring, int* ancestor, int numParticles) { int idx = blockIdx.x * blockDim.x + threadIdx.x; if(idx >= numParticles || idx < 0) return; int start = idx == 0 ? 0 : cumulativeOffspring[idx - 1]; int numCurrentOffspring = cumulativeOffspring[idx] - start; for(int j = 0; j < numCurrentOffspring; j++) ancestor[start+j] = idx; }
10,820
#include "includes.h" __global__ void UseForceKernel( float *force, float forceFactor, float *pointsCoordinates, int maxCells ) { int threadId = blockDim.x*blockIdx.y*gridDim.x //rows preceeding current row in grid + blockDim.x*blockIdx.x //blocks preceeding current block + threadIdx.x; if(threadId < maxCells * 3) { pointsCoordinates[threadId] += forceFactor * force[threadId]; } }
10,821
#include "includes.h" // wrapper pour une option d'achat __global__ void mc_kernel_call(float * d_s, float T, float K, float S0, float sigma, float mu, float r, float dt, float * d_normals, unsigned N_STEPS, unsigned N_PATHS) { const unsigned tid = threadIdx.x; // id du thread dans le bloc const unsigned bid = blockIdx.x; // id du bloc const unsigned bsz = blockDim.x; // taille du bloc int s_idx = tid + bid * bsz; int n_idx = tid + bid * bsz; float s_curr = S0; if (s_idx < N_PATHS) { int n = 0; do { s_curr = s_curr + mu*s_curr*dt + sigma*s_curr*d_normals[n_idx]; n_idx++; n++; } while (n < N_STEPS); double payoff = (s_curr>K ? s_curr - K : 0.0); __syncthreads(); // on attend que tous les threads aient fini avant de passer à la prochaine simulation d_s[s_idx] = exp(-r*T) * payoff; } }
10,822
#include "includes.h" __global__ void MotionVec(float *new_image_dev, float *old_image_dev, uchar4 *Image_dev, int w, int h ) { const int ix = blockDim.x * blockIdx.x + threadIdx.x; const int iy = blockDim.y * blockIdx.y + threadIdx.y; const float x = (float)ix + 0.5f; const float y = (float)iy + 0.5f; float diff = 0; diff = old_image_dev[w*iy + ix] - new_image_dev[w*iy + ix]; diff *= diff; float threshold = 5000; if (diff > threshold) { Image_dev[w*iy + ix].x = 0; //B /* MODIFY CODE HERE*/ Image_dev[w*iy + ix].y = 0; //G /* MODIFY CODE HERE*/ Image_dev[w*iy + ix].z = 255; //R /* MODIFY CODE HERE*/ } }
10,823
#include "includes.h" __global__ void polynomial_expansion (float* poly, int degree, int n, float* array) { int index = blockIdx.x * blockDim.x + threadIdx.x; if( index < n ) { float out = 0.0; float xtothepowerof = 1.0; for ( int x = 0; x <= degree; ++x) { out += xtothepowerof * poly[x]; xtothepowerof *= array[index]; } array[index] = out; } }
10,824
#include "includes.h" __global__ void FillOnes(float *vec, int size) { int idx = blockIdx.x * blockDim.x + threadIdx.x; if (idx >= size) return; vec[idx] = 1.0f; }
10,825
#include "includes.h" __global__ void sumArraysZeroCopy(int *A, int *B, int *C, const int N) { int i = blockIdx.x * blockDim.x + threadIdx.x; if (i < N) C[i] = A[i] + B[i]; }
10,826
#include <stdio.h> __global__ void kernel(int* a) { int idx = blockIdx.x * blockDim.x + threadIdx.x; int answer = idx; int i = 0; if (answer != 0) { while (answer != 1) { if (answer % 2 == 0) { answer /= 2; } else { answer = 3 * answer + 1; } i++; } } a[idx] = i; } int main() { int dimx = 3907*256; int num_bytes = dimx * sizeof(int); int *d_a = 0, *h_a = 0; h_a = (int*)malloc(num_bytes); cudaMalloc((void**)&d_a, num_bytes); if (0==h_a || 0==d_a) { printf("can't allocate memory"); } cudaMemset(d_a, 0, num_bytes); cudaMemcpy(d_a, h_a, num_bytes, cudaMemcpyHostToDevice); cudaEvent_t start, stop; cudaEventCreate(&start); cudaEventCreate(&stop); cudaEventRecord(start, 0); kernel<<<3907, 256>>>(d_a); cudaEventRecord(stop, 0); cudaEventSynchronize(stop); float et; cudaEventElapsedTime(&et, start, stop); cudaEventDestroy(start); cudaEventDestroy(stop); printf("kernel execution time: %8.6fms\n", et); cudaMemcpy(h_a, d_a, num_bytes, cudaMemcpyDeviceToHost); int max = 0; for(int i=0; i<dimx; i++) { // printf("%d ", h_a[i]); if (h_a[i] > max) max = h_a[i]; } printf("max is %d\n", max); free(h_a); cudaFree(d_a); return 0; }
10,827
/* Erick Juarez CPSC 479 Sec 1 HOMEWORK 6 - 4/20/20 tested using nvcc - CUDA compiler driver release 9.0, V9.0.176 */ #include <stdio.h> #include <cuda.h> // device function: perform calculations to square a matrix __global__ void square(int *matrix, int *result, int matrix_size){ // solves 1 row per thread int row_id = threadIdx.x; for (int col_ix = 0; col_ix < matrix_size; col_ix++){ for (int row_ix = 0; row_ix < matrix_size; row_ix++){ result[row_id * matrix_size + col_ix] += matrix[row_id * matrix_size + row_ix] * matrix[row_ix * matrix_size + col_ix]; } } } // device function is used to initalize both matrices in parallel __global__ void init(int * matrix, int * result){ result[threadIdx.x] = 0; matrix[threadIdx.x] = threadIdx.x +1; } // Main program: initialization int main(int argc, char* argv[]){ const int RLEN = 32; // matrix of size RLEN x RLEN (matrix has to be a sqaure) const int MSIZE = RLEN * RLEN; // total number of elements in the matrix // Allocate memory on host for inpaut and output int *h_matrix, *h_result; // allocate memory on device for copy of input and output int *d_matrix, *d_result; int byte_size = MSIZE * sizeof(int); dim3 b_dim(MSIZE); cudaMalloc((void**) &d_matrix, byte_size); cudaMalloc((void**) &d_result, byte_size); // initialize device matrix and result matrix. Then square original matrix and place result in result matrix init<<<1, b_dim>>>(d_matrix, d_result); square<<<1, b_dim>>>(d_matrix, d_result, RLEN); // copy device ouput to host and cleanup h_matrix = (int *) malloc(byte_size); h_result = (int *) malloc(byte_size); cudaMemcpy(h_matrix, d_matrix, byte_size, cudaMemcpyDeviceToHost); cudaMemcpy(h_result, d_result, byte_size, cudaMemcpyDeviceToHost); cudaFree(d_matrix); cudaFree(d_result); // Print results printf("==========Original Matrix==========\n"); for (int i = 0; i < RLEN; i++){ for (int k = 0; k < RLEN; k++){ printf("[%d] ", h_matrix[RLEN * i + k]); } printf("\n"); } printf("==========Squared Matrix==========\n"); for (int i = 0; i < RLEN; i++){ for (int k = 0; k < RLEN; k++){ printf("[%d] ", h_result[RLEN * i + k]); } printf("\n"); } free(h_matrix); free(h_result); return 0; }
10,828
#include <iostream> #include <fstream> #include <math.h> #include <string> #include <ctime> #include <cstdlib> #include <cuda.h> #include <stdio.h> const int n_x = 784; //Rozmiar warstwy wejsciowej, jednoczesnie rozmiar badanych obrazow (28x28) const int n_h = 7; //Ilosc neuronow w warstwie ukrytej const int n_y = 1; //Ilosc neuronow na wyjsciu sieci float learning_rate = 0.075; //Predkosc uczenia const int train_samples = 209; //Ilosc obrazow przechodzacych przez siec w fazie treningu const int test_samples = 50; //Ilosc obrazow przechodzacych przez siec w fazie testu int num_samples = train_samples; //Aktualna ilosc obrazow przechodzacych przez siec const int iter_num = 100; //Ilosc przejsc zestawu treningowego przez siec const int print_freq = 20; //Czestotliwosc wyswietlania wartosci funkcji kosztu //Struktura przechowujaca poszczegolne gradienty struct grad_data { float dW1[n_h][n_x]; float dW2[n_y][n_h]; float db1; float db2; float dA0[n_x][train_samples]; float dA1[n_h][train_samples]; float dA2[n_y][train_samples]; float dZ1[n_h][train_samples]; float dZ2[n_y][train_samples]; }; //Struktura przechowujaca parametry oraz wyjscia poszczegolnych warstw struct param_data { float train_x[n_x][train_samples]; float test_x[n_x][test_samples]; float train_y[n_y][train_samples]; float test_y[n_y][test_samples]; float W1[n_h][n_x]; float W2[n_y][n_h]; float b1; float b2; float A1[n_h][train_samples]; float A2[n_y][train_samples]; float Z1[n_h][train_samples]; float Z2[n_y][train_samples]; //Tablice pomocnicze float AT0[train_samples][n_x]; float AT1[train_samples][n_h]; float WT1[n_x][n_h]; float WT2[n_h][n_y]; }; //Funkcje obslugujace struktury void load_data(param_data&, grad_data&); void delete_data(param_data&, grad_data&); //Funkcja obliczajaca Z1 oraz dW1 __global__ void kernel(float *arr1,float *arr2, float *arr_out, float *b, int *size_1, int *size_2) { int x = blockIdx.x; int y = blockIdx.y; if((*size_1) == train_samples) { //Wtedy liczone jest dW1 arr_out[x + (*size_2)*y] = 0; for(int i = 0; i < (*size_1) ; i++) arr_out[x +(*size_2)*y] += (arr1[i + (*size_1)*y] * arr2[x + (*size_2)*i]) / train_samples; } else { //Wtedy liczone jest Z1 arr_out[x + (*size_2)*y] = *b; for(int i = 0; i < (*size_1) ; i++) arr_out[x +(*size_2)*y] += (arr1[i + (*size_1)*y] * arr2[x + (*size_2)*i]); } } //Funkcja aktywacji warstwy wyjsciowej (Sigmoid) void sigmoid(param_data &parameters) { int rows = 1; int cols = num_samples; for (int i = 0; i < rows; i++) { for (int j = 0; j < cols; j++) { parameters.A2[i][j] = 1 / (1 + exp(-parameters.Z2[i][j])); } } } //Funkcja aktywacji warstwy ukrytej (ReLu) void relu(param_data &parameters) { int rows = n_h; int cols = num_samples; for (int i = 0; i < rows; i++) { for (int j = 0; j < cols; j++) { if (parameters.Z1[i][j] <= 0) parameters.A1[i][j] = 0; else parameters.A1[i][j] = parameters.Z1[i][j]; } } } //Funkcja wyznaczajaca dZ1 za pomoca pochodnej z nieliniowej funkcji aktywacji (ReLu) oraz dA1 void relu_backward(param_data &parameters, grad_data &grads) { int rows = n_h; int cols = num_samples; for (int i = 0; i < rows; i++) { for (int j = 0; j < cols; j++) { if (parameters.Z1[i][j] <= 0) grads.dZ1[i][j] = 0; else grads.dZ1[i][j] = grads.dA1[i][j]; } } } //Funkcja wyznaczajaca dZ2 za pomocą pochodnej z nieliniowej funkcji aktywacji (Sigmoid) oraz dA2 void sigmoid_backward(param_data &parameters, grad_data &grads) { int rows = 1; int cols = num_samples; float s; for (int i = 0; i < rows; i++) { for (int j = 0; j < cols; j++) { s = 1 / (1 + exp(-parameters.Z2[i][j])); grads.dZ2[i][j] = grads.dA2[i][j] * s * (1 - s); } } } //Z1 = W1*X + b1 <-- liczone na GPU void linear_forward_relu(param_data &parameters) { int rows_w = n_h; int cols_w = n_x; int rows_z = n_h; int cols_z = num_samples; float *arr1,*arr2,*arr_out, *b; int *size_1, *size_2; cudaMalloc((void **)&arr1, rows_w*cols_w*sizeof(float)); cudaMalloc((void **)&arr2, cols_w*cols_z*sizeof(float)); cudaMalloc((void **)&arr_out, rows_z*cols_z*sizeof(float)); cudaMalloc((void **)&b, sizeof(float)); cudaMalloc((void **)&size_1, sizeof(int)); cudaMalloc((void **)&size_2, sizeof(int)); cudaMemcpy(arr1,parameters.W1, rows_w*cols_w*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(arr2,parameters.train_x, n_x*cols_z*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(b,&parameters.b1, sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(size_1,&n_x, sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(size_2,&num_samples, sizeof(int), cudaMemcpyHostToDevice); dim3 grid(cols_z,rows_z); kernel<<<grid,1>>>(arr1,arr2,arr_out,b,size_1,size_2); //cudaDeviceSynchronize(); cudaMemcpy(parameters.Z1,arr_out,rows_z*cols_z*sizeof(float),cudaMemcpyDeviceToHost); cudaFree(arr1); cudaFree(arr2); cudaFree(arr_out); cudaFree(b); cudaFree(size_1); cudaFree(size_2); // Alternatywny kod na CPU: // for (int i_z = 0; i_z < rows_z; i_z++) { // for (int j_z = 0; j_z < cols_z; j_z++) { // for (int j_w = 0; j_w < cols_w; j_w++) { // parameters.Z1[i_z][j_z] += parameters.W1[i_z][j_w] * parameters.train_x[j_w][j_z]; // } // parameters.Z1[i_z][j_z] += parameters.b1; // } // } } //Z2 = W2*A1 + b2 void linear_forward_sigm(param_data &parameters) { int rows_w = n_y; int cols_w = n_h; int rows_z = n_y; int cols_z = num_samples; for (int i_z = 0; i_z < rows_z; i_z++) { for (int j_z = 0; j_z < cols_z; j_z++) { for (int j_w = 0; j_w < cols_w; j_w++) { parameters.Z2[i_z][j_z] += parameters.W2[i_z][j_w] * parameters.A1[j_w][j_z]; } parameters.Z2[i_z][j_z] += parameters.b2; } } } //Funkcja wybierajaca tryb aktywacji void linear_activation_forward(param_data &parameters, std::string activation) { if (activation.compare("sigmoid") == 0) { linear_forward_sigm(parameters); sigmoid(parameters); } else { linear_forward_relu(parameters); relu(parameters); } } //Funkcja obliczajaca wartosc kosztu po pojedynczym przejsciu zestawu treningowego przez siec float compute_cost(param_data &parameters) { float cost = 0; float m = train_samples; for (int i = 0; i < m; i++) { cost += (-1 / m) * ( parameters.train_y[0][i] * log(parameters.A2[0][i]) + (1 - parameters.train_y[0][i]) * log(1 - parameters.A2[0][i])); } return cost; } //dW1 = (dZ1 * X.T) / train_samples <-- liczone na GPU //dA0 = (W1).T * dZ1 <-- nie musi być liczone //db1 = sum(dZ1) / train_samples <-- liczone na CPU void linear_backward_relu(param_data &parameters, grad_data &grads) { int rows_dw1 = n_h; int cols_dw1 = n_x; int rows_dz1 = n_h; int cols_dz1 = train_samples; int rows_da0 = n_x; int cols_da0 = train_samples; int cols_wt1 = n_h; for (int i = 0; i < rows_da0; i++) { for (int j = 0; j < cols_da0; j++) { parameters.AT0[j][i] = parameters.train_x[i][j]; } } for (int i = 0; i < rows_dw1; i++) { for (int j = 0; j < cols_dw1; j++) { parameters.WT1[j][i] = parameters.W1[i][j]; } } for (int i = 0; i < rows_dz1; i++) { for (int j = 0; j < cols_dz1; j++) { grads.db1 += grads.dZ1[i][j] / train_samples; } } float *arr1,*arr2,*arr_out, *b; int *size_1, *size_2; cudaMalloc((void **)&arr1, rows_dz1*cols_dz1*sizeof(float)); cudaMalloc((void **)&arr2, cols_da0*rows_da0*sizeof(float)); cudaMalloc((void **)&arr_out, rows_dw1*cols_dw1*sizeof(float)); cudaMalloc((void **)&b, sizeof(float)); cudaMalloc((void **)&size_1, sizeof(int)); cudaMalloc((void **)&size_2, sizeof(int)); cudaMemcpy(arr1, grads.dZ1, rows_dz1*cols_dz1*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(arr2, parameters.AT0, cols_da0*rows_da0*sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(b, &parameters.b1, sizeof(float), cudaMemcpyHostToDevice); cudaMemcpy(size_1, &train_samples, sizeof(int), cudaMemcpyHostToDevice); cudaMemcpy(size_2, &n_x, sizeof(int), cudaMemcpyHostToDevice); dim3 grid(cols_dw1,rows_dw1); kernel<<<grid,1>>>(arr1,arr2,arr_out,b,size_1,size_2); // cudaDeviceSynchronize(); cudaMemcpy(grads.dW1,arr_out,rows_dw1*cols_dw1*sizeof(float),cudaMemcpyDeviceToHost); cudaFree(arr1); cudaFree(arr2); cudaFree(arr_out); cudaFree(b); cudaFree(size_1); cudaFree(size_2); //Alternatywna wersja na CPU: // for (int i_dw1 = 0; i_dw1 < rows_dw1; i_dw1++) { // for (int j_dw1 = 0; j_dw1 < cols_dw1; j_dw1++) { // for (int j_dz1 = 0; j_dz1 < cols_dz1; j_dz1++) { // grads.dW1[i_dw1][j_dw1] += (grads.dZ1[i_dw1][j_dz1] * parameters.AT0[j_dz1][j_dw1]) / float(train_samples); // } // } // } // for (int i_da0 = 0; i_da0 < rows_da0; i_da0++) { // for (int j_da0 = 0; j_da0 < cols_da0; j_da0++) { // for (int j_wt1 = 0; j_wt1 < rows_dz1; j_wt1++) { // grads.dA0[i_da0][j_da0] += parameters.WT1[i_da0][j_wt1] * grads.dZ1[j_wt1][j_da0]; // } // } // } } //dW2 = dZ2 * (A1).T / train_samples //dA1 = (W2).T * dZ2 <-- Wszystko liczone na CPU //db2 = sum(dZ2) /train_samples void linear_backward_sigm(param_data &parameters, grad_data &grads) { int rows_dw2 = n_y; int cols_dw2 = n_h; int rows_dz2 = n_y; int cols_dz2 = train_samples; int rows_da1 = n_h; int cols_da1 = train_samples; int cols_wt2 = n_y; for (int i = 0; i < rows_da1; i++) { for (int j = 0; j < cols_da1; j++) { parameters.AT1[j][i] = parameters.A1[i][j]; } } for (int i = 0; i < rows_dw2; i++) { for (int j = 0; j < cols_dw2; j++) { parameters.WT2[j][i] = parameters.W2[i][j]; } } for (int i = 0; i < rows_dz2; i++) { for (int j = 0; j < cols_dz2; j++) { grads.db2 += grads.dZ2[i][j] / train_samples; } } for (int i_dw2 = 0; i_dw2 < rows_dw2; i_dw2++) { for (int j_dw2 = 0; j_dw2 < cols_dw2; j_dw2++) { for (int j_dz2 = 0; j_dz2 < cols_dz2; j_dz2++) { grads.dW2[i_dw2][j_dw2] += (grads.dZ2[i_dw2][j_dz2] * parameters.AT1[j_dz2][j_dw2]) / train_samples; } } } for (int i_da1 = 0; i_da1 < rows_da1; i_da1++) { for (int j_da1 = 0; j_da1 < cols_da1; j_da1++) { for (int j_wt2 = 0; j_wt2 < cols_wt2; j_wt2++) { grads.dA1[i_da1][j_da1] += parameters.WT2[i_da1][j_wt2] * grads.dZ2[j_wt2][j_da1]; } } } } //Funkcja wybierajaca tryb obliczania gradientow void linear_activation_backward(param_data &parameters, grad_data &grads, std::string activation) { if (activation.compare("relu") == 0) { relu_backward(parameters, grads); linear_backward_relu(parameters, grads); } else { sigmoid_backward(parameters, grads); linear_backward_sigm(parameters, grads); } } //Aktualizowanie parametrów sieci po jednej iteracji void update_parameters(param_data &parameters, grad_data &grads) { int rows_W1 = n_h; int cols_W1 = n_x; int rows_W2 = n_y; int cols_W2 = n_h; for (int i = 0; i < rows_W1; i++) { for (int j = 0; j < cols_W1; j++) { parameters.W1[i][j] -= learning_rate * grads.dW1[i][j]; } } for (int i = 0; i < rows_W2; i++) { for (int j = 0; j < cols_W2; j++) { parameters.W2[i][j] -= learning_rate * grads.dW2[i][j]; } } parameters.b1 -= learning_rate * grads.db1; parameters.b2 -= learning_rate * grads.db2; } //Glowna funkcja przechodzaca przez siec void two_layer_model(param_data &parameters, grad_data &grads) { float cost = 0; for (int i = 0; i < iter_num + 1; i++) { delete_data(parameters, grads); linear_activation_forward(parameters, "relu"); linear_activation_forward(parameters, "sigmoid"); cost = compute_cost(parameters); for (int j = 0; j < train_samples; j++) { grads.dA2[0][j] = -((parameters.train_y[0][j] / parameters.A2[0][j]) - ((1. - parameters.train_y[0][j]) / (1. - parameters.A2[0][j]))); } linear_activation_backward(parameters, grads, "sigmoid"); linear_activation_backward(parameters, grads, "relu"); update_parameters(parameters, grads); if (i % print_freq == 0) { std::cout << "Koszt po iteracji " << i << ": " << cost << "\n\n"; } } } //Sprawdzanie skutecznosci sieci void accuracy_check_train(param_data &parameters){ float accuracy = 0; linear_activation_forward(parameters, "relu"); linear_activation_forward(parameters, "sigmoid"); for (int j = 0; j < train_samples; j++) { if(parameters.A2[0][j] >= 0.5 && parameters.train_y[0][j] == 1) accuracy += 1; else if(parameters.A2[0][j] < 0.5 && parameters.train_y[0][j] == 0) accuracy += 1; } std::cout << "Accuracy (training): " << accuracy / train_samples << "\n"; num_samples = test_samples; accuracy = 0; linear_activation_forward(parameters, "relu"); linear_activation_forward(parameters, "sigmoid"); for (int j = 0; j < test_samples; j++) { if(parameters.A2[0][j] >= 0.5 && parameters.test_y[0][j] == 1) accuracy += 1; else if(parameters.A2[0][j] < 0.5 && parameters.test_y[0][j] == 0) accuracy += 1; } std::cout << "Accuracy (test): " << accuracy / test_samples << "\n"; } int main() { param_data parameters; grad_data grads; load_data(parameters, grads); two_layer_model(parameters, grads); accuracy_check_train(parameters); return 0; } void load_data(param_data &parameters, grad_data &grads) { srand(time(NULL)); std::cout << "Ladowanie zestawu treningowego i testowego.\n"; std::string path = "train_x.txt"; std::ifstream input(path.c_str()); for (int i = 0; i < n_x; i++) for (int j = 0; j < train_samples; j++) input >> parameters.train_x[i][j]; path = "test_x.txt"; std::ifstream input2(path.c_str()); for (int i = 0; i < n_x; i++) for (int j = 0; j < test_samples; j++) input2 >> parameters.test_x[i][j]; std::cout << "Wczytano zestaw treningowy i testowy.\n"; path = "train_y.txt"; std::ifstream input3(path.c_str()); for (int j = 0; j < train_samples; j++) input3 >> parameters.train_y[0][j]; path = "test_y.txt"; std::ifstream input4(path.c_str()); for (int j = 0; j < test_samples; j++) input4 >> parameters.test_y[0][j]; std::cout << "Wczytano zestaw klas.\n"; for (int i = 0; i < n_h; i++) for (int j = 0; j < n_x; j++) parameters.W1[i][j] = (rand()%10000 - 5000) * 0.000001; for (int i = 0; i < n_y; i++) for (int j = 0; j < n_h; j++) parameters.W2[i][j] = (rand()%10000 - 5000) * 0.000001; parameters.b1 = 0; parameters.b2 = 0; grads.db1 = 0; grads.db2 = 0; for (int i = 0; i < n_h; i++) for (int j = 0; j < train_samples; j++) parameters.Z1[i][j] = 0; for (int i = 0; i < n_y; i++) for (int j = 0; j < train_samples; j++) parameters.Z2[i][j] = 0; } void delete_data(param_data& parameters, grad_data& grads) { for (int i = 0; i < n_h; i++) for (int j = 0; j < train_samples; j++) parameters.Z1[i][j] = 0; for (int i = 0; i < n_y; i++) for (int j = 0; j < train_samples; j++) parameters.Z2[i][j] = 0; for (int i = 0; i < n_h; i++) for (int j = 0; j < train_samples; j++) parameters.A1[i][j] = 0; for (int i = 0; i < n_y; i++) for (int j = 0; j < train_samples; j++) parameters.A2[i][j] = 0; for (int i = 0; i < n_h; i++) for (int j = 0; j < n_x; j++) grads.dW1[i][j] = 0; for (int i = 0; i < n_y; i++) for (int j = 0; j < n_h; j++) grads.dW2[i][j] = 0; for (int i = 0; i < n_h; i++) for (int j = 0; j < train_samples; j++) grads.dA1[i][j] = 0; for (int i = 0; i < n_y; i++) for (int j = 0; j < train_samples; j++) grads.dA2[i][j] = 0; for (int i = 0; i < n_h; i++) for (int j = 0; j < train_samples; j++) grads.dZ1[i][j] = 0; for (int i = 0; i < n_y; i++) for (int j = 0; j < train_samples; j++) grads.dZ2[i][j] = 0; for (int i = 0; i < n_x; i++) for (int j = 0; j < train_samples; j++) grads.dA0[i][j] = 0; grads.db1 = 0; grads.db2 = 0; }
10,829
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> #include <stdlib.h> #define D 1000 #define re -0.5 #define im 0.45 #define scale 1.5 __device__ int julia( float x, float y){ float xj = scale * (float)(D/2 - x)/(D/2); float yj = scale * (float)(D/2 - y)/(D/2); for( int i=0; i<200; ++i){ float a = xj; float b = yj; xj = a*a-b*b+re; yj = 2*a*b+im; } if( xj*xj + yj*yj < 4) return 1; else return 0; } __global__ void generuj( int * picture ){ //sprawdz czy pkt nalezy do zbioru julii int i = blockIdx.x; int j = threadIdx.x; if( julia(i, j ) ) picture[ i * D + j ] = 1; else picture[ i * D + j ] = 0; } int main() { FILE *fp; if ((fp=fopen("obraz.pbm", "w"))==NULL) { printf ("Nie mogê otworzyæ pliku test.txt do zapisu!\n"); exit(1); } fprintf( fp, "P1\n%d %d\n", D, D); //deklarujê tablicê na karcie graficznej int * dev_obraz; cudaMalloc( &dev_obraz, sizeof(int) * D *D ); printf("udalo sie zaalokowac na karcie graficznej\n\n\n"); //generacja obrazu generuj <<< D, D >>> ( dev_obraz ); printf("funkcja zakonczyla dzialanie\n\n\n"); //skopiowanie obrazu z karty graficznej int ** obraz; obraz = (int **) malloc( sizeof(int*)*D ); for( int i=0; i<D; ++i){ obraz[i] = (int *) malloc( sizeof(int)*D ); cudaMemcpy( obraz[i], dev_obraz+i*D, sizeof(int)*D, cudaMemcpyDeviceToHost); } printf("skopiowano z karty graficznej\n\n\n"); //zapisanie obrazu w formie pbm (P1) for(int i=0; i<D; ++i){ for(int j=0; j<D; ++j) fprintf(fp, "%d", obraz[i][j]); fprintf(fp, "\n"); } fclose(fp); return 0; }
10,830
#include "includes.h" __global__ void rgb2gray(unsigned char* d_Pin, unsigned char* d_Pout, int width, int height) { int Row = blockIdx.y*blockDim.y + threadIdx.y; int Col = blockIdx.x*blockDim.x + threadIdx.x; if((Row < height) && (Col < width)) { d_Pout[Row*width+Col] = d_Pin[(Row*width+Col)*3+BLUE]*0.114 + d_Pin[(Row*width+Col)*3+GREEN]*0.587 + d_Pin[(Row*width+Col)*3+RED]*0.299; } }
10,831
#include "includes.h" #define N 1000 // size of vectors #define T 10000// number of threads per block __global__ void vecAdd(int *A, int *B, int *C) { int i = blockIdx.x*blockDim.x + threadIdx.x; C[i] = A[i] * 10 + B[i]; }
10,832
// fermi // Avoid mangling of function names extern "C" { __global__ void kmeans(int npoints, int nclusters, int nfeatures, const float* points, const float* clusters, int* pointsCluster); } __global__ void kmeans(int npoints, int nclusters, int nfeatures, const float* points, const float* clusters, int* pointsCluster) { const int ttpid = threadIdx.x; const int wtpid = threadIdx.y; const int bpid = blockIdx.x; int ind = 0; float min_dist = 3.0E+38; for (int cluster = 0; cluster < nclusters; cluster++) { float dist = 0; for (int feature = 0; feature < nfeatures; feature++) { float diff = points[1024 * bpid + (32 * wtpid + ttpid) + feature * npoints] - clusters[feature + cluster * nfeatures]; dist = dist + diff * diff; } if (dist < min_dist) { min_dist = dist; ind = cluster; } } pointsCluster[1024 * bpid + (32 * wtpid + ttpid)] = ind; }
10,833
// incrementArray.cu #include <stdio.h> #include <time.h> #include <sys/types.h> #include <unistd.h> void incrementArrayOnHost(float *a, int N) { int i; for (i=0; i < N; i++) a[i] = a[i]+1.f; } __global__ void incrementArrayOnDevice(float *a, int N) { int idx = blockIdx.x*blockDim.x + threadIdx.x; if (idx<N) a[idx] = a[idx]+1.f; } void printarray(float *a, int n) { int i = 0; for (i = 0; i < n; i++) printf("%f ", a[i]); printf("\n"); } // http://www.concentric.net/~Ttwang/tech/inthash.htm unsigned long mix(unsigned long a, unsigned long b, unsigned long c) { a=a-b; a=a-c; a=a^(c >> 13); b=b-c; b=b-a; b=b^(a << 8); c=c-a; c=c-b; c=c^(b >> 13); a=a-b; a=a-c; a=a^(c >> 12); b=b-c; b=b-a; b=b^(a << 16); c=c-a; c=c-b; c=c^(b >> 5); a=a-b; a=a-c; a=a^(c >> 3); b=b-c; b=b-a; b=b^(a << 10); c=c-a; c=c-b; c=c^(b >> 15); return c; } int main(int argc, char** argv) { // program args if (argc < 2) { printf("usage: incrementArrayRandom [max_size] [repetitions]\n"); return EXIT_SUCCESS; } int max_size = atoi(argv[1]); int repetitions = atoi(argv[2]); // randomize within same run srand(mix(clock(), time(NULL), getpid())); float *a_h, *b_h; // pointers to host memory float *a_d; // pointer to device memory int i, epoch = 0; int N = 0; int total_success = 0; for (epoch = 0; epoch < repetitions; epoch++) { N = rand() % max_size; size_t size = N*sizeof(float); // allocate arrays on host a_h = (float *)malloc(size); b_h = (float *)malloc(size); // allocate array on device cudaMalloc((void **) &a_d, size); // initialization of host data for (i=0; i<N; i++) a_h[i] = (float)i; // copy data from host to device cudaMemcpy(a_d, a_h, sizeof(float)*N, cudaMemcpyHostToDevice); // do calculation on host incrementArrayOnHost(a_h, N); // printarray(a_h, N); // do calculation on device: // Part 1 of 2. Compute execution configuration int blockSize = 4; int nBlocks = N/blockSize + (N%blockSize == 0?0:1); // Part 2 of 2. Call incrementArrayOnDevice kernel incrementArrayOnDevice <<< nBlocks, blockSize >>> (a_d, N); // Retrieve result from device and store in b_h cudaMemcpy(b_h, a_d, sizeof(float)*N, cudaMemcpyDeviceToHost); // check results // printarray(b_h, N); int success = 1; for (i=0; i<N; i++) { if (a_h[i] != b_h[i]) { success = 0; break; } } printf("epoch %d a[%d] = %s\n", epoch, N, (success == 1) ? "true" : "false"); if (success == 1) total_success += 1; } printf("\nsuccess rate: %f%%\n", total_success / ((float)repetitions) * 100.0); // cleanup free(a_h); free(b_h); cudaFree(a_d); return EXIT_SUCCESS; }
10,834
#include <iostream> #include <cmath> #include <chrono> __global__ void add(int n, float *x, float *y) { for (int i = 0; i < n; i++) { y[i]+= x[i]; } } int main() { int N = 1 << 20; float *x, *y; cudaMallocManaged(&x, N*sizeof(float)); cudaMallocManaged(&y, N*sizeof(float)); for (int i = 0; i < N; i++) { x[i] = 1.0f; y[i] = 2.0f; } auto start = std::chrono::high_resolution_clock::now(); add<<<1,1>>>(N, x, y); cudaDeviceSynchronize(); auto stop = std::chrono::high_resolution_clock::now(); auto duration = std::chrono::duration_cast<std::chrono::microseconds>(stop - start); float max_err = 0.0f; for (int i = 0; i < N; i++) { max_err = std::fmax(max_err, std::fabs(y[i]-3.0f)); } std::cout << duration.count() << " microseconds" << std::endl; std::cout << "Max error: " << max_err << std::endl; cudaFree(x); cudaFree(y); }
10,835
__global__ void tanh_kernel(float* Y, int batch_size, int num){ int idx = blockIdx.x * blockDim.x + threadIdx.x; float tp; // load to register and then do tanh tp = Y[idx]; Y[idx] = tanh(tp); } void launch_tanh(float* Y, int batch_size, int num){ dim3 gridSize((batch_size * num + 1023)/ 1024); dim3 blockSize(1024); tanh_kernel<<<gridSize, blockSize>>>(Y, batch_size, num); }
10,836
#include <cuda_runtime.h> #include <stdio.h> #include <sys/time.h> /* Globale Variablen stehen in allen Funktionen zur Verfuegung. Achtung: Das gilt *nicht* fuer Kernel-Funktionen! */ int Nx, Ny, N, npts; int *active; /* * Dieses Beispiel demonstriert die Addition zweier Arrays. * addArrayGPU soll die Arbeit ueber CUDA Threads auf der GPU verteilen. * addArrayHost iteriert sequentiell durch die Vektorelemente auf dem Host. */ // Macro zur Fehlerauswertung #define CHECK(call) \ { \ const cudaError_t error = call; \ if (error != cudaSuccess) \ { \ fprintf(stderr, "Error: %s:%d, ", __FILE__, __LINE__); \ fprintf(stderr, "code: %d, reason: %s\n", error, \ cudaGetErrorString(error)); \ exit(1); \ } \ } double seconds() { struct timeval tp; struct timezone tzp; int i = gettimeofday(&tp, &tzp); return ((double)tp.tv_sec + (double)tp.tv_usec * 1.e-6); } /* Fuer die Koordinaten: i = 0,1,...,Nx+1 j = 0,1,...,Ny+1 wird der fortlaufenden Index berechnet */ int coord2index(int i, int j) { return j*(Nx+2) + i; } /* Das Flag-Array der aktiven/inneren Punkte wird gesetzt. */ void active_pts() { int idx,i,j; active=(int*)malloc(npts*sizeof(int)); idx=0; // fortlaufender Index for (j=0; j<Ny+2; j++) { for (i=0; i<Nx+2; i++) { if ((i==0)||(j==0)||(i==Nx+1)||(j==Ny+1)) active[idx]=0; // Randpunkt else active[idx]=1; // innerer Punkt idx+=1; } } } /* Der Vektor p wird im Inneren auf zufaellige Werte gesetzt */ void random_vector(float *p) { int idx; for(idx = 0; idx < npts; idx++) { if (active[idx]) p[idx] = (float)(rand() & 0xFF ) / 10.0; } } /* Das Flag-Array der aktiven/inneren Punkte wird als 2D Gitter ausgegeben. */ void print_active() { int i,j,idx; printf("active points:\n"); idx=0; for (j=0; j<Ny+2; j++) { printf(" "); for (i=0; i<Nx+2; i++) { printf("%d ",active[idx]); idx+=1; } printf("\n"); } } /* Norm-Quadrat vom Vektor v. */ float norm_sqr(float *v) { int idx; float r=0.0; for (idx=0; idx<npts; idx++) { r+=v[idx]*v[idx]; } return r; } /* Der Vektor p wird als 2D Gitter fuer i,j<=16 ausgegeben. Es werden innere/aktive und, falls flag>0, auch die aeusseren Punkte ausgegeben. */ void print_vector(char *name, float *p, int flag) { int i,j,idx; float nrm; printf("%s = \n",name); idx=0; for (j=0; j<Ny+2; j++) { if (j>16) { printf(" ...\n"); break; } printf(" "); for (i=0; i<Nx+2; i++) { if ((i<16)&&((flag>0)||(active[idx]))) printf("%.2f ",p[idx]); if (i==16) printf("..."); idx+=1; } printf("\n"); } nrm=norm_sqr(p); printf("||%s|| = %.8f\n",name,sqrt(nrm)); } __global__ void laplace2d_GPU(float *w, float *v, const int N) { /* calculate the id */ int blockOffset = (blockIdx.x + blockIdx.y * gridDim.x) * blockDim.x * blockDim.y; int threadId = threadIdx.x + threadIdx.y * blockDim.x; int idx = blockOffset + threadId; int first_coord = (int) (idx % N); int second_coord = (int) (idx / N); if ( (first_coord > 0) && (first_coord < (N-1)) && (second_coord > 0) && (second_coord < (N-1)) ) { w[idx] = -4. * v[idx] + v[idx-1] + v[idx+1] + v[idx-N] + v[idx+N]; /* printf("%d, %f, %f\n", idx, w[idx], v[idx]); */ } } /* * laplace_2d(float *w, float *v) * * Calculates the product A*v for a vector v and writes the result into w. * A is a laplacian. * */ void laplace2d_CPU(float *w, float *v) { int i, j, n = Nx+2; int npts = n * n; /* two loops over every element of the unknowns */ for (i=0; i<npts; i++) { for (j=0; j<npts; j++) { /* skip the boundary */ if ( (i % n == 0) || ((int) (i / n) == 0) || (i % n == n-1) || ((int) (i / n) == n-1) ) { continue; } /* diagonal */ if (i == j) { w[i] += -4 * v[j]; } /* first two off diagonals */ if ( ((i == j+1) && (i % n != 0)) || ((i == j-1) && (j % n != 0)) ) { w[i] += 1 * v[j]; } /* other two off diagonals */ if ((i == j+n) || (i == j-n)) { w[i] += 1 * v[j]; } } } } /* * vector_addition(float *v, float *u, float *w) * * Adds two vectors u and v together and writes the result into w. * */ void vector_addition_CPU(float *u, float *v, float *w) { int i; for (i=0; i<npts; i++) { w[i] += u[i] + v[i]; } } __global__ void vector_addition_GPU(float *u, float *v, float *w) { /* calculate the id */ int blockOffset = (blockIdx.x + blockIdx.y * gridDim.x) * blockDim.x * blockDim.y; int threadId = threadIdx.x + threadIdx.y * blockDim.x; int idx = blockOffset + threadId; w[idx] = u[idx] + v[idx]; } /* * scale_vector(float prefactor, float *v, float *w) * * Scales a vector v by a prefactor and writes the result into w. * */ void scale_vector_CPU(float prefactor, float *v, float *w) { int i; for (i=0; i<npts; i++) { w[i] = prefactor * v[i]; } } __global__ void scale_vector_GPU(float prefactor, float *v, float *w) { /* calculate the id */ int blockOffset = (blockIdx.x + blockIdx.y * gridDim.x) * blockDim.x * blockDim.y; int threadId = threadIdx.x + threadIdx.y * blockDim.x; int idx = blockOffset + threadId; w[idx] = prefactor * v[idx]; } void laplaceOneLoop(FILE* laplace_speedup_file, const int N, int gridX, int gridY, int threadX, int threadY) { int nBytes; float *h_w, *h_v, *h_u; // Globale Variablen setzen: // Anzahl der Inneren Punkte in x- und y-Richtung Nx = N; Ny = N; // Gesamtanzahl der Gitterpunkte npts=(Nx+2)*(Ny+2); // Aktive Punkte - Array active_pts(); // Speicherbedarf pro Vektor in Byte nBytes=npts*sizeof(float); // Speicher für Vektoren allozieren h_w = (float *) malloc(npts * sizeof(float)); h_v = (float *) malloc(npts * sizeof(float)); h_u = (float *) malloc(npts * sizeof(float)); // auf Null setzen memset(h_w, 0, nBytes); memset(h_v, 0, nBytes); memset(h_u, 0, nBytes); // Zufaelliger Vektor random_vector(h_v); /* print_vector("v",h_v,1); */ // Device-Speicher allozieren mit cudaMalloc float *d_v, *d_w; CHECK(cudaMalloc((float**)&d_v, nBytes)); CHECK(cudaMalloc((float**)&d_w, nBytes)); // kopieren Host -> Device mit cudaMemcpy CHECK(cudaMemcpy(d_v, h_v, nBytes, cudaMemcpyHostToDevice)); CHECK(cudaMemcpy(d_w, h_w, nBytes, cudaMemcpyHostToDevice)); dim3 block(gridX, gridY); dim3 grid(threadX, threadY); double t_GPU_start = seconds(); laplace2d_GPU <<<block,grid>>> (d_w, d_v, N+2); cudaDeviceSynchronize(); double t_GPU_end = seconds(); double t_GPU = t_GPU_end - t_GPU_start; /* kopieren Device -> Host mit cudaMemcpy */ CHECK(cudaMemcpy(h_w, d_w, nBytes, cudaMemcpyDeviceToHost)); /* print_vector("w_GPU",h_w,1); */ // Device-Speicher freigeben CHECK(cudaFree(d_v)); CHECK(cudaFree(d_w)); double t_CPU_start = seconds(); laplace2d_CPU(h_u, h_v); double t_CPU_end = seconds(); double t_CPU = t_CPU_end - t_CPU_start; /* print_vector("w_CPU",h_w,1); */ fprintf( laplace_speedup_file, "%lf, %lf, %lf,\n", t_GPU, t_CPU, t_CPU/t_GPU ); free(active); free(h_w); free(h_v); } void vectorScaleOneLoop(FILE* vector_scale_speedup_file, const int N, int gridX, int gridY, int threadX, int threadY) { int nBytes; float *h_w, *h_v, *h_u; // Globale Variablen setzen: // Anzahl der Inneren Punkte in x- und y-Richtung Nx = N; Ny = N; // Gesamtanzahl der Gitterpunkte npts=(Nx+2)*(Ny+2); // Aktive Punkte - Array active_pts(); // Speicherbedarf pro Vektor in Byte nBytes=npts*sizeof(float); // Speicher für Vektoren allozieren h_w = (float *) malloc(npts * sizeof(float)); h_v = (float *) malloc(npts * sizeof(float)); h_u = (float *) malloc(npts * sizeof(float)); // auf Null setzen memset(h_w, 0, nBytes); memset(h_v, 0, nBytes); memset(h_u, 0, nBytes); // Zufaelliger Vektor random_vector(h_v); /* print_vector("v",h_v,1); */ // Device-Speicher allozieren mit cudaMalloc float *d_v, *d_w; CHECK(cudaMalloc((float**)&d_v, nBytes)); CHECK(cudaMalloc((float**)&d_w, nBytes)); // kopieren Host -> Device mit cudaMemcpy CHECK(cudaMemcpy(d_v, h_v, nBytes, cudaMemcpyHostToDevice)); CHECK(cudaMemcpy(d_w, h_w, nBytes, cudaMemcpyHostToDevice)); dim3 block(gridX, gridY); dim3 grid(threadX, threadY); double t_GPU_start = seconds(); laplace2d_GPU <<<block,grid>>> (d_w, d_v, N+2); cudaDeviceSynchronize(); double t_GPU_end = seconds(); double t_GPU = t_GPU_end - t_GPU_start; /* kopieren Device -> Host mit cudaMemcpy */ CHECK(cudaMemcpy(h_w, d_w, nBytes, cudaMemcpyDeviceToHost)); /* print_vector("w_GPU",h_w,1); */ // Device-Speicher freigeben CHECK(cudaFree(d_v)); CHECK(cudaFree(d_w)); double t_CPU_start = seconds(); laplace2d_CPU(h_u, h_v); double t_CPU_end = seconds(); double t_CPU = t_CPU_end - t_CPU_start; /* print_vector("w_CPU",h_w,1); */ fprintf( vector_scale_speedup_file, "%lf, %lf, %lf,\n", t_GPU, t_CPU, t_CPU/t_GPU ); free(active); free(h_w); free(h_v); } void vectorAddOneLoop(FILE* vector_add_speedup_file, const int N, int gridX, int gridY, int threadX, int threadY) { int nBytes; float *h_w, *h_v, *h_u; // Globale Variablen setzen: // Anzahl der Inneren Punkte in x- und y-Richtung Nx = N; Ny = N; // Gesamtanzahl der Gitterpunkte npts=(Nx+2)*(Ny+2); // Aktive Punkte - Array active_pts(); // Speicherbedarf pro Vektor in Byte nBytes=npts*sizeof(float); // Speicher für Vektoren allozieren h_w = (float *) malloc(npts * sizeof(float)); h_v = (float *) malloc(npts * sizeof(float)); h_u = (float *) malloc(npts * sizeof(float)); // auf Null setzen memset(h_w, 0, nBytes); memset(h_v, 0, nBytes); memset(h_u, 0, nBytes); // Zufaelliger Vektor random_vector(h_v); /* print_vector("v",h_v,1); */ // Device-Speicher allozieren mit cudaMalloc float *d_v, *d_w; CHECK(cudaMalloc((float**)&d_v, nBytes)); CHECK(cudaMalloc((float**)&d_w, nBytes)); // kopieren Host -> Device mit cudaMemcpy CHECK(cudaMemcpy(d_v, h_v, nBytes, cudaMemcpyHostToDevice)); CHECK(cudaMemcpy(d_w, h_w, nBytes, cudaMemcpyHostToDevice)); dim3 block(gridX, gridY); dim3 grid(threadX, threadY); double t_GPU_start = seconds(); laplace2d_GPU <<<block,grid>>> (d_w, d_v, N+2); cudaDeviceSynchronize(); double t_GPU_end = seconds(); double t_GPU = t_GPU_end - t_GPU_start; /* kopieren Device -> Host mit cudaMemcpy */ CHECK(cudaMemcpy(h_w, d_w, nBytes, cudaMemcpyDeviceToHost)); /* print_vector("w_GPU",h_w,1); */ // Device-Speicher freigeben CHECK(cudaFree(d_v)); CHECK(cudaFree(d_w)); double t_CPU_start = seconds(); laplace2d_CPU(h_u, h_v); double t_CPU_end = seconds(); double t_CPU = t_CPU_end - t_CPU_start; /* print_vector("w_CPU",h_w,1); */ fprintf( vector_add_speedup_file, "%lf, %lf, %lf,\n", t_GPU, t_CPU, t_CPU/t_GPU ); free(active); free(h_w); free(h_v); } int main(int argc, char **argv) { printf("%s Starting...\n", argv[0]); typedef struct grid_parameters { int N; int gridX; int gridY; int threadX; int threadY; } grid_params_t; int data_points = 64; grid_params_t grid_data[data_points]; /* read in grid parameters */ FILE *f = fopen("../scripts/factorizations", "r"); int i; for (i = 0; i != data_points && fscanf(f, "%d, %d, %d, %d, %d,\n", &grid_data[i].N, &grid_data[i].gridX, &grid_data[i].gridY, &grid_data[i].threadX, &grid_data[i].threadY) != EOF; i++ ); fclose(f); /* overwrite results */ FILE *laplace_speedup = fopen("../scripts/laplace_speedup_results", "w"); FILE *vector_scale_speedup = fopen("../scripts/vector_scale_speedup_results", "w"); FILE *vector_add_speedup = fopen("../scripts/vector_add_speedup_results", "w"); fprintf( laplace_speedup, "t_GPU, t_CPU, t_CPU/t_GPU,\n" ); fprintf( vector_scale_speedup, "t_GPU, t_CPU, t_CPU/t_GPU,\n" ); fprintf( vector_add_speedup, "t_GPU, t_CPU, t_CPU/t_GPU,\n" ); int gridX, gridY, threadX, threadY; for (i = 0; i < data_points; i++) { N = grid_data[i].N; gridX = grid_data[i].gridX; gridY = grid_data[i].gridY; threadX = grid_data[i].threadX; threadY = grid_data[i].threadY; laplaceOneLoop( laplace_speedup, N, gridX, gridY, threadX, threadY ); vectorScaleOneLoop( vector_scale_speedup, N, gridX, gridY, threadX, threadY ); vectorAddOneLoop( vector_add_speedup, N, gridX, gridY, threadX, threadY ); printf("%d/%d\r", i+1, data_points); fflush(stdout); } printf("\n"); fclose(laplace_speedup); fclose(vector_scale_speedup); fclose(vector_add_speedup); return (0); }
10,837
#include "includes.h" __global__ void updF_SoA(float *f, float *z1, float *z2, float *g, float tf, float lambda, 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; if ((px<(nx - 1))) DIVZ += z1[idx]; if ((px>0)) DIVZ -= z1[idx - 1]; if ((py<(ny - 1))) DIVZ += z2[idx]; if ((py>0)) DIVZ -= z2[idx - nx]; // update f //f[idx] = (1.-tf*lambda)*f[idx] + tf * DIVZ + tf*lambda*g[idx]; f[idx] = (f[idx] + tf * DIVZ + tf*lambda*g[idx]) / (1 + tf*lambda); } }
10,838
template<typename T> __device__ void columnsIndices(const T* matrix, int* result, const int rows, const int cols) { int bx = blockIdx.x; int by = blockIdx.y; int tx = threadIdx.x; int ty = threadIdx.y; int row = by * blockDim.y + ty; int col = bx * blockDim.x + tx; if (row < rows && col < cols) { int ij = row * cols + col; result[ij] = col; } } template<typename T> __device__ void rowsIndices(const T* matrix, int* result, const int rows, const int cols) { int bx = blockIdx.x; int by = blockIdx.y; int tx = threadIdx.x; int ty = threadIdx.y; int row = by * blockDim.y + ty; int col = bx * blockDim.x + tx; if (row < rows && col < cols) { int ij = row * cols + col; result[ij] = row; } }
10,839
#include <stdio.h> #include <cuda_runtime.h> #define N 12 // tugas 1: alokasi memori dan transfer dari device ke host __global__ void kern(int *A) { int idx = blockDim.x * blockIdx.x + threadIdx.x; A[idx] = idx; } /** * Host main routine */ int main(void) { // alokasikan memori, dan salin nilainya int *A = (int *) malloc(N*sizeof(int)); //alokasi memori di host int *d_A; cudaMalloc(&d_A,N*sizeof(int)); //alokasi memori di device cudaMemcpy(d_A,A,N*sizeof(int),cudaMemcpyHostToDevice); dim3 grid,block; block.x = 4; grid.x = 12/block.x; kern<<<grid,block>>>(d_A); cudaMemcpy(A,d_A,N*sizeof(int),cudaMemcpyDeviceToHost); // copy result for(int i = 0;i < N;i++) printf("A[%d] = %d\n",i,A[i]); free(A); cudaFree(d_A); return 0; }
10,840
// assumes square matrices (M = K = N) // Note: A and B are source matrices // A is M rows by K columns // B is K rows by N columns // C is destination // C is M rows by N columns extern "C" __global__ void sgemm( const float* A, const float* B, float* C, int widthA, int widthB) { const int col = threadIdx.x + blockIdx.x * blockDim.x; const int row = threadIdx.y + blockIdx.y * blockDim.y; if (row < widthB && col < widthA) { float sum = 0.0f; for (int i = 0; i < widthA; i++) { sum += A[i + row * widthA] * B[col + i * widthB]; } C[col + row * widthB] = sum; } } extern "C" __global__ void dgemm( const double* A, const double* B, double* C, int widthA, int widthB) { const int col = threadIdx.x + blockIdx.x * blockDim.x; const int row = threadIdx.y + blockIdx.y * blockDim.y; if (row < widthB && col < widthA) { float sum = 0.0f; for (int i = 0; i < widthA; i++) { sum += A[i + row * widthA] * B[col + i * widthB]; } C[col + row * widthB] = sum; } }
10,841
#include <stdlib.h> #include <stdio.h> #include <stdint.h> #include <unistd.h> #include <sys/types.h> #include <sys/stat.h> #include <fcntl.h> #include <cufft.h> #include <math.h> cudaEvent_t t_start, t_stop; cufftHandle plan; __global__ void filter_block(float *input_buffer, float *taps, float *output_buffer, int N, int P) { float temp_output = 0; for (int x=0; x < P; x++) { temp_output += taps[threadIdx.x + x*N]*input_buffer[threadIdx.x + x*N + (blockIdx.x*blockDim.x)]; // input buffer of continuous voltage samples should be distributed amongst N channels // index into input buffer: // current thread index indicates which channel we are currently processing // x which tap we are getting data for // blockIdx.x * blockDim.x is our depth into the block of data in multiples of channel number // structure of taps vector is actually [N,P]. So each thread will read taps[channel_number:channel_number+8] for taps } output_buffer[threadIdx.x + (blockIdx.x*blockDim.x)] = temp_output; } int main(int argc, char **argv) { int write_block = 1024*1024; // 1 MB worth of data at a time... int N = 512; // # of frequency channels in the PFB int P = 8; // length of pre filtering int write_size = sizeof(float) * write_block; int tap_size = sizeof(float) * P * N; int fft_output_size = (write_block / N) * sizeof(cufftComplex) * (N/2 + 1); // hold BATCH (write_block / N) worth of complex FFT outputs int fh; unsigned int i=0; char *data_file; char *fir_taps_file; char *base_buffer; float *float_buffer; float et; // counter for elapsed time of cuda ops float *fir_taps; float *device_input_buffer; float *device_output_buffer; cufftComplex *fft_output_buffer; cufftComplex *fft_buffer; float *device_fir_taps; if (argc > 2) { data_file = argv[1]; fir_taps_file = argv[2]; } else { printf("Please supply both data and fir_taps filenames...\n"); return -1;} base_buffer = (char*)malloc(write_block); float_buffer = (float*)malloc(write_size); fir_taps = (float*)malloc(tap_size); fft_buffer = (cufftComplex*)malloc(fft_output_size); fh = open(fir_taps_file, O_RDONLY); read(fh, fir_taps, tap_size); // source of taps vector should be flattened [P,N] array close(fh); //for (i=0; i < P*N; i++) { fprintf(stderr,"%f ",(float)*(fir_taps+i));} fh = open(data_file, O_LARGEFILE); read(fh, base_buffer, write_block); // read in a write block worth of int8 cudaEventCreate(&t_start); cudaEventCreate(&t_stop); cudaMalloc((void**)&device_input_buffer, write_size); cudaMalloc((void**)&device_output_buffer, write_size); cudaMalloc((void**)&fft_output_buffer, fft_output_size); cudaMalloc((void**)&device_fir_taps, tap_size); // allocate the device storage cudaMemcpy(device_fir_taps, fir_taps, tap_size, cudaMemcpyHostToDevice); // copy the filter taps to the device int threadsPerBlock = N; int blocksPerGrid = write_block / N; fprintf(stderr,"Blocks per grid: %i, Threads per block: %i\n",blocksPerGrid, threadsPerBlock); cufftPlan1d(&plan, N, CUFFT_R2C, int(write_block/N)); fprintf(stderr,"FFT Plan has length %i with batch size %i\n",N, int(write_block/N)); for (i = 0; i < write_block; ++i) { *(float_buffer+i) = (float)*(base_buffer+i); } //write(1, float_buffer, write_size); // output the raw data for comparison.. //for (i=0; i < write_block; i++) { fprintf(stderr,"Base: %i, Float: %f\n",(int)*(base_buffer+i),(float)*(float_buffer+i)); } cudaEventRecord(t_start, 0); cudaMemcpy(device_input_buffer, float_buffer, write_size, cudaMemcpyHostToDevice); // copy the floats to the device filter_block<<<blocksPerGrid, threadsPerBlock>>>(device_input_buffer, device_fir_taps, device_output_buffer, N, P); // kernel applies pre filtering to entire block leaving it in device_output_buffer ready for FFT cudaMemcpy(float_buffer, device_output_buffer, write_size, cudaMemcpyDeviceToHost); // get the intermediate results... write(1, float_buffer, write_size); // output the intermediate results... cufftExecR2C(plan, (cufftReal*)device_output_buffer, (cufftComplex*)fft_output_buffer); // Do FFT's over the entire block, one column at a time cudaMemcpy(fft_buffer, fft_output_buffer, fft_output_size, cudaMemcpyDeviceToHost); // get the final block //for (i=0; i < 100 * (N/2 + 1); i++) { fprintf(stderr,"Complex value %i has x=%f, y=%f\n", i, fft_buffer[i].x, fft_buffer[i].y); } cudaEventRecord(t_stop, 0); cudaEventSynchronize(t_stop); cudaEventElapsedTime(&et, t_start, t_stop); fprintf(stderr,"Done. CUDA time is %f ms\n", et); write(1, fft_buffer, write_size); // emit to stdout (which has hopefully been redirected...) return 0; }
10,842
#include <math.h> #include <stdio.h> #include <stdlib.h> double wmcw_energy(int n,double2* r,int* sig, double l){ int i, j, k; double x, y, inverselx, inversely, h, wmcw_energy; double pi = 4.0*atan(1.0); inverselx = 2.0*pi/l; inversely = 2.0*pi/l; wmcw_energy = 0.0; for (i = 0;i<=n-2;i++){ for (j=i+1;j<=n-1;j++){ if (i != j){ x = fabs(r[i].x-r[j].x) * inverselx; y = fabs(r[i].y-r[j].y) * inversely; h = 0.0; for (k = -10;k<=10;k++){ h = h + log((cosh(x-2.0*pi*(double)k)-cos(y))/cosh(2.0*pi*(double)k)); } h = h - x*x/(2.0*pi); wmcw_energy = wmcw_energy - h* ((double)sig[i]) *( (double)sig[j]); } } } //Difference between Weiss & McWilliams and (?)Montgomery & Joyce wmcw_energy = wmcw_energy/(double)n-0.177021697969890; return wmcw_energy; }
10,843
#include "includes.h" __global__ void _drop32(int n, float *x, float *xmask, float dropout, float scale) { int i = threadIdx.x + blockIdx.x * blockDim.x; while (i < n) { if (xmask[i] < dropout) x[i] = 0; else x[i] *= scale; i += blockDim.x * gridDim.x; } }
10,844
#include <curand_kernel.h> extern "C" __global__ void addVector(int* a, int aLen0, int* b, int bLen0, int* c, int cLen0) { int x = blockIdx.x; c[x] = a[x] + b[x]; } extern "C" __global__ void subVector(int* a, int aLen0, int* b, int bLen0, int* c, int cLen0) { int x = blockIdx.x; c[x] = a[x] - b[x]; } extern "C" __global__ void mulVector(int* a, int aLen0, int* b, int bLen0, int n) { int x = blockIdx.x; b[x] = a[x] * n; } extern "C" __global__ void divVector(int* a, int aLen0, int* b, int bLen0, int n) { int x = blockIdx.x; b[x] = a[x] / n; }
10,845
#include "includes.h" __global__ void mat_transpose(const float *a, float *b, int n, int m){ const int TIlE_WIDTH = 8; __shared__ float temp[TIlE_WIDTH][TIlE_WIDTH]; int bx = blockIdx.x, by = blockIdx.y; int tx = threadIdx.x, ty = threadIdx.y; int i = TIlE_WIDTH * bx + tx; int j = TIlE_WIDTH * by + ty; int idxa = j * n + i; int idxb = i * n + j; temp[ty][tx] = a[idxa]; __syncthreads(); b[idxb] = temp[ty][tx]; // if(i < n and j < m){ // b[idxb] = a[idxa]; // } }
10,846
/******************************************************************************* * ******************************************************************************/
10,847
#include <chrono> #include <iostream> #include "thrust/device_vector.h" #include <thrust/sequence.h> #include <thrust/transform.h> #include "cuComplex.h" #include "cufft.h" #include "cuda.h" #define NACC 256 #define NFPGAS 48 #define NCHAN_COARSE 336 #define NCHAN_FINE_IN 32 #define NCHAN_FINE_OUT 27 #define NACCUMULATE 128 #define NPOL 2 #define NSAMPS 4 #define NCHAN_SUM 16 #define NSAMP_PER_PACKET 128 #define NCHAN_PER_PACKET 7 #define HEADLEN 64 #define NSAMPS_SUMMED 2 #define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); } using std::cerr; using std::cout; using std::endl; #define XSIZE 7 #define YSIZE 128 #define ZSIZE 48 inline void gpuAssert(cudaError_t code, const char *file, int line, bool abort=true) { if (code != cudaSuccess) { fprintf(stderr,"GPUassert: %s %s %d\n", cudaGetErrorString(code), file, line); if (abort) exit(code); } } /*void UnpackCpu(void) { #pragma unroll for (int chan = 0; chan < 7; chan++) { for (int sample = 0; sample < 128; sample++) { idx = (sample * 7 + chan) * BYTES_PER_WORD; // get the start of the word in the received data array idx2 = chan * 128 + sample + startidx; // get the position in the buffer h_pol[idx2].x = static_cast<float>(static_cast<short>(data[HEADLEN + idx + 7] | (data[HEADLEN + idx + 6] << 8))); h_pol[idx2].y = static_cast<float>(static_cast<short>(data[HEADLEN + idx + 5] | (data[HEADLEN + idx + 4] << 8))); h_pol[idx2 + d_in_size / 2].x = static_cast<float>(static_cast<short>(data[HEADLEN + idx + 3] | (data[HEADLEN + idx + 2] << 8))); h_pol[idx2 + d_in_size / 2].y = static_cast<float>(static_cast<short>(data[HEADLEN + idx + 1] | (data[HEADLEN + idx + 0] << 8))); } } } */ __global__ void unpack_original_tex(cudaTextureObject_t texObj, cufftComplex * __restrict__ out, unsigned int acc) { int xidx = blockIdx.x * blockDim.x + threadIdx.x; int yidx = blockIdx.y * 128; int chanidx = threadIdx.x + blockIdx.y * 7; int skip; int2 word; for (int ac = 0; ac < acc; ac++) { skip = 336 * 128 * 2 * ac; for (int sample = 0; sample < YSIZE; sample++) { word = tex2D<int2>(texObj, xidx, yidx + ac * 48 * 128 + sample); out[skip + chanidx * YSIZE * 2 + sample].x = static_cast<float>(static_cast<short>(((word.y & 0xff000000) >> 24) | ((word.y & 0xff0000) >> 8))); out[skip + chanidx * YSIZE * 2 + sample].y = static_cast<float>(static_cast<short>(((word.y & 0xff00) >> 8) | ((word.y & 0xff) << 8))); out[skip + chanidx * YSIZE * 2 + YSIZE + sample].x = static_cast<float>(static_cast<short>(((word.x & 0xff000000) >> 24) | ((word.x & 0xff0000) >> 8))); out[skip + chanidx * YSIZE * 2 + YSIZE + sample].y = static_cast<float>(static_cast<short>(((word.x & 0xff00) >> 8) | ((word.x & 0xff) << 8))); } } } __global__ void unpack_new(const unsigned int *__restrict__ in, cufftComplex * __restrict__ out) { int skip = 0; __shared__ unsigned int accblock[1792]; int chan = 0; int time = 0; int line = 0; cufftComplex cpol; int polint; int2 tmp; int outskip = 0; for (int iacc = 0; iacc < NACCUMULATE; ++iacc) { // NOTE: This is skipping whole words as in will be cast to int2 skip = iacc * NCHAN_COARSE * NSAMP_PER_PACKET + blockIdx.x * NCHAN_PER_PACKET * NSAMP_PER_PACKET; for (int ichunk = 0; ichunk < 7; ++ichunk) { line = ichunk * blockDim.x + threadIdx.x; chan = line % 7; time = line / 7; tmp = ((int2*)in)[skip + line]; accblock[chan * NSAMP_PER_PACKET + time] = tmp.y; accblock[NSAMP_PER_PACKET * NCHAN_PER_PACKET + chan * NSAMP_PER_PACKET + time] = tmp.x; } __syncthreads(); skip = NCHAN_COARSE * NSAMP_PER_PACKET * NACCUMULATE; outskip = blockIdx.x * 7 * NSAMP_PER_PACKET * NACCUMULATE + iacc * NSAMP_PER_PACKET; for (chan = 0; chan < NCHAN_PER_PACKET; ++chan) { /*polaint = accblock[ichan * NSAMP_PER_PACKET + threadIdx.x]; polbint = accblock[NSAMP_PER_PACKET * NCHAN_PER_PACKET + ichan * NSAMP_PER_PACKET + threadIdx.x]; pola.x = static_cast<float>(static_cast<short>( ((polaint & 0xff000000) >> 24) | ((polaint & 0xff0000) >> 8) )); pola.y = static_cast<float>(static_cast<short>( ((polaint & 0xff00) >> 8) | ((polaint & 0xff) << 8) )); polb.x = static_cast<float>(static_cast<short>( ((polbint & 0xff000000) >> 24) | ((polbint & 0xff0000) >> 8) )); polb.y = static_cast<float>(static_cast<short>( ((polbint & 0xff00) >> 8) | ((polbint & 0xff) << 8) )); */ polint = accblock[chan * NSAMP_PER_PACKET + threadIdx.x]; cpol.x = static_cast<float>(static_cast<short>( ((polint & 0xff000000) >> 24) | ((polint & 0xff0000) >> 8) )); cpol.y = static_cast<float>(static_cast<short>( ((polint & 0xff00) >> 8) | ((polint & 0xff) << 8) )); out[outskip + threadIdx.x] = cpol; polint = accblock[NSAMP_PER_PACKET * NCHAN_PER_PACKET + chan * NSAMP_PER_PACKET + threadIdx.x]; cpol.x = static_cast<float>(static_cast<short>( ((polint & 0xff000000) >> 24) | ((polint & 0xff0000) >> 8) )); cpol.y = static_cast<float>(static_cast<short>( ((polint & 0xff00) >> 8) | ((polint & 0xff) << 8) )); out[skip + outskip + threadIdx.x] = cpol; outskip += NSAMP_PER_PACKET * NACCUMULATE; } } } __global__ void unpack_new_int2(int2 *__restrict__ in, cufftComplex *__restrict__ out) { int skip = 0; __shared__ int2 accblock[896]; int chan = 0; int time = 0; int line = 0; cufftComplex cpol; int polint; int outskip = 0; for (int iacc = 0; iacc < NACCUMULATE; ++iacc) { // NOTE: This is skipping whole words as in will be cast to int2 // skip = iacc * NCHAN_COARSE * NSAMP_PER_PACKET + blockIdx.x * NCHAN_PER_PACKET * NSAMP_PER_PACKET; skip = blockIdx.x * NCHAN_PER_PACKET * NSAMP_PER_PACKET * NACCUMULATE + iacc * NCHAN_PER_PACKET * NSAMP_PER_PACKET; for (int ichunk = 0; ichunk < 7; ++ichunk) { line = ichunk * blockDim.x + threadIdx.x; chan = line % 7; time = line / 7; accblock[chan * NSAMP_PER_PACKET + time] = in[skip + line]; } __syncthreads(); skip = NCHAN_COARSE * NSAMP_PER_PACKET * NACCUMULATE; outskip = blockIdx.x * 7 * NSAMP_PER_PACKET * NACCUMULATE + iacc * NSAMP_PER_PACKET; for (chan = 0; chan < NCHAN_PER_PACKET; ++chan) { polint = accblock[chan * NSAMP_PER_PACKET + threadIdx.x].y; cpol.x = static_cast<float>(static_cast<short>( ((polint & 0xff000000) >> 24) | ((polint & 0xff0000) >> 8) )); cpol.y = static_cast<float>(static_cast<short>( ((polint & 0xff00) >> 8) | ((polint & 0xff) << 8) )); out[outskip + threadIdx.x] = cpol; polint = accblock[chan * NSAMP_PER_PACKET + threadIdx.x].x; cpol.x = static_cast<float>(static_cast<short>( ((polint & 0xff000000) >> 24) | ((polint & 0xff0000) >> 8) )); cpol.y = static_cast<float>(static_cast<short>( ((polint & 0xff00) >> 8) | ((polint & 0xff) << 8) )); out[skip + outskip + threadIdx.x] = cpol; outskip += NSAMP_PER_PACKET * NACCUMULATE; } } } __global__ void unpack_alt(const unsigned int *__restrict__ in, cufftComplex * __restrict__ out) { if (threadIdx.x == 1022 || threadIdx.x == 1023) return; __shared__ unsigned int accblock[2044]; int inskip = blockIdx.x * NCHAN_PER_PACKET * NSAMP_PER_PACKET * NPOL * NACCUMULATE; int outskip = blockIdx.x * NCHAN_PER_PACKET * NSAMP_PER_PACKET * NACCUMULATE; int time = 0; int chan = 0; int line = 0; cufftComplex pola, polb; int polaint; int polbint; // NOTE: That will leave last 224 lines unprocessed // This can fit in 7 full warps of 32 for (int iacc = 0; iacc < 113; ++iacc) { line = iacc * blockDim.y + threadIdx.y; if (line < NCHAN_PER_PACKET * NSAMP_PER_PACKET * NACCUMULATE) { chan = threadIdx.y % 7; time = threadIdx.y / 7; accblock[chan * 146 + time] = in[inskip + threadIdx.y * NPOL]; accblock[NCHAN_PER_PACKET * 146 + chan * 146 + time] = in[inskip + threadIdx.y * NPOL + 1]; inskip += 2044; __syncthreads(); polbint = accblock[threadIdx.y]; polaint = accblock[NCHAN_PER_PACKET * 146 + threadIdx.y]; pola.x = static_cast<float>(static_cast<short>( ((polaint & 0xff000000) >> 24) | ((polaint & 0xff0000) >> 8) )); pola.y = static_cast<float>(static_cast<short>( ((polaint & 0xff00) >> 8) | ((polaint & 0xff) << 8) )); polb.x = static_cast<float>(static_cast<short>( ((polbint & 0xff000000) >> 24) | ((polbint & 0xff0000) >> 8) )); polb.y = static_cast<float>(static_cast<short>( ((polbint & 0xff00) >> 8) | ((polbint & 0xff) << 8) )); chan = threadIdx.y / 146; time = threadIdx.y % 146; out[outskip + chan * NSAMP_PER_PACKET * NACCUMULATE + time] = pola; out[NCHAN_COARSE * NSAMP_PER_PACKET * NACCUMULATE + outskip + chan * NSAMP_PER_PACKET * NACCUMULATE + time] = polb; outskip += 146; } } } __global__ void powertimefreq_new_hardcoded( cuComplex* __restrict__ in, float* __restrict__ out) { __shared__ float freq_sum_buffer[NCHAN_FINE_OUT*NCHAN_COARSE]; int warp_idx = threadIdx.x >> 0x5; int lane_idx = threadIdx.x & 0x1f; if (lane_idx >= NCHAN_FINE_OUT) return; int offset = blockIdx.x * NCHAN_COARSE * NPOL * NSAMPS * NCHAN_FINE_IN; int out_offset = blockIdx.x * NCHAN_COARSE * NCHAN_FINE_OUT / NCHAN_SUM; for (int coarse_chan_idx = warp_idx; coarse_chan_idx < NCHAN_COARSE; coarse_chan_idx += warpSize) { float real = 0.0f; float imag = 0.0f; int coarse_chan_offset = offset + coarse_chan_idx * NPOL * NSAMPS * NCHAN_FINE_IN; for (int pol=0; pol<NPOL; ++pol) { int pol_offset = coarse_chan_offset + pol * NSAMPS * NCHAN_FINE_IN; for (int samp=0; samp<NSAMPS; ++samp) { int samp_offset = pol_offset + samp * NCHAN_FINE_IN; cuComplex val = in[samp_offset + lane_idx]; real += val.x * val.x; imag += val.y * val.y; } } int output_idx = coarse_chan_idx * NCHAN_FINE_OUT + lane_idx; freq_sum_buffer[output_idx] = real+imag; //scaling goes here __syncthreads(); for (int start_chan=threadIdx.x; start_chan<NCHAN_FINE_OUT*NCHAN_COARSE; start_chan*=blockDim.x) { if ((start_chan+NCHAN_SUM) > NCHAN_FINE_OUT*NCHAN_COARSE) return; float sum = freq_sum_buffer[start_chan]; for (int ii=0; ii<4; ++ii) { sum += freq_sum_buffer[start_chan + (1<<ii)]; __syncthreads(); } out[out_offset+start_chan/NCHAN_SUM]; } } return; } __global__ void DetectScrunchKernel(cuComplex* __restrict__ in, float* __restrict__ out, short nchans) { /** * This block is going to do 2 timesamples for all coarse channels. * The fine channels are dealt with by the lanes, but on the fine * channel read we perform an fft shift and exclude the band edges. */ // gridDim.x should be Nacc * 128 / (32 * nsamps_to_add) == 256 __shared__ float freq_sum_buffer[NCHAN_FINE_OUT*NCHAN_COARSE]; // 9072 elements int warp_idx = threadIdx.x >> 0x5; int lane_idx = threadIdx.x & 0x1f; int pol_offset = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE; int coarse_chan_offet = NACCUMULATE * NCHAN_FINE_IN * NSAMPS; int block_offset = NCHAN_FINE_IN * NSAMPS_SUMMED * blockIdx.x; int nwarps_per_block = blockDim.x/warpSize; //Here we calculate indexes for FFT shift. int offset_lane_idx = (lane_idx + 19)%32; //Here only first 27 lanes are active as we drop //5 channels due to the 32/27 oversampling ratio if (lane_idx < 27) { // This warp // first sample in inner dimension = (32 * 2 * blockIdx.x) // This warp will loop over coarse channels in steps of NWARPS per block coarse_chan_idx (0,335) for (int coarse_chan_idx = warp_idx; coarse_chan_idx < NCHAN_COARSE; coarse_chan_idx += nwarps_per_block) { float real = 0.0f; float imag = 0.0f; int base_offset = coarse_chan_offet * coarse_chan_idx + block_offset + offset_lane_idx; for (int pol_idx=0; pol_idx<NPOL; ++pol_idx) { int offset = base_offset + pol_offset * pol_idx; for (int sample_idx=0; sample_idx<NSAMPS_SUMMED; ++sample_idx) { //Get first channel // IDX = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE * pol_idx // + NACCUMULATE * NCHAN_FINE_IN * NSAMPS * coarse_chan_idx // + blockIdx.x * NCHAN_FINE_IN * NSAMPS_SUMMED // + NCHAN_FINE_IN * sample_idx // + lane_idx; cuComplex val = in[offset + (NCHAN_FINE_IN * sample_idx)]; // load frequencies in right order real += val.x * val.x; imag += val.y * val.y; } // 3 is the leading dead lane count // sketchy freq_sum_buffer[coarse_chan_idx*NCHAN_FINE_OUT + lane_idx] = real + imag; } } } __syncthreads(); int saveoff = blockIdx.x * nchans; if (threadIdx.x < (NCHAN_FINE_OUT * NCHAN_COARSE / NCHAN_SUM)) { float sum = 0.0; for (int chan_idx = threadIdx.x * NCHAN_SUM; chan_idx < (threadIdx.x+1) * NCHAN_SUM; ++chan_idx) { sum += freq_sum_buffer[chan_idx]; } out[saveoff + threadIdx.x] = sum; } return; } __global__ void DetectScrunchScaleKernel(cuComplex* __restrict__ in, float* __restrict__ out, short nchans, float *means, float *scales) { /** * This block is going to do 2 timesamples for all coarse channels. * The fine channels are dealt with by the lanes, but on the fine * channel read we perform an fft shift and exclude the band edges. */ // gridDim.x should be Nacc * 128 / (32 * nsamps_to_add) == 256 __shared__ float freq_sum_buffer[NCHAN_FINE_OUT*NCHAN_COARSE]; // 9072 elements int warp_idx = threadIdx.x >> 0x5; int lane_idx = threadIdx.x & 0x1f; int pol_offset = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE; int coarse_chan_offet = NACCUMULATE * NCHAN_FINE_IN * NSAMPS; int block_offset = NCHAN_FINE_IN * NSAMPS_SUMMED * blockIdx.x; int nwarps_per_block = blockDim.x/warpSize; //Here we calculate indexes for FFT shift. int offset_lane_idx = (lane_idx + 19)%32; //Here only first 27 lanes are active as we drop //5 channels due to the 32/27 oversampling ratio if (lane_idx < 27) { // This warp // first sample in inner dimension = (32 * 2 * blockIdx.x) // This warp will loop over coarse channels in steps of NWARPS per block coarse_chan_idx (0,335) for (int coarse_chan_idx = warp_idx; coarse_chan_idx < NCHAN_COARSE; coarse_chan_idx += nwarps_per_block) { float real = 0.0f; float imag = 0.0f; int base_offset = coarse_chan_offet * coarse_chan_idx + block_offset + offset_lane_idx; for (int pol_idx=0; pol_idx<NPOL; ++pol_idx) { int offset = base_offset + pol_offset * pol_idx; for (int sample_idx=0; sample_idx<NSAMPS_SUMMED; ++sample_idx) { //Get first channel // IDX = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE * pol_idx // + NACCUMULATE * NCHAN_FINE_IN * NSAMPS * coarse_chan_idx // + blockIdx.x * NCHAN_FINE_IN * NSAMPS_SUMMED // + NCHAN_FINE_IN * sample_idx // + lane_idx; cuComplex val = in[offset + (NCHAN_FINE_IN * sample_idx)]; // load frequencies in right order real += val.x * val.x; imag += val.y * val.y; } // 3 is the leading dead lane count // sketchy freq_sum_buffer[coarse_chan_idx*NCHAN_FINE_OUT + lane_idx] = real + imag; } } } __syncthreads(); int saveoff = blockIdx.x * nchans; if (threadIdx.x < (NCHAN_FINE_OUT * NCHAN_COARSE / NCHAN_SUM)) { float sum = 0.0; int scaled = 0; for (int chan_idx = threadIdx.x * NCHAN_SUM; chan_idx < (threadIdx.x+1) * NCHAN_SUM; ++chan_idx) { sum += freq_sum_buffer[chan_idx]; } scaled = __float2int_ru((sum - means[threadIdx.x]) * scales[threadIdx.x] + 64.5f); if (scaled > 255) { scaled = 255; } else if (scaled < 0) { scaled = 0; } //out[saveoff + threadIdx.x] = (unsigned char)scaled; // NOTE: That puts the highest frequency first (OUTCHANS - 1 - threadIdx.x) out[saveoff + threadIdx.x] = (unsigned char)scaled; } return; } __global__ void DetectScrunchScaleRevKernel(cuComplex* __restrict__ in, float* __restrict__ out, short nchans, float *means, float *scales) { /** * This block is going to do 2 timesamples for all coarse channels. * The fine channels are dealt with by the lanes, but on the fine * channel read we perform an fft shift and exclude the band edges. */ // gridDim.x should be Nacc * 128 / (32 * nsamps_to_add) == 256 __shared__ float freq_sum_buffer[NCHAN_FINE_OUT*NCHAN_COARSE]; // 9072 elements int warp_idx = threadIdx.x >> 0x5; int lane_idx = threadIdx.x & 0x1f; int pol_offset = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE; int coarse_chan_offet = NACCUMULATE * NCHAN_FINE_IN * NSAMPS; int block_offset = NCHAN_FINE_IN * NSAMPS_SUMMED * blockIdx.x; int nwarps_per_block = blockDim.x/warpSize; //Here we calculate indexes for FFT shift. int offset_lane_idx = (lane_idx + 19)%32; //Here only first 27 lanes are active as we drop //5 channels due to the 32/27 oversampling ratio if (lane_idx < 27) { // This warp // first sample in inner dimension = (32 * 2 * blockIdx.x) // This warp will loop over coarse channels in steps of NWARPS per block coarse_chan_idx (0,335) for (int coarse_chan_idx = warp_idx; coarse_chan_idx < NCHAN_COARSE; coarse_chan_idx += nwarps_per_block) { float real = 0.0f; float imag = 0.0f; int base_offset = coarse_chan_offet * coarse_chan_idx + block_offset + offset_lane_idx; for (int pol_idx=0; pol_idx<NPOL; ++pol_idx) { int offset = base_offset + pol_offset * pol_idx; for (int sample_idx=0; sample_idx<NSAMPS_SUMMED; ++sample_idx) { //Get first channel // IDX = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE * pol_idx // + NACCUMULATE * NCHAN_FINE_IN * NSAMPS * coarse_chan_idx // + blockIdx.x * NCHAN_FINE_IN * NSAMPS_SUMMED // + NCHAN_FINE_IN * sample_idx // + lane_idx; cuComplex val = in[offset + (NCHAN_FINE_IN * sample_idx)]; // load frequencies in right order real += val.x * val.x; imag += val.y * val.y; } // 3 is the leading dead lane count // sketchy freq_sum_buffer[coarse_chan_idx*NCHAN_FINE_OUT + lane_idx] = real + imag; } } } __syncthreads(); int saveoff = blockIdx.x * nchans; if (threadIdx.x < (NCHAN_FINE_OUT * NCHAN_COARSE / NCHAN_SUM)) { float sum = 0.0; int scaled = 0; for (int chan_idx = threadIdx.x * NCHAN_SUM; chan_idx < (threadIdx.x+1) * NCHAN_SUM; ++chan_idx) { sum += freq_sum_buffer[chan_idx]; } scaled = __float2int_ru((sum - means[566 - threadIdx.x]) * scales[566 - threadIdx.x] + 64.5f); if (scaled > 255) { scaled = 255; } else if (scaled < 0) { scaled = 0; } //out[saveoff + threadIdx.x] = (unsigned char)scaled; // NOTE: That puts the highest frequency first (OUTCHANS - 1 - threadIdx.x) out[saveoff + 566 - threadIdx.x] = (unsigned char)scaled; } return; } __global__ void DetectScrunchScaleTruncKernel(cuComplex* __restrict__ in, float* __restrict__ out, short nchans, float *means, float *scales) { /** * This block is going to do 2 timesamples for all coarse channels. * The fine channels are dealt with by the lanes, but on the fine * channel read we perform an fft shift and exclude the band edges. */ // gridDim.x should be Nacc * 128 / (32 * nsamps_to_add) == 256 __shared__ float freq_sum_buffer[NCHAN_FINE_OUT*NCHAN_COARSE]; // 9072 elements int warp_idx = threadIdx.x >> 0x5; int lane_idx = threadIdx.x & 0x1f; int pol_offset = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE; int coarse_chan_offet = NACCUMULATE * NCHAN_FINE_IN * NSAMPS; int block_offset = NCHAN_FINE_IN * NSAMPS_SUMMED * blockIdx.x; int nwarps_per_block = blockDim.x/warpSize; //Here we calculate indexes for FFT shift. int offset_lane_idx = (lane_idx + 19)%32; //Here only first 27 lanes are active as we drop //5 channels due to the 32/27 oversampling ratio if (lane_idx < 27) { // This warp // first sample in inner dimension = (32 * 2 * blockIdx.x) // This warp will loop over coarse channels in steps of NWARPS per block coarse_chan_idx (0,335) for (int coarse_chan_idx = warp_idx; coarse_chan_idx < NCHAN_COARSE; coarse_chan_idx += nwarps_per_block) { float real = 0.0f; float imag = 0.0f; int base_offset = coarse_chan_offet * coarse_chan_idx + block_offset + offset_lane_idx; for (int pol_idx=0; pol_idx<NPOL; ++pol_idx) { int offset = base_offset + pol_offset * pol_idx; for (int sample_idx=0; sample_idx<NSAMPS_SUMMED; ++sample_idx) { //Get first channel // IDX = NCHAN_COARSE * NSAMPS * NCHAN_FINE_IN * NACCUMULATE * pol_idx // + NACCUMULATE * NCHAN_FINE_IN * NSAMPS * coarse_chan_idx // + blockIdx.x * NCHAN_FINE_IN * NSAMPS_SUMMED // + NCHAN_FINE_IN * sample_idx // + lane_idx; cuComplex val = in[offset + (NCHAN_FINE_IN * sample_idx)]; // load frequencies in right order real += val.x * val.x; imag += val.y * val.y; } // 3 is the leading dead lane count // sketchy freq_sum_buffer[coarse_chan_idx*NCHAN_FINE_OUT + lane_idx] = real + imag; } } } __syncthreads(); int saveoff = blockIdx.x * nchans; int skipbottom = 28 * NCHAN_SUM; if (threadIdx.x < 512) { float sum = 0.0; int scaled = 0; for (int chan_idx = threadIdx.x * NCHAN_SUM; chan_idx < (threadIdx.x+1) * NCHAN_SUM; ++chan_idx) { sum += freq_sum_buffer[skipbottom + chan_idx]; } scaled = __float2int_ru((sum - means[511 - threadIdx.x]) * scales[511 - threadIdx.x] + 64.5f); if (scaled > 255) { scaled = 255; } else if (scaled < 0) { scaled = 0; } //out[saveoff + threadIdx.x] = (unsigned char)scaled; // NOTE: That puts the highest frequency first (OUTCHANS - 1 - threadIdx.x) out[saveoff + 511 - threadIdx.x] = (unsigned char)scaled; } return; } __global__ void GetPowerAddTimeKernel(cufftComplex* __restrict__ in, float* __restrict__ out, unsigned int jump, unsigned int factort, unsigned int acc) { int idx1, idx2; int outidx; int skip1, skip2; float power1, power2; for (int iac = 0; iac < acc; iac++) { skip1 = iac * 336 * 128 * 2; skip2 = iac * 336 * 27; for (int ichan = 0; ichan < 7; ichan++) { outidx = skip2 + 7 * 27 * blockIdx.x + ichan * 27 + threadIdx.x; out[outidx] = (float)0.0; out[outidx + jump] = (float)0.0; out[outidx + 2 * jump] = (float)0.0; out[outidx + 3 * jump] = (float)0.0; idx1 = skip1 + 256 * (blockIdx.x * 7 + ichan); for (int itime = 0; itime < factort; itime++) { idx2 = threadIdx.x + itime * 32; power1 = (in[idx1 + idx2].x * in[idx1 + idx2].x + in[idx1 + idx2].y * in[idx1 + idx2].y); power2 = (in[idx1 + 128 + idx2].x * in[idx1 + 128 + idx2].x + in[idx1 + 128 + idx2].y * in[idx1 + 128 + idx2].y); out[outidx] += (power1 + power2); out[outidx + jump] += (power1 - power2); out[outidx + 2 * jump] += (2 * (in[idx1 + idx2].x * in[idx1 + 128 + idx2].x + in[idx1 + idx2].y * in[idx1 + 128 + idx2].y)); out[outidx + 3 * jump] += (2 * (in[idx1 + idx2].x * in[idx1 + 128 + idx2].y - in[idx1 + idx2].y * in[idx1 + 128 + idx2].x)); } } } } __global__ void GetScaleFactorsKernel(float *indata, float *base, float *stdev, float *factors, int nchans, int processed) { /* // NOTE: Filterbank file format coming in //float mean = indata[threadIdx.x]; float mean = 0.0f; // NOTE: Depending whether I save STD or VAR at the end of every run // float estd = stdev[threadIdx.x]; float estd = stdev[threadIdx.x] * stdev[threadIdx.x] * (processed - 1.0f); float oldmean = base[threadIdx.x]; //float estd = 0.0f; //float oldmean = 0.0; float val = 0.0f; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { val = indata[isamp * nchans + threadIdx.x]; diff = val - oldmean; mean = oldmean + diff * factors[processed + isamp + 1]; estd += diff * (val - mean); oldmean = mean; } base[threadIdx.x] = mean; stdev[threadIdx.x] = sqrtf(estd / (float)(processed + 2 * NACCUMULATE - 1.0f)); // stdev[threadIdx.x] = estd; */ float chmean = 0.0f; float chestd = 0.0f; float val = 0.0; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { val = indata[isamp * nchans + threadIdx.x]; diff = val - chmean; chmean += diff * factors[isamp + 1]; chestd += diff * (val - chmean); } float oldmean = base[threadIdx.x]; float oldestd = stdev[threadIdx.x] * stdev[threadIdx.x] * (processed - 1.0f); float newestd = 0.0f; diff = chmean - oldmean; base[threadIdx.x] = oldmean + diff * (float)(2.0f * NACCUMULATE) / (float)(processed + 2.0 * NACCUMULATE); newestd = oldestd + chestd + diff * diff * (float)(2.0f * NACCUMULATE) * (float)processed / (float)(processed + 2.0 * NACCUMULATE); stdev[threadIdx.x] = sqrt(newestd / (float)(processed + 2 * NACCUMULATE - 1.0f)); } __global__ void GetScaleFactorsDivKernel(float *indata, float *base, float *stdev, int nchans, int processed) { /* // NOTE: Filterbank file format coming in //float mean = indata[threadIdx.x]; float mean = 0.0f; // NOTE: Depending whether I save STD or VAR at the end of every run // float estd = stdev[threadIdx.x]; float estd = stdev[threadIdx.x] * stdev[threadIdx.x] * (processed - 1.0f); float oldmean = base[threadIdx.x]; //float estd = 0.0f; //float oldmean = 0.0; float val = 0.0f; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { val = indata[isamp * nchans + threadIdx.x]; diff = val - oldmean; mean = oldmean + diff * factors[processed + isamp + 1]; estd += diff * (val - mean); oldmean = mean; } base[threadIdx.x] = mean; stdev[threadIdx.x] = sqrtf(estd / (float)(processed + 2 * NACCUMULATE - 1.0f)); // stdev[threadIdx.x] = estd; */ float chmean = 0.0f; float chestd = 0.0f; float val = 0.0; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { val = indata[isamp * nchans + threadIdx.x]; diff = val - chmean; chmean += diff / (isamp + 1); chestd += diff * (val - chmean); } float oldmean = base[threadIdx.x]; float oldestd = stdev[threadIdx.x] * stdev[threadIdx.x] * (processed - 1.0f); float newestd = 0.0f; diff = chmean - oldmean; base[threadIdx.x] = oldmean + diff * (float)(2.0f * NACCUMULATE) / (float)(processed + 2.0 * NACCUMULATE); newestd = oldestd + chestd + diff * diff * (float)(2.0f * NACCUMULATE) * (float)processed / (float)(processed + 2.0 * NACCUMULATE); stdev[threadIdx.x] = sqrt(newestd / (float)(processed + 2 * NACCUMULATE - 1.0f)); } __global__ void GetScaleFactorsDoubleKernel(float *indata, float *base, float *stdev, int nchans) { /* // NOTE: Filterbank file format coming in //float mean = indata[threadIdx.x]; float mean = 0.0f; // NOTE: Depending whether I save STD or VAR at the end of every run // float estd = stdev[threadIdx.x]; float estd = stdev[threadIdx.x] * stdev[threadIdx.x] * (processed - 1.0f); float oldmean = base[threadIdx.x]; //float estd = 0.0f; //float oldmean = 0.0; float val = 0.0f; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { val = indata[isamp * nchans + threadIdx.x]; diff = val - oldmean; mean = oldmean + diff * factors[processed + isamp + 1]; estd += diff * (val - mean); oldmean = mean; } base[threadIdx.x] = mean; stdev[threadIdx.x] = sqrtf(estd / (float)(processed + 2 * NACCUMULATE - 1.0f)); // stdev[threadIdx.x] = estd; */ /* float chmean = 0.0f; float chestd = 0.0f; float val = 0.0; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { val = indata[isamp * nchans + threadIdx.x]; diff = val - chmean; chmean += diff / (isamp + 1); chestd += diff * (val - chmean); } float oldmean = base[threadIdx.x]; float oldestd = stdev[threadIdx.x] * stdev[threadIdx.x] * (processed - 1.0f); float newestd = 0.0f; diff = chmean - oldmean; base[threadIdx.x] = oldmean + diff * (float)(2.0f * NACCUMULATE) / (float)(processed + 2.0 * NACCUMULATE); newestd = oldestd + chestd + diff * diff * (float)(2.0f * NACCUMULATE) * (float)processed / (float)(processed + 2.0 * NACCUMULATE); stdev[threadIdx.x] = sqrt(newestd / (float)(processed + 2 * NACCUMULATE - 1.0f)); */ float sum = indata[threadIdx.x]; for (int isamp = 1; isamp < 2 * NACCUMULATE; ++isamp) { sum += indata[isamp * nchans + threadIdx.x]; } float mean = sum / (float)(2.0 * NACCUMULATE); base[threadIdx.x] = mean; float sumsq = 0.0f; float diff = 0.0; for (int isamp = 0; isamp < 2 * NACCUMULATE; ++isamp) { diff = indata[isamp * nchans + threadIdx.x] - mean; sumsq += diff * diff; } stdev[threadIdx.x] = sqrt(sumsq / (float)(NACCUMULATE - 1.0)); } struct FactorFunctor { __host__ __device__ float operator()(float val) { return val != 0.0f ? 1.0f/val : val; } }; int main(int argc, char* argv[]) { unsigned char *rawbuffer = new unsigned char[7168 * NFPGAS * NACC]; thrust::device_vector<unsigned char> rawdata(7168 * NFPGAS * NACCUMULATE); // NOTE: 336 coarse channels * 32 fine channels * 4 time samples * 2 polarisations thrust::device_vector<cuComplex> input(336*32*4*2*NACCUMULATE); // NOTE: 336 coarse channels * 27 fine channels thrust::device_vector<float> output(336*27*NACCUMULATE); thrust::device_vector<float> means(567); thrust::device_vector<float> scales(567); thrust::device_vector<float> factors(NACCUMULATE * 2); thrust::sequence(factors.begin(), factors.end()); thrust::transform(factors.begin(), factors.end(), factors.begin(), FactorFunctor()); // NOTE: Benchmarking the unpacker kernel cudaArray *rawarray; cudaChannelFormatDesc cdesc; cdesc = cudaCreateChannelDesc<int2>(); cudaMallocArray(&rawarray, &cdesc, 7, 48 * 128 * NACCUMULATE); cudaResourceDesc rdesc; memset(&rdesc, 0, sizeof(cudaResourceDesc)); rdesc.resType = cudaResourceTypeArray; rdesc.res.array.array = rawarray; cudaTextureDesc tdesc; memset(&tdesc, 0, sizeof(cudaTextureDesc)); tdesc.addressMode[0] = cudaAddressModeClamp; tdesc.filterMode = cudaFilterModePoint; tdesc.readMode = cudaReadModeElementType; cudaTextureObject_t texObj = 0; cudaCreateTextureObject(&texObj, &rdesc, &tdesc, NULL); cudaMemcpyToArray(rawarray, 0, 0, rawbuffer, 7168 * 48 * NACCUMULATE * sizeof(unsigned char), cudaMemcpyHostToDevice); dim3 rearrange_b(1,48,1); dim3 rearrange_t(7,1,1); dim3 unpackt(2, 128, 1); dim3 unpacka(1, 1024, 1); dim3 unpackb(48, 2, 1); // ################################## // ### UNPACK KERNEL BENCHMARKING ### // ################################## std::chrono::time_point<std::chrono::system_clock> unpackstart, unpackend; std::chrono::duration<double> unpackelapsed; unpackstart = std::chrono::system_clock::now(); // NOTE: Unpacking with texture memory for (int ii = 0; ii < 32; ++ii) { unpack_original_tex<<<rearrange_b, rearrange_t, 0>>>(texObj, thrust::raw_pointer_cast(input.data()), NACCUMULATE); gpuErrchk(cudaDeviceSynchronize()); } unpackend = std::chrono::system_clock::now(); unpackelapsed = unpackend - unpackstart; cout << "Unpacking with texture memory: " << unpackelapsed.count() / 32.0 << "s" << endl; // NOTE: Unpacking with shared memory unpackstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { unpack_new<<<48, 128, 0>>>(reinterpret_cast<unsigned int*>(thrust::raw_pointer_cast(rawdata.data())), thrust::raw_pointer_cast(input.data())); gpuErrchk(cudaDeviceSynchronize()); } unpackend = std::chrono::system_clock::now(); unpackelapsed = unpackend - unpackstart; cout << "Unpacking with shared memory: " << unpackelapsed.count() / 32.0 << "s" << endl; // NOTE: Unpacking with shared memory with cast to int2 in the function call unpackstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { unpack_new_int2<<<48, 128, 0>>>(reinterpret_cast<int2*>(thrust::raw_pointer_cast(rawdata.data())), thrust::raw_pointer_cast(input.data())); gpuErrchk(cudaDeviceSynchronize()); } unpackend = std::chrono::system_clock::now(); unpackelapsed = unpackend - unpackstart; cout << "Unpacking with shared memory with cast to int2 in the function call: " << unpackelapsed.count() / 32.0 << "s" << endl; // NOTE: Unpacking with shared memory with the alternative incoming buffer arrangement unpackstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { unpack_alt<<<48, unpacka, 0>>>(reinterpret_cast<unsigned int*>(thrust::raw_pointer_cast(rawdata.data())), thrust::raw_pointer_cast(input.data())); gpuErrchk(cudaDeviceSynchronize()); } unpackend = std::chrono::system_clock::now(); unpackelapsed = unpackend - unpackstart; cout << "Unpacking with shared memory with the alternative incoming buffer arrangement: " << unpackelapsed.count() / 32.0 << "s" << endl; std::chrono::time_point<std::chrono::system_clock> powerstart, powerend; std::chrono::duration<double> powerelapsed; // ################################# // ### POWER KERNEL BENCHMARKING ### // ################################# // NOTE: Unoptimised power kernel with time averaging only powerstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { GetPowerAddTimeKernel<<<48, 27, 0>>>(thrust::raw_pointer_cast(input.data()), thrust::raw_pointer_cast(output.data()), 0, NSAMPS, NACCUMULATE); gpuErrchk(cudaDeviceSynchronize()); } powerend = std::chrono::system_clock::now(); powerelapsed = powerend - powerstart; cout << "Unoptimised power kernel: " << powerelapsed.count() / 32.0 << "s" << endl; // NOTE: Optimised power kernel with shared memory with time and frequency averaging powerstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { DetectScrunchKernel<<<NACCUMULATE,1024,0>>>(thrust::raw_pointer_cast(input.data()),thrust::raw_pointer_cast(output.data()), 567); gpuErrchk(cudaDeviceSynchronize()); } powerend = std::chrono::system_clock::now(); powerelapsed = powerend - powerstart; cout << "Optimised power kernel: " << powerelapsed.count() / 32.0 << "s" << endl; // NOTE: Optimised power kernel with scaling powerstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { DetectScrunchScaleKernel<<<NACCUMULATE,1024,0>>>(thrust::raw_pointer_cast(input.data()),thrust::raw_pointer_cast(output.data()), 567, thrust::raw_pointer_cast(means.data()), thrust::raw_pointer_cast(scales.data())); gpuErrchk(cudaDeviceSynchronize()); } powerend = std::chrono::system_clock::now(); powerelapsed = powerend - powerstart; cout << "Optimised power kernel with scaling: " << powerelapsed.count() / 32.0 << "s" << endl; // NOTE: Optimised power kernel with scaling and reversed frequency powerstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { DetectScrunchScaleRevKernel<<<NACCUMULATE,1024,0>>>(thrust::raw_pointer_cast(input.data()),thrust::raw_pointer_cast(output.data()), 567, thrust::raw_pointer_cast(means.data()), thrust::raw_pointer_cast(scales.data())); gpuErrchk(cudaDeviceSynchronize()); } powerend = std::chrono::system_clock::now(); powerelapsed = powerend - powerstart; cout << "Optimised power kernel with scaling and reversed frequency: " << powerelapsed.count() / 32.0 << "s" << endl; // NOTE: Optimised power kernel with scaling truncated and reversed frequency powerstart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { DetectScrunchScaleTruncKernel<<<NACCUMULATE,1024,0>>>(thrust::raw_pointer_cast(input.data()),thrust::raw_pointer_cast(output.data()), 512, thrust::raw_pointer_cast(means.data()), thrust::raw_pointer_cast(scales.data())); gpuErrchk(cudaDeviceSynchronize()); } powerend = std::chrono::system_clock::now(); powerelapsed = powerend - powerstart; cout << "Optimised power kernel with scaling, truncated and reversed frequency: " << powerelapsed.count() / 32.0 << "s" << endl; // ################################# // ### SCALE KERNEL BENCHMARKING ### // ################################# std::chrono::time_point<std::chrono::system_clock> scalestart, scaleend; std::chrono::duration<double> scaleelapsed; // NOTE: Double-pass scaling factors kernel scalestart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { GetScaleFactorsDoubleKernel<<<1, 512, 0>>>(thrust::raw_pointer_cast(output.data()), thrust::raw_pointer_cast(means.data()), thrust::raw_pointer_cast(scales.data()), 512); gpuErrchk(cudaDeviceSynchronize()); } scaleend = std::chrono::system_clock::now(); scaleelapsed = scaleend - scalestart; cout << "Double pass scaling factors kernel with division: " << scaleelapsed.count() / 32.0 << "s" << endl; // NOTE: Unoptimised scaling factors kernel with division scalestart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { GetScaleFactorsDivKernel<<<1, 512, 0>>>(thrust::raw_pointer_cast(output.data()), thrust::raw_pointer_cast(means.data()), thrust::raw_pointer_cast(scales.data()), 512, 0); gpuErrchk(cudaDeviceSynchronize()); } scaleend = std::chrono::system_clock::now(); scaleelapsed = scaleend - scalestart; cout << "Unoptimised scaling factors kernel with division: " << scaleelapsed.count() / 32.0 << "s" << endl; // NOTE: Optimised scaling factors kernel scalestart = std::chrono::system_clock::now(); for (int ii = 0; ii < 32; ++ii) { GetScaleFactorsKernel<<<1, 512, 0>>>(thrust::raw_pointer_cast(output.data()), thrust::raw_pointer_cast(means.data()), thrust::raw_pointer_cast(scales.data()), thrust::raw_pointer_cast(factors.data()), 512, 0); gpuErrchk(cudaDeviceSynchronize()); } scaleend = std::chrono::system_clock::now(); scaleelapsed = scaleend - scalestart; cout << "Optimised scaling factors kernel: " << scaleelapsed.count() / 32.0 << "s" << endl; }
10,848
#include <bitset> #include <iomanip> #include <ios> #include <iostream> #include <sstream> #include <string> std::string get_UUID_as_String(const cudaUUID_t& uuid){ std::stringstream result; result << "GPU-"; size_t cnt = 0; for(auto& c: uuid.bytes) { std::bitset<8> bits(c); if(cnt == 4 || cnt == 6 || cnt == 8 || cnt == 10) result << "-"; result << std::hex << bits.to_ulong() ; cnt++; } return result.str(); } void print_device_information(const int deviceId) { cudaDeviceProp deviceProp{}; cudaGetDeviceProperties(&deviceProp, deviceId); std::cout << "================ DeviceId: " << deviceId << " ================ \n"; std::cout << "--> General Information: \n" << "\tDevice name: " << deviceProp.name << "\n" << "\tUUID: " << get_UUID_as_String(deviceProp.uuid) << "\n" << "\tIntegrated: " << deviceProp.integrated << "\n" << "\tClock rate (kHz): " << deviceProp.clockRate << "\n"; std::cout << "\n--> Computation: \n" << "\tComputer capability: " << deviceProp.major << "." << deviceProp.minor << "\n" << "\t# of SMs: " << deviceProp.multiProcessorCount << "\n" << "\tWarp size: " << deviceProp.warpSize << "\n" << "\tmax block dim: (" << deviceProp.maxThreadsDim[0] << ", " << deviceProp.maxThreadsDim[1] << ", " << deviceProp.maxThreadsDim[2] << ")\n" << "\tmax threads/block: " << deviceProp.maxThreadsPerBlock << "\n" << "\tmax threads/SM: " << deviceProp.maxThreadsPerMultiProcessor << "\n" << "\tSingle/Double precision ration: " << deviceProp.singleToDoublePrecisionPerfRatio << "\n" << "\n"; std::cout << "--> Memory: \n" << "\tUnified addressing: " << deviceProp.unifiedAddressing << "\n" << "\tSupports managed memory: " << deviceProp.managedMemory << "\n" << "\tTotal global memory (Gb): " << std::setprecision(3) << std::fixed << static_cast<float>(deviceProp.totalGlobalMem) / (1024. * 1024. * 1024.) << "\n" << "\tTotal constant memory (kb): " << deviceProp.totalConstMem / 1024 << "\n" << "\tsMem/block (kb): " << deviceProp.sharedMemPerBlock / 1024 << "\n" << "\tsMem/SM (kb): " << deviceProp.sharedMemPerMultiprocessor << "\n" << "\n"; } int main() { int deviceCount; cudaGetDeviceCount(&deviceCount); std::cout << "Detected " << deviceCount << " GPU devices.\n"; for (int device = 0; device < deviceCount; ++device) { print_device_information(device); } return 0; }
10,849
#include "cuda.h" __global__ void kernel_saxpy( int n, float * Masse, float * PosX, float * PosY, float * PosZ, float * VelX, float * VelY, float * VelZ, float * accX, float * accY, float * accZ ) { int i = blockIdx.x * blockDim.x + threadIdx.x; if ( i < n ) { float aX = 0; float aY = 0; float aZ = 0; int j; for(j=0; j<n; j++){ if(!(i==j)){ float deltaX = (PosX[j]-PosX[i]); float deltaY = (PosY[j]-PosY[i]); float deltaZ = (PosZ[j]-PosZ[i]); float Dij = sqrtf((deltaX*deltaX) + (deltaY*deltaY) + (deltaZ*deltaZ)); if ( Dij < 1.0 ) Dij = 1.0; float coef = 10 * 1 * (1/(Dij*Dij*Dij)) * Masse[j]; aX += deltaX * coef; aY += deltaY * coef; aZ += deltaZ * coef; } } accX[i] = aX; accY[i] = aY; accZ[i] = aZ; } } void saxpy( int nblocks, int nthreads, int n, float * deviceSrc1, float * deviceSrc2, float * deviceSrc3, float * deviceSrc4, float * deviceSrc5, float * deviceSrc6, float * deviceSrc7, float * deviceSrc8, float * deviceSrc9, float * deviceSrc10 ) { kernel_saxpy<<<nblocks, nthreads>>>( n, deviceSrc1, deviceSrc2, deviceSrc3, deviceSrc4, deviceSrc5, deviceSrc6, deviceSrc7, deviceSrc8, deviceSrc9, deviceSrc10 ); }
10,850
#include <stdlib.h> #include <stdio.h> #define N 64 #define TPB 32 float scale(int i, int n) { return ((float)i/(n-1)); } __device__ float distance(float x1, float x2) { return sqrt((x2-x1)*(x2-x1)); } __global__ void distanceKernel(float* d_out, float* d_in, float ref) { const int i = blockIdx.x * blockDim.x + threadIdx.x; const float x = d_in[i]; d_out[i] = distance(x, ref); printf("i = %2d: dist from %f to %f is %f.\n", i, ref, x, d_out[i]); } int main() { const float ref = 0.5f; float* in = NULL; float* out = NULL; cudaMallocManaged(&in, N * sizeof(float)); cudaMallocManaged(&out, N * sizeof(float)); for (int i = 0; i < N; ++i) { in[i] = scale(i, N); } int banks = (N+TPB-1)/TPB; distanceKernel<<<banks, TPB>>>(out, in, ref); cudaDeviceSynchronize(); cudaFree(in); cudaFree(out); return 0; }
10,851
/* * Copyright (C) 2002-2021 the Network-Based Computing Laboratory * (NBCL), The Ohio State University. * * Contact: Dr. D. K. Panda (panda@cse.ohio-state.edu) * * For detailed copyright and licensing information, please refer to the * copyright file COPYRIGHT in the top level OMB directory. */ __global__ void compute_kernel(float a, float *x, float *y, int N) { int i = blockIdx.x * blockDim.x + threadIdx.x; int count = 0; if (i < N) { for (count = 0; count < (N / 8); count++) { y[i] = a * x[i] + y[i]; } } } __global__ void touch_managed_kernel(char *buf, size_t len) { int i; i = blockIdx.x * blockDim.x + threadIdx.x; if (i < len) { buf[i] = buf[i] + 1; } } __global__ void empty_kernel(char *buf, size_t len) {} extern "C" void call_kernel(float a, float *d_x, float *d_y, int N, cudaStream_t *stream) { compute_kernel<<<(N + 255) / 256, 256, 0, *stream>>>(a, d_x, d_y, N); } extern "C" void call_touch_managed_kernel(char *buf, size_t length, cudaStream_t *stream) { touch_managed_kernel<<<(length + 255) / 256, 256, 0, *stream>>>(buf, length); } extern "C" void call_empty_kernel(char *buf, size_t length, cudaStream_t *stream) { empty_kernel<<<(length + 255) / 256, 256, 0, *stream>>>(buf, length); }
10,852
#include <stdio.h> #include <cstdlib> #include <cmath> #include <iostream> #include <sys/time.h> #define TIME_RESOLUTION 1000000 // time measuring resolution (us) #define NUM_BLOCKS 128 // this is wrong #define NUM_THREADS_PER_BLOCK 256 // this is wrong #define SIZE 2048 #define TILE_SIZE 16 using namespace std; float m1[SIZE][SIZE], m2[SIZE][SIZE], result[SIZE][SIZE], _result[SIZE][SIZE]; long long unsigned cpu_time; timeval t; void startTime (void) { gettimeofday(&t, NULL); cpu_time = t.tv_sec * TIME_RESOLUTION + t.tv_usec; } void stopTime (void) { gettimeofday(&t, NULL); long long unsigned final_time = t.tv_sec * TIME_RESOLUTION + t.tv_usec; final_time -= cpu_time; cout << final_time << " us have elapsed for the CPU execution" << endl; } void fillMatrices (void) { for (unsigned i = 0; i < SIZE; ++i) { for (unsigned j = 0; j < SIZE; ++j) { result[i][j] = 0.0; _result[i][j] = 0.0; m1[i][j] = ((float) rand()) / ((float) RAND_MAX); m2[i][j] = ((float) rand()) / ((float) RAND_MAX); } } } void checkCUDAError (const char *msg) { cudaError_t err = cudaGetLastError(); if( cudaSuccess != err) { cerr << "Cuda error: " << msg << ", " << cudaGetErrorString( err) << endl; exit(-1); } } // You need to optimize AND parallelize this first void regularMatrixMult (void) { for (unsigned i = 0; i < SIZE; ++i) { for (unsigned j = 0; j < SIZE; ++j) { result[i][j] = 0; for (unsigned k = 0; k < SIZE; ++k) { result[i][j] += m1[i][k] * m2[k][j]; } } } } void tiledMatrixMult (void) { for (unsigned m = 0; m < SIZE; m += TILE_SIZE) { for (unsigned n = 0; n < SIZE; n += TILE_SIZE) { //... } } } // Fill the input parameters and kernel qualifier __global__ void matrixMultKernel (double *dev_m1, double *dev_m2, double *dev_res) { int id = blockIdx.x*blockDim.x+threadIdx.x; } // Fill with the code required for the GPU stencil (mem allocation, transfers, kernel launch of stencilKernel) double* matrixMultGPU (void) { // you can either: // 1 - use 2D matrices, as in CPU // 2 - use 1D matrices, but YOU have to convert them here return NULL; } int main (int argc, char** argv) { fillMatrices (); // GPU stuff matrixMultGPU (); // CPU stuff startTime(); regularMatrixMult (); stopTime(); return 0; }
10,853
#include <iostream> #include <cstdlib> #include <stdio.h> #include <ctime> #include <assert.h> __global__ void allPrefixSums (long int* A_gpu, long int* arr, int N) { int id = threadIdx.x + (blockIdx.x * blockDim.x); if (id == 0) return; if (id > N-1) return; for (int i = 0; i < id; i++) { A_gpu[id] += arr[i]; } } int main(int argc, char* argv[]) { if (argc != 2) { std::cerr << "Incorrect input style, please do ./homework4 N" << std::endl; return 2; } int N = atoi(argv[1]); long int* arr = new long int[N]; for (int i = 0; i < N; i++) { arr[i] = rand() % 1000 + 1; } long int* A_cpu = new long int[N]; // Sequential Code for all prefix sum A_cpu[0] = 0; for (int i = 1; i < N; i++) { A_cpu[i] += (arr[i-1] + A_cpu[i-1]); } long int* deviceA; cudaMalloc(&deviceA, N * sizeof(long int)); long int* deviceArr; cudaMalloc(&deviceArr, N*sizeof(long int)); cudaMemcpy(deviceArr, arr, N*sizeof(long int), cudaMemcpyHostToDevice); dim3 threadsPerBlock(1024, 1, 1); dim3 numBlocks(N / 1024 + 1, 1, 1); // Make the parallel call allPrefixSums<<<numBlocks, threadsPerBlock>>>(deviceA, deviceArr, N); long int* A_gpu = new long int[N];; cudaMemcpy(A_gpu, deviceA, N*sizeof(long int), cudaMemcpyDeviceToHost); for (int i = 0; i < N; i++) { assert(A_gpu[i] == A_cpu[i]); } printf("GPU Output Matches CPU Output\n"); return 0; }
10,854
/* * ABCTE.cpp * * Created on: 01 февр. 2016 г. * Author: aleksandr */ #include "ABCTE.h" #include <iostream> ABCTE::ABCTE(GridTE* _grid) : EyLeft((_grid->sizeY-1)*6, 0), EyRight((_grid->sizeY-1)*6, 0), ExTop((_grid->sizeX-1)*6, 0), ExBottom((_grid->sizeX-1)*6, 0), coeff0(0), coeff1(0), coeff2(0), grid(_grid), coeffDevice(3, 0) { float temp1 = grid->S; float temp2 = 1.0 / temp1 + 2.0 + temp1; coeff0 = -(1.0 / temp1 - 2.0 + temp1) / temp2; coeff1 = -2.0 * (temp1 - 1.0 / temp1) / temp2; coeff2 = 4.0 * (temp1 + 1.0 / temp1) / temp2; int sizeX = grid->sizeX; int sizeY = grid->sizeY; std::vector<float> coeffHost(3, 0); coeffHost[0] = coeff0; coeffHost[1] = coeff1; coeffHost[2] = coeff2; coeffDevice = coeffHost; leftUpdater.setParams(grid->Ey.getDevicePtr(), EyLeft.getDevicePtr(), coeffDevice.data(), sizeX, sizeY); rightUpdater.setParams(grid->Ey.getDevicePtr(), EyRight.getDevicePtr(), coeffDevice.data(), sizeX, sizeY); topUpdater.setParams(grid->Ex.getDevicePtr(), ExTop.getDevicePtr(), coeffDevice.data(), sizeX, sizeY); bottomUpdater.setParams(grid->Ex.getDevicePtr(), ExBottom.getDevicePtr(), coeffDevice.data(), sizeX, sizeY); std::cout << "Absorption boundary conditions initialized \n"; }
10,855
#include <stdio.h> #include <stdlib.h> #include <cuda.h> #define N 8 #define TAMBLOCK 2 #define LIMIT 10 __global__ void jacobi2d(float *A, float *B){ int ind = (((blockIdx.y*blockDim.y+threadIdx.y)+1)*N)+((blockIdx.x*blockDim.x+threadIdx.x)+1); *(B+ind)=*(A+ind)+*(A+ind+1)+*(A+ind-1)+*(A+ind+N)+*(A+ind-N); } int main(){ int memsize=N*N*sizeof(float); float *A = (float *)malloc(memsize); //Matriz creada float *B = (float *)malloc(memsize); //2da Matriz, que usaremos para calcular valores. float *d_a, *d_b; cudaMalloc(&d_a, memsize); cudaMalloc(&d_b, memsize); //Primero vamos a inicializar la matriz A, dandole valores for(int i=0; i<N; ++i) for(int j=0;j<N;++j){ if((i==0||i==N)&&(j==0||j==N)) *(A+(i*N)+j)=150.0f; else *(A+(i*N)+j)=(float) (rand()%10); } cudaMemcpy(d_a, A, memsize, cudaMemcpyHostToDevice); cudaMemcpy(d_b, B, memsize, cudaMemcpyHostToDevice); dim3 block((N-2)/TAMBLOCK, (N-2)/TAMBLOCK); dim3 thread(TAMBLOCK, TAMBLOCK); //Forma 2 Jacobi, extremos = 150 y solo sumo por el interior. for(int k=0; k<LIMIT;++k){ jacobi2d <<<block, thread>>> (d_a, d_b); float *aux=d_a; d_a=d_b; d_b=aux; } cudaMemcpy(A, d_a, memsize, cudaMemcpyDeviceToHost); printf("\nValor del array A: \n"); for(int i=0;i<N*N;++i){ printf("%f ,",*(A+i)); if(i!=0 && (i+1)%8==0) printf("\n"); } //Este es el valor la matriz que contiene los valores anteriores. /*printf("\nValor del array B: \n"); for(int i=0;i<ROW*COL;++i){ printf("%f ,",*(B+i)); if(i!=0 && (i+1)%5==0) printf("\n"); }*/ }
10,856
#include <stdio.h> #include <stdlib.h> #include <assert.h> #include <time.h> #include <string.h> #include <iostream> #define BLOCOS 1 //#define THREAD #define CHECK_ERROR(call) do { \ if( cudaSuccess != call) { \ std::cerr << std::endl << "CUDA ERRO: " << \ cudaGetErrorString(call) << " in file: " << __FILE__ \ << " in line: " << __LINE__ << std::endl; \ exit(0); \ } } while (0) using namespace std; typedef struct automato { char letra; automato *prox; automato *ant; automato *inf; int final; } Automato; __global__ void pfac(Automato* at, int *matches, char *frase){ int x = blockDim.x * blockIdx.x + threadIdx.x; } Automato* newAutomato(Automato* ant) { Automato *nv = (Automato*) malloc(sizeof(Automato)); nv->prox = NULL; nv->inf = NULL; nv->ant = ant; return nv; } Automato* addAlgarismo(Automato *at, char algm, int first) { if (at != NULL && at->letra == algm && first == 1) { return at; } // Caso algarismo novo seja diferente do algarismo da raiz else if (at != NULL && at->letra != algm && first == 1) { Automato *pt = at->inf; Automato *ant = pt; while (pt != NULL) { if (pt->letra == algm) { return pt; } else { if (pt != NULL) { ant = pt; pt = pt->inf; } } } Automato *nv = newAutomato(at); nv->letra = algm; if (ant != NULL) { ant->inf = nv; return ant->inf; } else { at->inf = nv; return at->inf; } } else if(at != NULL && first == 0) { Automato *pt = at->prox; Automato *ant = NULL; while (pt != NULL) { if (pt->letra == algm) { return pt; } else { ant = pt; pt = pt->inf; } } Automato *nv = newAutomato(at); nv->letra = algm; if (ant != NULL) { ant->inf = nv; } else { at->prox = nv; } return nv; } else { Automato *nv = newAutomato(NULL); nv->letra = algm; return nv; } } void imprimir(Automato *at) { Automato *temp = at; while (temp != NULL) { printf("%c ", temp->letra); imprimir(temp->prox); temp = temp->inf; printf("\n"); } } /*Automato* mallocGPU(Automato *at) { Automato *temp = at; while (temp != NULL) { imprimir(temp->prox); temp = temp->inf; } }*/ int main (int argc, char **argv) { int GPU = 0; Automato *at = newAutomato(NULL); at->letra = 'a'; at->prox = NULL; char frase[255] = "ab abg bede ef"; //"abc acd abb agd acc"; int THREADS = strlen(frase); Automato *temp = at; int i = 0; int first = 1; while(frase[i] != '\0') { if(frase[i] != ' ') { temp = addAlgarismo(temp, frase[i], first); first = 0; //printf("Letra: %c\n", temp->letra); } else { temp->final = 1; temp = at; first = 1; } i++; } imprimir(at); // CPU char h_fita[255] = "ab abg bede ef"; int *h_matches = (int*) malloc(sizeof(int)); // GPU Automato *d_at = NULL; char *d_fita = NULL; int *d_matches = NULL; CHECK_ERROR(cudaSetDevice(GPU)); *h_matches = 0; //Reset na GPU selecionada CHECK_ERROR(cudaDeviceReset()); CHECK_ERROR(cudaMalloc((void**) &d_at, sizeof(Automato*))); CHECK_ERROR(cudaMalloc((void**) &d_fita, 255*sizeof(char))); CHECK_ERROR(cudaMalloc((void**) &d_matches, sizeof(int))); //Copiando CPU --> GPU CHECK_ERROR(cudaMemcpy(d_at, at, sizeof(Automato*), cudaMemcpyHostToDevice)); CHECK_ERROR(cudaMemcpy(d_fita, h_fita, 255*sizeof(char), cudaMemcpyHostToDevice)); CHECK_ERROR(cudaMemcpy(d_matches, h_matches, sizeof(int), cudaMemcpyHostToDevice)); pfac <<<BLOCOS, THREADS>>> (d_at, d_matches, d_fita); //Copiando GPU --> CPU CHECK_ERROR(cudaMemcpy(at, d_at, sizeof(Automato*), cudaMemcpyDeviceToHost)); CHECK_ERROR(cudaMemcpy(h_fita, d_fita, 255*sizeof(char), cudaMemcpyDeviceToHost)); CHECK_ERROR(cudaMemcpy(h_matches, d_matches, sizeof(int), cudaMemcpyDeviceToHost)); // Liberando memória na GPU CHECK_ERROR(cudaFree(d_at)); CHECK_ERROR(cudaFree(d_fita)); CHECK_ERROR(cudaFree(d_matches)); // Liberando memória na CPU free(at); free(h_matches); free(h_fita); return EXIT_SUCCESS; }
10,857
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> #include <stdlib.h> #include <time.h> #define SIZE 128*1024*1024 #define BLOCK_SIZE 1024 __global__ void offsetCopy(float *odata, const float *idata, int offset) { int xid = blockIdx.x * blockDim.x + threadIdx.x + offset; odata[xid] = idata[xid]; } __global__ void strideCopy(float *odata, const float *idata, int stride) { int xid = (blockIdx.x * blockDim.x + threadIdx.x) * stride; odata[xid] = idata[xid]; } int main() { srand((unsigned)time(NULL)); int i; float* cpu_A = (float*)malloc(sizeof(float)*SIZE); float* cpu_B = (float*)malloc(sizeof(float)*SIZE); memset(cpu_B, 0, sizeof(float)*SIZE); for (i = 0; i < SIZE; ++i){ cpu_A[i] = (float)(rand() / RAND_MAX); } float *dev_a; float *dev_b; cudaSetDevice(0); cudaMalloc((void**)&dev_a, SIZE * sizeof(float) * 2); cudaMalloc((void**)&dev_b, SIZE * sizeof(float) * 2); cudaMemcpy(dev_a, cpu_A, SIZE * sizeof(float), cudaMemcpyHostToDevice); cudaEvent_t start, stop; float elapsedTime; cudaEventCreate(&start); cudaEventCreate(&stop); for (i = 1; i <= 1024; i *= 2) { cudaEventRecord(start, 0); int blocks = SIZE / BLOCK_SIZE; int threads = BLOCK_SIZE; offsetCopy<<<blocks, threads>>>(dev_b, dev_a, i-1); //strideCopy << <SIZE / BLOCK_SIZE, BLOCK_SIZE >> >(dev_b, dev_a, 0); cudaThreadSynchronize(); cudaEventRecord(stop, 0); cudaEventSynchronize(stop); cudaEventElapsedTime(&elapsedTime, start, stop); printf("GPU use time: %f (ms), Step: %d \n", elapsedTime, i-1); } cudaFree(dev_a); cudaFree(dev_b); }
10,858
/* File: vec_add.cu * Purpose: Implement vector addition on a gpu using cuda * * Compile: nvcc [-g] [-G] -o vec_add vec_add.cu * Run: ./vec_add */ #include <stdio.h> #include <unistd.h> #include <stdlib.h> #include <math.h> __global__ void Vec_add(double x[], double y[], double z[], int n) { int thread_id = threadIdx.x; if (thread_id < n){ z[thread_id] = x[thread_id] + y[thread_id]; } } extern "C" void axpb_cpp_cuda(double h_x[], double h_y[], double h_z[], int n) { double *d_x, *d_y, *d_z; size_t size; /* Define vector length */ size = n*sizeof(double); /* Allocate vectors in device memory */ cudaMalloc(&d_x, size); cudaMalloc(&d_y, size); cudaMalloc(&d_z, size); /* Copy vectors from host memory to device memory */ cudaMemcpy(d_x, h_x, size, cudaMemcpyHostToDevice); cudaMemcpy(d_y, h_y, size, cudaMemcpyHostToDevice); /* Kernel Call */ Vec_add<<<1,1000>>>(d_x, d_y, d_z, n); cudaThreadSynchronize(); cudaMemcpy(h_z, d_z, size, cudaMemcpyDeviceToHost); /* Free device memory */ cudaFree(d_x); cudaFree(d_y); cudaFree(d_z); } /* main */
10,859
// #CSCS CUDA Training // // #Example 3.1 - transpose matrix // // #Author: Ugo Varetto // // #Goal: compute the transpose of a matrix and time operation using // GPU's on-board performance counters through streams; print the result in ms (10^-3 s) // // #Rationale: shows how to time GPU computation // // #Solution: straightworwad, simply compute the thread id associated with the element // and copy the transposed data into the output matrix; wrap kernel calls with event // recording and print time information // // #Code: typical flow + timing: // 1) compute launch grid configuration // 2) allocate data on host(cpu) and device(gpu) // 3) initialize data directly on the GPU // 4) create events // 5) record start time // 6) launch kernel // 7) synchronize events to guarantee that kernel execution is finished // 8) record stop time // 9) read data back // 10) print timing information as stop - start time // 11) delete events // 12) free memory // The code uses the default stream 0; streams are used to sychronize operations // to guarantee that all operations in the same stream are executed sequentially. // // #Compilation: nvcc -arch=sm_13 3_1_transpose-timing.cu -o transpose-timing // // #Execution: ./transpose-timing // // #Note: kernel invocations ( foo<<<...>>>(...) ) are *always* asynchronous and a call to // cudaThreadSynchronize() is required to wait for the end of kernel execution from // a host thread; in case of synchronous copy operations like cudaMemcpy(...,cudaDeviceToHost) // kernel execution is guaranteed to be terminated before data are copied // // #Note: the code is C++ also because the default compilation mode for CUDA is C++, all functions // are named with C++ convention and the syntax is checked by default against C++ grammar rules // // #Note: -arch=sm_13 allows the code to run on every card with hw architecture GT200 (gtx 2xx) or better // // #Note: -arch=sm_13 is the lowest architecture version that supports double precision // // #Note: the example can be extended to read configuration data and matrix size from the command line //#include <cuda_runtime.h> // automatically added by nvcc #include <vector> #include <iostream> typedef float real_t; __global__ void transpose( const real_t* in, real_t *out, int num_rows, int num_columns ) { const int col = blockIdx.x * blockDim.x + threadIdx.x; const int row = blockIdx.y * blockDim.y + threadIdx.y; const int input_index = row * num_columns + col; const int output_index = col * num_rows + row; out[ output_index ] = in[ input_index ]; } __global__ void init_matrix( real_t* in ) { const int c = threadIdx.x + blockDim.x * blockIdx.x; const int r = threadIdx.y + blockDim.y * blockIdx.y; const int idx = c + gridDim.x * blockDim.x * r; in[ idx ] = (real_t) idx; } void print_matrix( const real_t* m, int r, int c, int stride ) { for( int i = 0; i != r; ++i ) { for( int j = 0; j != c; ++j ) std::cout << m[ i * stride + j ] << ' '; std::cout << '\n'; } std::cout << std::endl; } //------------------------------------------------------------------------------ int main(int argc, char** argv ) { const dim3 BLOCKS( 512, 512 ); const dim3 THREADS_PER_BLOCK( 16, 16 ); const int ROWS = 512 * 16; // 8192 const int COLUMNS = 512 * 16; // 8192 const size_t SIZE = ROWS * COLUMNS * sizeof( real_t ); // device storage real_t* dev_in = 0; real_t* dev_out = 0; cudaMalloc( &dev_in, SIZE ); cudaMalloc( &dev_out, SIZE ); // host storage std::vector< real_t > outmatrix( ROWS * COLUMNS ); // initialize matrix with kernel; much faster than using // for loops on the cpu init_matrix<<<dim3( COLUMNS, ROWS ), 1>>>( dev_in ); cudaMemcpy( &outmatrix[ 0 ], dev_in, SIZE, cudaMemcpyDeviceToHost ); // print upper 4x4 left corner of input matrix std::cout << "INPUT MATRIX - " << ROWS << " rows, " << COLUMNS << " columns" << std::endl; print_matrix( &outmatrix[ 0 ], 4, 4, COLUMNS ); // create events for timing execution cudaEvent_t start = cudaEvent_t(); cudaEvent_t stop = cudaEvent_t(); cudaEventCreate( &start ); cudaEventCreate( &stop ); // record time into start event cudaEventRecord( start, 0 ); // 0 is the default stream id // execute kernel transpose<<<BLOCKS, THREADS_PER_BLOCK>>>( dev_in, dev_out, ROWS, COLUMNS ); // issue request to record time into stop event cudaEventRecord( stop, 0 ); // synchronize stop event to wait for end of kernel execution on stream 0 cudaEventSynchronize( stop ); // compute elapsed time (done by CUDA run-time) float elapsed = 0.f; cudaEventElapsedTime( &elapsed, start, stop ); std::cout << "Elapsed time (ms): " << elapsed << std::endl; // copy output data from device(gpu) to host(cpu) cudaMemcpy( &outmatrix[ 0 ], dev_out, SIZE, cudaMemcpyDeviceToHost ); // print upper 4x4 corner of transposed matrix std::cout << "\nOUTPUT MATRIX - " << COLUMNS << " rows, " << ROWS << " columns" << std::endl; print_matrix( &outmatrix[ 0 ], 4, 4, ROWS ); // free memory cudaFree( dev_in ); cudaFree( dev_out ); // release events cudaEventDestroy( start ); cudaEventDestroy( stop ); return 0; }
10,860
#include <stdio.h> #include <stdlib.h> #include <math.h> #include <string.h> #define R 4 #define C 40 /* * It returns the length of a string pointed by pointer s, * It acts like the cpu strlen() function */ __device__ int gpu_strlen(char * s) { int i = 0; while(s[i] != '\0') { i++; } return i; } /* * It returns 0 if input character ch is NOT an alphabetical letter * Otherwise, it returns one. */ __device__ int gpu_isAlpha(char ch) { //potentially think about using isalpha() function if(ch == 'a' || ch == 'b' || ch == 'c' || ch == 'd' || ch == 'e' || ch == 'f' || ch == 'g' || ch == 'h' || ch == 'i' || ch == 'j' || ch == 'k' || ch == 'l' || ch == 'm' || ch == 'n' || ch == 'o' || ch == 'p' || ch == 'q' || ch == 'r' || ch == 's' || ch == 't' || ch == 'u' || ch == 'v' || ch == 'w' || ch == 'x' || ch == 'y' || ch == 'z') return 1; else if(ch == 'A' || ch == 'B' || ch == 'C' || ch == 'D' || ch == 'E' || ch == 'F' || ch == 'G' || ch == 'H' || ch == 'I' || ch == 'J' || ch == 'K' || ch == 'L' || ch == 'M' || ch == 'N' || ch == 'O' || ch == 'P' || ch == 'Q' || ch == 'R' || ch == 'S' || ch == 'T' || ch == 'U' || ch == 'V' || ch == 'W' || ch == 'X' || ch == 'Y' || ch == 'Z') return 1; else return 0; } /* Cuda kernel to count number of words in each line of text pointed by a. * The output is stored back in 'out' array. * numLine specifies the num of lines in a, maxLineLen specifies the maximal * num of characters in one line of text. */ __global__ void wordCount( char **a, int **out, int numLine, int maxLineLen ) { int ix = blockIdx.x*blockDim.x + threadIdx.x; int iy = blockIdx.y*blockDim.y + threadIdx.y; int currLen = gpu_strlen(a[iy]); //each thread process one character within a line if( iy < numLine && ix < currLen && gpu_isAlpha(a[iy][ix]) != 1 ) { out[iy][ix] += 1; } __syncthreads(); if(out[iy][ix] == 1 && ix < currLen) out[iy][ix + 1] = 0; /* int col = blockIdx.x * blockDim.x + threadIdx.x; int row = blockIdx.y * blockDim.y + threadIdx.y; if( row < maxLineLen || col < numLine) return; out[iy][ix] = 1; int rowLen = gpu_strlen(a[row]); if(col < rowLen) return; if(gpu_isAlpha(a[row][col]) == 1) { out[row][col] = 1; } else { out[row][col] = 1; } __syncthreads(); if(col != 1 && out[row][col - 1] == 1) { out[row][col] = 0; } */ } /* Print out the all lines of text in a on stdout */ void printArr( char **a, int lines ) { int i; for(i=0; i<lines; i++) { printf("%s\n", a[i]); } } int main() { int i; char **d_in, **h_in, **h_out; int h_count_in[R][C], **h_count_out, **d_count_in; //allocate h_in = (char **)malloc(R * sizeof(char *)); h_out = (char **)malloc(R * sizeof(char *)); h_count_out = (int **)malloc(R * sizeof(int *)); cudaMalloc((void ***)&d_in, sizeof(char *) * R); cudaMalloc((void ***)&d_count_in, sizeof(int *) * R); //alocate for string data for(i = 0; i < R; ++i) { cudaMalloc((void **) &h_out[i],C * sizeof(char)); h_in[i]=(char *)calloc(C, sizeof(char));//allocate or connect the input data to it //!!!!!!!!!!!!!!!!! strcpy(h_in[i], "for you:: he "); cudaMemcpy(h_out[i], h_in[i], strlen(h_in[i]) + 1, cudaMemcpyHostToDevice); } cudaMemcpy(d_in, h_out, sizeof(char *) * R,cudaMemcpyHostToDevice); //alocate for output occurrence for(i = 0; i < R; ++i) { cudaMalloc((void **) &h_count_out[i], C * sizeof(int)); cudaMemset(h_count_out[i], 0, C * sizeof(int)); } cudaMemcpy(d_count_in, h_count_out, sizeof(int *) * R,cudaMemcpyHostToDevice); printArr(h_in, R); printf("\n\n"); //set up kernel configuartion variables dim3 grid, block; block.x = 2; block.y = 2; grid.x = ceil((float)C / block.x); grid.y = ceil((float)R / block.y); //careful must be type cast into float, otherwise, integer division used //printf("grid.x = %d, grid.y=%d\n", grid.x, grid.y ); //launch kernel wordCount<<<grid, block>>>( d_in, d_count_in, R, C); //copy data back from device to host for(i = 0; i < R; ++i) { cudaMemcpy(h_count_in[i], h_count_out[i], sizeof(int) * C,cudaMemcpyDeviceToHost); } printf("Occurrence array obtained from device:\n"); for(i = 0; i < R; i ++) { for(int j = 0; j < C; j ++) printf("%4d", h_count_in[i][j]); printf("\n"); } return 0; }
10,861
#include "includes.h" __global__ void vecAdd (int *a, int *b, int *c) { int index = blockIdx.x * blockDim.x + threadIdx.x; if(index < N){ c[index] = a[index] + b[index]; } }
10,862
/* * Compute beta parameters. * * The so-called beta parameters were developed in SNO as a measure of event * isotropy: * * The lth beta parameter, beta_l, is defined as the average value of the * Legendre polynomial, P_l, of the cosine of the angle between each pair * PMT hits in the event. * * beta_l = <P_l(cos(theta_ik)> where i != k * * Again, the angle is taken with respect to the fitted vertex position. * The combination beta_14 = beta_1 + 4 * beta_4 was selected by the SNO * collaboration for use in signal extraction due to the good separability * it provides and the ease of parameterisation of the Gaussian-like * distribution. * * - Measurement of the 8B Solar Neutrino Energy Spectrum at the Sudbury * Neutrino Observatory, Jeanne R. Wilson, p. 179 (Ph.D. Thesis) */ #include <map> #include <string> #include <cuda.h> #include <stdio.h> __global__ void get_coeffs(int2* pairs, float3* hit_pos, float3* fit_pos, float* p1, float* p4, int n) { int i = threadIdx.x + blockIdx.x * blockDim.x; if (i < n) { int2 pair = pairs[i]; float3 A = hit_pos[pair.x]; float3 B = hit_pos[pair.y]; float3 C = *fit_pos; float c = sqrtf((A.x-B.x)*(A.x-B.x) + (A.y-B.y)*(A.y-B.y) + (A.z-B.z)*(A.z-B.z)); float a = sqrtf((C.x-B.x)*(C.x-B.x) + (C.y-B.y)*(C.y-B.y) + (C.z-B.z)*(C.z-B.z)); float b = sqrtf((C.x-A.x)*(C.x-A.x) + (C.y-A.y)*(C.y-A.y) + (C.z-A.z)*(C.z-A.z)); // dust off that trig! p1[i] = (float)(-0.5) * (c*c - a*a - b*b) / (a*b); // Legendre P4(x) = (1/8) (25x**4 - 30x**2 + 3) p4[i] = (float)(0.125) * (25*p1[i]*p1[i]*p1[i]*p1[i] - 30*p1[i]*p1[i] + 3); } } float get_beta14(float3 &fit_pos, float3* hit_pos, int nhits) { // compute list of pmt pairs (this isn't really necessary) const int npairs = nhits * (nhits - 1) / 2; int2* pairs = (int2*) malloc(npairs * sizeof(int2)); unsigned pidx = 0; for (unsigned i=0; i<nhits-1; i++) { for (unsigned j=i+1; j<nhits; j++) { pairs[pidx] = make_int2(i, j); pidx++; } } // allocate device memory and copy over int2* pairs_device; float* p1_device; float* p4_device; float3* hit_pos_device; float3* fit_pos_device; cudaMalloc(&pairs_device, npairs * sizeof(int2)); cudaMalloc(&p1_device, npairs * sizeof(float)); cudaMalloc(&p4_device, npairs * sizeof(float)); cudaMalloc(&hit_pos_device, nhits * sizeof(float3)); cudaMalloc(&fit_pos_device, sizeof(float3)); cudaMemcpy(pairs_device, pairs, npairs * sizeof(int2), cudaMemcpyHostToDevice); cudaMemcpy(hit_pos_device, hit_pos, nhits * sizeof(float3), cudaMemcpyHostToDevice); cudaMemcpy(fit_pos_device, &fit_pos, sizeof(float3), cudaMemcpyHostToDevice); // execute kernel and retreive results int blocksize = 512; int nblocks = npairs / blocksize + 1; get_coeffs<<<nblocks, blocksize>>>(pairs_device, hit_pos_device, fit_pos_device, p1_device, p4_device, npairs); float* p1 = (float*) malloc(npairs * sizeof(float)); float* p4 = (float*) malloc(npairs * sizeof(float)); cudaMemcpy(p1, p1_device, npairs * sizeof(float), cudaMemcpyDeviceToHost); cudaMemcpy(p4, p4_device, npairs * sizeof(float), cudaMemcpyDeviceToHost); // compute average float p1ave = 0.0; float p4ave = 0.0; for (unsigned i=0; i<npairs; i++) { p1ave += p1[i]; p4ave += p4[i]; } p1ave /= npairs; p4ave /= npairs; float beta14 = p1ave + 4.0 * p4ave; free(pairs); free(p1); free(p4); cudaFree(pairs_device); cudaFree(hit_pos_device); cudaFree(fit_pos_device); cudaFree(p1_device); cudaFree(p4_device); return beta14; }
10,863
#include <cuda_runtime.h> #include <device_launch_parameters.h> #include <stdio.h> #include <stdlib.h> #include <iostream> __global__ void sum_array_gpu(int *a,int *b,int *c,int size) { int gid = blockDim.x * blockDim.y * gridDim.x * blockIdx.y + blockDim.x * blockDim.y * blockIdx.x + blockDim.x * threadIdx.y + threadIdx.x; if (gid < size) { c[gid] = a[gid] + b[gid]; } //printf("gid : %d, a[gid] : %d, b[gid] : %d, c[gid] : %d\n", gid, a[gid], b[gid], c[gid]); } 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]; } } bool checkResult(int *a, int *b, int size) { for(int i=0;i<size;i++){ if(a[i]!=b[i]){ printf("the %d th current value of a[i] and b[i] is: %d, %d\n",i,a[i],b[i]); return false; } //printf("the current value of a[i] and b[i] are the same\n"); } return true; } int main(int argc, char *argv[]) { int size = 1000; int dim_x = 2; int dim_y = 2; int block_x = 16; int block_y = 16; int byte_size = size * sizeof(int); int *a_input,*b_input,*c_output,*gpu_output; a_input = (int*)malloc(byte_size); b_input = (int*)malloc(byte_size); c_output = (int*)malloc(byte_size); gpu_output = (int*)malloc(byte_size); for(int i=0;i<size;i++) { a_input[i] = i; b_input[i] = i*2; } //cpu matrix sum calculation sum_array_cpu(a_input,b_input,c_output,size); int * a_gpu_input, * b_gpu_input, *c_gpu_output; cudaMalloc((void**)&a_gpu_input, byte_size); cudaMalloc((void**)&b_gpu_input, byte_size); cudaMalloc((void**)&c_gpu_output, byte_size); cudaMemcpy(a_gpu_input,a_input,byte_size,cudaMemcpyHostToDevice); cudaMemcpy(b_gpu_input,b_input,byte_size,cudaMemcpyHostToDevice); dim3 block(block_x,block_y); dim3 grid(dim_x,dim_y); printf("dimension of each block is: %d, %d\n", block.x, block.y); printf("dimension of grid is: %d, %d\n", grid.x, grid.y); sum_array_gpu<<<grid,block>>>(a_gpu_input,b_gpu_input,c_gpu_output,size); cudaDeviceSynchronize(); //memory transfer back to host cudaMemcpy(gpu_output,c_gpu_output,byte_size,cudaMemcpyDeviceToHost); bool test = checkResult(c_output,gpu_output,size); if(test==true){ printf("the result is true\n"); }else{ printf("the result is false\n"); } cudaFree(a_gpu_input); cudaFree(b_gpu_input); cudaFree(c_gpu_output); free(a_input); free(b_input); free(c_output); cudaDeviceReset(); return 0; }
10,864
#include "includes.h" __device__ float activateRandomly(float probability, float random) { return random < probability; } __global__ void RBMRandomActivationKernel( float *outputPtr, float *randomPtr, int size ) { int i = blockDim.x * blockIdx.y * gridDim.x //rows preceeding current row in grid + blockDim.x * blockIdx.x //blocks preceeding current block + threadIdx.x; if (i < size) { outputPtr[i] = activateRandomly(outputPtr[i], randomPtr[i]); } }
10,865
#include <stdbool.h> #define BLOCK_DIM 32 #define N_BLOCKS (48 * 2) // bug?? #define N_BLOCKS (48 * 4) #define N_WORKERS (N_BLOCKS * BLOCK_DIM) #define N_INIT_DISTRIBUTION (N_WORKERS * 4) #define N_INPUTS (N_WORKERS * 8) #define PLAN_LEN_MAX 255 #define BLACKLIST_BITS 10 #define STATE_WIDTH 4 #define STATE_N (STATE_WIDTH * STATE_WIDTH) typedef unsigned char uchar; typedef signed char Direction; #define dir_reverse(dir) ((Direction)(3 - (dir))) #define DIR_N 4 #define DIR_FIRST 0 #define DIR_UP 0 #define DIR_RIGHT 1 #define DIR_LEFT 2 #define DIR_DOWN 3 /* stack implementation */ __device__ __shared__ static struct dir_stack_tag { uchar i, j; int init_depth; uchar buf[PLAN_LEN_MAX]; } stack[BLOCK_DIM]; #define STACK (stack[threadIdx.x]) typedef struct search_stat_tag { bool solved; int len; long long nodes_expanded; int nodes_pruned; } search_stat; typedef struct input_tag { uchar tiles[STATE_N]; int init_depth; Direction parent_dir; } Input; __host__ static inline long long korf_hash(uchar tiles[]) { long long h = tiles[0]; for (int i = 1; i < STATE_N; ++i) h += h*3 + tiles[i]; return h; } __device__ static inline long long korf_hash_dev(uchar tiles[]) { long long h = tiles[0]; for (long long i = 1; i < STATE_N; ++i) h += h*3 + tiles[i]; return h; } typedef struct blacklist_entry_tag { uchar tiles[STATE_N]; long long hash; int g_value; } BlackListEntry; __shared__ BlackListEntry blacklist_dev[1 << BLACKLIST_BITS]; __device__ static inline void stack_init(Input input) { STACK.i = 0; STACK.j = 0; STACK.init_depth = input.init_depth; } __device__ static inline void stack_put(Direction dir) { STACK.buf[STACK.i] = dir; ++STACK.i; } __device__ static inline bool stack_is_empty(void) { return STACK.i == STACK.j; } __device__ static inline Direction stack_pop(void) { --STACK.i; return STACK.buf[STACK.i]; } __device__ static inline Direction stack_peak(void) { return STACK.buf[STACK.i - 1]; } /* state implementation */ static char assert_state_width_is_four[STATE_WIDTH == 4 ? 1 : -1]; #define POS_X(pos) ((pos) &3) #define POS_Y(pos) ((pos) >> 2) /* * goal: [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15] */ __device__ __shared__ static struct state_tag { uchar tile[STATE_N]; uchar empty; uchar h_value; /* ub of h_value is 6*16 */ } state[BLOCK_DIM]; #define STATE_TILE(i) (state[threadIdx.x].tile[(i)]) #define STATE_EMPTY (state[threadIdx.x].empty) #define STATE_HVALUE (state[threadIdx.x].h_value) #define distance(i, j) ((i) > (j) ? (i) - (j) : (j) - (i)) #define H_DIFF(opponent, empty, empty_dir) \ h_diff_table_shared[opponent][empty][empty_dir] __device__ __shared__ static signed char h_diff_table_shared[STATE_N][STATE_N] [DIR_N]; __device__ static void state_init_hvalue(void) { uchar from_x[STATE_N], from_y[STATE_N]; STATE_HVALUE = 0; for (int i = 0; i < STATE_N; ++i) { from_x[STATE_TILE(i)] = POS_X(i); from_y[STATE_TILE(i)] = POS_Y(i); } for (int i = 1; i < STATE_N; ++i) { STATE_HVALUE += distance(from_x[i], POS_X(i)); STATE_HVALUE += distance(from_y[i], POS_Y(i)); } } __device__ static void state_tile_fill(Input input) { for (int i = 0; i < STATE_N; ++i) { if (input.tiles[i] == 0) STATE_EMPTY = i; STATE_TILE(i) = input.tiles[i]; } } __device__ static inline bool state_is_goal(void) { return STATE_HVALUE == 0; } __device__ static char assert_direction2 [DIR_UP == 0 && DIR_RIGHT == 1 && DIR_LEFT == 2 && DIR_DOWN == 3 ? 1 : -1]; __device__ __shared__ static bool movable_table_shared[STATE_N][DIR_N]; __device__ static inline bool state_movable(Direction dir) { return movable_table_shared[STATE_EMPTY][dir]; } __device__ static char assert_direction [DIR_UP == 0 && DIR_RIGHT == 1 && DIR_LEFT == 2 && DIR_DOWN == 3 ? 1 : -1]; __device__ __constant__ const static int pos_diff_table[DIR_N] = { -STATE_WIDTH, 1, -1, +STATE_WIDTH}; __device__ static inline bool state_move_with_limit(Direction dir, unsigned int f_limit) { int new_empty = STATE_EMPTY + pos_diff_table[dir]; int opponent = STATE_TILE(new_empty); int new_h_value = STATE_HVALUE + H_DIFF(opponent, new_empty, dir); if (STACK.i + STACK.init_depth + 1 + new_h_value > f_limit) return false; STATE_HVALUE = new_h_value; STATE_TILE(STATE_EMPTY) = opponent; STATE_EMPTY = new_empty; return true; } __device__ static inline void state_move(Direction dir) { int new_empty = STATE_EMPTY + pos_diff_table[dir]; int opponent = STATE_TILE(new_empty); STATE_HVALUE += H_DIFF(opponent, new_empty, dir); STATE_TILE(STATE_EMPTY) = opponent; STATE_EMPTY = new_empty; } __device__ static inline bool blacklist_query(int *g_value) { long long hash = korf_hash_dev(state[threadIdx.x].tile); BlackListEntry e = blacklist_dev[hash & ((1u<<BLACKLIST_BITS) - 1)]; if (e.hash != hash) return false; for (int i = 0; i < STATE_N; ++i) if (e.tiles[i] != STATE_TILE(i)) return false; return true; } /* * solver implementation */ __device__ static void idas_internal(int f_limit, Input *input, int *input_ends, search_stat *stat) { int tid = threadIdx.x; int bid = blockIdx.x; int id = tid + bid * blockDim.x; for (int i = id == 0 ? 0 : input_ends[id - 1]; i < input_ends[id]; ++i) { long long nodes_expanded = 0; int nodes_pruned = 0; uchar dir = 0; int blacklist_g; Input this_input = input[i]; stack_init(this_input); state_tile_fill(this_input); state_init_hvalue(); for (;;) { if (state_is_goal()) asm("trap;"); /* solution found */ /* if continue search until solution found, just return true */ if (((stack_is_empty() && dir_reverse(dir) != this_input.parent_dir) || stack_peak() != dir_reverse(dir)) && state_movable(dir)) { ++nodes_expanded; /* sometimes check idle here */ /* and load balance if needed */ if (state_move_with_limit(dir, f_limit)) { if (!(blacklist_query(&blacklist_g) && blacklist_g <= this_input.init_depth + STACK.i)) { stack_put(dir); dir = 0; continue; } else ++nodes_pruned; } } while (++dir == DIR_N) { if (stack_is_empty()) goto END_THIS_NODE; dir = stack_pop(); state_move(dir_reverse(dir)); } } END_THIS_NODE: stat[i].nodes_expanded = nodes_expanded; stat[i].nodes_pruned = nodes_pruned; /* if my own works have done, get other's work here */ /* if every work finished, then return*/ } } __global__ void idas_kernel(Input *input, int *input_ends, signed char *plan, BlackListEntry *blacklist, search_stat *stat, int f_limit, signed char *h_diff_table, bool *movable_table) { int tid = threadIdx.x; for (int dir = 0; dir < DIR_N; ++dir) for (int i = tid; i < STATE_N; i += blockDim.x) if (i < STATE_N) movable_table_shared[i][dir] = movable_table[i * DIR_N + dir]; for (int i = 0; i < STATE_N * DIR_N; ++i) for (int j = tid; j < STATE_N; j += blockDim.x) if (j < STATE_N) h_diff_table_shared[j][i / DIR_N][i % DIR_N] = h_diff_table[j * STATE_N * DIR_N + i]; for (int i = tid; i < 1<<BLACKLIST_BITS; i += blockDim.x) blacklist_dev[i] = blacklist[i]; __syncthreads(); idas_internal(f_limit, input, input_ends, stat); } /* host library implementation */ #include <errno.h> #include <limits.h> #include <stddef.h> #include <stdio.h> #include <stdlib.h> #ifndef UNABLE_LOG #define elog(...) fprintf(stderr, __VA_ARGS__) #else #define elog(...) ; #endif void * palloc(size_t size) { void *ptr = malloc(size); if (!ptr) elog("malloc failed\n"); return ptr; } void * repalloc(void *old_ptr, size_t new_size) { void *ptr = realloc(old_ptr, new_size); if (!ptr) elog("realloc failed\n"); return ptr; } void pfree(void *ptr) { if (!ptr) elog("empty ptr\n"); free(ptr); } #include <assert.h> #include <stdbool.h> #include <stdlib.h> #include <string.h> typedef unsigned char idx_t; /* * [0,0] [1,0] [2,0] [3,0] * [0,1] [1,1] [2,1] [3,1] * [0,2] [1,2] [2,2] [3,2] * [0,3] [1,3] [2,3] [3,3] */ /* * goal state is * [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15] */ typedef struct state_tag_cpu { int depth; /* XXX: needed? */ uchar pos[STATE_WIDTH][STATE_WIDTH]; idx_t i, j; /* pos of empty */ Direction parent_dir; int h_value; } * State; #define v(state, i, j) ((state)->pos[i][j]) #define ev(state) (v(state, state->i, state->j)) #define lv(state) (v(state, state->i - 1, state->j)) #define dv(state) (v(state, state->i, state->j + 1)) #define rv(state) (v(state, state->i + 1, state->j)) #define uv(state) (v(state, state->i, state->j - 1)) static uchar from_x[STATE_WIDTH * STATE_WIDTH], from_y[STATE_WIDTH * STATE_WIDTH]; static inline void fill_from_xy(State from) { for (idx_t x = 0; x < STATE_WIDTH; ++x) for (idx_t y = 0; y < STATE_WIDTH; ++y) { from_x[v(from, x, y)] = x; from_y[v(from, x, y)] = y; } } static char assert_state_width_is_four2[STATE_WIDTH == 4 ? 1 : -1]; static inline int heuristic_manhattan_distance(State from) { int h_value = 0; fill_from_xy(from); for (idx_t i = 1; i < STATE_N; ++i) { h_value += distance(from_x[i], i & 3); h_value += distance(from_y[i], i >> 2); } return h_value; } bool state_is_goal(State state) { return state->h_value == 0; } static inline State state_alloc(void) { return (State) palloc(sizeof(struct state_tag_cpu)); } static inline void state_free(State state) { pfree(state); } State state_init(uchar v_list[STATE_WIDTH * STATE_WIDTH], int init_depth) { State state = state_alloc(); int cnt = 0; state->depth = init_depth; state->parent_dir = (Direction) -1; for (idx_t j = 0; j < STATE_WIDTH; ++j) for (idx_t i = 0; i < STATE_WIDTH; ++i) { if (v_list[cnt] == 0) { state->i = i; state->j = j; } v(state, i, j) = v_list[cnt++]; } state->h_value = heuristic_manhattan_distance(state); return state; } void state_fini(State state) { state_free(state); } State state_copy(State src) { State dst = state_alloc(); memcpy(dst, src, sizeof(*src)); return dst; } static inline bool state_left_movable(State state) { return state->i != 0; } static inline bool state_down_movable(State state) { return state->j != STATE_WIDTH - 1; } static inline bool state_right_movable(State state) { return state->i != STATE_WIDTH - 1; } static inline bool state_up_movable(State state) { return state->j != 0; } bool state_movable(State state, Direction dir) { return (dir != DIR_LEFT || state_left_movable(state)) && (dir != DIR_DOWN || state_down_movable(state)) && (dir != DIR_RIGHT || state_right_movable(state)) && (dir != DIR_UP || state_up_movable(state)); } #define h_diff(who, from_i, from_j, dir) \ (h_diff_table[((who) << 6) + ((from_j) << 4) + ((from_i) << 2) + (dir)]) static int h_diff_table[STATE_N * STATE_N * DIR_N] = { 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, 1, 1, -1, 1, -1, 1, 1, 1, -1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1}; void state_move(State state, Direction dir) { idx_t who; assert(state_movable(state, dir)); switch (dir) { case DIR_LEFT: who = ev(state) = lv(state); state->i--; break; case DIR_DOWN: who = ev(state) = dv(state); state->j++; break; case DIR_RIGHT: who = ev(state) = rv(state); state->i++; break; case DIR_UP: who = ev(state) = uv(state); state->j--; break; default: elog("unexpected direction"); assert(false); } state->h_value = state->h_value + h_diff(who, state->i, state->j, dir_reverse(dir)); state->parent_dir = dir; } bool state_pos_equal(State s1, State s2) { for (idx_t i = 0; i < STATE_WIDTH; ++i) for (idx_t j = 0; j < STATE_WIDTH; ++j) if (v(s1, i, j) != v(s2, i, j)) return false; return true; } size_t state_hash(State state) { /* FIXME: for A* */ size_t hash_value = 0; for (idx_t i = 0; i < STATE_WIDTH; ++i) for (idx_t j = 0; j < STATE_WIDTH; ++j) hash_value ^= (v(state, i, j) << ((i * 3 + j) << 2)); return hash_value; } int state_get_hvalue(State state) { return state->h_value; } int state_get_depth(State state) { return state->depth; } #include <stddef.h> #include <stdint.h> #include <string.h> #ifndef SIZE_MAX #define SIZE_MAX ((size_t) -1) #endif typedef enum { HT_SUCCESS = 0, HT_FAILED_FOUND, HT_FAILED_NOT_FOUND, } HTStatus; /* XXX: hash function for State should be surveyed */ inline static size_t hashfunc(State key) { return state_hash(key); } typedef struct ht_entry_tag *HTEntry; struct ht_entry_tag { HTEntry next; State key; int value; }; static HTEntry ht_entry_init(State key) { HTEntry entry = (HTEntry) palloc(sizeof(*entry)); entry->key = state_copy(key); entry->next = NULL; return entry; } static void ht_entry_fini(HTEntry entry) { pfree(entry); } typedef struct ht_tag { size_t n_bins; size_t n_elems; HTEntry *bin; } * HT; static bool ht_rehash_required(HT ht) { return ht->n_bins <= ht->n_elems; /* TODO: local policy is also needed */ } static size_t calc_n_bins(size_t required) { /* NOTE: n_bins is used for mask and hence it should be pow of 2, fon now */ size_t size = 1; assert(required > 0); while (required > size) size <<= 1; return size; } HT ht_init(size_t init_size_hint) { size_t n_bins = calc_n_bins(init_size_hint); HT ht = (HT) palloc(sizeof(*ht)); ht->n_bins = n_bins; ht->n_elems = 0; assert(sizeof(*ht->bin) <= SIZE_MAX / n_bins); ht->bin = (HTEntry *) palloc(sizeof(*ht->bin) * n_bins); memset(ht->bin, 0, sizeof(*ht->bin) * n_bins); return ht; } static void ht_rehash(HT ht) { HTEntry *new_bin; size_t new_size = ht->n_bins << 1; assert(ht->n_bins<SIZE_MAX>> 1); new_bin = (HTEntry *) palloc(sizeof(*new_bin) * new_size); memset(new_bin, 0, sizeof(*new_bin) * new_size); for (size_t i = 0; i < ht->n_bins; ++i) { HTEntry entry = ht->bin[i]; while (entry) { HTEntry next = entry->next; size_t idx = hashfunc(entry->key) & (new_size - 1); entry->next = new_bin[idx]; new_bin[idx] = entry; entry = next; } } pfree(ht->bin); ht->n_bins = new_size; ht->bin = new_bin; } void ht_fini(HT ht) { for (size_t i = 0; i < ht->n_bins; ++i) { HTEntry entry = ht->bin[i]; while (entry) { HTEntry next = entry->next; state_fini(entry->key); ht_entry_fini(entry); entry = next; } } pfree(ht->bin); pfree(ht); } HTStatus ht_insert(HT ht, State key, int **value) { size_t i; HTEntry entry, new_entry; if (ht_rehash_required(ht)) ht_rehash(ht); i = hashfunc(key) & (ht->n_bins - 1); entry = ht->bin[i]; while (entry) { if (state_pos_equal(key, entry->key)) { *value = &entry->value; return HT_FAILED_FOUND; } entry = entry->next; } new_entry = ht_entry_init(key); new_entry->next = ht->bin[i]; ht->bin[i] = new_entry; *value = &new_entry->value; assert(ht->n_elems < SIZE_MAX); ht->n_elems++; return HT_SUCCESS; } /* * Priority Queue implementation */ #include <assert.h> #include <stdint.h> typedef struct pq_entry_tag { State state; int f, g; } PQEntryData; typedef PQEntryData *PQEntry; /* tiebreaking is done comparing g value */ static inline bool pq_entry_higher_priority(PQEntry e1, PQEntry e2) { return e1->f < e2->f || (e1->f == e2->f && e1->g >= e2->g); } /* * NOTE: * This priority queue is implemented doubly reallocated array. * It will only extend and will not shrink, for now. * It may be improved by using array of layers of iteratively widened array */ typedef struct pq_tag { size_t n_elems; size_t capa; PQEntryData *array; } * PQ; static inline size_t calc_init_capa(size_t capa_hint) { size_t capa = 1; assert(capa_hint > 0); while (capa < capa_hint) capa <<= 1; return capa - 1; } PQ pq_init(size_t init_capa_hint) { PQ pq = (PQ) palloc(sizeof(*pq)); pq->n_elems = 0; pq->capa = calc_init_capa(init_capa_hint); assert(pq->capa <= SIZE_MAX / sizeof(PQEntryData)); pq->array = (PQEntryData *) palloc(sizeof(PQEntryData) * pq->capa); return pq; } void pq_fini(PQ pq) { for (size_t i = 0; i < pq->n_elems; ++i) state_fini(pq->array[i].state); pfree(pq->array); pfree(pq); } static inline bool pq_is_full(PQ pq) { assert(pq->n_elems <= pq->capa); return pq->n_elems == pq->capa; } static inline void pq_extend(PQ pq) { pq->capa = (pq->capa << 1) + 1; assert(pq->capa <= SIZE_MAX / sizeof(PQEntryData)); pq->array = (PQEntryData *) repalloc(pq->array, sizeof(PQEntryData) * pq->capa); } static inline void pq_swap_entry(PQ pq, size_t i, size_t j) { PQEntryData tmp = pq->array[i]; pq->array[i] = pq->array[j]; pq->array[j] = tmp; } static inline size_t pq_up(size_t i) { /* NOTE: By using 1-origin, it may be written more simply, i >> 1 */ return (i - 1) >> 1; } static inline size_t pq_left(size_t i) { return (i << 1) + 1; } static void heapify_up(PQ pq) { for (size_t i = pq->n_elems; i > 0;) { size_t ui = pq_up(i); assert(i > 0); if (!pq_entry_higher_priority(&pq->array[i], &pq->array[ui])) break; pq_swap_entry(pq, i, ui); i = ui; } } void pq_put(PQ pq, State state, int f, int g) { if (pq_is_full(pq)) pq_extend(pq); pq->array[pq->n_elems].state = state_copy(state); pq->array[pq->n_elems].f = f; /* this may be abundant */ pq->array[pq->n_elems].g = g; heapify_up(pq); ++pq->n_elems; } static void heapify_down(PQ pq) { size_t sentinel = pq->n_elems; for (size_t i = 0;;) { size_t ri, li = pq_left(i); if (li >= sentinel) break; ri = li + 1; if (ri >= sentinel) { if (pq_entry_higher_priority(&pq->array[li], &pq->array[i])) pq_swap_entry(pq, i, li); /* Reached the bottom */ break; } /* NOTE: If p(ri) == p(li), it may be good to go right * since the filling order is left-first */ if (pq_entry_higher_priority(&pq->array[li], &pq->array[ri])) { if (!pq_entry_higher_priority(&pq->array[li], &pq->array[i])) break; pq_swap_entry(pq, i, li); i = li; } else { if (!pq_entry_higher_priority(&pq->array[ri], &pq->array[i])) break; pq_swap_entry(pq, i, ri); i = ri; } } } State pq_pop(PQ pq) { State ret_state; if (pq->n_elems == 0) return NULL; ret_state = pq->array[0].state; --pq->n_elems; pq->array[0] = pq->array[pq->n_elems]; heapify_down(pq); return ret_state; } void pq_dump(PQ pq) { elog("%s: n_elems=%zu, capa=%zu\n", __func__, pq->n_elems, pq->capa); for (size_t i = 0, cr_required = 1; i < pq->n_elems; i++) { if (i == cr_required) { elog("\n"); cr_required = (cr_required << 1) + 1; } elog("%d,", pq->array[i].f); elog("%d ", pq->array[i].g); } elog("\n"); } static HT closed, BL1, BL2; bool distribute_astar(State init_state, Input input[], int input_ends[], int distr_n, int *cnt_inputs, int *min_fvalue) { int cnt = 0; State state; PQ q = pq_init(distr_n + 10); HTStatus ht_status; int * ht_value; bool solved = false; closed = ht_init(10000); BL1 = ht_init(1000); BL2 = ht_init(1000); ht_status = ht_insert(closed, init_state, &ht_value); *ht_value = 0; pq_put(q, state_copy(init_state), state_get_hvalue(init_state), 0); ++cnt; while ((state = pq_pop(q))) { --cnt; if (state_is_goal(state)) { solved = true; break; } ht_status = ht_insert(closed, state, &ht_value); if (ht_status == HT_FAILED_FOUND && *ht_value < state_get_depth(state)) { state_fini(state); continue; } else *ht_value = state_get_depth(state); for (int dir = 0; dir < DIR_N; ++dir) { if (state->parent_dir != dir_reverse(dir) && state_movable(state, (Direction) dir)) { State next_state = state_copy(state); state_move(next_state, (Direction) dir); next_state->depth++; ht_status = ht_insert(closed, next_state, &ht_value); if (ht_status == HT_FAILED_FOUND && *ht_value <= state_get_depth(next_state)) state_fini(next_state); else { ++cnt; *ht_value = state_get_depth(next_state); pq_put(q, next_state, *ht_value + state_get_hvalue(next_state), *ht_value); } } } state_fini(state); if (cnt >= distr_n) break; } *cnt_inputs = cnt; if (!solved) { int minf = INT_MAX; for (int id = 0; id < cnt; ++id) { State state = pq_pop(q); assert(state); for (int i = 0; i < STATE_N; ++i) input[id].tiles[i] = state->pos[i%STATE_WIDTH][i/STATE_WIDTH]; input[id].tiles[state->i + (state->j * STATE_WIDTH)] = 0; input[id].init_depth = state_get_depth(state); input[id].parent_dir = state->parent_dir; if (minf > state_get_depth(state) + state_get_hvalue(state)) minf = state_get_depth(state) + state_get_hvalue(state); } *min_fvalue = minf; printf("distr_n=%d, n_worers=%d, cnt=%d\n", distr_n, N_WORKERS, cnt); for (int id = 0; id < N_WORKERS; ++id) input_ends[id] = (distr_n / N_WORKERS) * (id + 1) - 1; input_ends[N_WORKERS - 1] = cnt; } pq_fini(q); return solved; } static int input_devide(Input input[], search_stat stat[], int i, int devide_n, int tail) { int cnt = 0; int * ht_value; State state = state_init(input[i].tiles, input[i].init_depth); state->parent_dir = input[i].parent_dir; PQ pq = pq_init(32); HTStatus ht_status = ht_insert(BL1, state, &ht_value); if (ht_status == HT_FAILED_FOUND && *ht_value > state_get_depth(state)) *ht_value = state_get_depth(state); pq_put(pq, state, state_get_hvalue(state), 0); ++cnt; while ((state = pq_pop(pq))) { --cnt; if (state_is_goal(state)) { /* It may not be optimal goal */ pq_put(pq, state, state_get_depth(state) + state_get_hvalue(state), state_get_depth(state)); ++cnt; break; } ht_status = ht_insert(closed, state, &ht_value); if (ht_status == HT_FAILED_FOUND && *ht_value < state_get_depth(state)) { state_fini(state); continue; } else *ht_value = state_get_depth(state); for (int dir = 0; dir < DIR_N; ++dir) { if (state->parent_dir != dir_reverse(dir) && state_movable(state, (Direction) dir)) { State next_state = state_copy(state); state_move(next_state, (Direction) dir); next_state->depth++; ht_status = ht_insert(closed, next_state, &ht_value); if (ht_status == HT_FAILED_FOUND && *ht_value < state_get_depth(next_state)) state_fini(next_state); else { ++cnt; *ht_value = state_get_depth(next_state); pq_put(pq, next_state, *ht_value + state_get_hvalue(next_state), *ht_value); } } } state_fini(state); if (cnt >= devide_n) break; } for (int id = 0; id < cnt; ++id) { int estimation_after_devision = stat[i].nodes_expanded / cnt; int ofs = id == 0 ? i : tail - 1 + id; State state = pq_pop(pq); assert(state); ht_status = ht_insert(BL2, state, &ht_value); if (ht_status == HT_FAILED_FOUND && *ht_value > state_get_depth(state)) *ht_value = state_get_depth(state); for (int j = 0; j < STATE_N; ++j) input[ofs].tiles[j] = state->pos[j%STATE_WIDTH][j/STATE_WIDTH]; input[ofs].tiles[state->i + (state->j * STATE_WIDTH)] = 0; input[ofs].init_depth = state_get_depth(state); input[ofs].parent_dir = state->parent_dir; stat[ofs].nodes_expanded = estimation_after_devision; } pq_fini(pq); return cnt == 0 ? 0 : cnt - 1; } static void fill_blacklist_internal(BlackListEntry blacklist[], HT bl) { for (unsigned int i = 0; i < bl->n_bins; ++i) { HTEntry e = bl->bin[i]; while (e) { HTEntry next = e->next; State s = e->key; uchar til[STATE_N]; for (int j = 0; j < STATE_N; ++j) til[j] = s->pos[j%STATE_WIDTH][j/STATE_WIDTH]; int hash = korf_hash(til); BlackListEntry *ble = &blacklist[hash & (1u << BLACKLIST_BITS)]; ble->hash = hash; for (int p = 0; p < STATE_N; ++p) ble->tiles[p] = til[p]; ble->g_value = state_get_depth(s); e = next; } } } static void fill_blacklist(BlackListEntry blacklist[]) { fill_blacklist_internal(blacklist, BL2); fill_blacklist_internal(blacklist, BL1); } /* main */ #include <errno.h> #include <stdio.h> #include <stdlib.h> #define exit_failure(...) \ do \ { \ printf(__VA_ARGS__); \ exit(EXIT_FAILURE); \ } while (0) static int pop_int_from_str(const char *str, char **end_ptr) { long int rv = strtol(str, end_ptr, 0); errno = 0; if (errno != 0) exit_failure("%s: %s cannot be converted into long\n", __func__, str); else if (end_ptr && str == *end_ptr) exit_failure("%s: reach end of string", __func__); if (rv > INT_MAX || rv < INT_MIN) exit_failure("%s: too big number, %ld\n", __func__, rv); return (int) rv; } #define MAX_LINE_LEN 100 static void load_state_from_file(const char *fname, uchar *s) { FILE *fp; char str[MAX_LINE_LEN]; char *str_ptr = str, *end_ptr; fp = fopen(fname, "r"); if (!fp) exit_failure("%s: %s cannot be opened\n", __func__, fname); if (!fgets(str, MAX_LINE_LEN, fp)) exit_failure("%s: fgets failed\n", __func__); for (int i = 0; i < STATE_N; ++i) { s[i] = pop_int_from_str(str_ptr, &end_ptr); str_ptr = end_ptr; } fclose(fp); } #undef MAX_LINE_LEN #define CUDA_CHECK(call) \ do \ { \ const cudaError_t e = call; \ if (e != cudaSuccess) \ exit_failure("Error: %s:%d code:%d, reason: %s\n", __FILE__, \ __LINE__, e, cudaGetErrorString(e)); \ } while (0) #define h_d_t(op, i, dir) \ (h_diff_table[(op) *STATE_N * DIR_N + (i) *DIR_N + (dir)]) __host__ static void init_mdist(signed char h_diff_table[]) { for (int opponent = 0; opponent < STATE_N; ++opponent) { int goal_x = POS_X(opponent), goal_y = POS_Y(opponent); for (int i = 0; i < STATE_N; ++i) { int from_x = POS_X(i), from_y = POS_Y(i); for (uchar dir = 0; dir < DIR_N; ++dir) { if (dir == DIR_LEFT) h_d_t(opponent, i, dir) = goal_x > from_x ? -1 : 1; if (dir == DIR_RIGHT) h_d_t(opponent, i, dir) = goal_x < from_x ? -1 : 1; if (dir == DIR_UP) h_d_t(opponent, i, dir) = goal_y > from_y ? -1 : 1; if (dir == DIR_DOWN) h_d_t(opponent, i, dir) = goal_y < from_y ? -1 : 1; } } } } #undef h_d_t #define m_t(i, d) (movable_table[(i) *DIR_N + (d)]) __host__ static void init_movable_table(bool movable_table[]) { for (int i = 0; i < STATE_N; ++i) for (unsigned int d = 0; d < DIR_N; ++d) { if (d == DIR_RIGHT) m_t(i, d) = (POS_X(i) < STATE_WIDTH - 1); else if (d == DIR_LEFT) m_t(i, d) = (POS_X(i) > 0); else if (d == DIR_DOWN) m_t(i, d) = (POS_Y(i) < STATE_WIDTH - 1); else if (d == DIR_UP) m_t(i, d) = (POS_Y(i) > 0); } } #undef m_t static void avoid_unused_static_assertions(void) { (void) assert_direction[0]; (void) assert_direction2[0]; (void) assert_state_width_is_four[0]; (void) assert_state_width_is_four2[0]; } static char dir_char[] = {'U', 'R', 'L', 'D'}; int main(int argc, char *argv[]) { int cnt_inputs; int input_size = sizeof(Input) * N_INPUTS; Input input[N_INPUTS]; Input *d_input; int input_ends_size = sizeof(int) * N_WORKERS; int input_ends[N_WORKERS]; int *d_input_ends; int plan_size = sizeof(signed char) * PLAN_LEN_MAX * N_INPUTS; signed char plan[PLAN_LEN_MAX * N_INPUTS]; signed char *d_plan; int stat_size = sizeof(search_stat) * N_INPUTS; search_stat stat[N_INPUTS]; search_stat *d_stat; int blacklist_size = sizeof(BlackListEntry) * (1 << BLACKLIST_BITS); BlackListEntry h_blacklist[N_INPUTS]; BlackListEntry *d_blacklist; bool movable_table[STATE_N * DIR_N]; bool * d_movable_table; int movable_table_size = sizeof(bool) * STATE_N * DIR_N; signed char h_diff_table[STATE_N * STATE_N * DIR_N]; signed char *d_h_diff_table; int h_diff_table_size = sizeof(signed char) * STATE_N * STATE_N * DIR_N; int min_fvalue = 0; if (argc < 2) { printf("usage: bin/cumain <ifname>\n"); exit(EXIT_FAILURE); } load_state_from_file(argv[1], input[0].tiles); { State init_state = state_init(input[0].tiles, 0); if (distribute_astar(init_state, input, input_ends, N_INIT_DISTRIBUTION, &cnt_inputs, &min_fvalue)) { puts("solution is found by distributor"); return 0; } } init_mdist(h_diff_table); init_movable_table(movable_table); CUDA_CHECK(cudaMalloc((void **) &d_input, input_size)); CUDA_CHECK(cudaMalloc((void **) &d_input_ends, input_ends_size)); CUDA_CHECK(cudaMalloc((void **) &d_plan, plan_size)); CUDA_CHECK(cudaMalloc((void **) &d_stat, stat_size)); CUDA_CHECK(cudaMalloc((void **) &d_blacklist, blacklist_size)); CUDA_CHECK(cudaMalloc((void **) &d_movable_table, movable_table_size)); CUDA_CHECK(cudaMalloc((void **) &d_h_diff_table, h_diff_table_size)); CUDA_CHECK(cudaMemcpy(d_movable_table, movable_table, movable_table_size, cudaMemcpyHostToDevice)); CUDA_CHECK(cudaMemcpy(d_h_diff_table, h_diff_table, h_diff_table_size, cudaMemcpyHostToDevice)); CUDA_CHECK(cudaMemset(d_input, 0, input_size)); CUDA_CHECK(cudaMemset(d_plan, 0, plan_size)); CUDA_CHECK(cudaMemset(d_stat, 0, stat_size)); CUDA_CHECK(cudaMemset(d_blacklist, 0, blacklist_size)); for (uchar f_limit = min_fvalue;; f_limit += 2) { elog("f=%d\n", (int) f_limit); CUDA_CHECK( cudaMemcpy(d_input, input, input_size, cudaMemcpyHostToDevice)); CUDA_CHECK(cudaMemcpy(d_input_ends, input_ends, input_ends_size, cudaMemcpyHostToDevice)); fill_blacklist(h_blacklist); CUDA_CHECK(cudaMemcpy(d_blacklist, h_blacklist, blacklist_size, cudaMemcpyHostToDevice)); elog("BL1: %zu, BL2: %zu\n", BL1->n_elems, BL2->n_elems); elog("kernel(block=%d, thread=%d)\n", N_BLOCKS, BLOCK_DIM); idas_kernel<<<N_BLOCKS, BLOCK_DIM>>>(d_input, d_input_ends, d_plan, d_blacklist, d_stat, f_limit, d_h_diff_table, d_movable_table); CUDA_CHECK(cudaGetLastError()); CUDA_CHECK(cudaMemcpy(plan, d_plan, plan_size, cudaMemcpyDeviceToHost)); CUDA_CHECK(cudaMemcpy(stat, d_stat, stat_size, cudaMemcpyDeviceToHost)); for (int i = 0; i < cnt_inputs; ++i) if (stat[i].solved) { elog("core id = %d\n", i); printf("cpu len=%d: \n", input[i].init_depth); /* CPU side output */ // FIXME: Not implemented, for now. It is easy to search path // from init state to this root. /* GPU side output */ printf("gpu len=%d: ", stat[i].len); for (int j = 0; j < stat[i].len; ++j) printf("%c ", dir_char[(int) plan[i * PLAN_LEN_MAX + j]]); putchar('\n'); goto solution_found; } long long int sum_of_expansion = 0; for (int i = 0; i < cnt_inputs; ++i) sum_of_expansion += stat[i].nodes_expanded; long long int sum_of_pruned = 0; for (int i = 0; i < cnt_inputs; ++i) sum_of_pruned += stat[i].nodes_pruned; long long int increased = 0; long long int avarage_expected_load = sum_of_expansion / N_WORKERS; int stat_cnt[10] = {0, 0, 0, 0, 0, 0, 0}; for (int i = 0; i < cnt_inputs; ++i) { if (stat[i].nodes_expanded < avarage_expected_load) stat_cnt[0]++; else if (stat[i].nodes_expanded < 2 * avarage_expected_load) stat_cnt[1]++; else if (stat[i].nodes_expanded < 4 * avarage_expected_load) stat_cnt[2]++; else if (stat[i].nodes_expanded < 8 * avarage_expected_load) stat_cnt[3]++; else if (stat[i].nodes_expanded < 16 * avarage_expected_load) stat_cnt[4]++; else if (stat[i].nodes_expanded < 32 * avarage_expected_load) stat_cnt[5]++; else stat_cnt[6]++; int policy = stat[i].nodes_expanded / (avarage_expected_load + 1) + 1; if (policy > 1 && stat[i].nodes_expanded > 20) increased += input_devide(input, stat, i, policy, cnt_inputs + increased); } elog("STAT: sum of expanded nodes: %lld\n", sum_of_expansion); elog("STAT: avarage expanded nodes: %lld\n", avarage_expected_load); elog("STAT: av=%d, 2av=%d, 4av=%d, 8av=%d, 16av=%d, 32av=%d, more=%d\n", stat_cnt[0], stat_cnt[1], stat_cnt[2], stat_cnt[3], stat_cnt[4], stat_cnt[5], stat_cnt[6]); elog("STAT: sum of pruned nodes: %lld\n", sum_of_pruned); if (cnt_inputs + increased > N_INPUTS) { elog("cnt_inputs too large"); abort(); } cnt_inputs += increased; elog("input count: %d\n", cnt_inputs); /* NOTE: optionally sort here by expected cost or g/h-value */ int id = 0; for (int i = 0, load = 0; i < cnt_inputs; ++i) { load += stat[i].nodes_expanded; if (load >= avarage_expected_load) { load = 0; input_ends[id++] = i; } } while (id < N_WORKERS) input_ends[id++] = cnt_inputs; } solution_found: CUDA_CHECK(cudaFree(d_input)); CUDA_CHECK(cudaFree(d_input_ends)); CUDA_CHECK(cudaFree(d_plan)); CUDA_CHECK(cudaFree(d_stat)); CUDA_CHECK(cudaFree(d_movable_table)); CUDA_CHECK(cudaFree(d_h_diff_table)); CUDA_CHECK(cudaDeviceReset()); avoid_unused_static_assertions(); return 0; }
10,866
#define N 96 #include <stdio.h> __global__ void helloFromGPU(){ printf("HelloFrom GPU! %d %d\n", blockIdx.x, threadIdx.x); } int main() { printf("Hello World from CPU\n"); helloFromGPU<<<3, 32>>>(); cudaDeviceReset(); return 0; }
10,867
#include <stdio.h> /* The "__global__" tag tells nvcc that the function will execute on the device but will be called from the host. Notice that we must use pointers! */ __global__ void add_int( int *a, int *b, int *c){ *c = *a + *b; printf("blockIdx: %d\n",blockIdx.x); } // Main program int main(void){ //host memory != device memory, must allocate differently //device pointers point to GPU Memory //host pointers point to CPU memory int a, b, c; //host copies int *dev_a, *dev_b, *dev_c; //device copies int size = sizeof( int ); //size of an interger //allocate space on device cudaMalloc( (void**)&dev_a, size ); cudaMalloc( (void**)&dev_b, size ); cudaMalloc( (void**)&dev_c, size ); a = 2; //storing values in host b = 7; // now we need the values to be copied to the device cudaMemcpy( dev_a, &a, size, cudaMemcpyHostToDevice ); cudaMemcpy( dev_b, &b, size, cudaMemcpyHostToDevice ); // launch the add_int kernel on the GPU add_int<<<3,1>>>(dev_a, dev_b, dev_c); //now we want the values back on the CPU cudaMemcpy( &c, dev_c, size, cudaMemcpyDeviceToHost ); printf("C: %d\n",c); cudaFree(dev_a); cudaFree(dev_b); cudaFree(dev_c); // your basic hello world program printf("Hello, World!\n"); return 0; }
10,868
// kernel.cu // Alex Getz #include <cstddef> #include <stdexcept> #include <memory> #include <cuda_profiler_api.h> #include <fstream> #include <cuda.h> #include <curand.h> #include <curand_kernel.h> #include <cuda_runtime.h> #include <stdio.h> #include <stdlib.h> #include <assert.h> #include <algorithm> #include <iostream> #include <sstream> #include <cstring> #include <string> #include <random> #include <functional> #include <math.h> #include <algorithm> #include <array> #define MAX_ENTRIES 11897026 #define B_SIZE 2000 #define TPB 128 #define SHMEM_SIZE 1024 * sizeof(int) /////////////////////////////////////////////////////////////////////////////////////// inline cudaError_t checkCuda(cudaError_t result) { #if defined(DEBUG) || defined(_DEBUG) if(result!=cudaSuccess){ fprintf(stderr,"CUDA Runtime Error: %s\n", cudaGetErrorString(result)); assert(result==cudaSuccess); } #endif return result; } #define CUDA_CALL(x) do { if((x) != cudaSuccess) { \ printf("Error at %s:%d\n",__FILE__,__LINE__); \ return EXIT_FAILURE;}} /*while(0)*/ #define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__);} inline void gpuAssert(cudaError_t code, const char *file, int line, bool abort=true){ if (code != cudaSuccess){ fprintf(stderr,"GPUassert: %s %s %d\n", cudaGetErrorString(code), file, line); if(abort) exit(code); } } //////////////////////////////////////////////////////////////////////////////////////// /* /// DEVICE FUNCTIONS /// */ __device__ float getnextrand(curandState *state){ return (float)(curand_uniform(state)); } __device__ int getnextrandscaled(curandState *state, unsigned long int scale){ return (unsigned long int) scale * getnextrand(state); } /* /// DEVICE KERNELS /// */ __global__ void initCurand(curandState *state, unsigned long seed){ int idx = threadIdx.x + blockIdx.x * blockDim.x; curand_init(idx+seed, 0, 0, &state[idx]); } __global__ void bootstrap(double *output_mean, int *d_sample, curandState *state) { __shared__ int partial_Sums[SHMEM_SIZE]; unsigned int idx = threadIdx.x + (blockIdx.x*blockDim.x); unsigned int tNum = threadIdx.x; int bNum = blockIdx.x; int bSize = blockDim.x; unsigned long int ts = 0; long long int tSum = 0; int count = 0; for(unsigned int i=tNum; i<MAX_ENTRIES; i+=bSize){ ts = getnextrandscaled(&state[idx], MAX_ENTRIES); tSum += d_sample[ts]; count++; } partial_Sums[tNum] = tSum / count; __syncthreads(); // Perform Sum reduction across all the threads of the block for(int s=(bSize/2); s>0; s >>= 1){ // Each thread does work unless the index goes off the block if(tNum<s){ partial_Sums[tNum] += partial_Sums[tNum+s]; } __syncthreads(); } // Use first thread of each block to write results back to main mem if(tNum==0){ output_mean[bNum] = (double)partial_Sums[0] / (double)bSize; } } /* /// HOST GLOBAL FUNCTIONS /// */ void throw_error(cudaError_t err){ if(err != cudaSuccess) throw std::runtime_error(cudaGetErrorString(err)); } /* /// HOST GLOBAL STRUCTS & VARIABLES /// */ struct cuda_free_deleter_t{ void operator()(void* ptr) const { cudaFree(ptr); } }; template <typename T> auto cudaAllocBuffer(std::size_t size){ void *ptr; throw_error(cudaMalloc(&ptr, size*sizeof(T))); return std::unique_ptr<T, cuda_free_deleter_t> { static_cast<T*>(ptr) }; } int main(){ int *BaseSample, *d_Base; double *d_mean, *h_mean; //curandState *devStates; checkCuda( cudaMallocHost((void**)&BaseSample,MAX_ENTRIES*sizeof(int))); //checkCuda( cudaMalloc((void**)&devStates,2048*1024*sizeof(curandState))); std::string line; uintmax_t m_numLines = 0; std::ifstream fs("../data/allCountries.txt"); if(!fs){ std::cout<<"ERROR\n"; }else{ while (std::getline(fs, line)) { int counter=0; std::stringstream ss; std::string temp; // std::cout<<"\n"<<line<<"\n"; ss << line; std::getline(ss,temp,'\t'); // std::cout<<temp<<", position: "<<++counter<<"\n"; while(std::getline(ss,temp,'\t')){ if(temp.length() == 4){ BaseSample[m_numLines] = std::atoi(temp.c_str()); // std::cout<<temp<<", position: "<<++counter<<"\n"; break; } else{ ++counter; } } m_numLines++; // if(m_numLines==5){ break; } } } //std::cout << "m_numLines = " << m_numLines << "\nMoving on...\n\n"; fs.close(); //std::cout << "Element 300,000 of BaseSample: " << BaseSample[300000]<<std::endl; ///--- Calculating the Mean ---/// /////////////////////////////////////////////////////////////////////////// std::cout<<"Sample has been generated.\nCalculating the mean...\n"; long long int BaseSum = 0; for(int i=0;i<MAX_ENTRIES;i++){ BaseSum += BaseSample[i]; } double BaseMean = (double)BaseSum / (double)MAX_ENTRIES; std::cout<<"Mean has been calculated! Moving on...\n\n"; ///--- KERNEL OPERATIONS ---/// /////////////////////////////////////////////////////////////////////////// checkCuda( cudaMalloc((void**)&d_Base,MAX_ENTRIES*sizeof(int))); checkCuda( cudaMemcpy(d_Base,BaseSample,MAX_ENTRIES*sizeof(int),cudaMemcpyHostToDevice)); //checkCuda( cudaFreeHost(BaseSample) ); checkCuda( cudaMalloc((void**)&d_mean,2048*sizeof(double))); checkCuda( cudaMallocHost((void**)&h_mean,2048*sizeof(double))); std::cout<<"Launching initCurand Kernel now\n\n"; ////////////////////////////////////// //checkCuda( cudaProfilerStart() ); try{ constexpr int block_size = 512; constexpr int num_blocks = 4096; auto devStates = cudaAllocBuffer<curandState>(num_blocks * block_size); initCurand<<<num_blocks, block_size>>>(devStates.get(),1234); throw_error(cudaPeekAtLastError()); throw_error(cudaDeviceSynchronize()); std::cout<<"Curand Kernel Launch Try block SUCCESSFUL!\n"; std::cout<<"Launching Bootstrap Kernel now\n\n"; bootstrap<<<2048,1024>>>(d_mean,d_Base,devStates.get()); throw_error(cudaPeekAtLastError()); throw_error(cudaDeviceSynchronize()); std::cout<<"Bootstrap Kernel Launch Try Block SUCCESSFUL!\n"; } catch (const std::exception& e) { std::cerr << "Error: " << e.what() << '\n'; return -1; } catch (...) { std::cerr << "Unknown Exception"; return -1; } std::cout<<"Kernels appear complete, attempting to copy data back to Host\n"; checkCuda( cudaMemcpy(h_mean,d_mean,2048*sizeof(double),cudaMemcpyDeviceToHost) ); /* This loop is meant for testing the validity of the memcpy output for(int i=0;i<2048;++i){ std::cout<<"element "<<i<<" : "<<h_mean[i]<<std::endl; } */ // Standard Error of the bootstrap means int n = 2048; double SumOfMeans=0; for(int i=0;i<n;i++){ SumOfMeans+=h_mean[i]; } double MeanOfMeans = SumOfMeans / (double)n; std::cout<<"MeanofMeans: "<<MeanOfMeans<<"\n"; double SqrDiff=0; for(int i=0;i<n;i++){ SqrDiff += (h_mean[i]-MeanOfMeans) * (h_mean[i]-MeanOfMeans); } double SqrdVariance = SqrDiff / (n-1); double BootError = ((n-1)/(n*n)) * SqrdVariance; std::cout<<"SqrDiff, SqrdVariance, & BootError: "<<SqrDiff<<", "<<SqrdVariance<<", "<<BootError<<"\n"; //std::sort(h_mean,h_mean+2048); std::cout<<"\nStandard Error is: "<<BootError<<"\n\n\n"; double C_Arr[2048] = {}; for(int i=0;i<n;i++){ C_Arr[i]=h_mean[i]-BaseMean; } std::sort(std::begin(C_Arr),std::end(C_Arr)); double L = 2048.0 * 0.1; int Lower = (int)L; double H = 2048 * 0.9; int Higher = (int)H; int LowerBound = C_Arr[Lower]; int UpperBound = C_Arr[Higher]; double Left = BaseMean - (double)LowerBound; double Right = BaseMean - (double)UpperBound; std::cout<<"\n\n\n------------------------------------------------\n"; std::cout<<"The Confidence Interval is: 80%\n"; std::cout<<"The Standard Error is: "<<BootError<<"\n"; std::cout<<"This is on the 10th & 90th percentiles: ["<<Left<<", "<<Right<<"]\n"; checkCuda( cudaFree(d_Base) ); checkCuda( cudaFree(d_mean) ); //checkCuda( cudaFree(devStates) ); checkCuda( cudaFreeHost(BaseSample) ); checkCuda( cudaFreeHost(h_mean) ); printf("\n\n\nDONE\n"); return 0; } //////////////////// Calculate Statistics. Soon to be offloaded // int finalMean[400]={0}; // // std::vector<int> *meanVector; // int bnum = 0; // int sum1; // int sum2 = 0; // // int temp1; // for (int a=0;a<400;++a){ // sum1=0; // for(int b=0;b<100;++b){ // sum1+=h_mean[b+(100*bnum)]; // } // finalMean[a]=sum1/100; // // temp1 = sum1/100; // // meanVector[a].push_back( temp1 ); // sum2 += std::pow( (finalMean[a]-meanOriginal), 2 ); // bnum++; // // std::cout<<"Final Mean "<<a<<" : "<<finalMean[a]<<std::endl; // } // printf("\n\n\n"); // std::sort(finalMean,finalMean+SAMPLE_SIZE); // std::cout<<"sum2 is "<<sum2<<std::endl; // int div = 400; // std::cout<<"div is "<<div<<std::endl; // float stdDeviation = sqrt( (sum2/div) ); // std::cout<<"Standard Deviation is "<<stdDeviation<<std::endl; // float stdErrorFactor = ( 100.0 / (100.0-1.0) ); // std::cout<<"The Error Factor is "<<stdErrorFactor<<std::endl; // float stdError = sqrt( stdErrorFactor ) * stdDeviation; // std::cout<<"Standard Error is "<<stdError<<std::endl; // int tempA; int tempB; // float lowerCI = 400 * ( 0.05/2 ); // tempA = finalMean[(int)lowerCI]; // std::cout<<"Lower (5%) Confidence Interval is "<<tempA<<std::endl; // float higherCI = 400 * ( 1 - (0.05/2) ); // tempB = finalMean[(int)higherCI]; // std::cout<<"Higher (95%) Confidence Interval is "<<tempB<<std::endl;
10,869
/* *To find sum of two large arrays */ #include <stdio.h> const long long ARRAY_SIZE = 320000; const long long ARRAY_BYTES = ARRAY_SIZE * sizeof(float); //kernal __global__ void sum(float *d_sum, float *d_a, float *d_b) { int id = blockIdx.x * blockDim.x + threadIdx.x; d_sum[id] = d_a[id] + d_b[id]; } //to check the final result int checkSum(float *h_a, float *h_b, float *h_sum) { int flag = 1; for(int i = 0; i < ARRAY_SIZE; i++) { if(h_sum[i] != h_a[i] + h_b[i]) { flag = 0; break; } } return flag; } int main() { float h_a[ARRAY_SIZE], h_b[ARRAY_SIZE], h_sum[ARRAY_SIZE]; for(int i = 0; i < ARRAY_SIZE; i++) { h_a[i] = rand(); h_b[i] = rand(); } float *d_a, *d_b, *d_sum; cudaMalloc((void**) &d_a, ARRAY_BYTES); cudaMalloc((void**) &d_b, ARRAY_BYTES); cudaMalloc((void**) &d_sum, ARRAY_BYTES); cudaMemcpy(d_a, h_a, ARRAY_BYTES, cudaMemcpyHostToDevice); cudaMemcpy(d_b, h_b, ARRAY_BYTES, cudaMemcpyHostToDevice); sum<<<(int)ceil(ARRAY_SIZE/32.0), 32>>>(d_sum, d_a, d_b); cudaMemcpy(h_sum, d_sum, ARRAY_BYTES, cudaMemcpyDeviceToHost); if(checkSum(h_a, h_b, h_sum)) { printf("The result is computed correctly!"); } else { printf("The result is not computed correctly!"); } cudaFree(d_a); cudaFree(d_b); cudaFree(d_sum); return 0; }
10,870
#include "includes.h" __global__ void ApplyThreshold( float* probabilitiesInputs, float* binaryOutput, float* probability, int count ) { int id = blockDim.x*blockIdx.y*gridDim.x + blockDim.x*blockIdx.x + threadIdx.x; if (id < count) { if (probabilitiesInputs[id] < probability[0]) { binaryOutput[id] = 0.0f; } else { binaryOutput[id] = 1.0f; } } }
10,871
#include "includes.h" __global__ void compute_d_w_kernel(float *v, float *h, float *dw, bool is_init, int input_size, int lu_padding, int channel_num, int filter_num, int filter_size, int feature_map_size){ int imgIdx = blockIdx.y / (feature_map_size / 32); int filterIdx = blockIdx.x / (channel_num * feature_map_size / 32); int channelIdx = (blockIdx.x % (channel_num * feature_map_size / 32)) / (feature_map_size / 32); int tx = (blockIdx.x % (channel_num * feature_map_size / 32)) % (feature_map_size / 32) *32 + threadIdx.x; int ty = (blockIdx.y % (feature_map_size / 32)) * 32 + threadIdx.y; __shared__ float shV[32+MAX_FILETER_SIZE][32+MAX_FILETER_SIZE]; __shared__ float shH[32][32]; float sign; if(is_init){ sign = 1.0f; }else{ sign = -1.0f; } v = v + imgIdx * channel_num * input_size * input_size + channelIdx * input_size * input_size; h = h + imgIdx * filter_num * feature_map_size * feature_map_size + filterIdx * feature_map_size * feature_map_size; dw = dw + filterIdx * channel_num * filter_size * filter_size + channelIdx * filter_size * filter_size; float local_dw = 0.0f; for(int loadX = 0; loadX <= 32; loadX += filter_size){ for(int loadY = 0; loadY <= 32; loadY += filter_size){ if(loadX < 32 && loadY < 32){ //TODO:feature map overflow shH[threadIdx.y+loadY][threadIdx.x+loadX] = h[(ty+loadY)*feature_map_size + (tx+loadX)]; } if((tx+loadX) < lu_padding || (ty+loadY) < lu_padding || (tx+loadX) >= (input_size+lu_padding) || (ty+loadY) >= (input_size+lu_padding)){ shV[threadIdx.y+loadY][threadIdx.x+loadX] = 0; }else{ shV[threadIdx.y+loadY][threadIdx.x+loadX] = v[(ty+loadY-lu_padding)*input_size + (tx+loadX-lu_padding)]; } } } __syncthreads(); for(int i = 0; i < 32; i++){ for(int j = 0; j < 32; j++){ local_dw += shV[threadIdx.y+i][threadIdx.x+j] * shH[i][j]; } } atomicAdd(dw + threadIdx.y*filter_size + threadIdx.x, sign * local_dw); }
10,872
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <stdio.h> #define NUM_BLOCKS 32 #define BLOCK_WIDTH 1 __global__ void hello() { printf("Hello world! I'm a thread in block %d\n", blockIdx.x); } int main(int argc, char **argv) { // launch the kernel hello <<<NUM_BLOCKS, BLOCK_WIDTH >>>(); // force the printf()s to flush cudaDeviceSynchronize(); printf("That's all!\n"); // cudaDeviceReset must be called before exiting in order for profiling and // tracing tools such as Nsight and Visual Profiler to show complete traces. cudaDeviceReset(); getchar(); return 0; }
10,873
#include "includes.h" __global__ void convolution1d_tiles_constant_kernel(int *In, int *Out){ unsigned int index = blockIdx.x * blockDim.x + threadIdx.x; // Index 1d iterator. __shared__ int Tile[TILE_SIZE + Mask_size - 1]; int n = Mask_size/2; int halo_left_index = (blockIdx.x - 1 ) * blockDim.x + threadIdx.x; if (threadIdx.x >= blockDim.x - n ){ Tile[threadIdx.x - (blockDim.x - n )] = (halo_left_index < 0) ? 0 : In[halo_left_index]; } if(index<N_elements){Tile[n + threadIdx.x] = In[index]; }else{Tile[n + threadIdx.x] = 0;} int halo_right_index = (blockIdx.x + 1 ) * blockDim.x + threadIdx.x; if (threadIdx.x < n) { Tile[n + blockDim.x + threadIdx.x]= (halo_right_index >= N_elements) ? 0 : In[halo_right_index]; } __syncthreads(); int Value = 0; for (unsigned int j = 0; j < Mask_size; j ++) { Value += Tile[threadIdx.x + j] * Global_Mask[j]; } Out[index] = Value; }
10,874
#pragma warning (disable : 4267) #pragma warning (disable : 4244) #include <thrust/random.h> #include <thrust/iterator/counting_iterator.h> #include <thrust/functional.h> #include <thrust/transform_reduce.h> #include <thrust/random/normal_distribution.h> #include <thrust/device_ptr.h> #include <curand.h> #include <cuda_runtime.h> #include <iostream> #include <iomanip> #include <cmath> #include "Example_MC_BS.cuh" const unsigned int DEFAULT_RAND_N = 10000000; const unsigned int DEFAULT_SEED = 1; struct estimate_BS3 : public thrust::unary_function<float, float> { __host__ __device__ float operator()(float W) { float S0 = 20.0f; float sig = 0.28f; float r = 0.045f; float K = 21.0f; float T = 0.5f; float sqrtT = sqrtf(T); float sig2 = sig*sig; float ST = S0 * expf((r - 0.5f*sig2)*T + sig*sqrtT*W); float ST_at = S0 * expf((r - 0.5f*sig2)*T - sig*sqrtT*W); return expf(-r*T)*(((ST-K > 0.0f)? ST-K:0.0f) + ((ST_at-K > 0.0f)? ST_at-K:0.0f))/2.0f; } }; void exmpl_thrust_MC_BS2() { unsigned int M = 200; unsigned int rand_n = DEFAULT_RAND_N; unsigned int seed = DEFAULT_SEED; curandGenerator_t prngGPU; curandCreateGenerator(&prngGPU, CURAND_RNG_PSEUDO_MTGP32); curandSetPseudoRandomGeneratorSeed(prngGPU, seed); float estimate = 0.0f; float *d_rand; cudaMalloc((void **)&d_rand, rand_n * sizeof(float)); thrust::device_ptr<float> d_rand_b = thrust::device_pointer_cast(d_rand); thrust::device_ptr<float> d_rand_e = d_rand_b + rand_n; for (unsigned int i = 0; i < M; ++i) { curandGenerateNormal(prngGPU, (float *) d_rand, rand_n, 0.0f, 1.0f); estimate += thrust::transform_reduce( d_rand_b, d_rand_e, estimate_BS3(), 0.0f, thrust::plus<float>()); } estimate /= (rand_n*M); std::cout << std::setprecision(10); std::cout << "Option price is approximately " << estimate << std::endl; curandDestroyGenerator(prngGPU); cudaFree(d_rand); cudaDeviceReset(); };
10,875
#include <cuda.h> #include <stdio.h> void printMatrix(float *matrix, int rows, int columns) { for (int i = 0; i < rows; i++) { for (int j = 0; j < columns; j++) printf("%g ", matrix[i * rows + j]); printf("\n"); } printf("\n"); } #define CUDA_CHECK_RETURN(value)\ {\ cudaError_t _m_cudaStat = value;\ if (_m_cudaStat != cudaSuccess) {\ fprintf(stderr, "Error %s at line %d in file %s\n",\ cudaGetErrorString(_m_cudaStat), __LINE__, __FILE__);\ exit(1);\ }\ } __global__ void initMatrix_1D(float *matrix) { int i = threadIdx.x + blockDim.x * blockIdx.x; matrix[i] = i; } __global__ void initMatrix_2D_I(float *matrix) { int i = threadIdx.x + blockDim.x * blockIdx.x; int j = threadIdx.y + blockDim.y * blockIdx.y; int I = gridDim.x * blockDim.x; // int J = gridDim.y * blockDim.y; // matrix[j * I + i] = j * I + i; // matrix[i + j * I] = i; // matrix[i + j * I] = j; // matrix[i + j * I] = I; // matrix[i + j * I] = J; matrix[i + j * I] = threadIdx.x; // matrix[i + j * I] = threadIdx.y; // matrix[i + j * I] = gridDim.x; // matrix[i + j * I] = gridDim.y; // matrix[i + j * I] = blockDim.x; // matrix[i + j * I] = blockDim.y; // matrix[i + j * I] = blockIdx.x; // matrix[i + j * I] = blockIdx.y; } __global__ void initMatrix_2D_J(float *matrix) { int i = threadIdx.x + blockDim.x * blockIdx.x; int j = threadIdx.y + blockDim.y * blockIdx.y; // int I = gridDim.x * blockDim.x; int J = gridDim.y * blockDim.y; matrix[j + i * J] = j + i * J; } __global__ void transp(float *matrix1, float *matrix2, float *check) { int i = threadIdx.x + blockDim.x * blockIdx.x; int j = threadIdx.y + blockDim.y * blockIdx.y; int I = gridDim.x * blockDim.x; matrix2[i * I + j] = matrix1[j * I + i]; check[i * I + j] = j * I + i; } int main(int argc, char *argv[]) { int blocks = (argc > 1) ? atoi(argv[1]) : 8; int threads = (argc > 2) ? atoi(argv[2]) : 4; int rows = 8; int columns = 8; int size_matrix = rows * columns; float *dmatrix1, *hmatrix1; float *dmatrix2, *hmatrix2; float *dcheck, *hcheck; float *dmatrix3, *hmatrix3; cudaMalloc((void**) &dmatrix1, size_matrix * sizeof(float)); cudaMalloc((void**) &dmatrix2, size_matrix * sizeof(float)); cudaMalloc((void**) &dcheck, size_matrix * sizeof(float)); cudaMalloc((void**) &dmatrix3, size_matrix * sizeof(float)); hmatrix1 = (float*) calloc(size_matrix, sizeof(float)); hmatrix2 = (float*) calloc(size_matrix, sizeof(float)); hcheck = (float*) calloc(size_matrix, sizeof(float)); hmatrix3 = (float*) calloc(size_matrix, sizeof(float)); initMatrix_2D_I<<<dim3(2, 2), dim3(8)>>>(dmatrix3); cudaDeviceSynchronize(); cudaMemcpy(hmatrix3, dmatrix3, size_matrix * sizeof(float), cudaMemcpyDeviceToHost); printMatrix(hmatrix3, rows, columns); // initMatrix_1D<<<dim3(blocks), dim3(threads)>>>(dmatrix1); initMatrix_2D_I<<<dim3(blocks), dim3(2, 2)>>>(dmatrix1); cudaDeviceSynchronize(); cudaMemcpy(hmatrix1, dmatrix1, size_matrix * sizeof(float), cudaMemcpyDeviceToHost); printMatrix(hmatrix1, rows, columns); #if 0 initMatrix_2D_J<<<dim3(blocks), dim3(2, 2)>>>(dmatrix2); cudaDeviceSynchronize(); cudaMemcpy(hmatrix2, dmatrix2, size_matrix * sizeof(float), cudaMemcpyDeviceToHost); printMatrix(hmatrix2, rows, columns); #endif transp<<<dim3(blocks), dim3(2, 2)>>>(dmatrix1, dmatrix2, dcheck); cudaDeviceSynchronize(); cudaMemcpy(hmatrix2, dmatrix2, size_matrix * sizeof(float), cudaMemcpyDeviceToHost); printMatrix(hmatrix2, rows, columns); cudaMemcpy(hcheck, dcheck, size_matrix * sizeof(float), cudaMemcpyDeviceToHost); printMatrix(hcheck, rows, columns); cudaFree(dmatrix1); cudaFree(dmatrix2); cudaFree(dcheck); free(hmatrix1); free(hmatrix2); free(hcheck); return 0; }
10,876
extern "C" { __global__ void DYbinaryentropyXsigmoidY_32(const int lengthX, const float *x, const float *y, const float *t, float *z) { int i = threadIdx.x + blockIdx.x * blockDim.x; if (i<lengthX) { z[i] += t[0]*(1.0/(1.0+exp(-y[i]))-x[i])/lengthX; } } }
10,877
#include <stdio.h> __global__ void helloWorld() { const int i = blockIdx.x * blockDim.x + threadIdx.x; printf("Hello World! My ThreadId is %2d\n", i); } int main() { helloWorld<<<1, 256>>>(); cudaDeviceSynchronize(); return 0; }
10,878
#include "includes.h" __global__ void reduceUnrolling4 (int *g_idata, int *g_odata, unsigned int n) { // set thread ID unsigned int tid = threadIdx.x; unsigned int idx = blockIdx.x * blockDim.x * 4 + threadIdx.x; // convert global data pointer to the local pointer of this block int *idata = g_idata + blockIdx.x * blockDim.x * 4; // unrolling 4 if (idx + 3 * blockDim.x < n) { int a1 = g_idata[idx]; int a2 = g_idata[idx + blockDim.x]; int a3 = g_idata[idx + 2 * blockDim.x]; int a4 = g_idata[idx + 3 * blockDim.x]; g_idata[idx] = a1 + a2 + a3 + a4; // g_idata[idx] = g_idata[idx] + g_idata[idx + blockDim.x] + g_idata[idx + 2*blockDim.x] + g_idata[idx + 3*blockDim.x]; } __syncthreads(); // in-place reduction in global memory for (int stride = blockDim.x / 2; stride > 0; stride >>= 1) { if (tid < stride) { idata[tid] += idata[tid + stride]; } // synchronize within threadblock __syncthreads(); } // write result for this block to global mem if (tid == 0) g_odata[blockIdx.x] = idata[0]; }
10,879
#define d_vx(z,x) d_vx[(x)*(nz)+(z)] #define d_vz(z,x) d_vz[(x)*(nz)+(z)] #define d_sxx(z,x) d_sxx[(x)*(nz)+(z)] #define d_szz(z,x) d_szz[(x)*(nz)+(z)] #define d_sxz(z,x) d_sxz[(x)*(nz)+(z)] #define d_mem_dszz_dz(z,x) d_mem_dszz_dz[(x)*(nz)+(z)] #define d_mem_dsxz_dx(z,x) d_mem_dsxz_dx[(x)*(nz)+(z)] #define d_mem_dsxz_dz(z,x) d_mem_dsxz_dz[(x)*(nz)+(z)] #define d_mem_dsxx_dx(z,x) d_mem_dsxx_dx[(x)*(nz)+(z)] #define d_mem_dvz_dz(z,x) d_mem_dvz_dz[(x)*(nz)+(z)] #define d_mem_dvz_dx(z,x) d_mem_dvz_dx[(x)*(nz)+(z)] #define d_mem_dvx_dz(z,x) d_mem_dvx_dz[(x)*(nz)+(z)] #define d_mem_dvx_dx(z,x) d_mem_dvx_dx[(x)*(nz)+(z)] #define d_Lambda(z,x) d_Lambda[(x)*(nz)+(z)] #define d_Den(z,x) d_Den[(x)*(nz)+(z)] #define d_Mu(z,x) d_Mu[(x)*(nz)+(z)] #define d_ave_Mu(z,x) d_ave_Mu[(x)*(nz)+(z)] #define d_ave_Byc_a(z,x) d_ave_Byc_a[(x)*(nz)+(z)] #define d_ave_Byc_b(z,x) d_ave_Byc_b[(x)*(nz)+(z)] #include<stdio.h> __global__ void el_velocity_adj( float *d_vz, float *d_vx, float *d_szz, float *d_sxx, float *d_sxz, \ float *d_mem_dszz_dz, float *d_mem_dsxz_dx, float *d_mem_dsxz_dz, float *d_mem_dsxx_dx, \ float *d_mem_dvz_dz, float *d_mem_dvz_dx, float *d_mem_dvx_dz, float *d_mem_dvx_dx, \ float *d_Lambda, float *d_Mu, float *d_ave_Mu, float *d_Den, float *d_ave_Byc_a, float *d_ave_Byc_b, \ float *d_K_z_half, float *d_a_z_half, float *d_b_z_half, \ float *d_K_x_half, float *d_a_x_half, float *d_b_x_half, \ float *d_K_z, float *d_a_z, float *d_b_z, \ float *d_K_x, float *d_a_x, float *d_b_x, \ int nz, int nx, float dt, float dz, float dx, int nPml, int nPad){ int gidz = blockIdx.x*blockDim.x + threadIdx.x; int gidx = blockIdx.y*blockDim.y + threadIdx.y; float dpsixx_dx = 0.0; float dszz_dx = 0.0; float dsxx_dx = 0.0; float dpsixz_dz = 0.0; float dsxz_dz = 0.0; float dpsizz_dz = 0.0; float dszz_dz = 0.0; float dsxx_dz = 0.0; float dpsizx_dx = 0.0; float dsxz_dx = 0.0; float c1 = 9.0/8.0; float c2 = 1.0/24.0; float lambda = d_Lambda(gidz,gidx); float mu = d_Mu(gidz,gidx); if(gidz>=2 && gidz<=nz-nPad-3 && gidx>=2 && gidx<=nx-3) { // update vx dpsixx_dx = (-c1*(d_mem_dvx_dx(gidz,gidx+1)-d_mem_dvx_dx(gidz,gidx)) \ + c2*(d_mem_dvx_dx(gidz,gidx+2)-d_mem_dvx_dx(gidz,gidx-1)))/dx; dszz_dx = (-c1*(d_szz(gidz,gidx+1)-d_szz(gidz,gidx)) + c2*(d_szz(gidz,gidx+2)-d_szz(gidz,gidx-1)))/dx; dsxx_dx = (-c1*(d_sxx(gidz,gidx+1)-d_sxx(gidz,gidx)) + c2*(d_sxx(gidz,gidx+2)-d_sxx(gidz,gidx-1)))/dx; dpsixz_dz = (-c1*(d_mem_dvx_dz(gidz,gidx)-d_mem_dvx_dz(gidz-1,gidx)) \ + c2*(d_mem_dvx_dz(gidz+1,gidx)-d_mem_dvx_dz(gidz-2,gidx)))/dz; dsxz_dz = (-c1*(d_sxz(gidz,gidx)-d_sxz(gidz-1,gidx)) + c2*(d_sxz(gidz+1,gidx)-d_sxz(gidz-2,gidx)))/dz; d_vx(gidz, gidx) += (d_a_x[gidx]*dpsixx_dx + lambda*dszz_dx/d_K_x[gidx]*dt \ + (lambda+2.0*mu)*dsxx_dx/d_K_x[gidx]*dt + d_a_z_half[gidz]*dpsixz_dz \ + d_ave_Mu(gidz,gidx)/d_K_z_half[gidz]*dsxz_dz*dt); //update phi_xx_x and phi_xz_z if(gidx<nPml || gidx>nx-nPml-1){ d_mem_dsxx_dx(gidz, gidx) = d_b_x_half[gidx]*d_mem_dsxx_dx(gidz, gidx) + d_ave_Byc_b(gidz, gidx)*d_vx(gidz, gidx)*dt; } if(gidz<nPml || (gidz>nz-nPml-nPad-1)){ d_mem_dsxz_dz(gidz, gidx) = d_b_z[gidz]*d_mem_dsxz_dz(gidz, gidx) + d_ave_Byc_b(gidz, gidx)*d_vx(gidz, gidx)*dt; } // update vz dpsizz_dz = (-c1*(d_mem_dvz_dz(gidz+1,gidx)-d_mem_dvz_dz(gidz,gidx)) \ + c2*(d_mem_dvz_dz(gidz+2,gidx)-d_mem_dvz_dz(gidz-1,gidx)))/dz; dszz_dz = (-c1*(d_szz(gidz+1,gidx)-d_szz(gidz,gidx)) + c2*(d_szz(gidz+2,gidx)-d_szz(gidz-1,gidx)))/dz; dsxx_dz = (-c1*(d_sxx(gidz+1,gidx)-d_sxx(gidz,gidx)) + c2*(d_sxx(gidz+2,gidx)-d_sxx(gidz-1,gidx)))/dz; dpsizx_dx = (-c1*(d_mem_dvz_dx(gidz,gidx)-d_mem_dvz_dx(gidz,gidx-1)) \ + c2*(d_mem_dvz_dx(gidz,gidx+1)-d_mem_dvz_dx(gidz,gidx-2)))/dx; dsxz_dx = (-c1*(d_sxz(gidz,gidx)-d_sxz(gidz,gidx-1)) + c2*(d_sxz(gidz,gidx+1)-d_sxz(gidz,gidx-2)))/dx; d_vz(gidz, gidx) += (d_a_z[gidz]*dpsizz_dz + (lambda+2.0*mu)*dszz_dz/d_K_z[gidz]*dt \ + lambda*dsxx_dz/d_K_z[gidz]*dt + d_a_x_half[gidx]*dpsizx_dx \ + d_ave_Mu(gidz,gidx)/d_K_x_half[gidx]*dsxz_dx*dt); // update phi_xz_x and phi_zz_z if(gidx<nPml || gidx>nx-nPml-1){ d_mem_dsxz_dx(gidz, gidx) = d_b_x[gidx]*d_mem_dsxz_dx(gidz, gidx) + d_ave_Byc_a(gidz, gidx)*d_vz(gidz, gidx)*dt; } if(gidz<nPml || (gidz>nz-nPml-nPad-1)){ d_mem_dszz_dz(gidz, gidx) = d_b_z_half[gidz]*d_mem_dszz_dz(gidz, gidx) + d_ave_Byc_a(gidz, gidx)*d_vz(gidz, gidx)*dt; } } else { return; } }
10,880
#include <stdio.h> #include "orbit_integrator_cuda.cu" #define N 512 * 512 * 64 float x[N]; float *x_device; cudaError_t err; int main(int argc, char** argv) { for(int i = 0; i < N; i++) { x[i] = i; } printf("x[%d] = %f\n", N-1, x[N-1]); if( argc > 1) { cudaMalloc((void**) &x_device, sizeof(float)*N); err = cudaGetLastError (); printf("malloc: %s\n", cudaGetErrorString(err)); cudaMemcpy(x_device, x, sizeof(float)*N, cudaMemcpyHostToDevice); err = cudaGetLastError (); printf("copy: %s\n", cudaGetErrorString(err)); dim3 dimBlock(512,64); square<<<dimBlock, 512>>>(x_device); err = cudaGetLastError (); printf("call: %s\n", cudaGetErrorString(err)); cudaMemcpy(x, x_device, sizeof(float)*N, cudaMemcpyDeviceToHost); err = cudaGetLastError (); printf("cpy: %s\n", cudaGetErrorString(err)); } else { for(int i = 0; i < N; i++) { for(int j = 0; j < M; j++) x[i] = x[i] + M; } } for(int i = 0; i < 10; i++) { printf("x[%d] = %f\n", i, x[i]); } printf("x[%d] = %f\n", N-1, x[N-1]); return 0; }
10,881
#include "includes.h" __global__ void insertArray ( const int n, const int indx, const float *ss, float *zz ) { int i = threadIdx.x + blockDim.x * blockIdx.x; if ( i < n ) { zz[indx+i] = ss[i]; } }
10,882
#include "AntGPU.cuh" #include <stdio.h> __device__ AntGPU::AntGPU(int initialLocation, int matrixDim, curandState_t randState) : position(initialLocation), possibleLocations(new int[matrixDim]), goodnessNumerators(new double[matrixDim]), possibleLocationsLastIndex(matrixDim - 1), m_randomState(randState) { } __device__ void AntGPU::Venture(int* route, const double* distanceMatrix, const double* pheromoneMatrix, int matrixDim, double alpha, double beta) { while (possibleLocationsLastIndex >= 0) { int nextHop = SelectNextHop(distanceMatrix, pheromoneMatrix, matrixDim, alpha, beta); GoTo(nextHop, route, distanceMatrix, matrixDim); } //printf("Distance Traveled: %f\n", distance + distanceMatrix[position]); route[matrixDim] = route[0]; distance += distanceMatrix[route[matrixDim - 1] * matrixDim + route[0]]; } __device__ int AntGPU::SelectNextHop(const double* distance_matrix, const double* pheromoneMatrix, int matrixDim, double alpha, double beta) { double denominator = 0; for (int i = 0; i <= possibleLocationsLastIndex; ++i) { int possiblePosition = possibleLocations[i]; double goodnessNumerator = pow(pheromoneMatrix[position * matrixDim + possiblePosition], alpha) * pow(1.0 / distance_matrix[position * matrixDim + possiblePosition], beta); goodnessNumerators[possiblePosition] = goodnessNumerator; denominator += goodnessNumerator; } double sum = 0; double random = curand_uniform_double(&m_randomState); for (int i = 0; i <= possibleLocationsLastIndex; ++i) { int possiblePosition = possibleLocations[i]; double numerator = goodnessNumerators[possiblePosition]; double probability = numerator / denominator; if (random <= sum + probability) { possibleLocationsNextIndex = i; return possiblePosition; } sum += probability; } return -1; } __device__ void AntGPU::GoTo(int next, int* route, const double* distanceMatrix, int matrixDim) { route[routeIndex++] = next; possibleLocations[possibleLocationsNextIndex] = possibleLocations[possibleLocationsLastIndex--]; distance += distanceMatrix[position * matrixDim + next]; position = next; } __device__ void AntGPU::Reset(int* route, int initialLocation, int matrixDim) { routeIndex = 0; distance = 0; position = initialLocation; possibleLocationsLastIndex = matrixDim - 1; for (int i = 0; i < matrixDim; ++i) { possibleLocations[i] = i; } route[routeIndex++] = initialLocation; possibleLocations[initialLocation] = possibleLocations[possibleLocationsLastIndex--]; }
10,883
__global__ void add(float2 *in, float2 *out, float2 *out2) { int index = blockIdx.x; float2 a = in[index]; out[index] = a; //(float2) in->x = 0.0; in->y = 0.0; out->x = 0.0; out->y = 0.0; //*(float2)a.x = index; }
10,884
#include "includes.h" __global__ void fix_nan_and_inf_kernel(float *input, size_t size) { const int index = blockIdx.x*blockDim.x + threadIdx.x; if (index < size) { float val = input[index]; if (isnan(val) || isinf(val)) { input[index] = 1.0f / (fabs((float)index) + 1); // pseudo random value } } }
10,885
#include<stdio.h> #include<cuda.h> __global__ void mtrxVecMult(float* d_a, float* d_b, float* d_x) { int i = threadIdx.x; int j = blockIdx.x; int n = blockDim.x; float aij = d_a[j*n+i]; d_x[i] = d_b[j]*aij; } int main(int argc, char** argv) { const int n = 8; const int A_BYTES = n * n * sizeof(float); const int B_BYTES = n * sizeof(float); float h_a[n*n]; float h_b[n]; float h_x[n]; for (int i=0; i < n; i++) { for (int j=0; j < n; j++) h_a[i*n+j] = 3*j; h_b[i] = 2; } //declare GPU memory pointers float *d_a; float *d_b; float *d_x; //allocate memory on the device cudaMalloc((void**)&d_a,A_BYTES); cudaMalloc((void**)&d_b,B_BYTES); cudaMalloc((void**)&d_x,B_BYTES); //transfer the array to the GPU //destination, source, size, method cudaMemcpy(d_x,h_x,B_BYTES,cudaMemcpyHostToDevice); cudaMemcpy(d_b,h_b,B_BYTES,cudaMemcpyHostToDevice); cudaMemcpy(d_a,h_a,A_BYTES,cudaMemcpyHostToDevice); //launch the kernel mtrxVecMult<<<n,n>>>(d_a,d_b,d_x); cudaDeviceSynchronize(); cudaGetLastError(); //copy the results back onto the device //destination, source, size, method cudaMemcpy(h_x,d_x,B_BYTES,cudaMemcpyDeviceToHost); for (int i=0; i<n; i++) { printf("%.2f",h_x[i]); printf("\n"); } //free memory previously allocated on the device cudaFree(d_a); cudaFree(d_b); cudaFree(d_x); }
10,886
#include "includes.h" __global__ void CuKnlSetField( double xCells, double yCells, double* energy0, double* energy1) { const int gid = threadIdx.x+blockIdx.x*blockDim.x; energy1[gid] = energy0[gid]; }
10,887
#include "includes.h" __global__ void cu_getRange(const float *src, float* dst, const int xstart, const int xend, const int ystart, const int yend, const int colssrc, const int n){ int tid = threadIdx.x + blockIdx.x * blockDim.x; int stride = blockDim.x * gridDim.x; int colsdst = xend - xstart + 1; while(tid < n){ int cdst = tid % colsdst; int rdst = tid / colsdst; int rsrc = rdst + ystart; int csrc = cdst + xstart; dst[tid] = src[rsrc * colssrc + csrc]; tid += stride; } }
10,888
#include "includes.h" __global__ void atemp(double* A, double* y, double* tmp, int NX, int NY) { int j; int i = blockDim.x * blockIdx.x + threadIdx.x; // Α(T)*temp if (i <= NY){ for (j = 0; j < NX; j++) { y[i] = y[i] + A[i + j*NY] * tmp[j]; } } }
10,889
/* CUDA C++ Program to add 2 1-dimensional vectors of length N 1 blocks, N threads in block */ #include<iostream> #include<cuda_runtime.h> #define N 10 __global__ void vecadd(int* a, int* b, int* c) { int idx = threadIdx.x; c[idx] = a[idx] + b[idx]; } int main() { /* Set up variables on host*/ int* a = new int[N]; int* b = new int[N]; int* c = new int[N]; /* Input values on host*/ unsigned int size = N*sizeof(int); for(int i=0; i<N; ++i) { a[i] = 2*i; b[i] = 3*i+1; } /* Setting up variables on device. i.e. GPU */ int* d_a, *d_b, *d_c; cudaMalloc((void**)&d_a, size); cudaMalloc((void**)&d_b, size); cudaMalloc((void**)&d_c, size); /* Copy data from host to device */ cudaMemcpy(d_a, a, size, cudaMemcpyHostToDevice); cudaMemcpy(d_b, b, size, cudaMemcpyHostToDevice); /* Kernel Launch Grid contains 1 block That block has N threads Hence index of vector is thread index */ vecadd<<<1, N>>>(d_a, d_b, d_c); cudaDeviceSynchronize(); /* Copy result from GPU device to host */ cudaMemcpy(c, d_c, size, cudaMemcpyDeviceToHost); /* Print result */ for(int i=0; i<N; ++i) { std::cout<<c[i]<<' '; } std::cout<<'\n'; /* Cleanup device and host memory */ cudaFree(d_a); cudaFree(d_b); cudaFree(d_c); delete a; delete b; delete c; return 0; }
10,890
#include "cuda_runtime.h" #include "device_launch_parameters.h" #include <iostream> #include <stdio.h> #include <cmath> #include "matrix_operations.cuh" #define OptNofThreads 128 #define OptNofBlocks 128 #define OptNofBlocksX 32 #define OptNofBlocksY 32 template <typename T> __global__ void get_first_layer_dropout(T* first_layer_weights, T* dropped_first_layer, unsigned int* first_dropout_indices, int n_of_neurons, int n_for_first_dropout) { int tdx = threadIdx.x; int bdx = blockIdx.x; int rand_index = first_dropout_indices[bdx*n_for_first_dropout + tdx] % n_of_neurons; first_dropout_indices[bdx*n_for_first_dropout + tdx] = rand_index; dropped_first_layer[bdx*n_for_first_dropout + tdx] = first_layer_weights[bdx*(n_of_neurons + 1) + rand_index]; first_layer_weights[bdx*(n_of_neurons + 1) + rand_index] = 0.0f; } template <typename T> __global__ void get_second_layer_dropout(T* second_layer_weights, T* dropped_second_layer, unsigned int* second_dropout_indices, int n_of_second_layer_neurons, int n_for_second_dropout) { int tdx = threadIdx.x; int bdx = blockIdx.x; int rand_index = second_dropout_indices[bdx*n_for_second_dropout + tdx] % n_of_second_layer_neurons; second_dropout_indices[bdx*n_for_second_dropout + tdx] = rand_index; dropped_second_layer[bdx*n_for_second_dropout + tdx] = second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + rand_index]; second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + rand_index] = 0.0f; } template <typename T> __global__ void get_third_layer_dropout(T* third_layer_weights, T* dropped_third_layer, unsigned int* third_dropout_indices, int n_of_third_layer_neurons, int n_for_third_dropout) { int tdx = threadIdx.x; int bdx = blockIdx.x; int rand_index = third_dropout_indices[bdx*n_for_third_dropout + tdx] % n_of_third_layer_neurons; third_dropout_indices[bdx*n_for_third_dropout + tdx] = rand_index; dropped_third_layer[bdx*n_for_third_dropout + tdx] = third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + rand_index]; third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + rand_index] = 0.0f; } template <typename T> __global__ void take_first_layer_dropout(T* first_layer_weights, T* dropped_first_layer, unsigned int* first_dropout_indices, int n_of_neurons, int n_for_first_dropout) { int tdx = threadIdx.x; int bdx = blockIdx.x; first_layer_weights[bdx*(n_of_neurons + 1) + first_dropout_indices[bdx*n_for_first_dropout + tdx]] = dropped_first_layer[bdx*n_for_first_dropout + tdx]; } template <typename T> __global__ void take_second_layer_dropout(T* second_layer_weights, T* dropped_second_layer, unsigned int* second_dropout_indices, int n_of_second_layer_neurons, int n_for_second_dropout) { int tdx = threadIdx.x; int bdx = blockIdx.x; second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + second_dropout_indices[bdx*n_for_second_dropout + tdx]] = dropped_second_layer[bdx*n_for_second_dropout + tdx]; } template <typename T> __global__ void take_third_layer_dropout(T* third_layer_weights, T* dropped_third_layer, unsigned int* third_dropout_indices, int n_of_third_layer_neurons, int n_for_third_dropout) { int tdx = threadIdx.x; int bdx = blockIdx.x; third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + third_dropout_indices[bdx*n_for_third_dropout + tdx]] = dropped_third_layer[bdx*n_for_third_dropout + tdx]; } template<typename T> __global__ void get_first_layer_output(T* images, T* first_layer_weights, T* first_layer_output, T* sigmoid_multiplier, int n_of_images, int n_of_second_layer_neurons, int n_of_neurons) { int tdx = threadIdx.x; int idx = threadIdx.x; int bdx = blockIdx.x; while (tdx < n_of_images) { if (bdx < n_of_second_layer_neurons) { first_layer_output[bdx * n_of_images + tdx] = 0.0f; sigmoid_multiplier[bdx*n_of_images + idx] = 0.0f; } else { first_layer_output[bdx * n_of_images + tdx] = 1.0f; } if (bdx < n_of_second_layer_neurons) { for (int i = 0; i < n_of_neurons; ++i) { first_layer_output[bdx * n_of_images + tdx] += images[tdx*n_of_neurons + i] * first_layer_weights[bdx * n_of_neurons + i]; } first_layer_output[bdx * n_of_images + tdx] = (float)1.0f / (1 + exp(-first_layer_output[bdx * n_of_images + tdx])); } tdx += OptNofThreads; } } template <typename T> __global__ void get_second_layer_output(T* first_layer_output, T* second_layer_output, T* second_layer_weights, T* error_multiplier, int n_of_images, int n_of_second_layer_neurons, int n_of_third_layer_neurons) { int tdx = threadIdx.x; int idx = threadIdx.x; int bdx = blockIdx.x; __shared__ T second_layer_output_shared[OptNofThreads]; while (bdx < n_of_third_layer_neurons + 1) { while (idx < n_of_images) { if (bdx < n_of_third_layer_neurons) { second_layer_output_shared[tdx] = 0; error_multiplier[bdx*n_of_images + idx] = 0.0f; } else { second_layer_output_shared[tdx] = 1; } if (bdx < n_of_third_layer_neurons) { for (int i = 0; i < n_of_second_layer_neurons + 1; ++i) { second_layer_output_shared[tdx] += second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + i] * first_layer_output[i*n_of_images + idx]; } second_layer_output_shared[tdx] = 1.0f / (1 + exp(-second_layer_output_shared[tdx])); } second_layer_output[n_of_images*bdx + idx] = second_layer_output_shared[tdx]; idx += OptNofThreads; } if (idx >= n_of_images) { idx = threadIdx.x; } bdx += OptNofBlocks; } } template <typename T> __global__ void get_third_layer_output(T* third_layer_weights, T* second_layer_output, T* third_layer_output, T* third_layer_sums, int n_of_images, int n_of_third_layer_neurons) { int tdx = threadIdx.x; int idx = threadIdx.x; int bdx = blockIdx.x; __shared__ T output[OptNofThreads]; while (idx < n_of_images) { output[tdx] = 0.0f; for (int i = 0; i < n_of_third_layer_neurons + 1; ++i) { output[tdx] += second_layer_output[i*n_of_images + idx] * third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + i]; } output[tdx] = exp(output[tdx]); third_layer_output[bdx*n_of_images + idx] = output[tdx]; if (bdx == 0) { third_layer_sums[idx] = 0.0f; } idx += OptNofThreads; } } template <typename T> __global__ void get_posibilities(T* third_layer_output, T* third_layer_sums, T* posibilities, int n_of_images, int n_of_models) { int tdx = threadIdx.x; int bdx = blockIdx.x; int bldx = blockDim.x; __shared__ T sums_array[10]; __shared__ T second_layer_sum[1]; while (bdx < n_of_images) { sums_array[tdx] = third_layer_output[tdx*n_of_images + bdx]; if (tdx == 0) { second_layer_sum[0] = 0.0f; } atomicAdd(&second_layer_sum[0], sums_array[tdx]); posibilities[tdx*n_of_images + bdx] = (float)sums_array[tdx] / second_layer_sum[0]; bdx += OptNofBlocks; } } template <typename T> __global__ void get_errors(T* posibilities, T* labels, T* errors, T* square_error, int n_of_images, int n_of_models, int i) { int idx = threadIdx.x; int tdx = threadIdx.x; int bdx = blockIdx.x; if (tdx == 0) { square_error[bdx] = 0.0f; } __shared__ T error_sums[OptNofThreads]; while (tdx < n_of_images) { float posibility = posibilities[bdx*n_of_images + tdx]; float label = labels[bdx*n_of_images + tdx]; errors[bdx*n_of_images + tdx] = (label*(posibility - 1) + (1 - label)*posibility); error_sums[idx] = (label - posibility)*(label - posibility); atomicAdd(&square_error[0], error_sums[idx]); tdx += OptNofThreads; } } template <typename T> __global__ void get_third_layer_correction(T* errors, T* second_layer_output, T* third_layer_correction, T* third_layer_weights, T* error_multiplier, int n_of_images, int n_of_third_layer_neurons, int n_of_models) { int idx = threadIdx.x; int bdx = blockIdx.x; if (idx == 0) { for (int i = 0; i < n_of_third_layer_neurons + 1; ++i) { third_layer_correction[bdx*(n_of_third_layer_neurons + 1) + i] = 0.0f; } } while (idx < n_of_images) { float error = errors[bdx*n_of_images + idx]; for (int i = 0; i < n_of_third_layer_neurons + 1; ++i) { atomicAdd(&third_layer_correction[bdx*(n_of_third_layer_neurons + 1) + i], second_layer_output[i*n_of_images + idx] * error); atomicAdd(&error_multiplier[i*n_of_images + idx], error * third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + i]); } idx += OptNofThreads; } } template <typename T> __global__ void third_layer_weights_update(T* third_layer_weights, T* third_layer_correction, T* previous_third_layer, int n_of_images, int n_of_third_layer_neurons, float alpha) { int tdx = threadIdx.x; int bdx = blockIdx.x; while (tdx < n_of_third_layer_neurons + 1) { previous_third_layer[bdx*(n_of_third_layer_neurons + 1) + tdx] = third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + tdx]; third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + tdx] -= alpha * third_layer_correction[bdx*(n_of_third_layer_neurons + 1) + tdx]; tdx += OptNofThreads; } } template <typename T> __global__ void third_layer_weights_back(T* third_layer_weights, T* previous_third_layer, int n_of_third_layer_neurons) { int tdx = threadIdx.x; int bdx = blockIdx.x; while (tdx < n_of_third_layer_neurons + 1) { third_layer_weights[bdx*(n_of_third_layer_neurons + 1) + tdx] = previous_third_layer[bdx*(n_of_third_layer_neurons + 1) + tdx]; tdx += OptNofThreads; } } template <typename T> __global__ void get_second_layer_correction(T* errors, T* first_layer_output, T* second_layer_output, T* second_layer_weights, T* third_layer_weights, T* second_layer_correction, T* error_multiplier, T* sigmoid_multiplier, int n_of_images, int n_of_second_layer_neurons, int n_of_third_layer_neurons, int n_of_models) { int tdx = threadIdx.x; int idx = threadIdx.x; int bdx = blockIdx.x; int bdy = blockIdx.y; __shared__ T additions[OptNofThreads]; if (tdx == 0) { while (bdy < n_of_third_layer_neurons) { while (bdx < n_of_second_layer_neurons + 1) { second_layer_correction[bdy*(n_of_second_layer_neurons + 1) + bdx] = 0.0f; bdx += OptNofBlocksX; } if (bdx >= n_of_second_layer_neurons + 1) { bdx = blockIdx.x; } bdy += OptNofBlocksY; } } bdx = blockIdx.x; bdy = blockIdx.y; while (bdy < n_of_third_layer_neurons) { while (bdx < n_of_second_layer_neurons + 1) { while (idx < n_of_images) { additions[tdx] = 0; additions[tdx] = error_multiplier[bdy*n_of_images + idx] * (second_layer_output[bdy*n_of_images + idx] * (1 - second_layer_output[bdy*n_of_images + idx])) * first_layer_output[bdx*n_of_images + idx]; if (bdx < n_of_second_layer_neurons) { sigmoid_multiplier[bdx*n_of_images + idx] += error_multiplier[bdy*n_of_images + idx] * second_layer_output[bdy*n_of_images + idx] * (1 - second_layer_output[bdy*n_of_images + idx])*second_layer_weights[bdy*(n_of_second_layer_neurons + 1) + bdx]; } atomicAdd(&second_layer_correction[bdy*(n_of_second_layer_neurons + 1) + bdx], additions[tdx]); idx += OptNofThreads; } if (idx >= n_of_images) { idx = tdx; } bdx += OptNofBlocksX; } if (bdx >= n_of_second_layer_neurons + 1) { bdx = blockIdx.x; } bdy += OptNofBlocksY; } } template <typename T> __global__ void second_layer_weights_update(T* second_layer_weights, T* correction, T* previous_second_layer, float alpha, int n_of_second_layer_neurons) { int tdx = threadIdx.x; int bdx = blockIdx.x; previous_second_layer[bdx*(n_of_second_layer_neurons + 1) + tdx] = second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + tdx]; second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + tdx] -= alpha * correction[bdx*(n_of_second_layer_neurons + 1) + tdx]; } template <typename T> __global__ void second_layer_weights_back(T* second_layer_weights, T* previous_second_layer, int n_of_second_layer_neurons) { int tdx = threadIdx.x; int bdx = blockIdx.x; second_layer_weights[bdx*(n_of_second_layer_neurons + 1) + tdx] = previous_second_layer[bdx*(n_of_second_layer_neurons + 1) + tdx]; } template <typename T> __global__ void get_first_layer_correction(T* errors, T* images, T* first_layer_output, T* second_layer_output, T* third_layer_weights, T* second_layer_weights, T* first_layer_correction, T* sigmoid_multiplier, int n_of_images, int n_of_neurons, int n_of_second_layer_neurons, int n_of_third_layer_neurons, int n_of_models) { int idx = threadIdx.x; int tdx = threadIdx.x; int bdx = blockIdx.x; int bdy = blockIdx.y; if (tdx == 0) { while (bdy < n_of_second_layer_neurons) { while (bdx < n_of_neurons) { first_layer_correction[bdy * n_of_neurons + bdx] = 0.0f; bdx += OptNofBlocksX; } if (bdx >= n_of_neurons) { bdx = blockIdx.x; } bdy += OptNofBlocksY; } } __threadfence(); bdx = blockIdx.x; bdy = blockIdx.y; while (bdy < n_of_second_layer_neurons) { while (bdx < n_of_neurons) { while (idx < n_of_images) { float main_multiplier; main_multiplier = first_layer_output[bdy*n_of_images + idx] * (1 - first_layer_output[bdy*n_of_images + idx])* sigmoid_multiplier[bdy*n_of_images + idx]; atomicAdd(&first_layer_correction[bdy*n_of_neurons + bdx], main_multiplier*images[idx*n_of_neurons + bdx]); idx += OptNofThreads; } if (idx >= n_of_images) { idx = threadIdx.x; } bdx += OptNofBlocksX; } if (bdx >= n_of_neurons) { bdx = blockIdx.x; } bdy += OptNofBlocksY; } } template <typename T> __global__ void first_layer_weights_update(T* first_layer_weights, T* first_layer_correction, T* previous_second_layer, float alpha, int n_of_neurons, int n_of_second_layer_neurons) { int idx = threadIdx.x; int bdx = blockIdx.x; while (bdx < n_of_second_layer_neurons) { while (idx < n_of_neurons) { previous_second_layer[bdx*n_of_neurons + idx] = first_layer_weights[bdx*n_of_neurons + idx]; first_layer_weights[bdx*n_of_neurons + idx] -= alpha * first_layer_correction[bdx*n_of_neurons + idx]; idx += OptNofThreads; } if (idx >= n_of_neurons) { idx = threadIdx.x; } bdx += OptNofBlocks; } } template <typename T> __global__ void first_layer_weights_back(T* first_layer_weights, T* previous_second_layer, int n_of_neurons, int n_of_second_layer_neurons) { int idx = threadIdx.x; int bdx = blockIdx.x; while (bdx < n_of_second_layer_neurons) { while (idx < n_of_neurons) { first_layer_weights[bdx*n_of_neurons + idx] = previous_second_layer[bdx*n_of_neurons + idx]; idx += OptNofThreads; } if (idx >= n_of_neurons) { idx = threadIdx.x; } bdx += OptNofBlocks; } }
10,891
/****************************************************************** File : lcsRedistributeParticles.cu Author : Mingcheng Chen Last Update : January 29th, 2013 *******************************************************************/ #include <stdio.h> #define BLOCK_SIZE 512 __device__ inline int Sign(double a, double eps) { return a < -eps ? -1 : a > eps; } __device__ inline int GetLocalTetID(int blockID, int tetID, int *startOffsetsInLocalIDMap, int *blocksOfTets, int *localIDsOfTets) { // blockID and tetID are all global IDs. int offset = startOffsetsInLocalIDMap[tetID]; int endOffset = -1; while (1) { if (blocksOfTets[offset] == blockID) return localIDsOfTets[offset]; if (endOffset == -1) endOffset = startOffsetsInLocalIDMap[tetID + 1]; offset++; if (offset >= endOffset) return -1; } } __device__ inline int GetBlockID(int x, int y, int z, int numOfBlocksInY, int numOfBlocksInZ) { return (x * numOfBlocksInY + y) * numOfBlocksInZ + z; } __global__ void CollectActiveBlocksKernel(int *activeParticles, int *exitCells, double *placesOfInterest, int *localTetIDs, int *blockLocations, int *interestingBlockMap, int *startOffsetsInLocalIDMap, int *blocksOfTets, int *localIDsOfTets, int *interestingBlockMarks, int *activeBlocks, int *activeBlockIndices, int *numOfActiveBlocks, // Initially 0 int mark, int numOfActiveParticles, //int numOfStages, int numOfBlocksInX, int numOfBlocksInY, int numOfBlocksInZ, double globalMinX, double globalMinY, double globalMinZ, double blockSize, double epsilon) { int globalID = blockDim.x * blockIdx.x + threadIdx.x; if (globalID < numOfActiveParticles) { int particleID = activeParticles[globalID]; double posX = placesOfInterest[particleID * 3]; double posY = placesOfInterest[particleID * 3 + 1]; double posZ = placesOfInterest[particleID * 3 + 2]; int x = (int)((posX - globalMinX) / blockSize); int y = (int)((posY - globalMinY) / blockSize); int z = (int)((posZ - globalMinZ) / blockSize); // Intuitive block ID int blockID = GetBlockID(x, y, z, numOfBlocksInY, numOfBlocksInZ); int tetID = exitCells[particleID]; int localTetID = GetLocalTetID(blockID, tetID, startOffsetsInLocalIDMap, blocksOfTets, localIDsOfTets); /// DEBUG /// /* if (particleID == 303) printf("x, y, z: %d %d %d\n", x, y, z); */ if (localTetID == -1) { int dx[3], dy[3], dz[3]; int lx = 1, ly = 1, lz = 1; dx[0] = dy[0] = dz[0] = 0; double xLower = globalMinX + x * blockSize; double yLower = globalMinY + y * blockSize; double zLower = globalMinZ + z * blockSize; if (!Sign(xLower - posX, 10 * epsilon)) dx[lx++] = -1; if (!Sign(yLower - posY, 10 * epsilon)) dy[ly++] = -1; if (!Sign(zLower - posZ, 10 * epsilon)) dz[lz++] = -1; if (!Sign(xLower + blockSize - posX, 10 * epsilon)) dx[lx++] = 1; if (!Sign(yLower + blockSize - posY, 10 * epsilon)) dy[ly++] = 1; if (!Sign(zLower + blockSize - posZ, 10 * epsilon)) dz[lz++] = 1; // Check every necessary neightbor for (int i = 0; localTetID == -1 && i < lx; i++) for (int j = 0; localTetID == -1 && j < ly; j++) for (int k = 0; k < lz; k++) { if (i + j + k == 0) continue; int _x = x + dx[i]; int _y = y + dy[j]; int _z = z + dz[k]; if (_x < 0 || _y < 0 || _z < 0 || _x >= numOfBlocksInX || _y >= numOfBlocksInY || _z >= numOfBlocksInZ) continue; blockID = GetBlockID(_x, _y, _z, numOfBlocksInY, numOfBlocksInZ); /// DEBUG /// // if (particleID == 303 && tetID == 6825504) printf("_x = %d, _y = %d, _z = %d, blockID = %d\n", _x, _y, _z, blockID); localTetID = GetLocalTetID(blockID, tetID, startOffsetsInLocalIDMap, blocksOfTets, localIDsOfTets); if (localTetID != -1) break; } /// DEBUG /// if (localTetID == -1) { /* if (particleID == 303) { printf("%lf %lf %lf\n", posX, posY, posZ); printf("tetID = %d\n", tetID); printf("%lf %lf %lf\n", xLower, yLower, zLower); for (int i = 0; i < lx; i++) printf(" %d", dx[i]); printf("\n"); for (int i = 0; i < ly; i++) printf(" %d", dy[i]); printf("\n"); for (int i = 0; i < lz; i++) printf(" %d", dz[i]); printf("\n"); } return; */ while (1); } } // localTetID must not be -1 at that point. localTetIDs[particleID] = localTetID; int interestingBlockID = interestingBlockMap[blockID]; blockLocations[particleID] = interestingBlockID; int oldMark = atomicAdd(interestingBlockMarks + interestingBlockID, 0); int index; if (oldMark < mark) { int delta = mark - oldMark; int newMark = atomicAdd(interestingBlockMarks + interestingBlockID, delta); if (newMark >= mark) atomicAdd(interestingBlockMarks + interestingBlockID, -delta); else { index = atomicAdd(numOfActiveBlocks, 1); activeBlocks[index] = interestingBlockID; activeBlockIndices[interestingBlockID] = index; } } // This one cannot be calculated in that kernel //activeBlockOfParticles[particleID] = index; } } __global__ void GetNumOfParticlesByStageInBlocksKernel(int *numOfParticlesByStageInBlocks, int *particleOrders, int *stages, int *activeParticles, int *blockLocations, int *activeBlockIndices, int numOfStages, int numOfActiveParticles) { int globalID = blockDim.x * blockIdx.x + threadIdx.x; if (globalID < numOfActiveParticles) { int particleID = activeParticles[globalID]; int posi = activeBlockIndices[blockLocations[particleID]] * numOfStages + stages[particleID]; particleOrders[particleID] = atomicAdd(numOfParticlesByStageInBlocks + posi, 1); } } __global__ void CollectParticlesToBlocksKernel(int *numOfParticlesByStageInBlocks, // Now it is a prefix sum array. int *particleOrders, int *stages, int *activeParticles, int *blockLocations, int *activeBlockIndices, int *blockedParticleList, int numOfStages, int numOfActiveParticles ) { int globalID = blockDim.x * blockIdx.x + threadIdx.x; if (globalID < numOfActiveParticles) { int particleID = activeParticles[globalID]; int interestingBlockID = blockLocations[particleID]; int activeBlockID = activeBlockIndices[interestingBlockID]; int stage = stages[particleID]; int position = numOfParticlesByStageInBlocks[activeBlockID * numOfStages + stage] + particleOrders[particleID]; blockedParticleList[position] = particleID; } } extern "C" void CollectActiveBlocks(int *activeParticles, int *exitCells, double *placesOfInterest, int *localTetIDs, int *blockLocations, int *interestingBlockMap, int *startOffsetsInLocalIDMap, int *blocksOfTets, int *localIDsOfTets, int *interestingBlockMarks, int *activeBlocks, int *activeBlockIndices, int *numOfActiveBlocks, // Initially 0 int mark, int numOfActiveParticles, //int numOfStages, int numOfBlocksInX, int numOfBlocksInY, int numOfBlocksInZ, double globalMinX, double globalMinY, double globalMinZ, double blockSize, double epsilon) { dim3 dimBlock(BLOCK_SIZE, 1, 1); dim3 dimGrid((numOfActiveParticles - 1) / dimBlock.x + 1, 1, 1); CollectActiveBlocksKernel<<<dimGrid, dimBlock>>>(activeParticles, exitCells, placesOfInterest, localTetIDs, blockLocations, interestingBlockMap, startOffsetsInLocalIDMap, blocksOfTets, localIDsOfTets, interestingBlockMarks, activeBlocks, activeBlockIndices, numOfActiveBlocks, // Initially 0 mark, numOfActiveParticles, numOfBlocksInX, numOfBlocksInY, numOfBlocksInZ, globalMinX, globalMinY, globalMinZ, blockSize, epsilon); cudaError_t err = cudaDeviceSynchronize(); if (err) { cudaGetErrorString(err); exit(0); } } extern "C" void GetNumOfParticlesByStageInBlocks(int *numOfParticlesByStageInBlocks, int *particleOrders, int *stages, int *activeParticles, int *blockLocations, int *activeBlockIndices, int numOfStages, int numOfActiveParticles) { dim3 dimBlock(BLOCK_SIZE, 1, 1); dim3 dimGrid((numOfActiveParticles - 1) / dimBlock.x + 1, 1, 1); GetNumOfParticlesByStageInBlocksKernel<<<dimGrid, dimBlock>>>(numOfParticlesByStageInBlocks, particleOrders, stages, activeParticles, blockLocations, activeBlockIndices, numOfStages, numOfActiveParticles); cudaError_t err = cudaDeviceSynchronize(); if (err) { cudaGetErrorString(err); exit(0); } } extern "C" void CollectParticlesToBlocks(int *numOfParticlesByStageInBlocks, // Now it is a prefix sum array. int *particleOrders, int *stages, int *activeParticles, int *blockLocations, int *activeBlockIndices, int *blockedParticleList, int numOfStages, int numOfActiveParticles) { dim3 dimBlock(BLOCK_SIZE, 1, 1); dim3 dimGrid((numOfActiveParticles - 1 ) / dimBlock.x + 1, 1, 1); CollectParticlesToBlocksKernel<<<dimGrid, dimBlock>>>(numOfParticlesByStageInBlocks, // Now it is a prefix sum array. particleOrders, stages, activeParticles, blockLocations, activeBlockIndices, blockedParticleList, numOfStages, numOfActiveParticles); cudaError_t err = cudaDeviceSynchronize(); if (err) { cudaGetErrorString(err); exit(0); } }
10,892
#include <stdio.h> #include <string.h> #include <stdlib.h> #include <cuda.h> #include <cuda_runtime_api.h> #include <time.h> struct timespec start, end; long long int time_taken; __device__ int is_a_match(char *password_crack) { char password_number_1[] = "AA9996"; char password_number_2[] = "AS1234"; char password_number_3[] = "LK9091"; char password_number_4[] = "FD1223"; char *number_1 = password_crack; char *number_2 = password_crack; char *number_3 = password_crack; char *number_4 = password_crack; char *password_1 = password_number_1; char *password_2 = password_number_2; char *password_3 = password_number_3; char *password_4 = password_number_4; while(*number_1 == *password_1) { if(*number_1 == '\0') { printf("***match found*** %s\n",password_number_1); break; } number_1++; password_1++; } while(*number_2 == *password_2) { if(*number_2 == '\0') { printf("***match found*** %s\n",password_number_2); break; } number_2++; password_2++; } while(*number_3 == *password_3) { if(*number_3 == '\0') { printf("***match found*** %s\n",password_number_3); break; } number_3++; password_3++; } while(*number_4 == *password_4) { if(*number_4 == '\0') { printf("***match found*** %s\n",password_number_4); return 1; } number_4++; password_4++; } return 0; } __global__ void kernel() { char a,b,c,d; char password[7]; password[6] = '\0'; int i = blockIdx.x+65; int j = threadIdx.x+65; char firstValue = i; char secondValue = j; password[0] = firstValue; password[1] = secondValue; for(a='0'; a<='9'; a++){ for(b='0'; b<='9'; b++){ for(c='0';c<='9';c++){ for(d='0';d<='9';d++){ password[2] = a; password[3] = b; password[4]= c; password[5]=d; if(is_a_match(password)) { //printf("***match found***"); } else { //printf(" %s\n", password); } } } } } } int t_difference(struct timespec *start, struct timespec *end, long long int *difference) { long long int ds = end->tv_sec - start->tv_sec; long long int dn = end->tv_nsec - start->tv_nsec; if(dn < 0 ) { ds--; dn += 1000000000; } *difference = ds * 1000000000 + dn; return !(*difference > 0); } int main() { clock_gettime(CLOCK_MONOTONIC, &start); kernel <<<26,26>>>(); cudaDeviceSynchronize(); clock_gettime(CLOCK_MONOTONIC, &end); t_difference(&start, &end, &time_taken); printf("Total time taken was %lldns or %0.9lfs\n", time_taken, (time_taken/1.0e9)); return 0; }
10,893
#include <stdio.h> #include <assert.h> void checkCUDAError(const char *msg); __global__ void ranksort(int *d_a, int *d_b) { extern __shared__ float sdata[]; int tid=threadIdx.x; int current=d_a[blockDim.x*blockIdx.x+tid], current_order=0; for(int i=0; i<blockIdx.x; i++){ sdata[tid]=sdata[tid+blockDim.x]=d_a[tid+blockDim.x*i]; __syncthreads(); for(int j=0; j<blockDim.x; j++) /* note for duplicate number */ if(sdata[tid+j]<=current)current_order++; __syncthreads(); } for(int i=blockIdx.x+1; i<gridDim.x; i++){ sdata[tid]=sdata[tid+blockDim.x]=d_a[tid+blockDim.x*i]; __syncthreads(); for(int j=0; j<blockDim.x; j++) /* note for duplicate number */ if(sdata[tid+j]<current)current_order++; __syncthreads(); } sdata[tid]=sdata[tid+blockDim.x]=current; __syncthreads(); for(int j=0; j<blockDim.x; j++) /* note for duplicate number */ if(sdata[tid+j]<current || (sdata[tid+j]==current && tid+j>=blockDim.x)) current_order++; __syncthreads(); d_b[current_order]=current; } void vector_print(int *v, int n) { for(int i=0; i<n; i++) printf("%d ", v[i]); printf("\n"); } int main( int argc, char** argv) { int *h_a; int *d_a; int *d_b; int numBlocks = 256; int numThreadsPerBlock = 256; int numThreads = numBlocks*numThreadsPerBlock; size_t memSize = numBlocks * numThreadsPerBlock * sizeof(int); h_a = (int *) malloc(memSize); cudaMalloc((void**)&d_a, memSize); cudaMalloc((void**)&d_b, memSize); for(int i=0; i<numThreads; i++){ h_a[i]=numThreads-i; } dim3 dimGrid(numBlocks); dim3 dimBlock(numThreadsPerBlock); size_t sharedMemSize=numThreadsPerBlock*2*sizeof(int); cudaMemcpy(d_a, h_a, memSize, cudaMemcpyHostToDevice); ranksort<<<dimGrid, dimBlock, sharedMemSize>>>(d_a, d_b); cudaThreadSynchronize(); checkCUDAError("kernel execution"); cudaMemcpy(h_a, d_b, memSize, cudaMemcpyDeviceToHost); checkCUDAError("cudaMemcpy"); for(int i = 1; i < numBlocks*numThreadsPerBlock; i++){ assert(h_a[i] >= h_a[i-1]); } cudaFree(d_a); cudaFree(d_b); free(h_a); return 0; } void checkCUDAError(const char *msg) { cudaError_t err = cudaGetLastError(); if( cudaSuccess != err) { fprintf(stderr, "Cuda error: %s: %s.\n", msg, cudaGetErrorString( err) ); exit(-1); } }
10,894
#include "includes.h" __global__ void threshKernel(unsigned char * image, unsigned char* moddedimage, int size, int threshold) { int i = blockIdx.x * blockDim.x + threadIdx.x; if (i < size) { if (image[i] > threshold) { moddedimage[i] = 255; } else { moddedimage[i] = 0; } } }
10,895
#include "includes.h" __global__ void CudaPermuteCudnnToPV( float *dest, float *src, int outFeatures, int ny, int nx, int inFeatures, int manyScaleX, int manyScaleY) { // parameter dimensions are in dest PV format int srcNx = nx / manyScaleX; int srcNy = ny / manyScaleY; int srcInFeatures = inFeatures * manyScaleX * manyScaleY; int kDest = (blockIdx.x * blockDim.x) + threadIdx.x; if (kDest < outFeatures * ny * nx * inFeatures) { int kOF = kDest / (ny * nx * inFeatures); int kY = (kDest % (ny * nx * inFeatures)) / (nx * inFeatures); int kX = (kDest % (nx * inFeatures)) / inFeatures; int kIF = (kDest % inFeatures); // Recalculate x, y, and f based on manyScale kIF = kIF + inFeatures * (kX % manyScaleX + (kY % manyScaleY) * manyScaleX); kX = kX / manyScaleX; kY = kY / manyScaleY; int sOF = srcInFeatures * srcNy * srcNx; int sIF = srcNy * srcNx; int sY = srcNx; int kSrc = kOF * sOF + kIF * sIF + kY * sY + kX; dest[kDest] = src[kSrc]; } }
10,896
#include <cuda.h> #include <stdio.h> #include <stdlib.h> #include <assert.h> #include <utility> using namespace std; __global__ void jacobiKernel(float* A, float* b, int matrix_size, float* x_prev, float* x_now) { int thread_idx = blockIdx.x * blockDim.x + threadIdx.x; if (thread_idx < matrix_size) { float sum = b[thread_idx]; int A_idx = matrix_size * thread_idx; for (int i = 0; i < matrix_size; ++i) sum -= thread_idx != i ? A[A_idx + i] * x_prev[i] : 0; x_now[thread_idx] = sum / A[A_idx + thread_idx]; } } int main(int argc, char *argv[]) { FILE* matrix_file = fopen(argv[1], "r"); int matrix_size = 0, iter_count = 20000; fscanf(matrix_file, "%d", &matrix_size); float* A = new float[matrix_size * matrix_size]; float* b = new float[matrix_size]; float* x = new float[matrix_size]; float* A_cuda, *b_cuda, *x_prev_cuda, *x_now_cuda; assert( cudaSuccess == cudaMalloc((void **) &A_cuda, matrix_size * matrix_size * sizeof(float))); assert( cudaSuccess == cudaMalloc((void **) &b_cuda, matrix_size * sizeof(float))); assert( cudaSuccess == cudaMalloc((void **) &x_prev_cuda, matrix_size * sizeof(float))); assert( cudaSuccess == cudaMalloc((void **) &x_now_cuda, matrix_size * sizeof(float))); for (int i = 0; i < matrix_size; i++) { x[i] = 0; for (int j = 0; j < matrix_size; j++) { fscanf(matrix_file, "%f", &A[i* matrix_size + j]); } fscanf(matrix_file, "%f", &b[i]); } cudaMemcpy(A_cuda, A, sizeof(float) * matrix_size * matrix_size, cudaMemcpyHostToDevice); cudaMemcpy(b_cuda, b, sizeof(float) * matrix_size, cudaMemcpyHostToDevice); cudaMemcpy(x_prev_cuda, x, sizeof(float) * matrix_size, cudaMemcpyHostToDevice); cudaMemcpy(x_now_cuda, x, sizeof(float) * matrix_size, cudaMemcpyHostToDevice); int blocks_count = 32; int threads_count = matrix_size / blocks_count + 1; for (int i = 0; i < iter_count; i++) { jacobiKernel<<< blocks_count, threads_count >>>(A_cuda, b_cuda, matrix_size, x_prev_cuda, x_now_cuda); swap(x_prev_cuda, x_now_cuda); } cudaMemcpy(x, x_prev_cuda, sizeof(float) * matrix_size, cudaMemcpyDeviceToHost); for (int i = 0; i < matrix_size; i++) { printf("%f ", x[i]); } cudaFree(A_cuda); cudaFree(b_cuda); cudaFree(x_prev_cuda); cudaFree(x_now_cuda); delete[] A; delete[] b; delete[] x; return 0; }
10,897
#include <stdio.h> __device__ __constant__ char d_coef[2]; __global__ void add_arrays(char *a, int *b) { a[threadIdx.x] = d_coef[0]; } /* char g_coef[2]={'a','b'}; void pre() { cudaMemcpyToSymbol(d_coef,g_coef,sizeof(char)*2); } */ void pre(); #define N 7 int main() { pre(); char a[N] = "Hello "; int b[N] = {15, 10, 6, 0, -11, 1,0}; char *ad; int *bd; const int csize = N*sizeof(char); const int isize = N*sizeof(int); // print the contents of a[] //printf("%s", a); // Allocate and Transfer memory to the device cudaMalloc( (void**)&ad, csize ); cudaMalloc( (void**)&bd, isize ); cudaMemcpy( ad, a, csize, cudaMemcpyHostToDevice ); cudaMemcpy( bd, b, isize, cudaMemcpyHostToDevice ); // Perform the array addition dim3 dimBlock( N ); dim3 dimGrid ( 1 ); add_arrays<<<dimGrid, dimBlock>>>(ad, bd); // Copy the Contents from the GPU cudaMemcpy( a, ad, csize, cudaMemcpyDeviceToHost ); cudaFree( ad ); // Display the results printf("%s\n", a); return EXIT_SUCCESS; }
10,898
#include <stdio.h> #define BLOCKSIZE 16 //Edited matrix multiply to calculate distance __global__ void distKernel(float *devA, float *devB, float *devC, int rows, int cols, int K) { //Get the thread's x and y locations for its run int idx = threadIdx.x + blockIdx.x * blockDim.x; int idy = threadIdx.y + blockIdx.y * blockDim.y; //Allocate shared memory to hold parts of A and B __shared__ float tileA[BLOCKSIZE][BLOCKSIZE]; __shared__ float tileB[BLOCKSIZE][BLOCKSIZE]; //Use sum to get the result for a specific element float sum = 0.0; //Use iter to see if the loop should be run again int iter = 0; do{ //Check if the x thread falls within bounds of the matrices if ((idy < rows) && (threadIdx.x + BLOCKSIZE*iter < K)){ tileA[threadIdx.y][threadIdx.x] = devA[threadIdx.x + idy*K + BLOCKSIZE*iter]; } else { tileA[threadIdx.y][threadIdx.x] = 0.0; } //Check if the y thread falls within bounds of the matrices if ((threadIdx.y + BLOCKSIZE*iter < K) && (idx < cols)){ tileB[threadIdx.y][threadIdx.x] = devB[idx + (threadIdx.y + BLOCKSIZE*iter)*cols]; } else { tileB[threadIdx.y][threadIdx.x] = 0.0; } //Sync to ensure that all of the data has been grabbed for the tiles in this warp __syncthreads(); //Sum the squared distance between the terms for (int i = 0; i < BLOCKSIZE; i++){ sum += (tileA[threadIdx.y][i] - tileB[i][threadIdx.x])*(tileA[threadIdx.y][i] - tileB[i][threadIdx.x]); } //Iterate the number done iter++; //Sync the threads again to ensure they have all done their work before going through the loop to get data __syncthreads(); //Check if the tiles have covered all of C } while (BLOCKSIZE*iter < K); //If the thread falls within the matrix C, fill in its element, scaled by alpha and beta if ((idy < rows) && (idx < cols)){ devC[idx + idy*cols] = sum; } }
10,899
#include "includes.h" __global__ void vector_addition (int *a, int *b, int *c, int n) { for (int i=0; i<n; i++) { c[i] = a[i] + b[i]; } }
10,900
/* ********************************************** * CS314 Principles of Programming Languages * * Fall 2020 * ********************************************** */ #include <stdio.h> #include <stdlib.h> __global__ void markFilterEdges_gpu(int * src, int * dst, int * matches, int * keepEdges, int numEdges) { /** YOUR CODE GOES BELOW **/ int num_threads = blockDim.x * gridDim.x;; int tid = blockDim.x * blockIdx.x + threadIdx.x; for (int i = tid; i < numEdges; i+=num_threads) { if(matches[src[i]] == -1 && matches[dst[i]] == -1) keepEdges[i] = 1; else keepEdges[i] = 0; } /** YOUR CODE GOES ABOVE **/ }