I visited the gnu gsl website and i dont find the example there to solve a differential equation to be intuitive at all (especially because it is using 2nd order differential equation). https://www.gnu.org/software/gsl/manual/html_node/ODE-Example-programs.html#ODE-Example-programs
Can somebody guide about where to find a descriptive guide to how solve a very simple first order differetial equation.
For example, supoose my function is y'=x+2y (or any such function) then how do i write code in gsl to solve it with a given fixed step size and initial condition.
For y'=f(x,y)=x+2y the arrays have all dimension 1, which normally is something to avoid, but here it is instructional. For the explicit solvers, i.e., those not containing imp in the name, you do not need the Jacobian:
#include <stdio.h>
#include <gsl/gsl_errno.h>
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_odeiv2.h>
int odefunc (double x, const double y[], double f[], void *params)
{
f[0] = x+2*y[0];
return GSL_SUCCESS;
}
int * jac;
int main ()
{
int dim = 1;
gsl_odeiv2_system sys = {odefunc, NULL, dim, NULL};
gsl_odeiv2_driver * d = gsl_odeiv2_driver_alloc_y_new (&sys, gsl_odeiv2_step_rkf45, 1e-6, 1e-6, 0.0);
int i;
double x0 = 0.0, xf = 100.0; /* start and end of integration interval */
double x = x0;
double y[1] = { 1.0 }; /* initial value */
for (i = 1; i <= 100; i++)
{
double xi = x0 + i * (xf-x0) / 100.0;
int status = gsl_odeiv2_driver_apply (d, &x, xi, y);
if (status != GSL_SUCCESS)
{
printf ("error, return value=%d\n", status);
break;
}
printf ("%.8e %.8e\n", x, y[0]);
}
gsl_odeiv2_driver_free (d);
return 0;
}
You may want to look up the book "Introduction to Computational Modeling Using C and Open-Source Tools" by Jose M. Garrido.
Lutzl, Please review:
'#include <stdio.h>
#include <gsl/gsl_errno.h>
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_odeiv2.h>
int odefunc (double x, const double y[], double f[], void *params)
{
f[0] = x+2*y[0];
return GSL_SUCCESS;
}
int jac(double x , const double y[] ,double *dfdy , double dfdx[], void *params) {
gsl_matrix_view dfdy_mat= gsl_matrix_view_array(dfdy,1,1);
gsl_matrix *m= &dfdy_mat.matrix;
gsl_matrix_set(m,0,0,x);
dfdx[0]=2;
return GSL_SUCCESS;
}
int main ()
{
int dim =1;
gsl_odeiv2_system sys = {odefunc, jac, dim, NULL};
gsl_odeiv2_driver * d = gsl_odeiv2_driver_alloc_y_new (&sys, gsl_odeiv2_step_rk1imp,1e-7,1e-7, 0.0);
int i;
double x0 = 0.0, xf =1.0; /*al value */
while(x<xf)
{
double xi = x0 + 0.25;
int status = gsl_odeiv2_driver_apply (d, &x, xi, y);
if (status != GSL_SUCCESS)
{
printf ("error, return value=%d\n", status);
break;
}
printf ("%.8e %.8e\n", x, y[0]);
}
gsl_odeiv2_driver_free (d);
return 0;
}
'
Related
I've got an error, regarding calling JacobiSVD in my cuda function.
This is the part of the code that causing the error.
Eigen::JacobiSVD<Eigen::Matrix3d> svd( cov_e, Eigen::ComputeThinU | Eigen::ComputeThinV);
And this is the error message.
CUDA_voxel_building.cu(43): error: calling a __host__
function("Eigen::JacobiSVD , (int)2> ::JacobiSVD") from a __global__
function("kernel") is not allowed
I've used the following command to compile it.
nvcc -std=c++11 -D_MWAITXINTRIN_H_INCLUDED -D__STRICT_ANSI__ -ptx CUDA_voxel_building.cu
I'm using code 8.0 with eigen3 on ubuntu 16.04.
It seems like other functions such as eigen value decomposition also gives the same error.
Anyone knows a solution? I'm enclosing my code below.
//nvcc -ptx CUDA_voxel_building.cu
#include </usr/include/eigen3/Eigen/Core>
#include </usr/include/eigen3/Eigen/SVD>
/*
#include </usr/include/eigen3/Eigen/Sparse>
#include </usr/include/eigen3/Eigen/Dense>
#include </usr/include/eigen3/Eigen/Eigenvalues>
*/
__global__ void kernel(double *p, double *breaks,double *ind, double *mu, double *cov, double *e,double *v, int *n, char *isgood, int minpts, int maxgpu){
bool debuginfo = false;
int idx = threadIdx.x + blockIdx.x * blockDim.x;
if(debuginfo)printf("Thread %d got pointer\n",idx);
if( idx < maxgpu){
int s_ind = breaks[idx];
int e_ind = breaks[idx+1];
int diff = e_ind-s_ind;
if(diff >minpts){
int cnt = 0;
Eigen::MatrixXd local_p(3,diff) ;
for(int k = s_ind;k<e_ind;k++){
int temp_ind=ind[k];
//Eigen::Matrix<double, 3, diff> local_p;
local_p(1,cnt) = p[temp_ind*3];
local_p(2,cnt) = p[temp_ind*3+1];
local_p(3,cnt) = p[temp_ind*3+2];
cnt++;
}
Eigen::Matrix3d centered = local_p.rowwise() - local_p.colwise().mean();
Eigen::Matrix3d cov_e = (centered.adjoint() * centered) / double(local_p.rows() - 1);
Eigen::JacobiSVD<Eigen::Matrix3d> svd( cov_e, Eigen::ComputeThinU | Eigen::ComputeThinV);
/* Eigen::Matrix3d Cp = svd.matrixU() * svd.singularValues().asDiagonal() * svd.matrixV().transpose();
mu[idx]=p[ind[s_ind]*3];
mu[idx+1]=p[ind[s_ind+1]*3];
mu[idx+2]=p[ind[s_ind+2]*3];
e[idx]=svd.singularValues()(0);
e[idx+1]=svd.singularValues()(1);
e[idx+2]=svd.singularValues()(2);
n[idx] = diff;
isgood[idx] = 1;
for(int x = 0; x < 3; x++)
{
for(int y = 0; y < 3; y++)
{
v[x+ 3*y +idx*9]=svd.matrixV()(x, y);
cov[x+ 3*y +idx*9]=cov_e(x, y);
//if(debuginfo)printf("%f ",R[x+ 3*y +i*9]);
if(debuginfo)printf("%f ",Rm(x, y));
}
}
*/
} else {
mu[idx]=0;
mu[idx+1]=0;
mu[idx+2]=0;
e[idx]=0;
e[idx+1]=0;
e[idx+2]=0;
n[idx] = 0;
isgood[idx] = 0;
for(int x = 0; x < 3; x++)
{
for(int y = 0; y < 3; y++)
{
v[x+ 3*y +idx*9]=0;
cov[x+ 3*y +idx*9]=0;
}
}
}
}
}
First of all, Ubuntu 16.04 provides Eigen 3.3-beta1, which is not really recommended to be used. I would suggest upgrading to a more recent version. Furthermore, to include Eigen, write (e.g.):
#include <Eigen/Eigenvalues>
and compile with -I /usr/include/eigen3 (if you use the version provided by the OS), or better -I /path/to/local/eigen-version.
Then, as talonmies noted, you can't call host-functions from kernels, (I'm not sure at the moment, why JacobiSVD is not marked as device function), but in your case it would make much more sense to use Eigen::SelfAdjointEigenSolver, anyway. Since the matrix you are decomposing is fixed-size 3x3 you should actually use the optimized computeDirect method:
Eigen::SelfAdjointEigenSolver<Eigen::Matrix3d> eig; // default constructor
eig.computeDirect(cov_e); // works for 2x2 and 3x3 matrices, does not require loops
It seems the computeDirect even works on the beta version provided by Ubuntu (I'd still recommend to update).
Some unrelated notes:
The following is wrong, since you should start with index 0:
local_p(1,cnt) = p[temp_ind*3];
local_p(2,cnt) = p[temp_ind*3+1];
local_p(3,cnt) = p[temp_ind*3+2];
Also, you can write this in one line:
local_p.col(cnt) = Eigen::Vector3d::Map(p+temp_ind*3);
This line will not fit (unless diff==3):
Eigen::Matrix3d centered = local_p.rowwise() - local_p.colwise().mean();
What you probably mean is (local_p is actually 3xn not nx3)
Eigen::Matrix<double, 3, Eigen::Dynamic> centered = local_p.colwise() - local_p.rowwise().mean();
And when computing cov_e you need to .adjoint() the second factor, not the first.
You can avoid both 'big' matrices local_p and centered, by directly accumulating Eigen::Matrix3d sum2 and Eigen::Vector3d sum with sum2 += v*v.adjoint() and sum +=v and computing
Eigen::Vector3d mu = sum / diff;
Eigen::Matrix3d cov_e = (sum2 - mu*mu.adjoint()*diff)/(diff-1);
I'm building a CUDA kernel to compute the numerical N*N jacobian of a function, using finite differences; in the example I provided, it is the square function (each entry of the vector is squared). The host coded allocates in linear memory, while I'm using a 2-dimensional indexing in the kernel.
My issue is that I haven't found a way to sum on the diagonal of the matrices cudaMalloc'ed. My attempt has been to use the statement threadIdx.x == blockIdx.x as a condition for the diagonal, but instead it evaluates to true only for them both at 0.
Here is the kernel and EDIT: I posted the whole code as an answer, based on the suggestions in the comments (the main() is basically the same, while the kernel is not)
template <typename T>
__global__ void jacobian_kernel (
T * J,
const T t0,
const T tn,
const T h,
const T * u0,
const T * un,
const T * un_old)
{
T cgamma = 2 - sqrtf(2);
const unsigned int t = threadIdx.x;
const unsigned int b = blockIdx.x;
const unsigned int tid = t + b * blockDim.x;
/*__shared__*/ T temp_sx[BLOCK_SIZE][BLOCK_SIZE];
/*__shared__*/ T temp_dx[BLOCK_SIZE][BLOCK_SIZE];
__shared__ T sm_temp_du[BLOCK_SIZE];
T* temp_du = &sm_temp_du[0];
if (tid < N )
{
temp_sx[b][t] = un[t];
temp_dx[b][t] = un[t];
if ( t == b )
{
if ( tn == t0 )
{
temp_du[t] = u0[t]*0.001;
temp_sx[b][t] += temp_du[t]; //(*)
temp_dx[b][t] -= temp_du[t];
temp_sx[b][t] += ( abs( temp_sx[b][t] ) < 10e-6 ? 0.1 : 0 );
temp_dx[b][t] += ( abs( temp_dx[b][t] ) < 10e-6 ? 0.1 : 0 );
temp_sx[b][t] = ( temp_sx[b][t] == 0 ? 0.1 : temp_sx[b][t] );
temp_dx[b][t] = ( temp_dx[b][t] == 0 ? 0.1 : temp_dx[b][t] );
}
else
{
temp_du[t] = MAX( un[t] - un_old[t], 10e-6 );
temp_sx[b][t] += temp_du[t];
temp_dx[b][t] -= temp_du[t];
}
}
__syncthreads();
//J = f(tn, un + du)
d_func(tn, (temp_sx[b]), (temp_sx[b]), 1.f);
d_func(tn, (temp_dx[b]), (temp_dx[b]), 1.f);
__syncthreads();
J[tid] = (temp_sx[b][t] - temp_dx[b][t]) * powf((2 * temp_du[t]), -1);
//J[tid]*= - h*cgamma/2;
//J[tid]+= ( t == b ? 1 : 0);
//J[tid] = temp_J[tid];
}
}
The general procedure for computing the jacobian is
Copy un into every row of temp_sx and temp_dx
Compute du as a 0.01 magnitude from u0
Sum du to the diagonal of temp_sx, subtract du from the diagonal of temp_dx
Compute the square function on each entry of temp_sx and temp_dx
Subtract them and divide every entry by 2*du
This procedure can be summarized with (f(un + du*e_i) - f(un - du*e_i))/2*du.
My problem is to sum du to the diagonal of the matrices of temp_sx and temp_dx like I tried in (*). How can I achieve that?
EDIT: Now calling 1D blocks and threads; in fact, .y axis wasn't used at all in the kernel. I'm calling the kernel with a fixed amount of shared memory
Note that in int main() I'm calling the kernel with
#define REAL sizeof(float)
#define N 32
#define BLOCK_SIZE 16
#define NUM_BLOCKS ((N*N + BLOCK_SIZE - 1)/ BLOCK_SIZE)
...
dim3 dimGrid(NUM_BLOCKS,);
dim3 dimBlock(BLOCK_SIZE);
size_t shm_size = N*N*REAL;
jacobian_kernel <<< dimGrid, dimBlock, size_t shm_size >>> (...);
So that I attempt to deal with block-splitting the function calls. In the kernel to sum on the diagonal I used if(threadIdx.x == blockIdx.x){...}. Why isn't this correct? I'm asking it because while debugging and making the code print the statement, It only evaluates true if they both are 0. Thus du[0] is the only numerical value and the matrix becomes nan. Note that this approach worked with the first code I built, where instead I called the kernel with
jacobian_kernel <<< N, N >>> (...)
So that when threadIdx.x == blockIdx.x the element is on the diagonal. This approach doesn't fit anymore though, since now I need to deal with larger N (possibly larger than 1024, which is the maximum number of threads per block).
What statement should I put there that works even if the matrices are split into blocks and threads?
Let me know if I should share some other info.
Here is how I managed to solve my problem, based on the suggestion in the comments on the answer. The example is compilable, provided you put helper_cuda.h and helper_string.h in the same directory or you add -I directive to the CUDA examples include path, installed along with the CUDA toolkit. The relevant changes are only in the kernel; there's a minor change in the main() though, since I was calling double the resources to execute the kernel, but the .y axis of the grid of thread blocks wasn't even used at all, so it didn't generate any error.
#include <stdio.h>
#include <stdlib.h>
#include <iostream>
#include <math.h>
#include <assert.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include "helper_cuda.h"
#include "helper_string.h"
#include <fstream>
#ifndef MAX
#define MAX(a,b) ((a > b) ? a : b)
#endif
#define REAL sizeof(float)
#define N 128
#define BLOCK_SIZE 128
#define NUM_BLOCKS ((N*N + BLOCK_SIZE - 1)/ BLOCK_SIZE)
template <typename T>
inline void printmatrix( T mat, int rows, int cols);
template <typename T>
__global__ void jacobian_kernel ( const T * A, T * J, const T t0, const T tn, const T h, const T * u0, const T * un, const T * un_old);
template<typename T>
__device__ void d_func(const T t, const T u[], T res[], const T h = 1);
template<typename T>
int main ()
{
float t0 = 0.; //float tn = 0.;
float h = 0.1;
float* u0 = (float*)malloc(REAL*N); for(int i = 0; i < N; ++i){u0[i] = i+1;}
float* un = (float*)malloc(REAL*N); memcpy(un, u0, REAL*N);
float* un_old = (float*)malloc(REAL*N); memcpy(un_old, u0, REAL*N);
float* J = (float*)malloc(REAL*N*N);
float* A = (float*)malloc(REAL*N*N); host_heat_matrix(A);
float *d_u0;
float *d_un;
float *d_un_old;
float *d_J;
float *d_A;
checkCudaErrors(cudaMalloc((void**)&d_u0, REAL*N)); //printf("1: %p\n", d_u0);
checkCudaErrors(cudaMalloc((void**)&d_un, REAL*N)); //printf("2: %p\n", d_un);
checkCudaErrors(cudaMalloc((void**)&d_un_old, REAL*N)); //printf("3: %p\n", d_un_old);
checkCudaErrors(cudaMalloc((void**)&d_J, REAL*N*N)); //printf("4: %p\n", d_J);
checkCudaErrors(cudaMalloc((void**)&d_A, REAL*N*N)); //printf("4: %p\n", d_J);
checkCudaErrors(cudaMemcpy(d_u0, u0, REAL*N, cudaMemcpyHostToDevice)); assert(d_u0 != NULL);
checkCudaErrors(cudaMemcpy(d_un, un, REAL*N, cudaMemcpyHostToDevice)); assert(d_un != NULL);
checkCudaErrors(cudaMemcpy(d_un_old, un_old, REAL*N, cudaMemcpyHostToDevice)); assert(d_un_old != NULL);
checkCudaErrors(cudaMemcpy(d_J, J, REAL*N*N, cudaMemcpyHostToDevice)); assert(d_J != NULL);
checkCudaErrors(cudaMemcpy(d_A, A, REAL*N*N, cudaMemcpyHostToDevice)); assert(d_A != NULL);
dim3 dimGrid(NUM_BLOCKS); std::cout << "NUM_BLOCKS \t" << dimGrid.x << "\n";
dim3 dimBlock(BLOCK_SIZE); std::cout << "BLOCK_SIZE \t" << dimBlock.x << "\n";
size_t shm_size = N*REAL; //std::cout << shm_size << "\n";
//HERE IS A RELEVANT CHANGE OF THE MAIN, SINCE I WAS CALLING
//THE KERNEL WITH A 2D GRID BUT WITHOUT USING THE .y AXIS,
//WHILE NOW THE GRID IS 1D
jacobian_kernel <<< dimGrid, dimBlock, shm_size >>> (d_A, d_J, t0, t0, h, d_u0, d_un, d_un_old);
checkCudaErrors(cudaMemcpy(J, d_J, REAL*N*N, cudaMemcpyDeviceToHost)); //printf("4: %p\n", d_J);
printmatrix( J, N, N);
checkCudaErrors(cudaDeviceReset());
free(u0);
free(un);
free(un_old);
free(J);
}
template <typename T>
__global__ void jacobian_kernel (
const T * A,
T * J,
const T t0,
const T tn,
const T h,
const T * u0,
const T * un,
const T * un_old)
{
T cgamma = 2 - sqrtf(2);
const unsigned int t = threadIdx.x;
const unsigned int b = blockIdx.x;
const unsigned int tid = t + b * blockDim.x;
/*__shared__*/ T temp_sx[BLOCK_SIZE][BLOCK_SIZE];
/*__shared__*/ T temp_dx[BLOCK_SIZE][BLOCK_SIZE];
__shared__ T sm_temp_du;
T* temp_du = &sm_temp_du;
//HERE IS A RELEVANT CHANGE (*)
if ( t < BLOCK_SIZE && b < NUM_BLOCKS )
{
temp_sx[b][t] = un[t]; //printf("temp_sx[%d] = %f\n", t,(temp_sx[b][t]));
temp_dx[b][t] = un[t];
//printf("t = %d, b = %d, t + b * blockDim.x = %d \n",t, b, tid);
//HERE IS A NOTE (**)
if ( t == b )
{
//printf("t = %d, b = %d \n",t, b);
if ( tn == t0 )
{
*temp_du = u0[t]*0.001;
temp_sx[b][t] += *temp_du;
temp_dx[b][t] -= *temp_du;
temp_sx[b][t] += ( abs( temp_sx[b][t] ) < 10e-6 ? 0.1 : 0 );
temp_dx[b][t] += ( abs( temp_dx[b][t] ) < 10e-6 ? 0.1 : 0 );
temp_sx[b][t] = ( temp_sx[b][t] == 0 ? 0.1 : temp_sx[b][t] );
temp_dx[b][t] = ( temp_dx[b][t] == 0 ? 0.1 : temp_dx[b][t] );
}
else
{
*temp_du = MAX( un[t] - un_old[t], 10e-6 );
temp_sx[b][t] += *temp_du;
temp_dx[b][t] -= *temp_du;
}
;
}
//printf("du[%d] %f\n", tid, (*temp_du));
__syncthreads();
//printf("temp_sx[%d][%d] = %f\n", b, t, temp_sx[b][t]);
//printf("temp_dx[%d][%d] = %f\n", b, t, temp_dx[b][t]);
//d_func(tn, (temp_sx[b]), (temp_sx[b]), 1.f);
//d_func(tn, (temp_dx[b]), (temp_dx[b]), 1.f);
matvec_dev( tn, A, (temp_sx[b]), (temp_sx[b]), N, N, 1.f );
matvec_dev( tn, A, (temp_dx[b]), (temp_dx[b]), N, N, 1.f );
__syncthreads();
//printf("temp_sx_later[%d][%d] = %f\n", b, t, (temp_sx[b][t]));
//printf("temp_sx_later[%d][%d] - temp_dx_later[%d][%d] = %f\n", b,t,b,t, (temp_sx[b][t] - temp_dx[b][t]) / 2 * *temp_du);
//if (t == b ) printf( "2du[%d]^-1 = %f\n",t, powf((2 * *temp_du), -1));
J[tid] = (temp_sx[b][t] - temp_dx[b][t]) / (2 * *temp_du);
}
}
template<typename T>
__device__ void d_func(const T t, const T u[], T res[], const T h )
{
__shared__ float temp_u;
temp_u = u[threadIdx.x];
res[threadIdx.x] = h*powf( (temp_u), 2);
}
template <typename T>
inline void printmatrix( T mat, int rows, int cols)
{
std::ofstream matrix_out;
matrix_out.open( "heat_matrix.txt", std::ofstream::out);
for( int i = 0; i < rows; i++)
{
for( int j = 0; j <cols; j++)
{
double next = mat[i + N*j];
matrix_out << ( (next >= 0) ? " " : "") << next << " ";
}
matrix_out << "\n";
}
}
The relevant change is on (*). Before I used if (tid < N) which has two downsides:
First, it is wrong, since it should be tid < N*N, as my data is 2D, while tid is a global index which tracks all the data.
Even if I wrote tid < N*N, since I'm splitting the function calls into blocks, the t < BLOCK_SIZE && b < NUM_BLOCKS seems clearer to me in how the indexing is arranged in the code.
Moreover, the statement t == b in (**) is actually the right one to operate on the diagonal elements of the matrix. The fact that it was evaluated true only on 0 was because of my error right above.
Thanks for the suggestions!
Recently I am learning the examples in the book CUDA by JASON SANDERS.
the example of Juila Set makes a bad performance of 7032ms.
Here is the program:
#include <cuda.h>
#include <cuda_runtime.h>
#include <cpu_bitmap.h>
#include <book.h>
#define DIM 1024
struct cuComplex{
float r;
float i;
__device__ cuComplex(float a, float b) : r(a),i(b){
}
__device__ float magnitude2(void){
return r*r+i*i;
}
__device__ cuComplex operator *(const cuComplex& a){
return cuComplex(r*a.r-i*a.i, i*a.r+r*a.i);
}
__device__ cuComplex operator +(const cuComplex& a){
return cuComplex(r+a.r,i+a.i);
}
};
__device__ int julia(int x,int y){
const float scale = 1.5;
float jx = scale * (float)(DIM/2 - x)/(DIM/2);
float jy = scale * (float)(DIM/2 - y)/(DIM/2);
cuComplex c(-0.8,0.156);
cuComplex a(jx,jy);
int i = 0;
for(i = 0; i<200; i++){
a = a*a + c;
if(a.magnitude2() > 1000){
return 0;
}
}
return 1;
}
__global__ void kernel(unsigned char *ptr){
int x = blockIdx.x;
int y = blockIdx.y;
int offset = x + y*gridDim.x;
int juliaValue = julia(x,y);
ptr[offset*4 + 0] = 255*juliaValue;
ptr[offset*4 + 1] = 0;
ptr[offset*4 + 2] = 1;
ptr[offset*4 + 3] = 255;
}
int main(void){
CPUBitmap bitmap(DIM,DIM);
unsigned char * dev_bitmap;
dim3 grid(DIM,DIM);
dim3 blocks(DIM/16,DIM/16);
dim3 threads(16,16);
dim3 thread(DIM,DIM);
cudaEvent_t start,stop;
cudaEvent_t bitmapCpy_start,bitmapCpy_stop;
HANDLE_ERROR(cudaEventCreate(&start));
HANDLE_ERROR(cudaEventCreate(&stop));
HANDLE_ERROR(cudaEventCreate(&bitmapCpy_start));
HANDLE_ERROR(cudaEventCreate(&bitmapCpy_stop));
HANDLE_ERROR(cudaMalloc((void **)&dev_bitmap,bitmap.image_size()));
HANDLE_ERROR(cudaEventRecord(start,0));
kernel<<<grid,1>>>(dev_bitmap);
HANDLE_ERROR(cudaMemcpy(bitmap.get_ptr(),dev_bitmap,bitmap.image_size(),cudaMemcpyDeviceToHost));
//HANDLE_ERROR(cudaEventRecord(bitmapCpy_stop,0));
//HANDLE_ERROR(cudaEventSynchronize(bitmapCpy_stop));
// float copyTime;
// HANDLE_ERROR(cudaEventElapsedTime(©Time,bitmapCpy_start,bitmapCpy_stop));
HANDLE_ERROR(cudaEventRecord(stop,0));
HANDLE_ERROR(cudaEventSynchronize(stop));
float elapsedTime;
HANDLE_ERROR(cudaEventElapsedTime(&elapsedTime,start,stop));
//printf("Total time is %3.1f ms, time for copying is %3.1f ms \n",elapsedTime,copyTime);
printf("Total time is %3.1f ms\n",elapsedTime);
bitmap.display_and_exit();
HANDLE_ERROR(cudaEventDestroy(start));
HANDLE_ERROR(cudaEventDestroy(stop));
HANDLE_ERROR(cudaEventDestroy(bitmapCpy_start));
HANDLE_ERROR(cudaEventDestroy(bitmapCpy_stop));
HANDLE_ERROR(cudaFree(dev_bitmap));
}
I think the main factor that influences the performance is that the program above just run 1 thread in every block:
kernel<<<grid,1>>>(dev_bitmap);
so I change the kernel like the following:
__global__ void kernel(unsigned char *ptr){
int x = threadIdx.x + blockIdx.x*blockDim.x;
int y = threadIdx.y + blockIdx.y*blockDim.y;
int offset = x + y*gridDim.x*blockIdx.x;
int juliaValue = julia(x,y);
ptr[offset*4 + 0] = 255*juliaValue;
ptr[offset*4 + 1] = 0;
ptr[offset*4 + 2] = 1;
ptr[offset*4 + 3] = 255;
}
and call kernel:
dim3 blocks(DIM/16,DIM/16);
dim3 threads(16,16);
kernel<<<blocks,threads>>>(dev_bitmap);
I think this change is not a big deal, but when I ran it, it acted like that it ran into some endless loops, no image appeared and I couldn't do anything with my screen, just blocked there.
toolkit: cuda 5.5
system: ubuntu 12.04
When I run the original code you have posted here, I get a correct display and a time of ~340ms.
When I make your kernel change, I get an "unspecified launch error" on the kernel launch.
In your modified kernel, you have the following which is an incorrect computation:
int offset = x + y*gridDim.x*blockIdx.x;
When I change it to:
int offset = x + y*gridDim.x*blockDim.x;
I get normal execution and results, and an indicated time of ~10ms.
I'm attempting to compile the example code relating to the ODE solver, gsl/gsl_odeiv2, using g++. The code below is from their website :
http://www.gnu.org/software/gsl/manual/html_node/ODE-Example-programs.html
and compiles fine under gcc, but g++ throws the error
invalid conversion from 'void*' to 'int (*)(double, const double*, double*, double*,
void*)' [-fpermissive]
in the code :
#include <stdio.h>
#include <gsl/gsl_errno.h>
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_odeiv2.h>
int func (double t, const double y[], double f[], void *params)
{
double mu = *(double *)params;
f[0] = y[1];
f[1] = -y[0] - mu*y[1]*(y[0]*y[0] - 1);
return GSL_SUCCESS;
}
int * jac;
int main ()
{
double mu = 10;
gsl_odeiv2_system sys = {func, jac, 2, &mu};
gsl_odeiv2_driver * d = gsl_odeiv2_driver_alloc_y_new (&sys, gsl_odeiv2_step_rkf45, 1e-6, 1e-6, 0.0);
int i;
double t = 0.0, t1 = 100.0;
double y[2] = { 1.0, 0.0 };
for (i = 1; i <= 100; i++)
{
double ti = i * t1 / 100.0;
int status = gsl_odeiv2_driver_apply (d, &t, ti, y);
if (status != GSL_SUCCESS)
{
printf ("error, return value=%d\n", status);
break;
}
printf ("%.5e %.5e %.5e\n", t, y[0], y[1]);
}
gsl_odeiv2_driver_free (d);
return 0;
}
The error is given on the line
gsl_odeiv2_system sys = {func, jac, 2, &mu};
Any help in solving this issue would be fantastic. I'm hoping to include some stdlib elements, hence wanting to compile it as C++. Also, if I can get it to compile with g++-4.7, I could more easily multithread it using C++11's additions to the language. Thank you very much.
It looks like you have some problems with Jacobian. In your particular case you could just use NULL instead of jac in the definition of your system, i.e.
gsl_odeiv2_system sys = {func, NULL, 2, &mu};
In general you Jacobian must be a function with particular entries - see gsl manual - that is why your compiler is complaining.
Also, you may want to link the gsl library manually:
-L/usr/local/lib -lgsl
if you are on a linux system.
OK, so lets say I have an ( N x N ) matrix that I would like to process. This matrix is quite large for my computer, and if I try to send it to the device all at once I get a 'out of memory error.'
So is there a way to send sections of the matrix to the device? One way I can see to do it is copy portions of the matrix on the host, and then send these manageable copied portions from the host to the device, and then put them back together at the end.
Here is something I have tried, but the cudaMemcpy in the for loop returns error code 11, 'invalid argument.'
int h_N = 10000;
size_t h_size_m = h_N*sizeof(float);
h_A = (float*)malloc(h_size_m*h_size_m);
int d_N = 2500;
size_t d_size_m = d_N*sizeof(float);
InitializeMatrices(h_N);
int i;
int iterations = (h_N*h_N)/(d_N*d_N);
for( i = 0; i < iterations; i++ )
{
float* h_array_ref = h_A+(i*d_N*d_N);
cudasafe( cudaMemcpy(d_A, h_array_ref, d_size_m*d_size_m, cudaMemcpyHostToDevice), "cudaMemcpy");
cudasafe( cudaFree(d_A), "cudaFree(d_A)" );
}
What I'm trying to accomplish with the above code is this: instead of send the entire matrix to the device, I simply send a pointer to a place within that matrix and reserve enough space on the device to do the work, and then with the next iteration of the loop move the pointer forward within the matrix, etc. etc.
Not only can you do this (assuming your problem is easily decomposed this way into sub-arrays), it can be a very useful thing to do for performance; once you get the basic approach you've described working, you can start using asynchronous memory copies and double-buffering to overlap some of the memory transfer time with the time spent computing what is already on-card.
But first one gets the simple thing working. Below is a 1d example (multiplying a vector by a scalar and adding another scalar) but using a linearized 2d array would be the same; the key part is
CHK_CUDA( cudaMalloc(&xd, batchsize*sizeof(float)) );
CHK_CUDA( cudaMalloc(&yd, batchsize*sizeof(float)) );
tick(&gputimer);
int nbatches = 0;
for (int nstart=0; nstart < n; nstart+=batchsize) {
int size=batchsize;
if ((nstart + batchsize) > n) size = n - nstart;
CHK_CUDA( cudaMemcpy(xd, &(x[nstart]), size*sizeof(float), cudaMemcpyHostToDevice) );
blocksize = (size+nblocks-1)/nblocks;
cuda_saxpb<<<nblocks, blocksize>>>(xd, a, b, yd, size);
CHK_CUDA( cudaMemcpy(&(ycuda[nstart]), yd, size*sizeof(float), cudaMemcpyDeviceToHost) );
nbatches++;
}
gputime = tock(&gputimer);
CHK_CUDA( cudaFree(xd) );
CHK_CUDA( cudaFree(yd) );
You allocate the buffers at the start, and then loop through until you're done, each time doing the copy, starting the kernel, and then copying back. You free at the end.
The full code is
#include <stdio.h>
#include <stdlib.h>
#include <getopt.h>
#include <cuda.h>
#include <sys/time.h>
#include <math.h>
#define CHK_CUDA(e) {if (e != cudaSuccess) {fprintf(stderr,"Error: %s\n", cudaGetErrorString(e)); exit(-1);}}
__global__ void cuda_saxpb(const float *xd, const float a, const float b,
float *yd, const int n) {
int i = threadIdx.x + blockIdx.x*blockDim.x;
if (i<n) {
yd[i] = a*xd[i]+b;
}
return;
}
void cpu_saxpb(const float *x, float a, float b, float *y, int n) {
int i;
for (i=0;i<n;i++) {
y[i] = a*x[i]+b;
}
return;
}
int get_options(int argc, char **argv, int *n, int *s, int *nb, float *a, float *b);
void tick(struct timeval *timer);
double tock(struct timeval *timer);
int main(int argc, char **argv) {
int n=1000;
int nblocks=10;
int batchsize=100;
float a = 5.;
float b = -1.;
int err;
float *x, *y, *ycuda;
float *xd, *yd;
double abserr;
int blocksize;
int i;
struct timeval cputimer;
struct timeval gputimer;
double cputime, gputime;
err = get_options(argc, argv, &n, &batchsize, &nblocks, &a, &b);
if (batchsize > n) {
fprintf(stderr, "Resetting batchsize to size of vector, %d\n", n);
batchsize = n;
}
if (err) return 0;
x = (float *)malloc(n*sizeof(float));
if (!x) return 1;
y = (float *)malloc(n*sizeof(float));
if (!y) {free(x); return 1;}
ycuda = (float *)malloc(n*sizeof(float));
if (!ycuda) {free(y); free(x); return 1;}
/* run CPU code */
tick(&cputimer);
cpu_saxpb(x, a, b, y, n);
cputime = tock(&cputimer);
/* run GPU code */
/* only have to allocate once */
CHK_CUDA( cudaMalloc(&xd, batchsize*sizeof(float)) );
CHK_CUDA( cudaMalloc(&yd, batchsize*sizeof(float)) );
tick(&gputimer);
int nbatches = 0;
for (int nstart=0; nstart < n; nstart+=batchsize) {
int size=batchsize;
if ((nstart + batchsize) > n) size = n - nstart;
CHK_CUDA( cudaMemcpy(xd, &(x[nstart]), size*sizeof(float), cudaMemcpyHostToDevice) );
blocksize = (size+nblocks-1)/nblocks;
cuda_saxpb<<<nblocks, blocksize>>>(xd, a, b, yd, size);
CHK_CUDA( cudaMemcpy(&(ycuda[nstart]), yd, size*sizeof(float), cudaMemcpyDeviceToHost) );
nbatches++;
}
gputime = tock(&gputimer);
CHK_CUDA( cudaFree(xd) );
CHK_CUDA( cudaFree(yd) );
abserr = 0.;
for (i=0;i<n;i++) {
abserr += fabs(ycuda[i] - y[i]);
}
printf("Y = a*X + b, problemsize = %d\n", n);
printf("CPU time = %lg millisec.\n", cputime*1000.);
printf("GPU time = %lg millisec (done with %d batches of %d).\n",
gputime*1000., nbatches, batchsize);
printf("CUDA and CPU results differ by %lf\n", abserr);
free(x);
free(y);
free(ycuda);
return 0;
}
int get_options(int argc, char **argv, int *n, int *s, int *nb, float *a, float *b) {
const struct option long_options[] = {
{"nvals" , required_argument, 0, 'n'},
{"nblocks" , required_argument, 0, 'B'},
{"batchsize" , required_argument, 0, 's'},
{"a", required_argument, 0, 'a'},
{"b", required_argument, 0, 'b'},
{"help", no_argument, 0, 'h'},
{0, 0, 0, 0}};
char c;
int option_index;
int tempint;
while (1) {
c = getopt_long(argc, argv, "n:B:a:b:s:h", long_options, &option_index);
if (c == -1) break;
switch(c) {
case 'n': tempint = atoi(optarg);
if (tempint < 1 || tempint > 500000) {
fprintf(stderr,"%s: Cannot use number of points %s;\n Using %d\n", argv[0], optarg, *n);
} else {
*n = tempint;
}
break;
case 's': tempint = atoi(optarg);
if (tempint < 1 || tempint > 50000) {
fprintf(stderr,"%s: Cannot use number of points %s;\n Using %d\n", argv[0], optarg, *s);
} else {
*s = tempint;
}
break;
case 'B': tempint = atoi(optarg);
if (tempint < 1 || tempint > 1000 || tempint > *n) {
fprintf(stderr,"%s: Cannot use number of blocks %s;\n Using %d\n", argv[0], optarg, *nb);
} else {
*nb = tempint;
}
break;
case 'a': *a = atof(optarg);
break;
case 'b': *b = atof(optarg);
break;
case 'h':
puts("Calculates y[i] = a*x[i] + b on the GPU.");
puts("Options: ");
puts(" --nvals=N (-n N): Set the number of values in y,x.");
puts(" --batchsize=N (-s N): Set the number of values to transfer at a time.");
puts(" --nblocks=N (-B N): Set the number of blocks used.");
puts(" --a=X (-a X): Set the parameter a.");
puts(" --b=X (-b X): Set the parameter b.");
puts(" --niters=N (-I X): Set number of iterations to calculate.");
puts("");
return +1;
}
}
return 0;
}
void tick(struct timeval *timer) {
gettimeofday(timer, NULL);
}
double tock(struct timeval *timer) {
struct timeval now;
gettimeofday(&now, NULL);
return (now.tv_usec-timer->tv_usec)/1.0e6 + (now.tv_sec - timer->tv_sec);
}
Running this one gets:
$ ./batched-saxpb --nvals=10240 --batchsize=10240 --nblocks=20
Y = a*X + b, problemsize = 10240
CPU time = 0.072 millisec.
GPU time = 0.117 millisec (done with 1 batches of 10240).
CUDA and CPU results differ by 0.000000
$ ./batched-saxpb --nvals=10240 --batchsize=5120 --nblocks=20
Y = a*X + b, problemsize = 10240
CPU time = 0.066 millisec.
GPU time = 0.133 millisec (done with 2 batches of 5120).
CUDA and CPU results differ by 0.000000
$ ./batched-saxpb --nvals=10240 --batchsize=2560 --nblocks=20
Y = a*X + b, problemsize = 10240
CPU time = 0.067 millisec.
GPU time = 0.167 millisec (done with 4 batches of 2560).
CUDA and CPU results differ by 0.000000
The GPU time goes up in this case (we're doing more memory copies) but the answers stay the same.
Edited: The original version of this code had an option for running multiple iterations of the kernel for timing purposes, but that's unnecessarily confusing in this context so it's removed.