Implement sleep() in OpenCL C [duplicate]

Question

This question already has an answer here:

• Calculate run time of kernel code in OpenCL C 1 answer

I want to measure the performance of different devices viz CPU and GPUs. This is my kernel code:

__kernel void dataParallel(__global int* A)
{  
    sleep(10);
    A[0]=2;
    A[1]=3;
    A[2]=5;
    int pnp;//pnp=probable next prime
    int pprime;//previous prime
    int i,j;
    for(i=3;i<10;i++)
    {
        j=0;
        pprime=A[i-1];
        pnp=pprime+2;
        while((j<i) && A[j]<=sqrt((float)pnp))
        {
            if(pnp%A[j]==0)
                {
                    pnp+=2;
                    j=0;
                }
            j++;

        }
        A[i]=pnp;

    }
}

However the sleep() function doesnt work. I am getting the following error in buildlog:

<kernel>:4:2: warning: implicit declaration of function 'sleep' is      invalid in C99
    sleep(10);
builtins: link error: Linking globals named '__gpu_suld_1d_i8_trap': symbol multiply defined!

Is there any other way to implement the function. Also is there a way to record the time taken to execute this code snippet.

P.S. I have included #include <unistd.h> in my host code.

Answer 1

You dont need to use sleep in your kernel to measure the execution time.

There are two ways to measure the time. 1. Use opencl inherent profiling look here: cl api

2. get timestamps in your hostcode and compare them before and after execution. example:

       double start = getTimeInMS();
       //The kernel starts here
       clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL, &tasksize, &local_size_in, 0, NULL, NULL)
   //wait for kernel execution
   clFinish(command_queue);
   cout << "kernel execution time " << (getTimeInMS() - start) << endl;
   

Where getTimeinMs() is a function that returns a double value of miliseconds: (windows specific, override with other implementation if you dont use windows)

static inline double getTimeInMS(){

SYSTEMTIME st;
GetLocalTime(&st);

return (double)st.wSecond * (double)1000 + (double)st.wMilliseconds;}

Also you want to:

#include <time.h>

For Mac it would be (could work on Linux as well, not sure):

 static inline double getTime() {
    struct timeval starttime;
    gettimeofday(&starttime, 0x0);


    return (double)starttime.tv_sec * (double)1000 + (double)starttime.tv_usec / (double)1000;}