Measuring time between CUDA and sequencial software need help with the

Accelerated Computing CUDA CUDA Programming and Performance
1

What kind of resource can I use to measure time in Windows?

I’m trying to measure execution time between my CUDA algorithm and my Sequencial C algorithm.

The precision must be better then seconds. Milliseconds is fine.

I tried to use the GetSystemTime from an example on the internet but the precision was seconds.

2

Use the CUDA event API:

http://developer.download.nvidia.com/compu…854e6020a5.html

3

void startTimer(unsigned int timer, cudaEvent_t start)

{

cutilCheckError( cutStartTimer( timer)); // CPU timer

cutilSafeCall( cudaEventRecord( start, 0 ) ); //GPU timer

}

void stopTimer_DisplayElapsedTime(unsigned int timer, cudaEvent_t start, cudaEvent_t stop, float elapsedTimeInMs_GPU, float elapsedTimeInMs_CPU)

{

cutilCheckError( cutStopTimer( timer)); //CPU timer

cutilSafeCall( cudaEventRecord( stop, 0 ) ); // GPU timer

    cutilSafeCall( cudaThreadSynchronize() );

cutilSafeCall( cudaEventElapsedTime( &elapsedTimeInMs_GPU, start, stop ) );

elapsedTimeInMs_CPU = cutGetTimerValue( timer);

printf("elapsed time in ms : GPU counter = %lf , CPU counter = %lf \n", elapsedTimeInMs_GPU, elapsedTimeInMs_CPU);

}

4

Great! I’ll try that today! Thanks alot!