malloc & cudaMalloc confusion over initialization of the two

Accelerated Computing CUDA CUDA Programming and Performance
21

No sir,

its the execution time of the kernel.

22

No sir,

its the execution time of the kernel.

23

cudaThreadSynchronize measures the kernel time I suppose.

24

cudaThreadSynchronize measures the kernel time I suppose.

25

No, cudaThreadSynchronize() just stalls the host thread in a spinlock until the GPU is finished. If you are using host side timers, it is the preferred way to get a reasonable indication of execution times.

This measures only kernel launch time:

timerstartcode();

kernel <<>> ();

timestopcode();

whereas this measures both kernel launch and execution time:

timerstartcode();

kernel <<>> ();

cudaThreadSynchronize();

timestopcode();
26

No, cudaThreadSynchronize() just stalls the host thread in a spinlock until the GPU is finished. If you are using host side timers, it is the preferred way to get a reasonable indication of execution times.

This measures only kernel launch time:

timerstartcode();

kernel <<>> ();

timestopcode();

whereas this measures both kernel launch and execution time:

timerstartcode();

kernel <<>> ();

cudaThreadSynchronize();

timestopcode();
27

Sir, I am using this code to measure timing of the kernel, will this not work?

[b]cudaEventCreate(&start);

cudaEventCreate(&stop);

cudaEventRecord(start,0);

kernel<<<>>>

cudaEventRecord(stop,0);

cudaEventSynchronize(stop);

cudaEventElapsedTime(&time,start,stop);

printf(“\n\nProcessing time is:\t%f (ms)\n\n”,time);

cudaEventDestroy(start);

cudaEventDestroy(stop);

[/b]

28

Sir, I am using this code to measure timing of the kernel, will this not work?

[b]cudaEventCreate(&start);

cudaEventCreate(&stop);

cudaEventRecord(start,0);

kernel<<<>>>

cudaEventRecord(stop,0);

cudaEventSynchronize(stop);

cudaEventElapsedTime(&time,start,stop);

printf(“\n\nProcessing time is:\t%f (ms)\n\n”,time);

cudaEventDestroy(start);

cudaEventDestroy(stop);

[/b]