cudaMemcpy(dataDev, dataHost, mem_size, cudaMemcpyHostToDevice) execution time to long

Accelerated Computing CUDA CUDA Programming and Performance
#1

cudaMemcpy(dataDev, dataHost, mem_size, cudaMemcpyHostToDevice) execution time

this: http://openpaste.org/en/18964/ code

do this:

mem_size = 2098304

Processing time dev : 0.026260 (ms)

Processing time dev copy : 1023.973694 (ms)

Processing time host: 1487.460327 (ms)

Total Errors = 0

Press ENTER to exit...

output

“Processing time dev copy” not relative “mem_size”

but hard relative with “Processing time dev” and “Processing time host”

WTF ???

#2

Your timing is wrong. The kernel launch is asynchronous, so you need to do two things to fix this: use cudaThreadSynchronize() before stopping the timer…

…and wash you mouth out with soap.

#3

working well

mem_size = 2098304

Processing time dev : 1024.689697 (ms)

Processing time dev copy : 3.146769 (ms)

Processing time host: 1490.070679 (ms)

Total Errors = 0

Press ENTER to exit...

is anyone CUDA optimized version of linux crypt(3) function?