Weird function run time

vision
1

I got a weird function run time in my code:

start1 = time.time()
func1()
end1 = time.time()
print(end1-start1)

start2 = time.time()
func2()
end2 = time.time()
print(end2-start2)

I try my best to optimize my code. First of all, the run time of func1 is about 0.012, and func2 is 0.0008.
I rewrite func1 by some powerful torch api and the run time of func1 is about 8*10^-5, but the run time of func2 is about 0.012…
I don’t change any thing in func2.
Why???
Thanks for anyone’s reply!

2

Are the functions using some operations of the GPU?
Since CUDA calls are asynchronous, you should synchronize before calculating stopping the timer.

...
func1()
torch.cuda.synchronize()
end1 = time.time()
...
3

Thank u! I think I find the real slower function!