PyCUDA kernel timing errors

Question

Simple enough

start=cuda.Event()
func(args,block=blockdims)
cuda.memcpy_dtoh(d,h)
end=cuda.Event()

dur=start.time_till(end)
print dur

But I'm getting this error

File "gpu.py", line 161, in gpu_test
    dur=start.time_till(end)
pycuda._driver.LogicError: cuEventElapsedTime failed: invalid handle

This is as far as I can tell from the docs the correct usage. Anyone got any idea what I'm doing wrong?

jfs · Accepted Answer · 2011-04-18 16:57:56Z

1

Take a look at SimpleSpeedTest.py:

start=cuda.Event()
end=cuda.Event()

start.record() # start timing
func(args,block=blockdims)
cuda.memcpy_dtoh(d,h)
end.record() # end timing
# calculate the run length
end.synchronize()
millis = start.time_till(end)
print millis

answered Apr 18, 2011 at 16:57

jfs

417k210 gold badges1k silver badges1.7k bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

talonmies Over a year ago

This is how is should be done. A pair of record events need to be inserted in the stream around the action being timed. Synchronizing the stream will make the host wait until the stream is idle, which also ensures both record events have been processed by the driver.

Collectives™ on Stack Overflow

PyCUDA kernel timing errors

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related