Lecture 11

Question 1

cudaDeviceSynchronize()

Answer 1

is used from the host.

Wait until all current kernels finish

Question 2

cudaStreamSynchronize()

Answer 2

waits until all kernels

in a stream finish.

Question 3

__syncthreads()

Answer 3

is used inside a kernel.

Stop thread until all threads reach the location!

Question 4

cudaEventCreate

Answer 4

initialize an event variable

Question 5

cudaEventRecord

Answer 5

place a marker in the queue

Question 6

cudaEventSynchronize

Answer 6

wait until all markers

have received values

Question 7

cudaEventElapsedTime

Answer 7

get the time difference

between two events

Question 8

Coalescing memory

Answer 8

Always access global memory ”in order”
If threads access global memory in order of thread
numbers, performance will be improved!

Question 9

CUDA can be coupled closer to ________

Answer 9

OpenGL