In this exercise you will learn how to create a fixed data flow using an
in-order queue
.
queue_benchmarking
Your old coworker is lazy, and has enqueued all the kernels in the same in-order queue. Performance is poor as kernels cannot execute concurrently. Your goal is to improve it:
vector_add
Using the application from the “Data and Dependencies” exercise which uses the
buffer/accessor model convert the queue
to in-order using the
property::queue::in_order
property, converting the data flow graph to a fixed
execution order.
Feel free to have the kernel functions execute in any order you like providing the necessary dependencies described in the “Data and Dependencies” exercise are still met.
Now do the same using the application from the “Data and Dependencies” exercise
which uses the USM model, again converting the queue
to in-order, converting
the data flow graph to a fixed execution order.
Again feel free to have the kernel functions execute in any order you like providing the necessary dependencies described in the “Data and Dependencies” exercise are still met.
Note that in the USM model when using an in-order queue
it is no longer
necessary to chain commands using event
s.
For DevCloud via JupiterLab follow these instructions.
For DPC++: instructions.
For AdaptiveCpp: instructions.