-
Notifications
You must be signed in to change notification settings - Fork 75
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Parallel algorithms POC #559
base: master
Are you sure you want to change the base?
Conversation
try execution::par_unseq for inhibition in SP. Results: takes much longer than seq.
This reverts commit b27b40c.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please review if there's an obvious error or if the MNIST_SP.cpp settings may be too artificial.
Otherwise my parallelization effort using c++17 TS Parallel has hit the reality.
- The problem is memory bound
execution::par_unseq
really puts all threads to a good use- MNIST is still slower than single-thread!
- significantly: 60s (single) vs 240+s (parallel)
- likely broken cache locality
- or even CPU throttling (or boosting the only signle thread running)
@@ -844,19 +845,25 @@ void SpatialPooler::inhibitColumnsGlobal_(const vector<Real> &overlaps, | |||
// faster than a regular sort because it stops after it partitions the | |||
// elements about the Nth element, with all elements on their correct side of | |||
// the Nth element. | |||
std::nth_element( | |||
tNth.start(); | |||
std::nth_element(htm::parallel::mode, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
inhibition identified as one of the slowest methods in SP.
nth_element
as the most significant in global inh.
parallel TS
For #214 #254
This is for your interest, not intended for merging yet.