Skip to content
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

Accuracy tuning #189

Merged
merged 24 commits into from
Oct 3, 2023
Merged

Accuracy tuning #189

merged 24 commits into from
Oct 3, 2023

Conversation

stijnh
Copy link
Member

@stijnh stijnh commented Mar 22, 2023

This draft PR will integrate the ability to tune the arguments of kernels for different floating-point precision.

It is a draft for now. Let's see what the CI has to say.

@stijnh stijnh force-pushed the accuracy-tuning branch 2 times, most recently from c97dab9 to 095bf0d Compare May 9, 2023 10:06
@sonarcloud
Copy link

sonarcloud bot commented Jul 4, 2023

Kudos, SonarCloud Quality Gate passed!    Quality Gate passed

Bug A 0 Bugs
Vulnerability A 0 Vulnerabilities
Security Hotspot A 0 Security Hotspots
Code Smell A 16 Code Smells

No Coverage information No Coverage information
0.0% 0.0% Duplication

@benvanwerkhoven benvanwerkhoven marked this pull request as ready for review September 27, 2023 14:51
@sonarcloud
Copy link

sonarcloud bot commented Oct 2, 2023

Kudos, SonarCloud Quality Gate passed!    Quality Gate passed

Bug A 0 Bugs
Vulnerability A 0 Vulnerabilities
Security Hotspot A 0 Security Hotspots
Code Smell A 20 Code Smells

No Coverage information No Coverage information
0.0% 0.0% Duplication

@stijnh
Copy link
Member Author

stijnh commented Oct 3, 2023

This PR is ready to be merged! @benvanwerkhoven @fjwillemsen

In short, this PR adds the following:

  • A new type of observer called OutputObserver that is called with the output of the kernel once for each configuration.
  • A subclass of this observer called AccuracyObserver that can calculate an error on the output.
  • Built-in support for 10 types of error metrics (see error_metric_from_name).
  • A new data structure called Tunable that can be passed to a kernel to make one of its arguments tunable.
  • A subclass of this data structure called TunablePrecision that can be used to tune the precision of a kernel argument.
  • Changes in how metrics are calculated. Previously, cached entries were directly added to the results. Now, cached entries have their metrics recomputed in every run.
  • Adds an example of how to do accuracy tuning for CUDA.

@stijnh stijnh merged commit 2d5bff2 into KernelTuner:master Oct 3, 2023
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant