Skip to content

Conversation

griwes
Copy link
Contributor

@griwes griwes commented Oct 1, 2025

Description

Note: the "big keys shouldn't compile" case started compiling because we now actually do the proper math, presumably that's fine? I'd like some confirmation on that one.

Resolves #4362

Checklist

  • New or existing tests cover these changes.
  • The documentation is up to date with these changes.

@griwes griwes requested review from a team as code owners October 1, 2025 01:30
@griwes griwes requested review from NaderAlAwar and fbusato October 1, 2025 01:30
@github-project-automation github-project-automation bot moved this to Todo in CCCL Oct 1, 2025
@cccl-authenticator-app cccl-authenticator-app bot moved this from Todo to In Review in CCCL Oct 1, 2025
Copy link
Contributor

github-actions bot commented Oct 1, 2025

😬 CI Workflow Results

🟥 Finished in 6h 28m: Pass: 98%/185 | Total: 3d 12h | Max: 4h 09m | Hits: 94%/190718

See results here.

Copy link
Contributor

@NaderAlAwar NaderAlAwar left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good, we should keep that test instead of deleting it and just increase the size for now.


operation_t op = make_operation(
"op",
"struct large_key_pair { int a; char c[100]; };\n"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Important: instead of deleting this, let us keep it but increase the size of the dtype. Use char c[500] here instead. The reason the build succeeds after your changes is that the selected policy uses a smaller block size with less items per thread, which allows us to use a type of this size without running out of shared memory.

VSmem is still not supported, if you search for VSMemPerBlock in unique_by_key.cu you can see we return 0.

This issue is tracked here #3790

struct large_key_pair
{
int a;
char c[100];
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You will also want to update the size here

Copy link
Contributor

@bernhardmgruber bernhardmgruber left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

CUB part loooks good

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Status: In Review
Development

Successfully merging this pull request may close these issues.

Refactor unique_by_key in c.parallel to use json magic for reusing CUB tuning policies
3 participants