KI Sub-groups

[christiangnrd]

Includes #682

[github-actions]

Your PR no longer requires formatting changes. Thank you for your contribution!

[codecov]

Codecov Report

❌ Patch coverage is 57.77778% with 19 lines in your changes missing coverage. Please review.
✅ Project coverage is 52.24%. Comparing base (6c027c7) to head (82d32db).

Files with missing lines	Patch %	Lines
src/pocl/backend.jl	42.30%	15 Missing ⚠️
src/intrinsics.jl	0.00%	3 Missing ⚠️
src/pocl/compiler/compilation.jl	90.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #668      +/-   ##
==========================================
+ Coverage   52.18%   52.24%   +0.06%     
==========================================
  Files          22       22              
  Lines        1694     1736      +42     
==========================================
+ Hits          884      907      +23     
- Misses        810      829      +19     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[christiangnrd]

This is mostly ready except that for correctness of algorithms, I think it requires us to know the subgroup size of a device at compile time. I tried to force a known subgroup size for openCL/pocl, but that seems to have broken on-kernel rand. I'm not sure how to best approach this. I currently work around this in OpenCL.jl to test that the rest works by forcing a workgroup size in the KI launch interface so that the non-KI rand kernel doesn't encounter this issue, but that's a hack.

I think the ideal solution would be to have get_num_sub_groups always return at least 1 so we can force a sub-group size.