[Origami] Reapply the origami refactor with fixes

[minsukim-amd]

Testing Plan

• Azure CI (deprecated): Runs all the Catch2 and new Python tests. ✅
• TheRock CI (integration):
  • CI workflow: https://github.com/ROCm/TheRock/actions/runs/20316557318 ✅
  • TheRock branch: https://github.com/ROCm/TheRock/tree/users/neoblizz/minsukim-refactor ✅ (Some tests fail but build passed)
• TheRock CI (libraries: hipblaslt, etc.) ✅
• Math CI (performance/hipblaslt) ✅
• Reviews from older PRs are addressed with the few exceptions. ✅

PRs History

• Original PR: [origami] Major refactoring of codebase #1859
• Rebased PR: [origami] Major refactoring of codebase #2718
• Reverted PR: Revert "[origami] Major refactoring of codebase (#2718)" #3416 (comment)
• Hot-fix PR: [Origami] Hot-fix for TheRock #3417 (review) (TheRock CI Report)
• And now this!

Technical Details

Refactor

• New file types.hpp, consolidates various origami types.
• New structs to replace the growing tuple.
• API updates to make it scalable, down from 20+ parameters to a few.
• Lots of redundant code removal.
• Reorganize into types.hpp; data types (see hardware.hpp for what needs to be moved)
• Refactor extract APIs
• Add enums for transpose
• Refactor debug/log reporting
• Remove mutable and statics
• Decouple latency out of config_t
• Rebase develop into this branch

Python APIs & Unit Tests

• Update the tests
• Update rocroller
• hipblaslt/tensilelite/scripts to use the new API

Testing Infrastructure

• Replace YAML-based tests with Catch2 C++ tests
• Replace GTest with Catch2 framework
• Convert YAML-driven parameterized tests to pure C++ tests
• Update common.hpp with direct C++ helper functions
• Update CMakeLists.txt to use Catch2 instead of GTest
• Remove Boost dependency (was only for YAML)
• All test data now hardcoded in C++ for type safety
• Better IDE support, debugging, and error messages

Questions

• Rocroller: coordinate on the API design, should they be reused?
• Reuse dim3_t from hip_runtime.h (does HIP's dim3 have the same functionality)?
• Should transpose/data-types be part of config as well for precomputing?

Motivation

This PR reapplies Origami project's refactor.

1. Make it simpler to make model updates.
2. Well-defined, scalable, clean interface.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

Work Moved to Future PRs

• Logger APIs
• Runtime Options
• Move instruction map into separate file
• Additional tests: [Origami] Improve Origami's Test Coverage #3301
• Minor batch-count specific changes: [Origami] Origami change for batch_count > 0 #3289
• Move Origami's Azure CI -> TheRock CI @ibrahimw1

[stellaraccident]

In the future, please avoid this style of omnibus PR. It is fine to have a big PR that touches many LOC, but the bigger the PR, the more focused it should be on one concern (ie. Api changes, test framework changes, code refactor, functional changes). If you have the git skills, you can prepare a curated stack of commits for rebase merge, but typically if a dev is having trouble managing PR scope anyway, they lack the skills for a more advanced workflow. So I only recommend that if you have a solid process and experience.

Thank you for the additional PR details, although it is still pretty light (typical of such an omnibus PR).

What was the fix to the original issue?

[neoblizz]

Thank you for the additional PR details, although it is still pretty light (typical of such an omnibus PR).

I'll work on adding more details.

What was the fix to the original issue?

@stellaraccident It was an ODR violation fixed in this commit: 093e67f, there were two static data members that were inside the .cpp file. There were multiple libraries that included the .hpp file and compiled the .cpp file. Each shared library emitted its own definition of static member causing the ODR violation. At program termination the destructor then was called twice, once per shared object.

[stellaraccident]

Thank you for the additional PR details, although it is still pretty light (typical of such an omnibus PR).

I'll work on adding more details.

What was the fix to the original issue?

@stellaraccident It was an ODR violation fixed in this commit: 093e67f, there were two static data members that were inside the .cpp file. There were multiple libraries that included the .hpp file and compiled the .cpp file. Each shared library emitted its own definition of static member causing the ODR violation. At program termination the destructor then was called twice, once per shared object.

So trivially would have been caught by enabling host ASAN. Did valgrind catch it? You guys do know that you have the knobs to be enabling asan in your dev and CI flows, right?

[stellaraccident]

They were enabled months ago and David knows. But no one prioritizing work on the blas team is allocating time for quality improvements.

[bnemanich]

I did run an asan build locally, but it didn't actually show any issues. valgrind did actually show a small memory leak, but did not catch the actual issue either.