[origami] Major refactoring of codebase

[NaveenElumalaiAMD]

Technical Details

Rebase @neoblizz (Muhammad) changes (#1859) with latest develop

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

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

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

[math-ci]

perfci run on commit 5ac1a36

math-ci run

[Copilot]

Pull request overview

Copilot reviewed 40 out of 40 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[math-ci]

perfci run on commit ba05c37

math-ci run

[math-ci]

perfci run on commit 2d68ed0

math-ci run

[stellaraccident]

This patch needs to be reverted: ROCm/TheRock#2522 (comment)

Crashes the entire product and was clearly not tested. It cost devops a full day and a half to bisect and stalled everyone in the project for a week.

The level of detail around a commit of this magnitude is completely unacceptable and I will be calling a post mortem.

[jayhawk-commits]

It was determined that this commit introduces memory corruption issues with TheRock python packaging build. See ROCm/TheRock#2522 and need this commit reverted to move the rocm-libraries submodule pointer in TheRock forward.

Error

corrupted size vs. prev_size in fastbins

Affected Versions

• python 3.11, 3.12, 3.13
• pytorch 2.7, 2.8, 2.9, main branches
• TheRock with rocm-libraries submodule pointed to this commit or newer

Ways to Reproduce

1. Build ROCm via TheRock.
2. Build pytorch wheels via TheRock.
3. Run rocm-sdk test.
4. Run python ./external-builds/pytorch/run_pytorch_smoke_tests.py.

Error happens during steps 2-4. I have not seen an instance of a workflow getting past step 4.

Test Results A

• pytorch build workflow with rocm-libraries submodule at 283f438
• Run: https://github.com/ROCm/TheRock/actions/runs/20254886491/job/58154625946

Test Results B

• pytorch build workflow with rocm-libraries submodule at above SHA-1: e0fa9ee
• Run: https://github.com/ROCm/TheRock/actions/runs/20254276739/job/58154885187

Test Results A show corrupted size vs. prev_size in fastbins throughout the different python 3.11-3.13 runs.
Test Results B do not show the issue.