[Test] Tests for the compilation pipeline.

[xiaoyao-NKU]

PR Description: [Test] Compilation Pipeline Testing Framework and Sunmmio Operator Verification

This PR introduces a compilation pipeline testing framework for the sunmmio-a4e hardware target and adds a series of end-to-end compilation tests for core operators and communication primitives.

Summary

Established a test_config-driven compilation pipeline verification mechanism. By comparing IR strings generated at different compilation stages (e.g., LowerTileOp, InjectSunmmioSync, MergeSharedMemoryAllocationsSunmmio, DeviceMode), the correctness of compilation transformations is ensured.

Key Changes

• Test Framework Enhancements:
  • Added compile_pipeline.py: Defines a unified compile_test interface supporting multi-stage IR validation.
  • Added formal_verify_funcs.py: Provides formal verification utility functions for validating critical metadata such as memory allocation sizes.
• Synchronization and Barrier Verification:
  • Validated the InjectSunmmioSync pass in test_sync.py.
  • Ensured automatic insertion of synchronization primitives such as T.sync_token_id, T.barrier_init, and T.barrier_arrive_and_wait.
• Communication Primitives Verification:
  • Covered compilation transformations of T.comm.broadcast, T.comm.put, and T.comm.all_gather under Mesh topology in test_comm.py.
• End-to-End Tests for Complex Operators:
  • Added test cases for FlashAttention (test_flashattn.py), SUMMA (test_summa.py), and multi-level MMA (test_mma_3times.py).

Technical Impact

• Memory Management Optimization: Verified that MergeSharedMemoryAllocationsSunmmio correctly merges memory allocations across different scopes (asram, wsram, rsram), effectively improving on-chip storage utilization.
• Pipeline Stability: Prevents regression errors in subsequent compiler optimization passes through fine-grained IR expectation matching.
• Hardware Feature Coverage: Improved automated test coverage for DMA copy synchronization and multi-core interaction logic in the sunmmio-a4e architecture.

Verification

• Ran all new test scripts: python3 testing/python/compile_pipeline/test_*.py.
• All test cases passed IR expectation matching and formal verification of memory sizes.