modify some script and test and add build_llvm script

[XingerZhu]

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

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

[Copilot]

Pull request overview

This PR modernizes the rocDSL test infrastructure by standardizing path setup across test files, improving GPU compilation pipelines, and adding a new LLVM build script. The changes remove hardcoded developer-specific paths in favor of dynamic project root detection, update MLIR transform APIs to use newer functions with folding support, and implement workarounds for MLIR Python binding issues. Additionally, a new build_llvm.sh script simplifies the LLVM/MLIR build process for new contributors.

Key changes include:

• Standardized path setup pattern across 14 Python test files for portability
• Enhanced GPU compilation pipelines with explicit lowering pass ordering and reconciliation
• New build_llvm.sh script for automated LLVM/MLIR building
• C++ API updates to use applyPatternsAndFoldGreedily in transform passes
• Operation.create() workarounds for MLIR Python binding recursion bugs
• Polyfills for missing ROCDL operations in Python bindings

Reviewed changes

Copilot reviewed 25 out of 25 changed files in this pull request and generated 10 comments.

Show a summary per file

File	Description
tests/python/ir/test_rocir_coord_ops.py	Updated to use standardized project root detection for sys.path setup
tests/python/ir/test_product_divide.py	Updated to use standardized project root detection for sys.path setup
tests/python/ir/test_local_ops.py	Updated to use standardized project root detection for sys.path setup
tests/python/ir/test_basic_ops.py	Updated to use standardized project root detection for sys.path setup
tests/python/gpu/utils.py	Expanded GPU compilation pipeline with explicit control flow lowering passes before GPU-to-ROCDL conversion
tests/python/gpu/test_shared_working.py	Updated to use standardized project root detection for sys.path setup
tests/python/gpu/test_mfma_simple_working.py	Updated to use standardized project root detection for sys.path setup
tests/python/gpu/test_mfma_gemm_real_python_api.py	Added Operation.create() workaround for MFMA operations, enhanced compilation pipeline, added gpu.ModuleEndOp() terminator
tests/python/gpu/test_mfma_gemm_fp8_rocir.py	Added Operation.create() workaround for FP8 MFMA operations, enhanced compilation pipeline with additional lowering passes, added gpu.ModuleEndOp() terminator
tests/python/gpu/test_mfma_gemm_fp8_python_api.py	Added Operation.create() workaround for FP8 MFMA operations, enhanced compilation pipeline, added gpu.ModuleEndOp() terminator
tests/python/gpu/test_gpu_with_rocir_coords.py	Updated to use standardized project root detection for sys.path setup
tests/python/gpu/test_gpu_simple.py	Updated to use standardized project root detection for sys.path setup, added missing os import
tests/python/gpu/test_gpu_rocdsl.py	Updated to use standardized project root detection for sys.path setup
tests/python/gpu/test_gpu_layout.py	Updated to use standardized project root detection for sys.path setup
tests/python/examples/test_rocdl_ops.py	Updated to use standardized project root detection, added comments about potentially missing ROCDL operations
tests/python/conftest.py	Updated library paths to use dynamic project root detection instead of hardcoded absolute paths
python/rocdsl/rocir_opt_runner.py	Updated rocir-opt path to use dynamic project root detection
python/rocdsl/dialects/ext/rocir.py	Refactored operations to use Operation.create() API for crd2idx, size, cosize, and rank operations
python/rocdsl/dialects/ext/rocdl.py	Added polyfills for missing ROCDL operations (wavefrontsize, s_waitcnt, readlane, readfirstlane)
python/rocdsl/dialects/ext/gpu.py	Added automatic ModuleEndOp terminator to gpu.module region_op
python/rocdsl/compiler/pipeline.py	Updated rocir-opt path to use dynamic project root detection
lib/Transforms/RocirToStandard.cpp	Updated to use applyPatternsAndFoldGreedily instead of applyPatternsGreedily
lib/Transforms/RocirCoordLowering.cpp	Updated to use applyPatternsAndFoldGreedily instead of applyPatternsGreedily
build_llvm.sh	New script to automate LLVM/MLIR build process with release/19.x branch
README.md	Updated build instructions and test paths to reflect project structure changes

---

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

[Copilot]

The commented-out lines # ir.InsertionPoint.current.insert(op) are present but unnecessary. When using Operation.create() within an active InsertionPoint context, the operation is automatically inserted. These commented lines should be removed to clean up the code.

Suggested change

# ir.InsertionPoint.current.insert(op)

[Copilot]

The s_waitcnt polyfill returns .results (a list), while other polyfills like wavefrontsize, readlane, and readfirstlane return .results[0] (a single result). This inconsistency may cause issues if the function is expected to return a single result value. Consider returning .results[0] for consistency, or document why it returns a list.

Suggested change

	return Operation.create("rocdl.waitcnt", operands=[simm16]).results
	return Operation.create("rocdl.waitcnt", operands=[simm16]).results[0]

[Copilot]

The path contains a double slash // which is incorrect. Should be -DMLIR_DIR=$MLIR_PATH/lib/cmake/mlir with only a single slash separator.

Suggested change

	cmake .. -DMLIR_DIR=$MLIR_PATH//lib/cmake/mlir
	cmake .. -DMLIR_DIR=$MLIR_PATH/lib/cmake/mlir

[Copilot]

The insertion_point variable captured from the with statement is unused. Consider either removing the variable name (use just with ip or InsertionPoint.current:), or document why it's captured if there's a specific reason.

Suggested change

	with ip or InsertionPoint.current as insertion_point:
	with ip or InsertionPoint.current:

[Copilot]

The insertion_point variable captured from the with statement is unused. Consider either removing the variable name (use just with ip or InsertionPoint.current:), or document why it's captured if there's a specific reason.

[Copilot]

The commented-out line insertion_point.insert(op) suggests uncertainty about whether manual insertion is needed. When using Operation.create() within an active InsertionPoint context (from the with statement), the operation is automatically inserted. The comment should be removed for clarity, or if there was a specific reason to keep it commented, that reason should be documented.

Suggested change

# insertion_point.insert(op)

[Copilot]

The insertion_point variable captured from the with statement is unused. Consider either removing the variable name (use just with ip or InsertionPoint.current:), or document why it's captured if there's a specific reason.

[Copilot]

The insertion_point variable captured from the with statement is unused. Consider either removing the variable name (use just with ip or InsertionPoint.current:), or document why it's captured if there's a specific reason.

[Copilot]

The commented-out lines ir.InsertionPoint.current.insert(op) are present but unnecessary. When using Operation.create() within an active InsertionPoint context, the operation is automatically inserted. These commented lines should be removed to clean up the code.

Suggested change

# ir.InsertionPoint.current.insert(op)

[Copilot]

The commented-out lines # ir.InsertionPoint.current.insert(op) are present but unnecessary. When using Operation.create() within an active InsertionPoint context, the operation is automatically inserted. These commented lines should be removed to clean up the code.

Suggested change

# ir.InsertionPoint.current.insert(op)