SPO600 2025 Winter Project - Stage 3: Plan (Part1)

 As part of the final stage of the SPO600 project, I will be extending and polishing my custom GCC pass (skim) to support analyzing multiple clone groups and provide accurate [PRUNE] and [NO PRUNE] recommendations. Below is a structured plan for how I intend to proceed with Stage III, in line with the official requirements and challenges outlined by professor.

Objectives

1. Support multiple clone groups

   • Remove the assumption that only one base+clone pair exists.

   • Detect and compare multiple function clone groups independently.

2. Build realistic test cases

   • Write at least two FMV-cloned functions per test case (e.g. scale_samples, sum_data).

   • Design one function to produce a [PRUNE] and another to produce a [NO PRUNE] result.

3. Ensure cross-architecture compatibility

   • Validate pass functionality on both x86_64 and aarch64.

   • Confirm that PRUNE/NOPRUNE logic works across target-specific clones.

4. Improve diagnostics

   • Add BB count and statement count diff messages to aid debugging.

   • Log detailed clone comparisons per function group.

5. Finalize cleanup and usability

• Keep my implementation style, avoiding major structure rewrites.

• Fix any logic gaps that led to silent failures in Stage II.

Test Plan

• Create a test with two functions:

  1. A simple sum_data() to encourage pruning.

  2. A complex scale_samples() to avoid pruning.

• Use __attribute__((target_clones(...))) to generate FMV variants.

• Compile with custom-built GCC and inspect .skim dumps for correct classification.

Goal

By the end of Stage III, my pass should be able to:

• Analyze multiple clone groups.

• Output detailed logs in .skim files.

• Provide both [PRUNE] and [NO PRUNE] results within the same compilation.

• Be usable by others through shared Makefile and test sources.

This wraps up the technical plan. Implementation and testing will follow in the next blog update.