[FIRRTL] Fix mask repeat order in FlattenMemory

[Waryc]

Fixes #10105

Summary

Fix incorrect mask-bit repetition ordering in FlattenMemory when expanding aggregate memory masks.

The previous logic indexed repetition counts by FIRRTL bit index directly, while maskWidths is ordered MSB-to-LSB. This mismatch could assign repetition counts to the wrong mask bits for mixed-width aggregates.

Changes

• Use reverse index mapping when reading repetition count from maskWidths in FlattenMemory.cpp.
• Add and adjust flatten-memory.mlir checks to match corrected ordering.
• Add a MaskRepeatOrder regression case for a compact mixed-width aggregate.

Validation

• Built tools with Ninja: firtool, circt-opt, FileCheck.
• Verified targeted test via circt-opt + FileCheck on test/Dialect/FIRRTL/flatten-memory.mlir.
• Ran llvm-lit -sv test/Dialect/FIRRTL/flatten-memory.mlir.
• Re-ran firtool on m.fir (in the related issue) and confirmed corrected mask concatenation ordering in both FIR and Verilog outputs.

Assisted-by

• GPT-5.3-Codex: Created minimal reproducible examples, assisted in reading and guiding modifications to CIRCT code, contributed to regression tests, and drafted the bug report along with this pull request.
• DeepSeek-R1: Handled partial translation tasks.

My role: Reviewed and assumed responsibility for all AI-generated content, with minor textual refinements.

[Copilot]

Pull request overview

Fixes incorrect ordering when repeating/expanding aggregate memory mask bits during FlattenMemory, ensuring repetition counts align with FIRRTL bit ordering (MSB→LSB) rather than raw bit indices.

Changes:

• Adjust mask-bit repetition to use a reverse index mapping when looking up repetition counts.
• Update existing FileCheck patterns to reflect the corrected concatenation order.
• Add a regression test covering a compact mixed-width aggregate mask expansion case.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
test/Dialect/FIRRTL/flatten-memory.mlir	Updates existing checks for corrected mask concatenation order and adds a new regression module @MaskRepeatOrder.
lib/Dialect/FIRRTL/Transforms/FlattenMemory.cpp	Fixes repetition-count lookup by reversing the index used to access maskWidths during mask expansion.