CC cleanups and optimization

[jgarzik]

No description provided.

[Copilot]

Pull request overview

This PR refactors the cc (pcc) compiler to replace many string-based keyword/builtin/directive checks with deterministic StringId/tag lookups, and centralizes target-specific lowering decisions into a new “mapping” pass that expands complex IR ops into simpler sequences or rtlib calls.

Changes:

• Add a pre-interned keyword system (cc/kw.rs) with tag-based membership checks, and update parser/preprocessor/builtin checks to use StringId comparisons instead of &str.
• Introduce an architecture mapping pass (shared + per-arch mappers) to lower int128 ops, Float16 soft-float on x86-64, and long double rtlib calls on aarch64/Linux.
• Enhance developer tooling and tests: staged --dump-ir, lexer char-classification table + tests, and expanded codegen/IR tests for Float16 and int128 behaviors.

Reviewed changes

Copilot reviewed 29 out of 29 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
cc/token/preprocess.rs	Switch directive dispatch and __has_* evaluation to StringId/tag-based checks.
cc/token/lexer.rs	Replace match-based char classification with a compile-time lookup table; update tokenization and add unit tests.
cc/tests/codegen/misc.rs	Add large codegen regression tests for Float16 and int128 carry/shift/constant behaviors.
cc/strings.rs	Pre-intern all keyword strings at startup to ensure deterministic StringId assignments.
cc/rtlib.rs	Remove complex/longdouble/int128 rtlib name selection helpers and their tests (logic moved elsewhere).
cc/parse/parser.rs	Replace many keyword string comparisons with crate::kw StringId constants/tags.
cc/parse/mod.rs	Change nullability qualifier predicate to operate on StringId via tags.
cc/main.rs	Add kw module, recursion limit, staged --dump-ir with optional function filter, and integrate mapping pass in pipeline.
cc/lib.rs	Export kw module and set recursion limit for macro-heavy keyword generation.
cc/kw.rs	New keyword/tag system: deterministic IDs + bitmask tags + tests.
cc/ir/test_linearize.rs	Update Float16 conversion tests to expect IR conversion opcodes (mapping pass performs rtlib lowering).
cc/ir/mod.rs	Add new opcodes for int128 decomposition, add Instruction::call_with_abi, improve instruction display for conversions, add Function::create_reg_pseudo.
cc/ir/lower.rs	Use Function::create_reg_pseudo helper when creating temporaries.
cc/ir/linearize.rs	Remove direct rtlib emission/Float16 soft-float helpers and rely on mapping pass; route complex rtlib name selection through mapping utilities.
cc/builtins.rs	Add is_builtin_id fast-path using keyword tags.
cc/arch/x86_64/regalloc.rs	Adjust int128 pseudo tracking for new decomposition opcodes.
cc/arch/x86_64/mod.rs	Expose x86-64 mapping module.
cc/arch/x86_64/mapping.rs	New x86-64 mapper: expand int128 ops, map int128 div/mod + int128↔float, and expand Float16 soft-float sequences.
cc/arch/x86_64/expression.rs	Remove backend-handled int128 ops now expanded by mapping pass; add support for new decomposition ops and fix hi-half constant handling.
cc/arch/x86_64/codegen.rs	Dispatch new decomposition opcodes (Lo64/Hi64/Pair64/AddC/AdcC/SubC/SbcC/UMulHi).
cc/arch/mod.rs	Export the new shared mapping module.
cc/arch/codegen.rs	Fix hi-half emission of 128-bit immediates to avoid sign-extension artifacts.
cc/arch/aarch64/mod.rs	Expose aarch64 mapping module.
cc/arch/aarch64/mapping.rs	New aarch64 mapper: expand int128 ops, map int128 div/mod + int128↔float, and lower long double ops via rtlib on Linux.
cc/arch/aarch64/lir.rs	Remove AArch64 LIR ops that were used by backend int128 handling (now mapping-expanded).
cc/arch/aarch64/expression.rs	Limit backend int128 handling to shifts + truncation-from-128; add support for new decomposition ops.
cc/arch/aarch64/codegen.rs	Dispatch new decomposition opcodes (Lo64/Hi64/Pair64/AddC/AdcC/SubC/SbcC/UMulHi).