On Windows with LTO, required procedures with external linkage need to
be added to @llvm.used to survive linker-level dead code elimination.
LLVM may generate implicit calls to runtime builtins (e.g., __extendhfsf2
for f16 conversions) during instruction lowering, after the IR is
finalized. Without @llvm.used, the linker discards these procedures
before the implicit calls are generated.
This adds required procedures to @llvm.used at creation time. The fix
is Windows-specific; other platforms handle this correctly.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
lld-link doesn't recognize /lldltojobs:N as a standalone flag and
treats it as a file path. Use /opt:lldltojobs=N instead.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
With ThinLTO, the linker runs sanitizer passes at link time via
-fsanitize= flags, where it has whole-program visibility. Running
them at bitcode emission too double-instruments every module,
producing hundreds of "Redundant instrumentation detected" warnings.
Per-function sanitize/no_sanitize attributes are preserved in the
bitcode and respected by the linker's pass.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add -lto:thin and -lto:thin-files CLI flags with validation
- Emit LLVM bitcode (.bc) instead of object files when LTO is enabled
- Pass -flto=thin and -flto-jobs to clang/lld linkers
- Guard linkage corrections to skip declarations without definitions
(required for LTO where declarations appear across modules)
- Allow module-per-file with LTO even at higher optimization levels
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
If no `-minimum-os-version` is given, ignore the `override-module`
warnings. The user not using, `-minimum-os-version` opts you into linker
warnings about target versions.
If a `-minimum-os-version` is provided, normalize it to a full version
`11` to `11.0.0` for example. The linker seems to want that when doing
LTO.
