[uurl]

Summary

• Add 11 libm math families in both f32 and f64: erf, erfc, exp10, j0, lgamma, y0, y1, fdim, fmod, nextafter, and remainder.
• Generate typed scalar signatures and dispatch from the libdevice math shard. Coverage reaches 104 of libm 0.2.16's 124 f32/f64 exports.
• Validate 96 referenced libdevice signatures against CUDA 13.0 bitcode.
• Add an independent libm export fixture, lowering tests, and the math_aliases GPU example.

j1, jn, and yn remain unmapped: direct libdevice calls differ from libm for signed zero or negative orders. They require semantic adapters.

Validation

• cargo oxide run math_aliases: 416/416 checks passed on sm_120a.
• just check: passed.
• git diff --check: passed.

Stacked on #1649.

Refs #1224

Testing

• just -f cuda-oxide/Justfile check passes (the local mirror of CI: fmt, clippy, tests, guards, docs)
• cargo oxide run <example> passes, or cuda-oxide/scripts/smoketest.sh -o '^<example>$'
• New example added (if applicable)

Checklist

• All commits signed off (git commit -s)
• SPDX headers on new source files

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.