The sanitized lane's -fsanitize=address link needs the attribute to
instrument the emitted functions. Without a sanitizer pass it is not
inert: LLVM does not shrink-wrap functions that carry it, so every call
saved and restored all of a function's callee-saved registers on entry,
including on paths that return before any call, and it blocks
speculative loads. The LLVM helper already dropped it for builds that
import runtime bitcode (speed); release builds kept it.
- The new sanitizeAttributes option (default true for direct emission)
is set from --sanitize wherever objectAudit is.
- Without it and without a target CPU level, attribute group #0 holds a
string attribute LLVM ignores, since a group cannot be empty.
Release builds compile for the x86-64 baseline, so LLVM vectorizes with
SSE2 only. `--target-cpu=<baseline|x86-64-v2|x86-64-v3|x86-64-v4|native>`
compiles the program's code for a higher level: every emitted function
carries the level as its `target-cpu`, so LLVM can use AVX2, BMI2, FMA or
AVX-512. The executable then runs only on CPUs of that level.
- `native` resolves to the highest level the build machine supports, read
from /proc/cpuinfo on Linux and `sysctl machdep.cpu` on macOS.
- The level is part of the executable cache key; `baseline` and builds
without the option produce the same output and keys as before.
- The runtime pack keeps its baseline code.
- Targets other than x86-64 refuse a level (SC3002). The CLI's cached
bootstrap route leaves builds with the option to the full compiler.
- `bench:runtime` takes `--baseline-target-cpu` and `--candidate-target-cpu`.
Chris Tate