Force PIE on bootgen to fix PDI-generation segfault

[jgmelber]

Summary

Fixes a real, hardware-reproducible crash found while validating #3323 (ROCm mlir_aie wheel on Ryzen AI self-hosted runners): nearly every programming_examples/ test failed with

```
RuntimeError: [aiecc] Compilation failed with exit code 1:
Error: Command failed with exit code -2
Error message: Segmentation fault (core dumped)
Error generating PDI
```

on both `aie2-4col` and `aie2p-8col` — see full diagnosis in this comment.

Root cause: `bootgen` has no dependency on MLIR/LLVM (it only links `OpenSSL::SSL` + its own `bootgen-lib`), but it's configured as part of the same top-level CMake invocation as everything else, so it silently inherits whatever global compiler/linker defaults `find_package(MLIR)`/`include(HandleLLVMOptions)` set for that run. Building against the ROCm-built MLIR wheel flips `bootgen` from a PIE executable to a non-PIE one, while it still declares dynamic `NEEDED` libs and an `RPATH`. `auditwheel repair`'s `patchelf`-based RPATH rewrite (needed to point at vendored/renamed OpenSSL libs) then corrupts that non-standard layout — the binary `SIGSEGV`s (`SEGV_ACCERR`) jumping into its own non-executable `DYNAMIC` segment before `main()` even runs.

Confirmed via `strace`:
```
--- SIGSEGV {si_signo=SIGSEGV, si_code=SEGV_ACCERR, si_addr=0x4020c8} ---
```
`readelf -l` shows `0x4020c8` falls inside the `DYNAMIC` segment, which is mapped `RW` only — something is jumping there to execute code.

Fix: pin `POSITION_INDEPENDENT_CODE ON` directly on the `bootgen-lib` and `bootgen` CMake targets, so their ELF layout no longer depends on whatever the linked MLIR/LLVM package happens to set globally.

Verified locally:

• Before: `bootgen -help` segfaults immediately — even under `LD_TRACE_LOADED_OBJECTS=1` (what `ldd` uses internally), i.e. before `bootgen`'s own code runs. ELF type: `EXEC`.
• After (rebuilt `bootgen` target against the same ROCm MLIR content): ELF type is now `pie executable`, and `bootgen -help` runs correctly, printing the normal `Bootgen v2023.2` banner.
• Also rebuilt `aiecc` (which links the same `bootgen-lib`) to confirm no regression there.

This blocks #3323 and #3326 (which depends on #3323 passing) — once merged, #3323 should be re-run against a wheel built with this fix.

Test plan

• CI passes on this PR (should be a no-op for every other target)
• Re-run [TEST] ROCm mlir_aie wheel on Ryzen AI self-hosted runners #3323's wheel build + hardware test with this fix included and confirm programming_examples pass

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR hardens the build configuration for the bootgen host tool so its ELF type/layout is stable across different top-level MLIR/LLVM build flag configurations (notably ROCm wheel builds), preventing a reproducible segfault during PDI generation after auditwheel repair rewrites RPATHs.

Changes:

• Forces POSITION_INDEPENDENT_CODE ON on bootgen-lib to ensure it is compiled as PIC.
• Forces POSITION_INDEPENDENT_CODE ON on the bootgen executable target to ensure it is built as PIE (where supported by the toolchain/CMake).
• Adds in-file rationale documenting why target-local PIE/PIC is required (avoids inheriting MLIR/LLVM global flags).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[andrej]

Looks good. Did make the comment a bit more terse. I just left a similar comment on a recent PR from Erika -- I feel like AI tends to create too many/too verbose comments that fill up my human context window.