[llvm] [BOLT][WIP] Always treat function entry as code (PR #160161)

[Maksim Panchenko via llvm-commits]

maksfb wrote:

I think we can ignore x86 in the context of constant islands, since we don't create them for x86.

For aarch64, I checked openssl 1.1.1 and 3.x. Only in `libcrypto.so` from 1.1.1 there's a  "CI" at the start of a "function" (as identified by BOLT):
```
_vpaes_consts
_armv8_sha512_probe
ecp_nistz256_precomputed
iotas
```
Of those four, only `_armv8_sha512_probe` is marked with `ST_FUNC` and it is in fact code, just incorrectly specified with `.long` instead of `.inst`. You've fixed that in your local branch https://github.com/yavtuk/openssl/commit/28b80f26eed28cd0163e159663d870eafbb6804e

To summarize, it seems to be safe to resolve a conflict between `ST_FUNC` and `$d` in favor of code.

https://github.com/llvm/llvm-project/pull/160161