[NativeAOT] ObjWriter: Fix over-alignment of .eh_frame entries

[filipnavara]

Ref: #95876 (comment)
Ref: https:/dotnet/llvm-project/blob/d8bacb4031b1d1e290ab2792e8378560419ee0de/llvm/lib/MC/MCDwarf.cpp#L1759-L1764

The DWARF specification requires the CIE/FDE entries in the .eh_frame sections to be aligned to pointer size. However, I incorrectly implemented it as the platform target pointer size, while it's supposed to be the DWARF pointer size specified in the pointer encoding field. Since we use 4-byte relative pointers on all platforms this means the alignment is supposed to be on the 4-byte boundary, not on the 8-byte boundary.

Tested on https:/aspnet/Benchmarks/tree/main/src/BenchmarksApps/BasicMinimalApi where we now produce the same size of .eh_frame section as the legacy ObjWriter. Prior to the fix the additional alignment contributed 53,736 bytes to the section size (out of roughly ~1MiB).

[ghost]

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Ref: #95876 (comment)
Ref: https:/dotnet/llvm-project/blob/d8bacb4031b1d1e290ab2792e8378560419ee0de/llvm/lib/MC/MCDwarf.cpp#L1759-L1764

The DWARF specification requires the CIE/FDE entries in the .eh_frame sections to be aligned to pointer size. However, I incorrectly implemented it as the platform target pointer size, while it's supposed to be the DWARF pointer size specified in the pointer encoding field. Since we use 4-byte relative pointers on all platforms this means the alignment is supposed to be on the 4-byte boundary, and not on the 8-byte boundary.

Author:	filipnavara
Assignees:	-
Labels:	area-NativeAOT-coreclr
Milestone:	-

[MichalStrehovsky]

Thanks!

[am11]

Unfortunately this missed preview 1 by few hours (#97336), @carlossanlop any chance we can include this change in P1? ilc-in-c# is a new feature, it will be tested more thoroughly by masses once P1 is announced.

[carlossanlop]

Not sure if we still have time, let me ask Tactics what the build status is.

@MichalStrehovsky @agocke while we confirm there's still runway, can you please take a look and let me know if you'd like to backport this change to release/9.0-preview1?

[jkotas]

This is a fix for relatively minor perf regression. I do not think it is needed in Preview 1.

[am11]

Agreed with @jkotas, it was a non-critical suggestion (if we still had time). :)

[carlossanlop]

Since we are still taking some more changes, we have time to get this backported. Let's do it.

[carlossanlop]

/backport to release/9.0-preview1

[github-actions]

Started backporting to release/9.0-preview1: https:/dotnet/runtime/actions/runs/7643752342