diff options
author | Fangrui Song <i@maskray.me> | 2025-07-18 09:51:21 -0700 |
---|---|---|
committer | GitHub <noreply@github.com> | 2025-07-18 09:51:21 -0700 |
commit | 73e4b589ba9526c72f495ca6898ed18d730d2db4 (patch) | |
tree | cb47ed45a62ea47be7eb0aef515d044f9b7b9a0a /llvm/lib/Support/ModRef.cpp | |
parent | 01213141357e4a79d2d97187ff0cb89d8d173634 (diff) | |
download | llvm-73e4b589ba9526c72f495ca6898ed18d730d2db4.zip llvm-73e4b589ba9526c72f495ca6898ed18d730d2db4.tar.gz llvm-73e4b589ba9526c72f495ca6898ed18d730d2db4.tar.bz2 |
MC: Simplify fragment reuse determination
First, avoid checking MCSubtargetInfo by reducing unnecessary overhead
introduced in https://reviews.llvm.org/D44928 . That change passed STI
to both FT_Data and FT_Relaxable fragments, but STI is only necessary
for FT_Relaxable.
The use of STI in FT_Data was added for:
* Bundle alignment mode, which has been removed (#148781).
* ARM, which inappropriately uses STI in `ARMAsmBackend::applyFixup` due
to tech debt, unlike other targets. All tests passed even without the
`copySTI` change.
To ensure safety, `copySTI` now starts a new fragment to prevent mixed
STI values.
Second, avoid checking LinkerRelaxable by eagerly starting a new
fragment when a FT_Data/FT_Align fragment is marked linker-relaxable.
There is currently an extra empty FT_Data if an alignment immediately
follows a linker-relaxable fragment, which will be improved in the
future when FT_Align information is moved to the variable-tail.
Pull Request: https://github.com/llvm/llvm-project/pull/149471
Diffstat (limited to 'llvm/lib/Support/ModRef.cpp')
0 files changed, 0 insertions, 0 deletions