================
@@ -548,7 +548,15 @@ TranslationUnitDecl *Decl::getTranslationUnitDecl() {
}
ASTContext &Decl::getASTContext() const {
- return getTranslationUnitDecl()->getASTContext();
+ if (const auto *TUD = dyn_cast<TranslationUnitDecl>(this))
+ return TUD->getASTContext();
----------------
nico wrote:
Ah, that doesn't build, I need to get a DC from somewhere, and a
TranslationUnitDecl has a nullptr DC 😅 I did measure this:
```
ASTContext &Decl::getASTContext() const {
// This is called a lot, e.g. for every attribute lookup. Only walk up to the
// TranslationUnitDecl once per DeclContext.
const DeclContext *DC = getDeclContext();
if (LLVM_LIKELY(DC && DC->CachedASTContext))
return *DC->CachedASTContext;
if (const auto *TUD = dyn_cast<TranslationUnitDecl>(this))
return TUD->getASTContext();
DC->CachedASTContext = &getTranslationUnitDecl()->getASTContext();
return *DC->CachedASTContext;
}
```
It's a line of code more. It's one instruction less at least on aarch64. It
does retire 0.3% fewer instructions, but takes a little more wall time
(possibly due to the dependent load).
Happy to post disassembly or more detailed numbers, but I think I like the code
as-is the most.
https://github.com/llvm/llvm-project/pull/226612
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
