================
@@ -1353,6 +1353,19 @@ void CodeGenFunction::EmitLifetimeEnd(llvm::Value *Size,
llvm::Value *Addr) {
C->setDoesNotThrow();
}
+void CodeGenFunction::EmitFakeUse(Address Addr) {
+ // We do not emit a fake use if we want to apply optnone to this function,
+ // even if we might not apply it anyway due to minsize or similar attributes.
+ if (!CGM.getCodeGenOpts().DisableO0ImplyOptNone &&
+ CGM.getCodeGenOpts().OptimizationLevel == 0)
+ return;
----------------
SLTozer wrote:
The thoughts I have are that by default, in a typical `-O0` build,
`-fextend-lifetimes` is useless. But there may be atypical circumstances that
would make it useful - such as following the somewhat common pattern (in LLVM
development) of compiling with `-O0 -emit-llvm -disable-O0-optnone` to produce
unoptimized IR (except for compulsory optimizations) from clang to pass to
`opt` or some other tool that may modify it; in these cases if a user has
explicitly requested `-fextend-lifetimes`, it seems reasonable to me to still
emit fake uses. The argument _against_ emitting fake uses is that they have no
value, but given that the flag is opt-in only I think it's fine to trust that
the user isn't going to accidentally add a flag that they don't need.
https://github.com/llvm/llvm-project/pull/110102
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
