rsmith added inline comments.
================ Comment at: lib/CodeGen/CGBuiltin.cpp:1930-1931 + if (CGM.getCodeGenOpts().OptimizationLevel == 0) + // At -O0, we don't perform inlining, so we don't need to delay the + // processing. + return RValue::get(ConstantInt::get(Int32Ty, 0)); ---------------- Are there cases where a call to an `always_inline` function could change the outcome? ================ Comment at: lib/Sema/SemaExpr.cpp:4973-4974 + if ((ICEArguments & (1 << (ArgIx - 1))) != 0) + Arg = ConstantExpr::Create(Context, Arg); + ---------------- void wrote: > rsmith wrote: > > We should create this node as part of checking that the argument is an ICE > > (in `SemaBuiltinConstantArg`). > It turns out that `SemaBuiltinConstantArg` isn't called for > `__builtin_constant_p()`. Its argument type is `.`, which...I'm not sure what > that means. It looks like a vararg, but that doesn't make sense for the > builtin. > > Should I change its signature in `Builtins.def`? It's also marked `t`, which means "signature is meaningless, uses custom type checking". The argument to `__builtin_constant_p` isn't required to be an ICE, though, so we shouldn't be calling `SemaBuiltinConstantArg` for it / creating a `ConstantExpr` wrapping it, as far as I can tell. Repository: rC Clang https://reviews.llvm.org/D54355 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits