Auto merge of #118770 - saethlin:fix-inline-never-uses, r=nnethercote

Fix cases where std accidentally relied on inline(never)

This PR increases the power of `-Zcross-crate-inline-threshold=always` so that it applies through `#[inline(never)]`. Note that though this is called "cross-crate-inlining" in this case especially it is _just_ lazy per-CGU codegen. The MIR inliner and LLVM still respect the attribute as much as they ever have.

Trying to bootstrap with the new `-Zcross-crate-inline-threshold=always` change revealed two bugs:

We have special intrinsics `assert_inhabited`, `assert_zero_valid`, and `assert_mem_uniniitalized_valid` which codegen backends will lower to nothing or a call to `panic_nounwind`.  Since we may not have any call to `panic_nounwind` in MIR but emit one anyway, we need to specially tell `MirUsedCollector` about this situation.

`#[lang = "start"]` is special-cased already so that `MirUsedCollector` will collect it, but then when we make it cross-crate-inlinable it is only assigned to a CGU based on whether `MirUsedCollector` saw a call to it, which of course we didn't.

---

I started looking into this because https://github.com/rust-lang/rust/pull/118683 revealed a case where we were accidentally relying on a function being `#[inline(never)]`, and cranking up cross-crate-inlinability seems like a way to find other situations like that.

r? `@nnethercote` because I don't like what I'm doing to the CGU partitioning code here but I can't come up with something much better

compiler/rustc_mir_transform/src/cross_crate_inline.rs

@@ -22,6 +22,18 @@ fn cross_crate_inlinable(tcx: TyCtxt<'_>, def_id: LocalDefId) -> bool {
         return false;
     }
 
+    // This just reproduces the logic from Instance::requires_inline.
+    match tcx.def_kind(def_id) {
+        DefKind::Ctor(..) | DefKind::Closure => return true,
+        DefKind::Fn | DefKind::AssocFn => {}
+        _ => return false,
+    }
+
+    // From this point on, it is valid to return true or false.
+    if tcx.sess.opts.unstable_opts.cross_crate_inline_threshold == InliningThreshold::Always {
+        return true;
+    }
+
     // Obey source annotations first; this is important because it means we can use
     // #[inline(never)] to force code generation.
     match codegen_fn_attrs.inline {
@@ -30,13 +42,6 @@ fn cross_crate_inlinable(tcx: TyCtxt<'_>, def_id: LocalDefId) -> bool {
         _ => {}
     }
 
-    // This just reproduces the logic from Instance::requires_inline.
-    match tcx.def_kind(def_id) {
-        DefKind::Ctor(..) | DefKind::Closure => return true,
-        DefKind::Fn | DefKind::AssocFn => {}
-        _ => return false,
-    }
-
     // Don't do any inference when incremental compilation is enabled; the additional inlining that
     // inference permits also creates more work for small edits.
     if tcx.sess.opts.incremental.is_some() {

compiler/rustc_monomorphize/src/collector.rs

@@ -176,6 +176,7 @@
 use rustc_middle::mir::{self, Location};
 use rustc_middle::query::TyCtxtAt;
 use rustc_middle::ty::adjustment::{CustomCoerceUnsized, PointerCoercion};
+use rustc_middle::ty::layout::ValidityRequirement;
 use rustc_middle::ty::print::with_no_trimmed_paths;
 use rustc_middle::ty::{
     self, AssocKind, GenericParamDefKind, Instance, InstanceDef, Ty, TyCtxt, TypeFoldable,
@@ -923,6 +924,21 @@ fn visit_instance_use<'tcx>(
         return;
     }
 
+    // The intrinsics assert_inhabited, assert_zero_valid, and assert_mem_uninitialized_valid will
+    // be lowered in codegen to nothing or a call to panic_nounwind. So if we encounter any
+    // of those intrinsics, we need to include a mono item for panic_nounwind, else we may try to
+    // codegen a call to that function without generating code for the function itself.
+    if let ty::InstanceDef::Intrinsic(def_id) = instance.def {
+        let name = tcx.item_name(def_id);
+        if let Some(_requirement) = ValidityRequirement::from_intrinsic(name) {
+            let def_id = tcx.lang_items().get(LangItem::PanicNounwind).unwrap();
+            let panic_instance = Instance::mono(tcx, def_id);
+            if should_codegen_locally(tcx, &panic_instance) {
+                output.push(create_fn_mono_item(tcx, panic_instance, source));
+            }
+        }
+    }
+
     match instance.def {
         ty::InstanceDef::Virtual(..) | ty::InstanceDef::Intrinsic(_) => {
             if !is_direct_call {

compiler/rustc_monomorphize/src/partitioning.rs

@@ -212,11 +212,17 @@ fn place_mono_items<'tcx, I>(cx: &PartitioningCx<'_, 'tcx>, mono_items: I) -> Pl
     let cgu_name_cache = &mut FxHashMap::default();
 
     for mono_item in mono_items {
-        // Handle only root items directly here. Inlined items are handled at
+        // Handle only root (GloballyShared) items directly here. Inlined (LocalCopy) items
-        // the bottom of the loop based on reachability.
+        // are handled at the bottom of the loop based on reachability, with one exception.
+        // The #[lang = "start"] item is the program entrypoint, so there are no calls to it in MIR.
+        // So even if its mode is LocalCopy, we need to treat it like a root.
         match mono_item.instantiation_mode(cx.tcx) {
             InstantiationMode::GloballyShared { .. } => {}
-            InstantiationMode::LocalCopy => continue,
+            InstantiationMode::LocalCopy => {
+                if Some(mono_item.def_id()) != cx.tcx.lang_items().start_fn() {
+                    continue;
+                }
+            }
         }
 
         let characteristic_def_id = characteristic_def_id_of_mono_item(cx.tcx, mono_item);

library/std/src/rt.rs

@@ -155,7 +155,6 @@ fn lang_start_internal(
 }
 
 #[cfg(not(test))]
-#[inline(never)]
 #[lang = "start"]
 fn lang_start<T: crate::process::Termination + 'static>(
     main: fn() -> T,