4 files changed, 30 insertions, 12 deletions
diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs
index 9f2d37d39d8..573c51a9539 100644
--- a/compiler/rustc_codegen_llvm/src/attributes.rs
+++ b/compiler/rustc_codegen_llvm/src/attributes.rs
@@ -296,6 +296,19 @@ pub(crate) fn tune_cpu_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribu
         .map(|tune_cpu| llvm::CreateAttrStringValue(cx.llcx, "tune-cpu", tune_cpu))
 }
 
+/// Get the `target-features` LLVM attribute.
+pub(crate) fn target_features_attr<'ll>(
+    cx: &CodegenCx<'ll, '_>,
+    function_features: Vec<String>,
+) -> Option<&'ll Attribute> {
+    let global_features = cx.tcx.global_backend_features(()).iter().map(String::as_str);
+    let function_features = function_features.iter().map(String::as_str);
+    let target_features =
+        global_features.chain(function_features).intersperse(",").collect::<String>();
+    (!target_features.is_empty())
+        .then(|| llvm::CreateAttrStringValue(cx.llcx, "target-features", &target_features))
+}
+
 /// Get the `NonLazyBind` LLVM attribute,
 /// if the codegen options allow skipping the PLT.
 pub(crate) fn non_lazy_bind_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribute> {
@@ -523,14 +536,7 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
         }
     }
 
-    let global_features = cx.tcx.global_backend_features(()).iter().map(|s| s.as_str());
-    let function_features = function_features.iter().map(|s| s.as_str());
-    let target_features: String =
-        global_features.chain(function_features).intersperse(",").collect();
-
-    if !target_features.is_empty() {
-        to_add.push(llvm::CreateAttrStringValue(cx.llcx, "target-features", &target_features));
-    }
+    to_add.extend(target_features_attr(cx, function_features));
 
     attributes::apply_to_llfn(llfn, Function, &to_add);
 }
diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs
index 326b876e7e6..ad2e722cfef 100644
--- a/compiler/rustc_codegen_llvm/src/back/lto.rs
+++ b/compiler/rustc_codegen_llvm/src/back/lto.rs
@@ -11,7 +11,7 @@ use object::{Object, ObjectSection};
 use rustc_codegen_ssa::back::lto::{SerializedModule, ThinModule, ThinShared};
 use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput};
 use rustc_codegen_ssa::traits::*;
-use rustc_codegen_ssa::{ModuleCodegen, looks_like_rust_object_file};
+use rustc_codegen_ssa::{ModuleCodegen, ModuleKind, looks_like_rust_object_file};
 use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::memmap::Mmap;
 use rustc_errors::DiagCtxtHandle;
@@ -225,9 +225,15 @@ fn fat_lto(
     // All the other modules will be serialized and reparsed into the new
     // context, so this hopefully avoids serializing and parsing the largest
     // codegen unit.
+    //
+    // Additionally use a regular module as the base here to ensure that various
+    // file copy operations in the backend work correctly. The only other kind
+    // of module here should be an allocator one, and if your crate is smaller
+    // than the allocator module then the size doesn't really matter anyway.
     let costliest_module = in_memory
         .iter()
         .enumerate()
+        .filter(|&(_, module)| module.kind == ModuleKind::Regular)
         .map(|(i, module)| {
             let cost = unsafe { llvm::LLVMRustModuleCost(module.module_llvm.llmod()) };
             (cost, i)
diff --git a/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs b/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs
index 1280ab1442a..0737a18384b 100644
--- a/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs
+++ b/compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs
@@ -193,7 +193,7 @@ fn gen_define_handling<'ll>(
     // reference) types.
     let num_ptr_types = types
         .iter()
-        .map(|&x| matches!(cx.type_kind(x), rustc_codegen_ssa::common::TypeKind::Pointer))
+        .filter(|&x| matches!(cx.type_kind(x), rustc_codegen_ssa::common::TypeKind::Pointer))
         .count();
 
     // We do not know their size anymore at this level, so hardcode a placeholder.
diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs
index 257c7b95666..a69fa54a54a 100644
--- a/compiler/rustc_codegen_llvm/src/context.rs
+++ b/compiler/rustc_codegen_llvm/src/context.rs
@@ -853,7 +853,7 @@ impl<'ll, 'tcx> MiscCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
     fn declare_c_main(&self, fn_type: Self::Type) -> Option<Self::Function> {
         let entry_name = self.sess().target.entry_name.as_ref();
         if self.get_declared_value(entry_name).is_none() {
-            Some(self.declare_entry_fn(
+            let llfn = self.declare_entry_fn(
                 entry_name,
                 llvm::CallConv::from_conv(
                     self.sess().target.entry_abi,
@@ -861,7 +861,13 @@ impl<'ll, 'tcx> MiscCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
                 ),
                 llvm::UnnamedAddr::Global,
                 fn_type,
-            ))
+            );
+            attributes::apply_to_llfn(
+                llfn,
+                llvm::AttributePlace::Function,
+                attributes::target_features_attr(self, vec![]).as_slice(),
+            );
+            Some(llfn)
         } else {
             // If the symbol already exists, it is an error: for example, the user wrote
             // #[no_mangle] extern "C" fn main(..) {..}