diff --git a/compiler/rustc_borrowck/src/diagnostics/mod.rs b/compiler/rustc_borrowck/src/diagnostics/mod.rs
index 11561539f6ddb..0106e285604d6 100644
--- a/compiler/rustc_borrowck/src/diagnostics/mod.rs
+++ b/compiler/rustc_borrowck/src/diagnostics/mod.rs
@@ -4,7 +4,6 @@ use crate::session_diagnostics::{
     CaptureArgLabel, CaptureReasonLabel, CaptureReasonNote, CaptureReasonSuggest, CaptureVarCause,
     CaptureVarKind, CaptureVarPathUseCause, OnClosureNote,
 };
-use itertools::Itertools;
 use rustc_errors::{Applicability, Diag};
 use rustc_hir as hir;
 use rustc_hir::def::{CtorKind, Namespace};
@@ -226,16 +225,15 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
                         }
                     } else {
                         if autoderef_index.is_none() {
-                            autoderef_index =
-                                match place.projection.into_iter().rev().find_position(|elem| {
-                                    !matches!(
-                                        elem,
-                                        ProjectionElem::Deref | ProjectionElem::Downcast(..)
-                                    )
-                                }) {
-                                    Some((index, _)) => Some(place.projection.len() - index),
-                                    None => Some(0),
-                                };
+                            autoderef_index = match place.projection.iter().rposition(|elem| {
+                                !matches!(
+                                    elem,
+                                    ProjectionElem::Deref | ProjectionElem::Downcast(..)
+                                )
+                            }) {
+                                Some(index) => Some(index + 1),
+                                None => Some(0),
+                            };
                         }
                         if index >= autoderef_index.unwrap() {
                             buf.insert(0, '*');
diff --git a/compiler/rustc_codegen_gcc/src/asm.rs b/compiler/rustc_codegen_gcc/src/asm.rs
index 9b679019e96ce..06b14a1f118a3 100644
--- a/compiler/rustc_codegen_gcc/src/asm.rs
+++ b/compiler/rustc_codegen_gcc/src/asm.rs
@@ -541,7 +541,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
             let builtin_unreachable: RValue<'gcc> = unsafe {
                 std::mem::transmute(builtin_unreachable)
             };
-            self.call(self.type_void(), None, None, builtin_unreachable, &[], None);
+            self.call(self.type_void(), None, None, builtin_unreachable, &[], None, None);
         }
 
         // Write results to outputs.
diff --git a/compiler/rustc_codegen_gcc/src/builder.rs b/compiler/rustc_codegen_gcc/src/builder.rs
index f5cda81f6ab86..43cc46cfe682f 100644
--- a/compiler/rustc_codegen_gcc/src/builder.rs
+++ b/compiler/rustc_codegen_gcc/src/builder.rs
@@ -25,7 +25,7 @@ use rustc_middle::ty::layout::{
     FnAbiError, FnAbiOfHelpers, FnAbiRequest, HasParamEnv, HasTyCtxt, LayoutError, LayoutOfHelpers,
     TyAndLayout,
 };
-use rustc_middle::ty::{ParamEnv, Ty, TyCtxt};
+use rustc_middle::ty::{ParamEnv, Ty, TyCtxt, Instance};
 use rustc_span::def_id::DefId;
 use rustc_span::Span;
 use rustc_target::abi::{
@@ -592,12 +592,13 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         then: Block<'gcc>,
         catch: Block<'gcc>,
         _funclet: Option<&Funclet>,
+        instance: Option<Instance<'tcx>>,
     ) -> RValue<'gcc> {
         let try_block = self.current_func().new_block("try");
 
         let current_block = self.block.clone();
         self.block = try_block;
-        let call = self.call(typ, fn_attrs, None, func, args, None); // TODO(antoyo): use funclet here?
+        let call = self.call(typ, fn_attrs, None, func, args, None, instance); // TODO(antoyo): use funclet here?
         self.block = current_block;
 
         let return_value =
@@ -1667,6 +1668,7 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         func: RValue<'gcc>,
         args: &[RValue<'gcc>],
         funclet: Option<&Funclet>,
+        _instance: Option<Instance<'tcx>>,
     ) -> RValue<'gcc> {
         // FIXME(antoyo): remove when having a proper API.
         let gcc_func = unsafe { std::mem::transmute(func) };
diff --git a/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs b/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
index a6c8b72e851b2..cebd45c09aa39 100644
--- a/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
+++ b/compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
@@ -133,6 +133,7 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
                     func,
                     &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
                     None,
+                    None,
                 )
             }
             sym::likely => self.expect(args[0].immediate(), true),
@@ -401,7 +402,7 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
     fn abort(&mut self) {
         let func = self.context.get_builtin_function("abort");
         let func: RValue<'gcc> = unsafe { std::mem::transmute(func) };
-        self.call(self.type_void(), None, None, func, &[], None);
+        self.call(self.type_void(), None, None, func, &[], None, None);
     }
 
     fn assume(&mut self, value: Self::Value) {
@@ -1103,7 +1104,7 @@ fn try_intrinsic<'a, 'b, 'gcc, 'tcx>(
     dest: RValue<'gcc>,
 ) {
     if bx.sess().panic_strategy() == PanicStrategy::Abort {
-        bx.call(bx.type_void(), None, None, try_func, &[data], None);
+        bx.call(bx.type_void(), None, None, try_func, &[data], None, None);
         // Return 0 unconditionally from the intrinsic call;
         // we can never unwind.
         let ret_align = bx.tcx.data_layout.i32_align.abi;
@@ -1177,21 +1178,21 @@ fn codegen_gnu_try<'gcc>(
         let zero = bx.cx.context.new_rvalue_zero(bx.int_type);
         let ptr = bx.cx.context.new_call(None, eh_pointer_builtin, &[zero]);
         let catch_ty = bx.type_func(&[bx.type_i8p(), bx.type_i8p()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], None);
+        bx.call(catch_ty, None, None, catch_func, &[data, ptr], None, None);
         bx.ret(bx.const_i32(1));
 
         // NOTE: the blocks must be filled before adding the try/catch, otherwise gcc will not
         // generate a try/catch.
         // FIXME(antoyo): add a check in the libgccjit API to prevent this.
         bx.switch_to_block(current_block);
-        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None);
+        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None, None);
     });
 
     let func = unsafe { std::mem::transmute(func) };
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, func, &[try_func, data, catch_func], None);
+    let ret = bx.call(llty, None, None, func, &[try_func, data, catch_func], None, None);
     let i32_align = bx.tcx().data_layout.i32_align.abi;
     bx.store(ret, dest, i32_align);
 }
diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs
index 74539d4d49570..500904ce18805 100644
--- a/compiler/rustc_codegen_llvm/src/asm.rs
+++ b/compiler/rustc_codegen_llvm/src/asm.rs
@@ -466,11 +466,11 @@ pub(crate) fn inline_asm_call<'ll>(
 
             let call = if !labels.is_empty() {
                 assert!(catch_funclet.is_none());
-                bx.callbr(fty, None, None, v, inputs, dest.unwrap(), labels, None)
+                bx.callbr(fty, None, None, v, inputs, dest.unwrap(), labels, None, None)
             } else if let Some((catch, funclet)) = catch_funclet {
-                bx.invoke(fty, None, None, v, inputs, dest.unwrap(), catch, funclet)
+                bx.invoke(fty, None, None, v, inputs, dest.unwrap(), catch, funclet, None)
             } else {
-                bx.call(fty, None, None, v, inputs, None)
+                bx.call(fty, None, None, v, inputs, None, None)
             };
 
             // Store mark in a metadata node so we can map LLVM errors
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
index 63e59ea13fc35..a5a5ae73d77a1 100644
--- a/compiler/rustc_codegen_llvm/src/builder.rs
+++ b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -19,9 +19,12 @@ use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrs;
 use rustc_middle::ty::layout::{
     FnAbiError, FnAbiOfHelpers, FnAbiRequest, LayoutError, LayoutOfHelpers, TyAndLayout,
 };
-use rustc_middle::ty::{self, Ty, TyCtxt};
+use rustc_middle::ty::{self, Instance, Ty, TyCtxt};
 use rustc_span::Span;
-use rustc_symbol_mangling::typeid::{kcfi_typeid_for_fnabi, typeid_for_fnabi, TypeIdOptions};
+use rustc_symbol_mangling::typeid::{
+    kcfi_typeid_for_fnabi, kcfi_typeid_for_instance, typeid_for_fnabi, typeid_for_instance,
+    TypeIdOptions,
+};
 use rustc_target::abi::{self, call::FnAbi, Align, Size, WrappingRange};
 use rustc_target::spec::{HasTargetSpec, SanitizerSet, Target};
 use smallvec::SmallVec;
@@ -221,6 +224,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         then: &'ll BasicBlock,
         catch: &'ll BasicBlock,
         funclet: Option<&Funclet<'ll>>,
+        instance: Option<Instance<'tcx>>,
     ) -> &'ll Value {
         debug!("invoke {:?} with args ({:?})", llfn, args);
 
@@ -233,10 +237,10 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         }
 
         // Emit CFI pointer type membership test
-        self.cfi_type_test(fn_attrs, fn_abi, llfn);
+        self.cfi_type_test(fn_attrs, fn_abi, instance, llfn);
 
         // Emit KCFI operand bundle
-        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, llfn);
+        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, instance, llfn);
         let kcfi_bundle = kcfi_bundle.as_ref().map(|b| &*b.raw);
         if let Some(kcfi_bundle) = kcfi_bundle {
             bundles.push(kcfi_bundle);
@@ -1231,6 +1235,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         llfn: &'ll Value,
         args: &[&'ll Value],
         funclet: Option<&Funclet<'ll>>,
+        instance: Option<Instance<'tcx>>,
     ) -> &'ll Value {
         debug!("call {:?} with args ({:?})", llfn, args);
 
@@ -1243,10 +1248,10 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         }
 
         // Emit CFI pointer type membership test
-        self.cfi_type_test(fn_attrs, fn_abi, llfn);
+        self.cfi_type_test(fn_attrs, fn_abi, instance, llfn);
 
         // Emit KCFI operand bundle
-        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, llfn);
+        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, instance, llfn);
         let kcfi_bundle = kcfi_bundle.as_ref().map(|b| &*b.raw);
         if let Some(kcfi_bundle) = kcfi_bundle {
             bundles.push(kcfi_bundle);
@@ -1468,7 +1473,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
 
     pub(crate) fn call_intrinsic(&mut self, intrinsic: &str, args: &[&'ll Value]) -> &'ll Value {
         let (ty, f) = self.cx.get_intrinsic(intrinsic);
-        self.call(ty, None, None, f, args, None)
+        self.call(ty, None, None, f, args, None, None)
     }
 
     fn call_lifetime_intrinsic(&mut self, intrinsic: &str, ptr: &'ll Value, size: Size) {
@@ -1526,7 +1531,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
             format!("llvm.{instr}.sat.i{int_width}.f{float_width}")
         };
         let f = self.declare_cfn(&name, llvm::UnnamedAddr::No, self.type_func(&[src_ty], dest_ty));
-        self.call(self.type_func(&[src_ty], dest_ty), None, None, f, &[val], None)
+        self.call(self.type_func(&[src_ty], dest_ty), None, None, f, &[val], None, None)
     }
 
     pub(crate) fn landing_pad(
@@ -1554,6 +1559,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         default_dest: &'ll BasicBlock,
         indirect_dest: &[&'ll BasicBlock],
         funclet: Option<&Funclet<'ll>>,
+        instance: Option<Instance<'tcx>>,
     ) -> &'ll Value {
         debug!("invoke {:?} with args ({:?})", llfn, args);
 
@@ -1566,10 +1572,10 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         }
 
         // Emit CFI pointer type membership test
-        self.cfi_type_test(fn_attrs, fn_abi, llfn);
+        self.cfi_type_test(fn_attrs, fn_abi, instance, llfn);
 
         // Emit KCFI operand bundle
-        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, llfn);
+        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, instance, llfn);
         let kcfi_bundle = kcfi_bundle.as_ref().map(|b| &*b.raw);
         if let Some(kcfi_bundle) = kcfi_bundle {
             bundles.push(kcfi_bundle);
@@ -1601,6 +1607,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         &mut self,
         fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
+        instance: Option<Instance<'tcx>>,
         llfn: &'ll Value,
     ) {
         let is_indirect_call = unsafe { llvm::LLVMRustIsNonGVFunctionPointerTy(llfn) };
@@ -1622,7 +1629,11 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
                 options.insert(TypeIdOptions::NORMALIZE_INTEGERS);
             }
 
-            let typeid = typeid_for_fnabi(self.tcx, fn_abi, options);
+            let typeid = if let Some(instance) = instance {
+                typeid_for_instance(self.tcx, &instance, options)
+            } else {
+                typeid_for_fnabi(self.tcx, fn_abi, options)
+            };
             let typeid_metadata = self.cx.typeid_metadata(typeid).unwrap();
 
             // Test whether the function pointer is associated with the type identifier.
@@ -1644,6 +1655,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         &mut self,
         fn_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
+        instance: Option<Instance<'tcx>>,
         llfn: &'ll Value,
     ) -> Option<llvm::OperandBundleDef<'ll>> {
         let is_indirect_call = unsafe { llvm::LLVMRustIsNonGVFunctionPointerTy(llfn) };
@@ -1665,7 +1677,12 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
                 options.insert(TypeIdOptions::NORMALIZE_INTEGERS);
             }
 
-            let kcfi_typeid = kcfi_typeid_for_fnabi(self.tcx, fn_abi, options);
+            let kcfi_typeid = if let Some(instance) = instance {
+                kcfi_typeid_for_instance(self.tcx, &instance, options)
+            } else {
+                kcfi_typeid_for_fnabi(self.tcx, fn_abi, options)
+            };
+
             Some(llvm::OperandBundleDef::new("kcfi", &[self.const_u32(kcfi_typeid)]))
         } else {
             None
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index 2409b2e78d736..ab135e3ed6444 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -181,6 +181,7 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     simple_fn,
                     &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
                     None,
+                    Some(instance),
                 )
             }
             sym::likely => {
@@ -539,7 +540,7 @@ fn catch_unwind_intrinsic<'ll>(
 ) {
     if bx.sess().panic_strategy() == PanicStrategy::Abort {
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.call(try_func_ty, None, None, try_func, &[data], None);
+        bx.call(try_func_ty, None, None, try_func, &[data], None, None);
         // Return 0 unconditionally from the intrinsic call;
         // we can never unwind.
         let ret_align = bx.tcx().data_layout.i32_align.abi;
@@ -640,7 +641,7 @@ fn codegen_msvc_try<'ll>(
         let ptr_align = bx.tcx().data_layout.pointer_align.abi;
         let slot = bx.alloca(bx.type_ptr(), ptr_align);
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None);
+        bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None, None);
 
         bx.switch_to_block(normal);
         bx.ret(bx.const_i32(0));
@@ -684,7 +685,7 @@ fn codegen_msvc_try<'ll>(
         let funclet = bx.catch_pad(cs, &[tydesc, flags, slot]);
         let ptr = bx.load(bx.type_ptr(), slot, ptr_align);
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet));
+        bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet), None);
         bx.catch_ret(&funclet, caught);
 
         // The flag value of 64 indicates a "catch-all".
@@ -692,7 +693,7 @@ fn codegen_msvc_try<'ll>(
         let flags = bx.const_i32(64);
         let null = bx.const_null(bx.type_ptr());
         let funclet = bx.catch_pad(cs, &[null, flags, null]);
-        bx.call(catch_ty, None, None, catch_func, &[data, null], Some(&funclet));
+        bx.call(catch_ty, None, None, catch_func, &[data, null], Some(&funclet), None);
         bx.catch_ret(&funclet, caught);
 
         bx.switch_to_block(caught);
@@ -701,7 +702,7 @@ fn codegen_msvc_try<'ll>(
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None);
+    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
     let i32_align = bx.tcx().data_layout.i32_align.abi;
     bx.store(ret, dest, i32_align);
 }
@@ -750,7 +751,7 @@ fn codegen_wasm_try<'ll>(
         //   }
         //
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None);
+        bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None, None);
 
         bx.switch_to_block(normal);
         bx.ret(bx.const_i32(0));
@@ -766,7 +767,7 @@ fn codegen_wasm_try<'ll>(
         let _sel = bx.call_intrinsic("llvm.wasm.get.ehselector", &[funclet.cleanuppad()]);
 
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet));
+        bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet), None);
         bx.catch_ret(&funclet, caught);
 
         bx.switch_to_block(caught);
@@ -775,7 +776,7 @@ fn codegen_wasm_try<'ll>(
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None);
+    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
     let i32_align = bx.tcx().data_layout.i32_align.abi;
     bx.store(ret, dest, i32_align);
 }
@@ -818,7 +819,7 @@ fn codegen_gnu_try<'ll>(
         let data = llvm::get_param(bx.llfn(), 1);
         let catch_func = llvm::get_param(bx.llfn(), 2);
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None);
+        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None, None);
 
         bx.switch_to_block(then);
         bx.ret(bx.const_i32(0));
@@ -836,13 +837,13 @@ fn codegen_gnu_try<'ll>(
         bx.add_clause(vals, tydesc);
         let ptr = bx.extract_value(vals, 0);
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, ptr], None);
+        bx.call(catch_ty, None, None, catch_func, &[data, ptr], None, None);
         bx.ret(bx.const_i32(1));
     });
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None);
+    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
     let i32_align = bx.tcx().data_layout.i32_align.abi;
     bx.store(ret, dest, i32_align);
 }
@@ -882,7 +883,7 @@ fn codegen_emcc_try<'ll>(
         let data = llvm::get_param(bx.llfn(), 1);
         let catch_func = llvm::get_param(bx.llfn(), 2);
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
-        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None);
+        bx.invoke(try_func_ty, None, None, try_func, &[data], then, catch, None, None);
 
         bx.switch_to_block(then);
         bx.ret(bx.const_i32(0));
@@ -920,13 +921,13 @@ fn codegen_emcc_try<'ll>(
         bx.store(is_rust_panic, catch_data_1, i8_align);
 
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
-        bx.call(catch_ty, None, None, catch_func, &[data, catch_data], None);
+        bx.call(catch_ty, None, None, catch_func, &[data, catch_data], None, None);
         bx.ret(bx.const_i32(1));
     });
 
     // Note that no invoke is used here because by definition this function
     // can't panic (that's what it's catching).
-    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None);
+    let ret = bx.call(llty, None, None, llfn, &[try_func, data, catch_func], None, None);
     let i32_align = bx.tcx().data_layout.i32_align.abi;
     bx.store(ret, dest, i32_align);
 }
@@ -1439,6 +1440,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             f,
             &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
             None,
+            None,
         );
         Ok(c)
     }
@@ -1607,6 +1609,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             f,
             &[args[1].immediate(), alignment, mask, args[0].immediate()],
             None,
+            None,
         );
         return Ok(v);
     }
@@ -1706,6 +1709,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             f,
             &[args[1].immediate(), alignment, mask, args[2].immediate()],
             None,
+            None,
         );
         return Ok(v);
     }
@@ -1799,6 +1803,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             f,
             &[args[2].immediate(), args[1].immediate(), alignment, mask],
             None,
+            None,
         );
         return Ok(v);
     }
@@ -1904,6 +1909,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             f,
             &[args[0].immediate(), args[1].immediate(), alignment, mask],
             None,
+            None,
         );
         return Ok(v);
     }
@@ -2352,11 +2358,12 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
                 f,
                 &[args[0].immediate(), bx.const_int(bx.type_i1(), 0)],
                 None,
+                None,
             ))
         } else {
             let fn_ty = bx.type_func(&[vec_ty], vec_ty);
             let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-            Ok(bx.call(fn_ty, None, None, f, &[args[0].immediate()], None))
+            Ok(bx.call(fn_ty, None, None, f, &[args[0].immediate()], None, None))
         };
     }
 
@@ -2409,7 +2416,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
 
         let fn_ty = bx.type_func(&[vec_ty, vec_ty], vec_ty);
         let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-        let v = bx.call(fn_ty, None, None, f, &[lhs, rhs], None);
+        let v = bx.call(fn_ty, None, None, f, &[lhs, rhs], None, None);
         return Ok(v);
     }
 
diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs
index 13809ef72ec7b..f7f2bfca838ea 100644
--- a/compiler/rustc_codegen_ssa/src/base.rs
+++ b/compiler/rustc_codegen_ssa/src/base.rs
@@ -462,27 +462,34 @@ pub fn maybe_create_entry_wrapper<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
         let ptr_ty = cx.type_ptr();
         let (arg_argc, arg_argv) = get_argc_argv(cx, &mut bx);
 
-        let (start_fn, start_ty, args) = if let EntryFnType::Main { sigpipe } = entry_type {
+        let (start_fn, start_ty, args, instance) = if let EntryFnType::Main { sigpipe } = entry_type
+        {
             let start_def_id = cx.tcx().require_lang_item(LangItem::Start, None);
-            let start_fn = cx.get_fn_addr(ty::Instance::expect_resolve(
+            let start_instance = ty::Instance::expect_resolve(
                 cx.tcx(),
                 ty::ParamEnv::reveal_all(),
                 start_def_id,
                 cx.tcx().mk_args(&[main_ret_ty.into()]),
-            ));
+            );
+            let start_fn = cx.get_fn_addr(start_instance);
 
             let i8_ty = cx.type_i8();
             let arg_sigpipe = bx.const_u8(sigpipe);
 
             let start_ty = cx.type_func(&[cx.val_ty(rust_main), isize_ty, ptr_ty, i8_ty], isize_ty);
-            (start_fn, start_ty, vec![rust_main, arg_argc, arg_argv, arg_sigpipe])
+            (
+                start_fn,
+                start_ty,
+                vec![rust_main, arg_argc, arg_argv, arg_sigpipe],
+                Some(start_instance),
+            )
         } else {
             debug!("using user-defined start fn");
             let start_ty = cx.type_func(&[isize_ty, ptr_ty], isize_ty);
-            (rust_main, start_ty, vec![arg_argc, arg_argv])
+            (rust_main, start_ty, vec![arg_argc, arg_argv], None)
         };
 
-        let result = bx.call(start_ty, None, None, start_fn, &args, None);
+        let result = bx.call(start_ty, None, None, start_fn, &args, None, instance);
         if cx.sess().target.os.contains("uefi") {
             bx.ret(result);
         } else {
diff --git a/compiler/rustc_codegen_ssa/src/mir/block.rs b/compiler/rustc_codegen_ssa/src/mir/block.rs
index dcc27a4f0e568..8c668597a43b3 100644
--- a/compiler/rustc_codegen_ssa/src/mir/block.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/block.rs
@@ -232,6 +232,7 @@ impl<'a, 'tcx> TerminatorCodegenHelper<'tcx> {
                 ret_llbb,
                 unwind_block,
                 self.funclet(fx),
+                instance,
             );
             if fx.mir[self.bb].is_cleanup {
                 bx.apply_attrs_to_cleanup_callsite(invokeret);
@@ -247,7 +248,8 @@ impl<'a, 'tcx> TerminatorCodegenHelper<'tcx> {
             }
             MergingSucc::False
         } else {
-            let llret = bx.call(fn_ty, fn_attrs, Some(fn_abi), fn_ptr, llargs, self.funclet(fx));
+            let llret =
+                bx.call(fn_ty, fn_attrs, Some(fn_abi), fn_ptr, llargs, self.funclet(fx), instance);
             if fx.mir[self.bb].is_cleanup {
                 bx.apply_attrs_to_cleanup_callsite(llret);
             }
@@ -502,7 +504,6 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         let ty = location.ty(self.mir, bx.tcx()).ty;
         let ty = self.monomorphize(ty);
         let drop_fn = Instance::resolve_drop_in_place(bx.tcx(), ty);
-        let instance = drop_fn.clone();
 
         if let ty::InstanceDef::DropGlue(_, None) = drop_fn.def {
             // we don't actually need to drop anything.
@@ -518,7 +519,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             args1 = [place.llval];
             &args1[..]
         };
-        let (drop_fn, fn_abi) =
+        let (drop_fn, fn_abi, drop_instance) =
             match ty.kind() {
                 // FIXME(eddyb) perhaps move some of this logic into
                 // `Instance::resolve_drop_in_place`?
@@ -550,6 +551,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                         meth::VirtualIndex::from_index(ty::COMMON_VTABLE_ENTRIES_DROPINPLACE)
                             .get_fn(bx, vtable, ty, fn_abi),
                         fn_abi,
+                        virtual_drop,
                     )
                 }
                 ty::Dynamic(_, _, ty::DynStar) => {
@@ -592,9 +594,14 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                         meth::VirtualIndex::from_index(ty::COMMON_VTABLE_ENTRIES_DROPINPLACE)
                             .get_fn(bx, meta.immediate(), ty, fn_abi),
                         fn_abi,
+                        virtual_drop,
                     )
                 }
-                _ => (bx.get_fn_addr(drop_fn), bx.fn_abi_of_instance(drop_fn, ty::List::empty())),
+                _ => (
+                    bx.get_fn_addr(drop_fn),
+                    bx.fn_abi_of_instance(drop_fn, ty::List::empty()),
+                    drop_fn,
+                ),
             };
         helper.do_call(
             self,
@@ -605,7 +612,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             Some((ReturnDest::Nothing, target)),
             unwind,
             &[],
-            Some(instance),
+            Some(drop_instance),
             mergeable_succ,
         )
     }
@@ -1699,7 +1706,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         } else {
             let fn_ty = bx.fn_decl_backend_type(fn_abi);
 
-            let llret = bx.call(fn_ty, None, Some(fn_abi), fn_ptr, &[], funclet.as_ref());
+            let llret = bx.call(fn_ty, None, Some(fn_abi), fn_ptr, &[], funclet.as_ref(), None);
             bx.apply_attrs_to_cleanup_callsite(llret);
         }
 
diff --git a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs
index 0e8c4abf21264..0af84ff067af4 100644
--- a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs
@@ -709,7 +709,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                     } else {
                         None
                     };
-                    bx.call(fn_ty, fn_attrs, Some(fn_abi), fn_ptr, &[], None)
+                    bx.call(fn_ty, fn_attrs, Some(fn_abi), fn_ptr, &[], None, Some(instance))
                 } else {
                     bx.get_static(def_id)
                 };
diff --git a/compiler/rustc_codegen_ssa/src/size_of_val.rs b/compiler/rustc_codegen_ssa/src/size_of_val.rs
index e2e95cede60a5..c250cc2682323 100644
--- a/compiler/rustc_codegen_ssa/src/size_of_val.rs
+++ b/compiler/rustc_codegen_ssa/src/size_of_val.rs
@@ -70,7 +70,15 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
             // (But we are in good company, this code is duplicated plenty of times.)
             let fn_ty = bx.fn_decl_backend_type(fn_abi);
 
-            bx.call(fn_ty, /* fn_attrs */ None, Some(fn_abi), llfn, &[msg.0, msg.1], None);
+            bx.call(
+                fn_ty,
+                /* fn_attrs */ None,
+                Some(fn_abi),
+                llfn,
+                &[msg.0, msg.1],
+                None,
+                None,
+            );
 
             // This function does not return so we can now return whatever we want.
             let size = bx.const_usize(layout.size.bytes());
diff --git a/compiler/rustc_codegen_ssa/src/traits/builder.rs b/compiler/rustc_codegen_ssa/src/traits/builder.rs
index 7bc9dee3a8955..6c8dcc5b69001 100644
--- a/compiler/rustc_codegen_ssa/src/traits/builder.rs
+++ b/compiler/rustc_codegen_ssa/src/traits/builder.rs
@@ -17,7 +17,7 @@ use crate::MemFlags;
 
 use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrs;
 use rustc_middle::ty::layout::{HasParamEnv, TyAndLayout};
-use rustc_middle::ty::Ty;
+use rustc_middle::ty::{Instance, Ty};
 use rustc_session::config::OptLevel;
 use rustc_span::Span;
 use rustc_target::abi::call::FnAbi;
@@ -82,6 +82,7 @@ pub trait BuilderMethods<'a, 'tcx>:
         then: Self::BasicBlock,
         catch: Self::BasicBlock,
         funclet: Option<&Self::Funclet>,
+        instance: Option<Instance<'tcx>>,
     ) -> Self::Value;
     fn unreachable(&mut self);
 
@@ -389,6 +390,7 @@ pub trait BuilderMethods<'a, 'tcx>:
         llfn: Self::Value,
         args: &[Self::Value],
         funclet: Option<&Self::Funclet>,
+        instance: Option<Instance<'tcx>>,
     ) -> Self::Value;
     fn zext(&mut self, val: Self::Value, dest_ty: Self::Type) -> Self::Value;
 
diff --git a/compiler/rustc_hir_analysis/src/check/intrinsicck.rs b/compiler/rustc_hir_analysis/src/check/intrinsicck.rs
index 6cb15708a42c9..df4db3ec3fbd3 100644
--- a/compiler/rustc_hir_analysis/src/check/intrinsicck.rs
+++ b/compiler/rustc_hir_analysis/src/check/intrinsicck.rs
@@ -6,7 +6,9 @@ use rustc_session::lint;
 use rustc_span::def_id::LocalDefId;
 use rustc_span::Symbol;
 use rustc_target::abi::FieldIdx;
-use rustc_target::asm::{InlineAsmReg, InlineAsmRegClass, InlineAsmRegOrRegClass, InlineAsmType};
+use rustc_target::asm::{
+    InlineAsmReg, InlineAsmRegClass, InlineAsmRegOrRegClass, InlineAsmType, ModifierInfo,
+};
 
 pub struct InlineAsmCtxt<'a, 'tcx> {
     tcx: TyCtxt<'tcx>,
@@ -251,8 +253,11 @@ impl<'a, 'tcx> InlineAsmCtxt<'a, 'tcx> {
         }
 
         // Check whether a modifier is suggested for using this type.
-        if let Some((suggested_modifier, suggested_result)) =
-            reg_class.suggest_modifier(asm_arch, asm_ty)
+        if let Some(ModifierInfo {
+            modifier: suggested_modifier,
+            result: suggested_result,
+            size: suggested_size,
+        }) = reg_class.suggest_modifier(asm_arch, asm_ty)
         {
             // Search for any use of this operand without a modifier and emit
             // the suggestion for them.
@@ -266,8 +271,11 @@ impl<'a, 'tcx> InlineAsmCtxt<'a, 'tcx> {
                 }
             }
             if !spans.is_empty() {
-                let (default_modifier, default_result) =
-                    reg_class.default_modifier(asm_arch).unwrap();
+                let ModifierInfo {
+                    modifier: default_modifier,
+                    result: default_result,
+                    size: default_size,
+                } = reg_class.default_modifier(asm_arch).unwrap();
                 self.tcx.node_span_lint(
                     lint::builtin::ASM_SUB_REGISTER,
                     expr.hir_id,
@@ -276,10 +284,10 @@ impl<'a, 'tcx> InlineAsmCtxt<'a, 'tcx> {
                     |lint| {
                         lint.span_label(expr.span, "for this argument");
                         lint.help(format!(
-                            "use `{{{idx}:{suggested_modifier}}}` to have the register formatted as `{suggested_result}`",
+                            "use `{{{idx}:{suggested_modifier}}}` to have the register formatted as `{suggested_result}` (for {suggested_size}-bit values)",
                         ));
                         lint.help(format!(
-                            "or use `{{{idx}:{default_modifier}}}` to keep the default formatting of `{default_result}`",
+                            "or use `{{{idx}:{default_modifier}}}` to keep the default formatting of `{default_result}` (for {default_size}-bit values)",
                         ));
                     },
                 );
diff --git a/compiler/rustc_symbol_mangling/src/typeid.rs b/compiler/rustc_symbol_mangling/src/typeid.rs
index e8763e49e624b..3bf564a4a16dd 100644
--- a/compiler/rustc_symbol_mangling/src/typeid.rs
+++ b/compiler/rustc_symbol_mangling/src/typeid.rs
@@ -36,7 +36,7 @@ pub fn typeid_for_instance<'tcx>(
     instance: &Instance<'tcx>,
     options: TypeIdOptions,
 ) -> String {
-    typeid_itanium_cxx_abi::typeid_for_instance(tcx, instance, options)
+    typeid_itanium_cxx_abi::typeid_for_instance(tcx, *instance, options)
 }
 
 /// Returns a KCFI type metadata identifier for the specified FnAbi.
@@ -61,6 +61,6 @@ pub fn kcfi_typeid_for_instance<'tcx>(
     // A KCFI type metadata identifier is a 32-bit constant produced by taking the lower half of the
     // xxHash64 of the type metadata identifier. (See llvm/llvm-project@cff5bef.)
     let mut hash: XxHash64 = Default::default();
-    hash.write(typeid_itanium_cxx_abi::typeid_for_instance(tcx, instance, options).as_bytes());
+    hash.write(typeid_itanium_cxx_abi::typeid_for_instance(tcx, *instance, options).as_bytes());
     hash.finish() as u32
 }
diff --git a/compiler/rustc_symbol_mangling/src/typeid/typeid_itanium_cxx_abi.rs b/compiler/rustc_symbol_mangling/src/typeid/typeid_itanium_cxx_abi.rs
index 07a382d161d74..367fec0e8fcb7 100644
--- a/compiler/rustc_symbol_mangling/src/typeid/typeid_itanium_cxx_abi.rs
+++ b/compiler/rustc_symbol_mangling/src/typeid/typeid_itanium_cxx_abi.rs
@@ -178,14 +178,14 @@ fn encode_fnsig<'tcx>(
     // Encode the return type
     let transform_ty_options = TransformTyOptions::from_bits(options.bits())
         .unwrap_or_else(|| bug!("encode_fnsig: invalid option(s) `{:?}`", options.bits()));
-    let ty = transform_ty(tcx, fn_sig.output(), transform_ty_options);
+    let ty = transform_ty(tcx, fn_sig.output(), &mut Vec::new(), transform_ty_options);
     s.push_str(&encode_ty(tcx, ty, dict, encode_ty_options));
 
     // Encode the parameter types
     let tys = fn_sig.inputs();
     if !tys.is_empty() {
         for ty in tys {
-            let ty = transform_ty(tcx, *ty, transform_ty_options);
+            let ty = transform_ty(tcx, *ty, &mut Vec::new(), transform_ty_options);
             s.push_str(&encode_ty(tcx, ty, dict, encode_ty_options));
         }
 
@@ -767,11 +767,12 @@ fn transform_predicates<'tcx>(
 fn transform_args<'tcx>(
     tcx: TyCtxt<'tcx>,
     args: GenericArgsRef<'tcx>,
+    parents: &mut Vec<Ty<'tcx>>,
     options: TransformTyOptions,
 ) -> GenericArgsRef<'tcx> {
     let args = args.iter().map(|arg| match arg.unpack() {
         GenericArgKind::Type(ty) if ty.is_c_void(tcx) => Ty::new_unit(tcx).into(),
-        GenericArgKind::Type(ty) => transform_ty(tcx, ty, options).into(),
+        GenericArgKind::Type(ty) => transform_ty(tcx, ty, parents, options).into(),
         _ => arg,
     });
     tcx.mk_args_from_iter(args)
@@ -781,9 +782,12 @@ fn transform_args<'tcx>(
 // c_void types into unit types unconditionally, generalizes pointers if
 // TransformTyOptions::GENERALIZE_POINTERS option is set, and normalizes integers if
 // TransformTyOptions::NORMALIZE_INTEGERS option is set.
-fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptions) -> Ty<'tcx> {
-    let mut ty = ty;
-
+fn transform_ty<'tcx>(
+    tcx: TyCtxt<'tcx>,
+    mut ty: Ty<'tcx>,
+    parents: &mut Vec<Ty<'tcx>>,
+    options: TransformTyOptions,
+) -> Ty<'tcx> {
     match ty.kind() {
         ty::Float(..) | ty::Str | ty::Never | ty::Foreign(..) | ty::CoroutineWitness(..) => {}
 
@@ -843,17 +847,20 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
         _ if ty.is_unit() => {}
 
         ty::Tuple(tys) => {
-            ty = Ty::new_tup_from_iter(tcx, tys.iter().map(|ty| transform_ty(tcx, ty, options)));
+            ty = Ty::new_tup_from_iter(
+                tcx,
+                tys.iter().map(|ty| transform_ty(tcx, ty, parents, options)),
+            );
         }
 
         ty::Array(ty0, len) => {
             let len = len.eval_target_usize(tcx, ty::ParamEnv::reveal_all());
 
-            ty = Ty::new_array(tcx, transform_ty(tcx, *ty0, options), len);
+            ty = Ty::new_array(tcx, transform_ty(tcx, *ty0, parents, options), len);
         }
 
         ty::Slice(ty0) => {
-            ty = Ty::new_slice(tcx, transform_ty(tcx, *ty0, options));
+            ty = Ty::new_slice(tcx, transform_ty(tcx, *ty0, parents, options));
         }
 
         ty::Adt(adt_def, args) => {
@@ -862,7 +869,8 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
             } else if options.contains(TransformTyOptions::GENERALIZE_REPR_C) && adt_def.repr().c()
             {
                 ty = Ty::new_adt(tcx, *adt_def, ty::List::empty());
-            } else if adt_def.repr().transparent() && adt_def.is_struct() {
+            } else if adt_def.repr().transparent() && adt_def.is_struct() && !parents.contains(&ty)
+            {
                 // Don't transform repr(transparent) types with an user-defined CFI encoding to
                 // preserve the user-defined CFI encoding.
                 if let Some(_) = tcx.get_attr(adt_def.did(), sym::cfi_encoding) {
@@ -881,38 +889,48 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
                     // Generalize any repr(transparent) user-defined type that is either a pointer
                     // or reference, and either references itself or any other type that contains or
                     // references itself, to avoid a reference cycle.
+
+                    // If the self reference is not through a pointer, for example, due
+                    // to using `PhantomData`, need to skip normalizing it if we hit it again.
+                    parents.push(ty);
                     if ty0.is_any_ptr() && ty0.contains(ty) {
                         ty = transform_ty(
                             tcx,
                             ty0,
+                            parents,
                             options | TransformTyOptions::GENERALIZE_POINTERS,
                         );
                     } else {
-                        ty = transform_ty(tcx, ty0, options);
+                        ty = transform_ty(tcx, ty0, parents, options);
                     }
+                    parents.pop();
                 } else {
                     // Transform repr(transparent) types without non-ZST field into ()
                     ty = Ty::new_unit(tcx);
                 }
             } else {
-                ty = Ty::new_adt(tcx, *adt_def, transform_args(tcx, args, options));
+                ty = Ty::new_adt(tcx, *adt_def, transform_args(tcx, args, parents, options));
             }
         }
 
         ty::FnDef(def_id, args) => {
-            ty = Ty::new_fn_def(tcx, *def_id, transform_args(tcx, args, options));
+            ty = Ty::new_fn_def(tcx, *def_id, transform_args(tcx, args, parents, options));
         }
 
         ty::Closure(def_id, args) => {
-            ty = Ty::new_closure(tcx, *def_id, transform_args(tcx, args, options));
+            ty = Ty::new_closure(tcx, *def_id, transform_args(tcx, args, parents, options));
         }
 
         ty::CoroutineClosure(def_id, args) => {
-            ty = Ty::new_coroutine_closure(tcx, *def_id, transform_args(tcx, args, options));
+            ty = Ty::new_coroutine_closure(
+                tcx,
+                *def_id,
+                transform_args(tcx, args, parents, options),
+            );
         }
 
         ty::Coroutine(def_id, args) => {
-            ty = Ty::new_coroutine(tcx, *def_id, transform_args(tcx, args, options));
+            ty = Ty::new_coroutine(tcx, *def_id, transform_args(tcx, args, parents, options));
         }
 
         ty::Ref(region, ty0, ..) => {
@@ -924,9 +942,9 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
                 }
             } else {
                 if ty.is_mutable_ptr() {
-                    ty = Ty::new_mut_ref(tcx, *region, transform_ty(tcx, *ty0, options));
+                    ty = Ty::new_mut_ref(tcx, *region, transform_ty(tcx, *ty0, parents, options));
                 } else {
-                    ty = Ty::new_imm_ref(tcx, *region, transform_ty(tcx, *ty0, options));
+                    ty = Ty::new_imm_ref(tcx, *region, transform_ty(tcx, *ty0, parents, options));
                 }
             }
         }
@@ -940,9 +958,9 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
                 }
             } else {
                 if ty.is_mutable_ptr() {
-                    ty = Ty::new_mut_ptr(tcx, transform_ty(tcx, *ptr_ty, options));
+                    ty = Ty::new_mut_ptr(tcx, transform_ty(tcx, *ptr_ty, parents, options));
                 } else {
-                    ty = Ty::new_imm_ptr(tcx, transform_ty(tcx, *ptr_ty, options));
+                    ty = Ty::new_imm_ptr(tcx, transform_ty(tcx, *ptr_ty, parents, options));
                 }
             }
         }
@@ -955,9 +973,9 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
                     .skip_binder()
                     .inputs()
                     .iter()
-                    .map(|ty| transform_ty(tcx, *ty, options))
+                    .map(|ty| transform_ty(tcx, *ty, parents, options))
                     .collect();
-                let output = transform_ty(tcx, fn_sig.skip_binder().output(), options);
+                let output = transform_ty(tcx, fn_sig.skip_binder().output(), parents, options);
                 ty = Ty::new_fn_ptr(
                     tcx,
                     ty::Binder::bind_with_vars(
@@ -987,6 +1005,7 @@ fn transform_ty<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>, options: TransformTyOptio
             ty = transform_ty(
                 tcx,
                 tcx.normalize_erasing_regions(ty::ParamEnv::reveal_all(), ty),
+                parents,
                 options,
             );
         }
@@ -1037,7 +1056,7 @@ pub fn typeid_for_fnabi<'tcx>(
     // Encode the return type
     let transform_ty_options = TransformTyOptions::from_bits(options.bits())
         .unwrap_or_else(|| bug!("typeid_for_fnabi: invalid option(s) `{:?}`", options.bits()));
-    let ty = transform_ty(tcx, fn_abi.ret.layout.ty, transform_ty_options);
+    let ty = transform_ty(tcx, fn_abi.ret.layout.ty, &mut Vec::new(), transform_ty_options);
     typeid.push_str(&encode_ty(tcx, ty, &mut dict, encode_ty_options));
 
     // Encode the parameter types
@@ -1049,7 +1068,7 @@ pub fn typeid_for_fnabi<'tcx>(
         let mut pushed_arg = false;
         for arg in fn_abi.args.iter().filter(|arg| arg.mode != PassMode::Ignore) {
             pushed_arg = true;
-            let ty = transform_ty(tcx, arg.layout.ty, transform_ty_options);
+            let ty = transform_ty(tcx, arg.layout.ty, &mut Vec::new(), transform_ty_options);
             typeid.push_str(&encode_ty(tcx, ty, &mut dict, encode_ty_options));
         }
         if !pushed_arg {
@@ -1062,7 +1081,8 @@ pub fn typeid_for_fnabi<'tcx>(
             if fn_abi.args[n].mode == PassMode::Ignore {
                 continue;
             }
-            let ty = transform_ty(tcx, fn_abi.args[n].layout.ty, transform_ty_options);
+            let ty =
+                transform_ty(tcx, fn_abi.args[n].layout.ty, &mut Vec::new(), transform_ty_options);
             typeid.push_str(&encode_ty(tcx, ty, &mut dict, encode_ty_options));
         }
 
@@ -1088,11 +1108,15 @@ pub fn typeid_for_fnabi<'tcx>(
 /// vendor extended type qualifiers and types for Rust types that are not used at the FFI boundary.
 pub fn typeid_for_instance<'tcx>(
     tcx: TyCtxt<'tcx>,
-    instance: &Instance<'tcx>,
+    mut instance: Instance<'tcx>,
     options: TypeIdOptions,
 ) -> String {
+    if matches!(instance.def, ty::InstanceDef::Virtual(..)) {
+        instance.args = strip_receiver_auto(tcx, instance.args)
+    }
+
     let fn_abi = tcx
-        .fn_abi_of_instance(tcx.param_env(instance.def_id()).and((*instance, ty::List::empty())))
+        .fn_abi_of_instance(tcx.param_env(instance.def_id()).and((instance, ty::List::empty())))
         .unwrap_or_else(|instance| {
             bug!("typeid_for_instance: couldn't get fn_abi of instance {:?}", instance)
         });
@@ -1138,3 +1162,23 @@ pub fn typeid_for_instance<'tcx>(
 
     typeid_for_fnabi(tcx, fn_abi, options)
 }
+
+fn strip_receiver_auto<'tcx>(
+    tcx: TyCtxt<'tcx>,
+    args: ty::GenericArgsRef<'tcx>,
+) -> ty::GenericArgsRef<'tcx> {
+    let ty = args.type_at(0);
+    let ty::Dynamic(preds, lifetime, kind) = ty.kind() else {
+        bug!("Tried to strip auto traits from non-dynamic type {ty}");
+    };
+    let filtered_preds =
+        if preds.principal().is_some() {
+            tcx.mk_poly_existential_predicates_from_iter(preds.into_iter().filter(|pred| {
+                !matches!(pred.skip_binder(), ty::ExistentialPredicate::AutoTrait(..))
+            }))
+        } else {
+            ty::List::empty()
+        };
+    let new_rcvr = Ty::new_dynamic(tcx, filtered_preds, *lifetime, *kind);
+    tcx.mk_args_trait(new_rcvr, args.into_iter().skip(1))
+}
diff --git a/compiler/rustc_target/src/asm/aarch64.rs b/compiler/rustc_target/src/asm/aarch64.rs
index 97132311a5c9a..70528c1222c44 100644
--- a/compiler/rustc_target/src/asm/aarch64.rs
+++ b/compiler/rustc_target/src/asm/aarch64.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use crate::spec::{RelocModel, Target};
 use rustc_data_structures::fx::FxIndexSet;
 use rustc_macros::HashStable_Generic;
@@ -27,32 +27,28 @@ impl AArch64InlineAsmRegClass {
         None
     }
 
-    pub fn suggest_modifier(
-        self,
-        _arch: InlineAsmArch,
-        ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    pub fn suggest_modifier(self, _arch: InlineAsmArch, ty: InlineAsmType) -> Option<ModifierInfo> {
         match self {
             Self::reg => match ty.size().bits() {
                 64 => None,
-                _ => Some(('w', "w0")),
+                _ => Some(('w', "w0", 32).into()),
             },
             Self::vreg | Self::vreg_low16 => match ty.size().bits() {
-                8 => Some(('b', "b0")),
-                16 => Some(('h', "h0")),
-                32 => Some(('s', "s0")),
-                64 => Some(('d', "d0")),
-                128 => Some(('q', "q0")),
+                8 => Some(('b', "b0", 8).into()),
+                16 => Some(('h', "h0", 16).into()),
+                32 => Some(('s', "s0", 32).into()),
+                64 => Some(('d', "d0", 64).into()),
+                128 => Some(('q', "q0", 128).into()),
                 _ => None,
             },
             Self::preg => None,
         }
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         match self {
-            Self::reg => Some(('x', "x0")),
-            Self::vreg | Self::vreg_low16 => Some(('v', "v0")),
+            Self::reg => Some(('x', "x0", 64).into()),
+            Self::vreg | Self::vreg_low16 => Some(('v', "v0", 128).into()),
             Self::preg => None,
         }
     }
diff --git a/compiler/rustc_target/src/asm/arm.rs b/compiler/rustc_target/src/asm/arm.rs
index 514e30ae0204d..f56dbac708b65 100644
--- a/compiler/rustc_target/src/asm/arm.rs
+++ b/compiler/rustc_target/src/asm/arm.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use crate::spec::{RelocModel, Target};
 use rustc_data_structures::fx::FxIndexSet;
 use rustc_macros::HashStable_Generic;
@@ -35,11 +35,11 @@ impl ArmInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/avr.rs b/compiler/rustc_target/src/asm/avr.rs
index 9a96a61f5b2cd..eab38a4a4f4a3 100644
--- a/compiler/rustc_target/src/asm/avr.rs
+++ b/compiler/rustc_target/src/asm/avr.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -29,11 +29,11 @@ impl AvrInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/bpf.rs b/compiler/rustc_target/src/asm/bpf.rs
index 3b03766a089b2..9bc94274675d5 100644
--- a/compiler/rustc_target/src/asm/bpf.rs
+++ b/compiler/rustc_target/src/asm/bpf.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -23,11 +23,11 @@ impl BpfInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/csky.rs b/compiler/rustc_target/src/asm/csky.rs
index db3d8106040b6..64607ee4b81db 100644
--- a/compiler/rustc_target/src/asm/csky.rs
+++ b/compiler/rustc_target/src/asm/csky.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -23,11 +23,11 @@ impl CSKYInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/hexagon.rs b/compiler/rustc_target/src/asm/hexagon.rs
index d20270ac9e95a..19da7b8084853 100644
--- a/compiler/rustc_target/src/asm/hexagon.rs
+++ b/compiler/rustc_target/src/asm/hexagon.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -22,11 +22,11 @@ impl HexagonInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/loongarch.rs b/compiler/rustc_target/src/asm/loongarch.rs
index 9d1a4f3eeeafa..15d0f54ce3b6e 100644
--- a/compiler/rustc_target/src/asm/loongarch.rs
+++ b/compiler/rustc_target/src/asm/loongarch.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -23,11 +23,11 @@ impl LoongArchInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/m68k.rs b/compiler/rustc_target/src/asm/m68k.rs
index 8c857550cf21a..ac94dcc03dc6b 100644
--- a/compiler/rustc_target/src/asm/m68k.rs
+++ b/compiler/rustc_target/src/asm/m68k.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -24,11 +24,11 @@ impl M68kInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/mips.rs b/compiler/rustc_target/src/asm/mips.rs
index 4e7c2eb1bf88e..0ac1a43ae183b 100644
--- a/compiler/rustc_target/src/asm/mips.rs
+++ b/compiler/rustc_target/src/asm/mips.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -23,11 +23,11 @@ impl MipsInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/mod.rs b/compiler/rustc_target/src/asm/mod.rs
index a11884bea268f..2e04dca98c57b 100644
--- a/compiler/rustc_target/src/asm/mod.rs
+++ b/compiler/rustc_target/src/asm/mod.rs
@@ -6,6 +6,18 @@ use rustc_span::Symbol;
 use std::fmt;
 use std::str::FromStr;
 
+pub struct ModifierInfo {
+    pub modifier: char,
+    pub result: &'static str,
+    pub size: u64,
+}
+
+impl From<(char, &'static str, u64)> for ModifierInfo {
+    fn from((modifier, result, size): (char, &'static str, u64)) -> Self {
+        Self { modifier, result, size }
+    }
+}
+
 macro_rules! def_reg_class {
     ($arch:ident $arch_regclass:ident {
         $(
@@ -512,11 +524,7 @@ impl InlineAsmRegClass {
     /// Such suggestions are useful if a type smaller than the full register
     /// size is used and a modifier can be used to point to the subregister of
     /// the correct size.
-    pub fn suggest_modifier(
-        self,
-        arch: InlineAsmArch,
-        ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    pub fn suggest_modifier(self, arch: InlineAsmArch, ty: InlineAsmType) -> Option<ModifierInfo> {
         match self {
             Self::X86(r) => r.suggest_modifier(arch, ty),
             Self::Arm(r) => r.suggest_modifier(arch, ty),
@@ -545,7 +553,7 @@ impl InlineAsmRegClass {
     /// This is only needed when the register class can suggest a modifier, so
     /// that the user can be shown how to get the default behavior without a
     /// warning.
-    pub fn default_modifier(self, arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, arch: InlineAsmArch) -> Option<ModifierInfo> {
         match self {
             Self::X86(r) => r.default_modifier(arch),
             Self::Arm(r) => r.default_modifier(arch),
diff --git a/compiler/rustc_target/src/asm/msp430.rs b/compiler/rustc_target/src/asm/msp430.rs
index a27d6390a72e8..439f3ba0b575b 100644
--- a/compiler/rustc_target/src/asm/msp430.rs
+++ b/compiler/rustc_target/src/asm/msp430.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -22,11 +22,11 @@ impl Msp430InlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/nvptx.rs b/compiler/rustc_target/src/asm/nvptx.rs
index 8e1e91e7c5f1f..57aa50fceb88b 100644
--- a/compiler/rustc_target/src/asm/nvptx.rs
+++ b/compiler/rustc_target/src/asm/nvptx.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 
@@ -23,11 +23,11 @@ impl NvptxInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/powerpc.rs b/compiler/rustc_target/src/asm/powerpc.rs
index d3ccb30350a27..4e8cbe34de948 100644
--- a/compiler/rustc_target/src/asm/powerpc.rs
+++ b/compiler/rustc_target/src/asm/powerpc.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -26,11 +26,11 @@ impl PowerPCInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/riscv.rs b/compiler/rustc_target/src/asm/riscv.rs
index dea6d50fe2ba8..2505d36f5b8a4 100644
--- a/compiler/rustc_target/src/asm/riscv.rs
+++ b/compiler/rustc_target/src/asm/riscv.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use crate::spec::{RelocModel, Target};
 use rustc_data_structures::fx::FxIndexSet;
 use rustc_macros::HashStable_Generic;
@@ -26,11 +26,11 @@ impl RiscVInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/s390x.rs b/compiler/rustc_target/src/asm/s390x.rs
index b8afeb824d847..6bc668454b191 100644
--- a/compiler/rustc_target/src/asm/s390x.rs
+++ b/compiler/rustc_target/src/asm/s390x.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 use std::fmt;
@@ -24,11 +24,11 @@ impl S390xInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/spirv.rs b/compiler/rustc_target/src/asm/spirv.rs
index 31073da10b21b..d13a6131f9ab4 100644
--- a/compiler/rustc_target/src/asm/spirv.rs
+++ b/compiler/rustc_target/src/asm/spirv.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 
@@ -21,11 +21,11 @@ impl SpirVInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/wasm.rs b/compiler/rustc_target/src/asm/wasm.rs
index f095b7c6e118c..eb0b23ef43dd8 100644
--- a/compiler/rustc_target/src/asm/wasm.rs
+++ b/compiler/rustc_target/src/asm/wasm.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use rustc_macros::HashStable_Generic;
 use rustc_span::Symbol;
 
@@ -21,11 +21,11 @@ impl WasmInlineAsmRegClass {
         self,
         _arch: InlineAsmArch,
         _ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    ) -> Option<ModifierInfo> {
         None
     }
 
-    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, _arch: InlineAsmArch) -> Option<ModifierInfo> {
         None
     }
 
diff --git a/compiler/rustc_target/src/asm/x86.rs b/compiler/rustc_target/src/asm/x86.rs
index 3902dac7ff654..3b5da8806ccc2 100644
--- a/compiler/rustc_target/src/asm/x86.rs
+++ b/compiler/rustc_target/src/asm/x86.rs
@@ -1,4 +1,4 @@
-use super::{InlineAsmArch, InlineAsmType};
+use super::{InlineAsmArch, InlineAsmType, ModifierInfo};
 use crate::spec::{RelocModel, Target};
 use rustc_data_structures::fx::FxIndexSet;
 use rustc_macros::HashStable_Generic;
@@ -53,32 +53,28 @@ impl X86InlineAsmRegClass {
         }
     }
 
-    pub fn suggest_modifier(
-        self,
-        arch: InlineAsmArch,
-        ty: InlineAsmType,
-    ) -> Option<(char, &'static str)> {
+    pub fn suggest_modifier(self, arch: InlineAsmArch, ty: InlineAsmType) -> Option<ModifierInfo> {
         match self {
             Self::reg => match ty.size().bits() {
-                16 => Some(('x', "ax")),
-                32 if arch == InlineAsmArch::X86_64 => Some(('e', "eax")),
+                16 => Some(('x', "ax", 16).into()),
+                32 if arch == InlineAsmArch::X86_64 => Some(('e', "eax", 32).into()),
                 _ => None,
             },
             Self::reg_abcd => match ty.size().bits() {
-                16 => Some(('x', "ax")),
-                32 if arch == InlineAsmArch::X86_64 => Some(('e', "eax")),
+                16 => Some(('x', "ax", 16).into()),
+                32 if arch == InlineAsmArch::X86_64 => Some(('e', "eax", 32).into()),
                 _ => None,
             },
             Self::reg_byte => None,
             Self::xmm_reg => None,
             Self::ymm_reg => match ty.size().bits() {
                 256 => None,
-                _ => Some(('x', "xmm0")),
+                _ => Some(('x', "xmm0", 128).into()),
             },
             Self::zmm_reg => match ty.size().bits() {
                 512 => None,
-                256 => Some(('y', "ymm0")),
-                _ => Some(('x', "xmm0")),
+                256 => Some(('y', "ymm0", 256).into()),
+                _ => Some(('x', "xmm0", 128).into()),
             },
             Self::kreg | Self::kreg0 => None,
             Self::mmx_reg | Self::x87_reg => None,
@@ -86,19 +82,19 @@ impl X86InlineAsmRegClass {
         }
     }
 
-    pub fn default_modifier(self, arch: InlineAsmArch) -> Option<(char, &'static str)> {
+    pub fn default_modifier(self, arch: InlineAsmArch) -> Option<ModifierInfo> {
         match self {
             Self::reg | Self::reg_abcd => {
                 if arch == InlineAsmArch::X86_64 {
-                    Some(('r', "rax"))
+                    Some(('r', "rax", 64).into())
                 } else {
-                    Some(('e', "eax"))
+                    Some(('e', "eax", 32).into())
                 }
             }
             Self::reg_byte => None,
-            Self::xmm_reg => Some(('x', "xmm0")),
-            Self::ymm_reg => Some(('y', "ymm0")),
-            Self::zmm_reg => Some(('z', "zmm0")),
+            Self::xmm_reg => Some(('x', "xmm0", 128).into()),
+            Self::ymm_reg => Some(('y', "ymm0", 256).into()),
+            Self::zmm_reg => Some(('z', "zmm0", 512).into()),
             Self::kreg | Self::kreg0 => None,
             Self::mmx_reg | Self::x87_reg => None,
             Self::tmm_reg => None,
diff --git a/compiler/stable_mir/src/mir/alloc.rs b/compiler/stable_mir/src/mir/alloc.rs
index 6645793343875..408e0bafa5896 100644
--- a/compiler/stable_mir/src/mir/alloc.rs
+++ b/compiler/stable_mir/src/mir/alloc.rs
@@ -55,7 +55,7 @@ impl IndexedVal for AllocId {
 /// Utility function used to read an allocation data into a unassigned integer.
 pub(crate) fn read_target_uint(mut bytes: &[u8]) -> Result<u128, Error> {
     let mut buf = [0u8; std::mem::size_of::<u128>()];
-    match MachineInfo::target_endianess() {
+    match MachineInfo::target_endianness() {
         Endian::Little => {
             bytes.read_exact(&mut buf[..bytes.len()])?;
             Ok(u128::from_le_bytes(buf))
@@ -70,7 +70,7 @@ pub(crate) fn read_target_uint(mut bytes: &[u8]) -> Result<u128, Error> {
 /// Utility function used to read an allocation data into an assigned integer.
 pub(crate) fn read_target_int(mut bytes: &[u8]) -> Result<i128, Error> {
     let mut buf = [0u8; std::mem::size_of::<i128>()];
-    match MachineInfo::target_endianess() {
+    match MachineInfo::target_endianness() {
         Endian::Little => {
             bytes.read_exact(&mut buf[..bytes.len()])?;
             Ok(i128::from_le_bytes(buf))
diff --git a/compiler/stable_mir/src/target.rs b/compiler/stable_mir/src/target.rs
index 3a9011a2ffe15..e00a418c54039 100644
--- a/compiler/stable_mir/src/target.rs
+++ b/compiler/stable_mir/src/target.rs
@@ -14,7 +14,7 @@ impl MachineInfo {
         with(|cx| cx.target_info())
     }
 
-    pub fn target_endianess() -> Endian {
+    pub fn target_endianness() -> Endian {
         with(|cx| cx.target_info().endian)
     }
 
diff --git a/library/core/src/intrinsics/simd.rs b/library/core/src/intrinsics/simd.rs
index b69f4f853b990..427a95f4665da 100644
--- a/library/core/src/intrinsics/simd.rs
+++ b/library/core/src/intrinsics/simd.rs
@@ -470,7 +470,7 @@ extern "rust-intrinsic" {
     /// No matter whether the output is an array or an unsigned integer, it is treated as a single
     /// contiguous list of bits. The bitmask is always packed on the least-significant side of the
     /// output, and padded with 0s in the most-significant bits. The order of the bits depends on
-    /// endianess:
+    /// endianness:
     ///
     /// * On little endian, the least significant bit corresponds to the first vector element.
     /// * On big endian, the least significant bit corresponds to the last vector element.
diff --git a/library/std/src/sys/pal/uefi/helpers.rs b/library/std/src/sys/pal/uefi/helpers.rs
index ba53ed88f3765..23aa4da14a763 100644
--- a/library/std/src/sys/pal/uefi/helpers.rs
+++ b/library/std/src/sys/pal/uefi/helpers.rs
@@ -10,14 +10,16 @@
 //! - More information about protocols can be found [here](https://edk2-docs.gitbook.io/edk-ii-uefi-driver-writer-s-guide/3_foundation/36_protocols_and_handles)
 
 use r_efi::efi::{self, Guid};
+use r_efi::protocols::{device_path, device_path_to_text};
 
+use crate::ffi::OsString;
+use crate::io::{self, const_io_error};
 use crate::mem::{size_of, MaybeUninit};
-use crate::os::uefi;
+use crate::os::uefi::{self, env::boot_services, ffi::OsStringExt};
 use crate::ptr::NonNull;
-use crate::{
-    io::{self, const_io_error},
-    os::uefi::env::boot_services,
-};
+use crate::slice;
+use crate::sync::atomic::{AtomicPtr, Ordering};
+use crate::sys_common::wstr::WStrUnits;
 
 const BOOT_SERVICES_UNAVAILABLE: io::Error =
     const_io_error!(io::ErrorKind::Other, "Boot Services are no longer available");
@@ -142,9 +144,74 @@ pub(crate) unsafe fn close_event(evt: NonNull<crate::ffi::c_void>) -> io::Result
 
 /// Get the Protocol for current system handle.
 /// Note: Some protocols need to be manually freed. It is the callers responsibility to do so.
-pub(crate) fn image_handle_protocol<T>(protocol_guid: Guid) -> Option<NonNull<T>> {
-    let system_handle = uefi::env::try_image_handle()?;
-    open_protocol(system_handle, protocol_guid).ok()
+pub(crate) fn image_handle_protocol<T>(protocol_guid: Guid) -> io::Result<NonNull<T>> {
+    let system_handle = uefi::env::try_image_handle().ok_or(io::const_io_error!(
+        io::ErrorKind::NotFound,
+        "Protocol not found in Image handle"
+    ))?;
+    open_protocol(system_handle, protocol_guid)
+}
+
+pub(crate) fn device_path_to_text(path: NonNull<device_path::Protocol>) -> io::Result<OsString> {
+    fn path_to_text(
+        protocol: NonNull<device_path_to_text::Protocol>,
+        path: NonNull<device_path::Protocol>,
+    ) -> io::Result<OsString> {
+        let path_ptr: *mut r_efi::efi::Char16 = unsafe {
+            ((*protocol.as_ptr()).convert_device_path_to_text)(
+                path.as_ptr(),
+                // DisplayOnly
+                r_efi::efi::Boolean::FALSE,
+                // AllowShortcuts
+                r_efi::efi::Boolean::FALSE,
+            )
+        };
+
+        // SAFETY: `convert_device_path_to_text` returns a pointer to a null-terminated UTF-16
+        // string, and that string cannot be deallocated prior to dropping the `WStrUnits`, so
+        // it's safe for `WStrUnits` to use.
+        let path_len = unsafe {
+            WStrUnits::new(path_ptr)
+                .ok_or(io::const_io_error!(io::ErrorKind::InvalidData, "Invalid path"))?
+                .count()
+        };
+
+        let path = OsString::from_wide(unsafe { slice::from_raw_parts(path_ptr.cast(), path_len) });
+
+        if let Some(boot_services) = crate::os::uefi::env::boot_services() {
+            let boot_services: NonNull<r_efi::efi::BootServices> = boot_services.cast();
+            unsafe {
+                ((*boot_services.as_ptr()).free_pool)(path_ptr.cast());
+            }
+        }
+
+        Ok(path)
+    }
+
+    static LAST_VALID_HANDLE: AtomicPtr<crate::ffi::c_void> =
+        AtomicPtr::new(crate::ptr::null_mut());
+
+    if let Some(handle) = NonNull::new(LAST_VALID_HANDLE.load(Ordering::Acquire)) {
+        if let Ok(protocol) = open_protocol::<device_path_to_text::Protocol>(
+            handle,
+            device_path_to_text::PROTOCOL_GUID,
+        ) {
+            return path_to_text(protocol, path);
+        }
+    }
+
+    let device_path_to_text_handles = locate_handles(device_path_to_text::PROTOCOL_GUID)?;
+    for handle in device_path_to_text_handles {
+        if let Ok(protocol) = open_protocol::<device_path_to_text::Protocol>(
+            handle,
+            device_path_to_text::PROTOCOL_GUID,
+        ) {
+            LAST_VALID_HANDLE.store(handle.as_ptr(), Ordering::Release);
+            return path_to_text(protocol, path);
+        }
+    }
+
+    Err(io::const_io_error!(io::ErrorKind::NotFound, "No device path to text protocol found"))
 }
 
 /// Get RuntimeServices
diff --git a/library/std/src/sys/pal/uefi/os.rs b/library/std/src/sys/pal/uefi/os.rs
index e6693db68e641..58838c5876ebd 100644
--- a/library/std/src/sys/pal/uefi/os.rs
+++ b/library/std/src/sys/pal/uefi/os.rs
@@ -1,4 +1,4 @@
-use super::{unsupported, RawOsError};
+use super::{helpers, unsupported, RawOsError};
 use crate::error::Error as StdError;
 use crate::ffi::{OsStr, OsString};
 use crate::fmt;
@@ -7,6 +7,7 @@ use crate::marker::PhantomData;
 use crate::os::uefi;
 use crate::path::{self, PathBuf};
 use crate::ptr::NonNull;
+use r_efi::efi::protocols::{device_path, loaded_image_device_path};
 use r_efi::efi::Status;
 
 pub fn errno() -> RawOsError {
@@ -164,7 +165,10 @@ impl fmt::Display for JoinPathsError {
 impl StdError for JoinPathsError {}
 
 pub fn current_exe() -> io::Result<PathBuf> {
-    unsupported()
+    let protocol = helpers::image_handle_protocol::<device_path::Protocol>(
+        loaded_image_device_path::PROTOCOL_GUID,
+    )?;
+    helpers::device_path_to_text(protocol).map(PathBuf::from)
 }
 
 pub struct Env(!);
diff --git a/library/unwind/src/lib.rs b/library/unwind/src/lib.rs
index 25bb5938b4cce..51d31a00afe91 100644
--- a/library/unwind/src/lib.rs
+++ b/library/unwind/src/lib.rs
@@ -4,6 +4,7 @@
 #![feature(staged_api)]
 #![feature(c_unwind)]
 #![feature(strict_provenance)]
+#![cfg_attr(target_arch = "wasm64", feature(simd_wasm64))]
 #![cfg_attr(not(target_env = "msvc"), feature(libc))]
 #![cfg_attr(
     all(target_family = "wasm", not(target_os = "emscripten")),
diff --git a/library/unwind/src/wasm.rs b/library/unwind/src/wasm.rs
index b06671bcb8309..f4ffac1ba16da 100644
--- a/library/unwind/src/wasm.rs
+++ b/library/unwind/src/wasm.rs
@@ -59,7 +59,10 @@ pub unsafe fn _Unwind_RaiseException(exception: *mut _Unwind_Exception) -> _Unwi
             wasm_throw(0, exception.cast())
         } else {
             let _ = exception;
-            core::arch::wasm32::unreachable()
+            #[cfg(target_arch = "wasm32")]
+            core::arch::wasm32::unreachable();
+            #[cfg(target_arch = "wasm64")]
+            core::arch::wasm64::unreachable();
         }
     }
 }
diff --git a/src/tools/miri/src/shims/intrinsics/simd.rs b/src/tools/miri/src/shims/intrinsics/simd.rs
index af98b38af8c96..6973c0e9c3571 100644
--- a/src/tools/miri/src/shims/intrinsics/simd.rs
+++ b/src/tools/miri/src/shims/intrinsics/simd.rs
@@ -457,7 +457,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                 let bitmask_len = u32::try_from(bitmask_len).unwrap();
 
                 // To read the mask, we transmute it to an integer.
-                // That does the right thing wrt endianess.
+                // That does the right thing wrt endianness.
                 let mask_ty = this.machine.layouts.uint(mask.layout.size).unwrap();
                 let mask = mask.transmute(mask_ty, this)?;
                 let mask: u64 = this.read_scalar(&mask)?.to_bits(mask_ty.size)?.try_into().unwrap();
@@ -509,7 +509,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                     }
                 }
                 // We have to change the type of the place to be able to write `res` into it. This
-                // transmutes the integer to an array, which does the right thing wrt endianess.
+                // transmutes the integer to an array, which does the right thing wrt endianness.
                 let dest =
                     dest.transmute(this.machine.layouts.uint(dest.layout.size).unwrap(), this)?;
                 this.write_int(res, &dest)?;
diff --git a/src/tools/miri/src/shims/unix/linux/fd/event.rs b/src/tools/miri/src/shims/unix/linux/fd/event.rs
index 1f17ffb88c886..49408fda3ae31 100644
--- a/src/tools/miri/src/shims/unix/linux/fd/event.rs
+++ b/src/tools/miri/src/shims/unix/linux/fd/event.rs
@@ -38,7 +38,7 @@ impl FileDescriptor for Event {
     }
 
     /// A write call adds the 8-byte integer value supplied in
-    /// its buffer (in native endianess) to the counter.  The maximum value that may be
+    /// its buffer (in native endianness) to the counter.  The maximum value that may be
     /// stored in the counter is the largest unsigned 64-bit value
     /// minus 1 (i.e., 0xfffffffffffffffe).  If the addition would
     /// cause the counter's value to exceed the maximum, then the
@@ -57,7 +57,7 @@ impl FileDescriptor for Event {
     ) -> InterpResult<'tcx, io::Result<usize>> {
         let v1 = self.val.get();
         let bytes: [u8; 8] = bytes.try_into().unwrap(); // FIXME fail gracefully when this has the wrong size
-        // Convert from target endianess to host endianess.
+        // Convert from target endianness to host endianness.
         let num = match tcx.sess.target.endian {
             Endian::Little => u64::from_le_bytes(bytes),
             Endian::Big => u64::from_be_bytes(bytes),
diff --git a/tests/codegen/sanitizer/cfi/emit-type-metadata-id-itanium-cxx-abi-repr-transparent-types.rs b/tests/codegen/sanitizer/cfi/emit-type-metadata-id-itanium-cxx-abi-repr-transparent-types.rs
index 6f47f5e335577..1332338b26ab9 100644
--- a/tests/codegen/sanitizer/cfi/emit-type-metadata-id-itanium-cxx-abi-repr-transparent-types.rs
+++ b/tests/codegen/sanitizer/cfi/emit-type-metadata-id-itanium-cxx-abi-repr-transparent-types.rs
@@ -34,6 +34,12 @@ pub struct Bar(i32);
 #[repr(transparent)]
 pub struct Type3<T>(T);
 
+// repr(transparent) wrapper which engages in self-reference
+#[repr(transparent)]
+pub struct Type4(Type4Helper<Type4>);
+#[repr(transparent)]
+pub struct Type4Helper<T>(*mut T);
+
 pub fn foo1(_: Type1) { }
 // CHECK: define{{.*}}4foo1{{.*}}!type ![[TYPE1:[0-9]+]] !type !{{[0-9]+}} !type !{{[0-9]+}} !type !{{[0-9]+}}
 pub fn foo2(_: Type1, _: Type1) { }
@@ -52,6 +58,13 @@ pub fn foo8(_: Type3<Bar>, _: Type3<Bar>) { }
 // CHECK: define{{.*}}4foo8{{.*}}!type ![[TYPE8:[0-9]+]] !type !{{[0-9]+}} !type !{{[0-9]+}} !type !{{[0-9]+}}
 pub fn foo9(_: Type3<Bar>, _: Type3<Bar>, _: Type3<Bar>) { }
 // CHECK: define{{.*}}4foo9{{.*}}!type ![[TYPE9:[0-9]+]] !type !{{[0-9]+}} !type !{{[0-9]+}} !type !{{[0-9]+}}
+pub fn foo10(_: Type4) { }
+// CHECK: define{{.*}}5foo10{{.*}}!type ![[TYPE10:[0-9]+]] !type !{{[0-9]+}} !type !{{[0-9]+}} !type !{{[0-9]+}}
+pub fn foo11(_: Type4, _: Type4) { }
+// CHECK: define{{.*}}5foo11{{.*}}!type ![[TYPE11:[0-9]+]] !type !{{[0-9]+}} !type !{{[0-9]+}} !type !{{[0-9]+}}
+pub fn foo12(_: Type4, _: Type4, _: Type4) { }
+// CHECK: define{{.*}}5foo12{{.*}}!type ![[TYPE12:[0-9]+]] !type !{{[0-9]+}} !type !{{[0-9]+}} !type !{{[0-9]+}}
+
 
 // CHECK: ![[TYPE1]] = !{i64 0, !"_ZTSFvu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}3FooE"}
 // CHECK: ![[TYPE2]] = !{i64 0, !"_ZTSFvu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}3FooS_E"}
@@ -62,3 +75,6 @@ pub fn foo9(_: Type3<Bar>, _: Type3<Bar>, _: Type3<Bar>) { }
 // CHECK: ![[TYPE7]] = !{i64 0, !"_ZTSFvu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}3BarE"}
 // CHECK: ![[TYPE8]] = !{i64 0, !"_ZTSFvu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}3BarS_E"}
 // CHECK: ![[TYPE9]] = !{i64 0, !"_ZTSFvu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}3BarS_S_E"}
+// CHECK: ![[TYPE10]] = !{i64 0, !"_ZTSFvPu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}5Type4E"}
+// CHECK: ![[TYPE11]] = !{i64 0, !"_ZTSFvPu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}5Type4S0_E"}
+// CHECK: ![[TYPE12]] = !{i64 0, !"_ZTSFvPu{{[0-9]+}}NtC{{[[:print:]]+}}_{{[[:print:]]+}}5Type4S0_S0_E"}
diff --git a/tests/ui/asm/aarch64/type-check-3.stderr b/tests/ui/asm/aarch64/type-check-3.stderr
index f710df2dcde96..9e37bb4c203f5 100644
--- a/tests/ui/asm/aarch64/type-check-3.stderr
+++ b/tests/ui/asm/aarch64/type-check-3.stderr
@@ -4,8 +4,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(reg) 0u8);
    |               ^^           --- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
    = note: `#[warn(asm_sub_register)]` on by default
 
 warning: formatting may not be suitable for sub-register argument
@@ -14,8 +14,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(reg) 0u16);
    |               ^^           ---- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:52:15
@@ -23,8 +23,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(reg) 0i32);
    |               ^^           ---- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:54:15
@@ -32,8 +32,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(reg) 0f32);
    |               ^^           ---- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:57:15
@@ -41,8 +41,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(vreg) 0i16);
    |               ^^            ---- for this argument
    |
-   = help: use `{0:h}` to have the register formatted as `h0`
-   = help: or use `{0:v}` to keep the default formatting of `v0`
+   = help: use `{0:h}` to have the register formatted as `h0` (for 16-bit values)
+   = help: or use `{0:v}` to keep the default formatting of `v0` (for 128-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:59:15
@@ -50,8 +50,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(vreg) 0f32);
    |               ^^            ---- for this argument
    |
-   = help: use `{0:s}` to have the register formatted as `s0`
-   = help: or use `{0:v}` to keep the default formatting of `v0`
+   = help: use `{0:s}` to have the register formatted as `s0` (for 32-bit values)
+   = help: or use `{0:v}` to keep the default formatting of `v0` (for 128-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:61:15
@@ -59,8 +59,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(vreg) 0f64);
    |               ^^            ---- for this argument
    |
-   = help: use `{0:d}` to have the register formatted as `d0`
-   = help: or use `{0:v}` to keep the default formatting of `v0`
+   = help: use `{0:d}` to have the register formatted as `d0` (for 64-bit values)
+   = help: or use `{0:v}` to keep the default formatting of `v0` (for 128-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:63:15
@@ -68,8 +68,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(vreg_low16) 0f64);
    |               ^^                  ---- for this argument
    |
-   = help: use `{0:d}` to have the register formatted as `d0`
-   = help: or use `{0:v}` to keep the default formatting of `v0`
+   = help: use `{0:d}` to have the register formatted as `d0` (for 64-bit values)
+   = help: or use `{0:v}` to keep the default formatting of `v0` (for 128-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:66:15
@@ -77,8 +77,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{0} {0}", in(reg) 0i16);
    |               ^^^ ^^^           ---- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:68:15
@@ -86,8 +86,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{0} {0:x}", in(reg) 0i16);
    |               ^^^                 ---- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
 
 error: type `i128` cannot be used with this register class
   --> $DIR/type-check-3.rs:73:28
diff --git a/tests/ui/asm/bad-template.aarch64.stderr b/tests/ui/asm/bad-template.aarch64.stderr
index b18946d7c6d85..5023cf317d7b0 100644
--- a/tests/ui/asm/bad-template.aarch64.stderr
+++ b/tests/ui/asm/bad-template.aarch64.stderr
@@ -194,8 +194,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{:foo}", in(reg) foo);
    |               ^^^^^^           --- for this argument
    |
-   = help: use `{0:w}` to have the register formatted as `w0`
-   = help: or use `{0:x}` to keep the default formatting of `x0`
+   = help: use `{0:w}` to have the register formatted as `w0` (for 32-bit values)
+   = help: or use `{0:x}` to keep the default formatting of `x0` (for 64-bit values)
    = note: `#[warn(asm_sub_register)]` on by default
 
 error: aborting due to 21 previous errors; 1 warning emitted
diff --git a/tests/ui/asm/bad-template.x86_64.stderr b/tests/ui/asm/bad-template.x86_64.stderr
index 2f584c30a3282..1b9775636f564 100644
--- a/tests/ui/asm/bad-template.x86_64.stderr
+++ b/tests/ui/asm/bad-template.x86_64.stderr
@@ -194,8 +194,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{:foo}", in(reg) foo);
    |               ^^^^^^           --- for this argument
    |
-   = help: use `{0:e}` to have the register formatted as `eax`
-   = help: or use `{0:r}` to keep the default formatting of `rax`
+   = help: use `{0:e}` to have the register formatted as `eax` (for 32-bit values)
+   = help: or use `{0:r}` to keep the default formatting of `rax` (for 64-bit values)
    = note: `#[warn(asm_sub_register)]` on by default
 
 error: aborting due to 21 previous errors; 1 warning emitted
diff --git a/tests/ui/asm/x86_64/type-check-3.stderr b/tests/ui/asm/x86_64/type-check-3.stderr
index 1baf50ff6e0c7..34bfcd71caca7 100644
--- a/tests/ui/asm/x86_64/type-check-3.stderr
+++ b/tests/ui/asm/x86_64/type-check-3.stderr
@@ -44,8 +44,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{0} {0}", in(reg) 0i16);
    |               ^^^ ^^^           ---- for this argument
    |
-   = help: use `{0:x}` to have the register formatted as `ax`
-   = help: or use `{0:r}` to keep the default formatting of `rax`
+   = help: use `{0:x}` to have the register formatted as `ax` (for 16-bit values)
+   = help: or use `{0:r}` to keep the default formatting of `rax` (for 64-bit values)
    = note: `#[warn(asm_sub_register)]` on by default
 
 warning: formatting may not be suitable for sub-register argument
@@ -54,8 +54,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{0} {0:x}", in(reg) 0i16);
    |               ^^^                 ---- for this argument
    |
-   = help: use `{0:x}` to have the register formatted as `ax`
-   = help: or use `{0:r}` to keep the default formatting of `rax`
+   = help: use `{0:x}` to have the register formatted as `ax` (for 16-bit values)
+   = help: or use `{0:r}` to keep the default formatting of `rax` (for 64-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:38:15
@@ -63,8 +63,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(reg) 0i32);
    |               ^^           ---- for this argument
    |
-   = help: use `{0:e}` to have the register formatted as `eax`
-   = help: or use `{0:r}` to keep the default formatting of `rax`
+   = help: use `{0:e}` to have the register formatted as `eax` (for 32-bit values)
+   = help: or use `{0:r}` to keep the default formatting of `rax` (for 64-bit values)
 
 warning: formatting may not be suitable for sub-register argument
   --> $DIR/type-check-3.rs:41:15
@@ -72,8 +72,8 @@ warning: formatting may not be suitable for sub-register argument
 LL |         asm!("{}", in(ymm_reg) 0i64);
    |               ^^               ---- for this argument
    |
-   = help: use `{0:x}` to have the register formatted as `xmm0`
-   = help: or use `{0:y}` to keep the default formatting of `ymm0`
+   = help: use `{0:x}` to have the register formatted as `xmm0` (for 128-bit values)
+   = help: or use `{0:y}` to keep the default formatting of `ymm0` (for 256-bit values)
 
 error: type `i8` cannot be used with this register class
   --> $DIR/type-check-3.rs:52:28
diff --git a/tests/ui/sanitizer/cfi-self-ref.rs b/tests/ui/sanitizer/cfi-self-ref.rs
new file mode 100644
index 0000000000000..32d0b10070265
--- /dev/null
+++ b/tests/ui/sanitizer/cfi-self-ref.rs
@@ -0,0 +1,33 @@
+// Check that encoding self-referential types works with #[repr(transparent)]
+
+//@ needs-sanitizer-cfi
+// FIXME(#122848) Remove only-linux once OSX CFI binaries work
+//@ only-linux
+//@ compile-flags: --crate-type=bin -Cprefer-dynamic=off -Clto -Zsanitizer=cfi
+//@ compile-flags: -C target-feature=-crt-static -C codegen-units=1 -C opt-level=0
+//@ run-pass
+
+use std::marker::PhantomData;
+
+struct X<T> {
+    _x: u8,
+    p: PhantomData<T>,
+}
+
+#[repr(transparent)]
+struct Y(X<Y>);
+
+trait Fooable {
+    fn foo(&self, y: Y);
+}
+
+struct Bar;
+
+impl Fooable for Bar {
+    fn foo(&self, _: Y) {}
+}
+
+fn main() {
+    let x = &Bar as &dyn Fooable;
+    x.foo(Y(X {_x: 0, p: PhantomData}));
+}
diff --git a/tests/ui/sanitizer/cfi-virtual-auto.rs b/tests/ui/sanitizer/cfi-virtual-auto.rs
new file mode 100644
index 0000000000000..7a0c246a41221
--- /dev/null
+++ b/tests/ui/sanitizer/cfi-virtual-auto.rs
@@ -0,0 +1,22 @@
+// Tests that calling a trait object method on a trait object with additional auto traits works.
+
+//@ needs-sanitizer-cfi
+// FIXME(#122848) Remove only-linux once OSX CFI binaries work
+//@ only-linux
+//@ compile-flags: --crate-type=bin -Cprefer-dynamic=off -Clto -Zsanitizer=cfi
+//@ compile-flags: -C target-feature=-crt-static -C codegen-units=1 -C opt-level=0
+//@ run-pass
+
+trait Foo {
+    fn foo(&self);
+}
+
+struct Bar;
+impl Foo for Bar {
+    fn foo(&self) {}
+}
+
+pub fn main() {
+    let x: &(dyn Foo + Send) = &Bar;
+    x.foo();
+}