about summary refs log tree commit diff
path: root/compiler
diff options
context:
space:
mode:
authorThe Miri Cronjob Bot <miri@cron.bot>2025-09-05 05:00:55 +0000
committerThe Miri Cronjob Bot <miri@cron.bot>2025-09-05 05:00:55 +0000
commit7c5fdb072e3d07674ba88153a4cac4f792c20fdc (patch)
tree78897a875065b81c78004b1cd0f83e78254fb11d /compiler
parent9a1eb85394e3612213e829ed871eb281079a2364 (diff)
parentb6f8824315367f73e43efe08f6e5a88cc3f117ce (diff)
downloadrust-7c5fdb072e3d07674ba88153a4cac4f792c20fdc.tar.gz
rust-7c5fdb072e3d07674ba88153a4cac4f792c20fdc.zip
Merge ref 'b3cfb8faf84c' from rust-lang/rust
Pull recent changes from https://github.com/rust-lang/rust via Josh.

Upstream ref: b3cfb8faf84c5f3b7909479a9f9b6a3290d5f4d7
Filtered ref: 8995aa7743caf019203bc853f27af6006705ae30
Upstream diff: https://github.com/rust-lang/rust/compare/9385c64c95d971329e62917adc4349c8ccdbafe0...b3cfb8faf84c5f3b7909479a9f9b6a3290d5f4d7

This merge was created using https://github.com/rust-lang/josh-sync.
Diffstat (limited to 'compiler')
-rw-r--r--compiler/rustc_ast/src/token.rs20
-rw-r--r--compiler/rustc_attr_parsing/messages.ftl2
-rw-r--r--compiler/rustc_codegen_cranelift/src/driver/aot.rs11
-rw-r--r--compiler/rustc_codegen_gcc/src/back/lto.rs2
-rw-r--r--compiler/rustc_codegen_llvm/src/attributes.rs31
-rw-r--r--compiler/rustc_codegen_llvm/src/back/lto.rs22
-rw-r--r--compiler/rustc_codegen_llvm/src/builder.rs29
-rw-r--r--compiler/rustc_codegen_llvm/src/context.rs4
-rw-r--r--compiler/rustc_codegen_llvm/src/intrinsic.rs26
-rw-r--r--compiler/rustc_codegen_ssa/messages.ftl14
-rw-r--r--compiler/rustc_codegen_ssa/src/back/link.rs17
-rw-r--r--compiler/rustc_codegen_ssa/src/back/linker.rs17
-rw-r--r--compiler/rustc_codegen_ssa/src/back/lto.rs22
-rw-r--r--compiler/rustc_codegen_ssa/src/back/symbol_export.rs52
-rw-r--r--compiler/rustc_codegen_ssa/src/back/write.rs137
-rw-r--r--compiler/rustc_codegen_ssa/src/base.rs91
-rw-r--r--compiler/rustc_codegen_ssa/src/codegen_attrs.rs9
-rw-r--r--compiler/rustc_codegen_ssa/src/lib.rs2
-rw-r--r--compiler/rustc_const_eval/messages.ftl2
-rw-r--r--compiler/rustc_driver_impl/messages.ftl4
-rw-r--r--compiler/rustc_errors/src/lib.rs2
-rw-r--r--compiler/rustc_expand/src/mbe/macro_parser.rs20
-rw-r--r--compiler/rustc_feature/src/unstable.rs2
-rw-r--r--compiler/rustc_hir_analysis/src/check/intrinsic.rs3
-rw-r--r--compiler/rustc_hir_analysis/src/check/region.rs8
-rw-r--r--compiler/rustc_interface/src/interface.rs6
-rw-r--r--compiler/rustc_lint_defs/src/builtin.rs54
-rw-r--r--compiler/rustc_metadata/src/creader.rs18
-rw-r--r--compiler/rustc_middle/src/middle/codegen_fn_attrs.rs2
-rw-r--r--compiler/rustc_middle/src/middle/region.rs39
-rw-r--r--compiler/rustc_middle/src/ty/context.rs4
-rw-r--r--compiler/rustc_middle/src/ty/rvalue_scopes.rs37
-rw-r--r--compiler/rustc_mir_transform/src/check_inline_always_target_features.rs88
-rw-r--r--compiler/rustc_mir_transform/src/errors.rs41
-rw-r--r--compiler/rustc_mir_transform/src/lib.rs4
-rw-r--r--compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs4
-rw-r--r--compiler/rustc_parse/messages.ftl2
-rw-r--r--compiler/rustc_parse/src/lib.rs20
-rw-r--r--compiler/rustc_passes/messages.ftl1
-rw-r--r--compiler/rustc_passes/src/errors.rs18
-rw-r--r--compiler/rustc_passes/src/liveness.rs50
-rw-r--r--compiler/rustc_query_system/messages.ftl4
-rw-r--r--compiler/rustc_resolve/messages.ftl2
-rw-r--r--compiler/rustc_resolve/src/diagnostics.rs172
-rw-r--r--compiler/rustc_resolve/src/errors.rs12
-rw-r--r--compiler/rustc_resolve/src/late.rs52
-rw-r--r--compiler/rustc_resolve/src/lib.rs4
-rw-r--r--compiler/rustc_session/src/options.rs71
-rw-r--r--compiler/rustc_span/src/analyze_source_file.rs109
-rw-r--r--compiler/rustc_span/src/lib.rs1
-rw-r--r--compiler/rustc_span/src/symbol.rs3
-rw-r--r--compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs2
-rw-r--r--compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs2
-rw-r--r--compiler/rustc_type_ir/src/interner.rs7
-rw-r--r--compiler/rustc_type_ir/src/solve/inspect.rs2
55 files changed, 1024 insertions, 356 deletions
diff --git a/compiler/rustc_ast/src/token.rs b/compiler/rustc_ast/src/token.rs
index ea98bebd305..6dc6d1026f6 100644
--- a/compiler/rustc_ast/src/token.rs
+++ b/compiler/rustc_ast/src/token.rs
@@ -22,8 +22,7 @@ pub enum CommentKind {
     Block,
 }
 
-// This type must not implement `Hash` due to the unusual `PartialEq` impl below.
-#[derive(Copy, Clone, Debug, Encodable, Decodable, HashStable_Generic)]
+#[derive(Copy, Clone, PartialEq, Debug, Encodable, Decodable, HashStable_Generic)]
 pub enum InvisibleOrigin {
     // From the expansion of a metavariable in a declarative macro.
     MetaVar(MetaVarKind),
@@ -45,20 +44,6 @@ impl InvisibleOrigin {
     }
 }
 
-impl PartialEq for InvisibleOrigin {
-    #[inline]
-    fn eq(&self, _other: &InvisibleOrigin) -> bool {
-        // When we had AST-based nonterminals we couldn't compare them, and the
-        // old `Nonterminal` type had an `eq` that always returned false,
-        // resulting in this restriction:
-        // https://doc.rust-lang.org/nightly/reference/macros-by-example.html#forwarding-a-matched-fragment
-        // This `eq` emulates that behaviour. We could consider lifting this
-        // restriction now but there are still cases involving invisible
-        // delimiters that make it harder than it first appears.
-        false
-    }
-}
-
 /// Annoyingly similar to `NonterminalKind`, but the slight differences are important.
 #[derive(Debug, Copy, Clone, PartialEq, Eq, Encodable, Decodable, Hash, HashStable_Generic)]
 pub enum MetaVarKind {
@@ -142,7 +127,8 @@ impl Delimiter {
         }
     }
 
-    // This exists because `InvisibleOrigin`s should be compared. It is only used for assertions.
+    // This exists because `InvisibleOrigin`s should not be compared. It is only used for
+    // assertions.
     pub fn eq_ignoring_invisible_origin(&self, other: &Delimiter) -> bool {
         match (self, other) {
             (Delimiter::Parenthesis, Delimiter::Parenthesis) => true,
diff --git a/compiler/rustc_attr_parsing/messages.ftl b/compiler/rustc_attr_parsing/messages.ftl
index b8a748563d5..7fa1293463c 100644
--- a/compiler/rustc_attr_parsing/messages.ftl
+++ b/compiler/rustc_attr_parsing/messages.ftl
@@ -151,7 +151,7 @@ attr_parsing_unrecognized_repr_hint =
 attr_parsing_unstable_cfg_target_compact =
     compact `cfg(target(..))` is experimental and subject to change
 
-attr_parsing_unstable_feature_bound_incompatible_stability = Item annotated with `#[unstable_feature_bound]` should not be stable
+attr_parsing_unstable_feature_bound_incompatible_stability = item annotated with `#[unstable_feature_bound]` should not be stable
     .help = If this item is meant to be stable, do not use any functions annotated with `#[unstable_feature_bound]`. Otherwise, mark this item as unstable with `#[unstable]`
 
 attr_parsing_unsupported_literal_cfg_boolean =
diff --git a/compiler/rustc_codegen_cranelift/src/driver/aot.rs b/compiler/rustc_codegen_cranelift/src/driver/aot.rs
index 7e77781dc2f..c3adb5e767e 100644
--- a/compiler/rustc_codegen_cranelift/src/driver/aot.rs
+++ b/compiler/rustc_codegen_cranelift/src/driver/aot.rs
@@ -12,9 +12,7 @@ use cranelift_object::{ObjectBuilder, ObjectModule};
 use rustc_codegen_ssa::assert_module_sources::CguReuse;
 use rustc_codegen_ssa::back::link::ensure_removed;
 use rustc_codegen_ssa::base::determine_cgu_reuse;
-use rustc_codegen_ssa::{
-    CodegenResults, CompiledModule, CrateInfo, ModuleKind, errors as ssa_errors,
-};
+use rustc_codegen_ssa::{CodegenResults, CompiledModule, CrateInfo, errors as ssa_errors};
 use rustc_data_structures::profiling::SelfProfilerRef;
 use rustc_data_structures::stable_hasher::{HashStable, StableHasher};
 use rustc_data_structures::sync::{IntoDynSyncSend, par_map};
@@ -363,7 +361,6 @@ fn emit_cgu(
         invocation_temp,
         prof,
         product.object,
-        ModuleKind::Regular,
         name.clone(),
         producer,
     )?;
@@ -372,7 +369,6 @@ fn emit_cgu(
         module_regular,
         module_global_asm: global_asm_object_file.map(|global_asm_object_file| CompiledModule {
             name: format!("{name}.asm"),
-            kind: ModuleKind::Regular,
             object: Some(global_asm_object_file),
             dwarf_object: None,
             bytecode: None,
@@ -389,7 +385,6 @@ fn emit_module(
     invocation_temp: Option<&str>,
     prof: &SelfProfilerRef,
     mut object: cranelift_object::object::write::Object<'_>,
-    kind: ModuleKind,
     name: String,
     producer_str: &str,
 ) -> Result<CompiledModule, String> {
@@ -430,7 +425,6 @@ fn emit_module(
 
     Ok(CompiledModule {
         name,
-        kind,
         object: Some(tmp_file),
         dwarf_object: None,
         bytecode: None,
@@ -485,7 +479,6 @@ fn reuse_workproduct_for_cgu(
     Ok(ModuleCodegenResult {
         module_regular: CompiledModule {
             name: cgu.name().to_string(),
-            kind: ModuleKind::Regular,
             object: Some(obj_out_regular),
             dwarf_object: None,
             bytecode: None,
@@ -495,7 +488,6 @@ fn reuse_workproduct_for_cgu(
         },
         module_global_asm: source_file_global_asm.map(|source_file| CompiledModule {
             name: cgu.name().to_string(),
-            kind: ModuleKind::Regular,
             object: Some(obj_out_global_asm),
             dwarf_object: None,
             bytecode: None,
@@ -651,7 +643,6 @@ fn emit_allocator_module(tcx: TyCtxt<'_>) -> Option<CompiledModule> {
             tcx.sess.invocation_temp.as_deref(),
             &tcx.sess.prof,
             product.object,
-            ModuleKind::Allocator,
             "allocator_shim".to_owned(),
             &crate::debuginfo::producer(tcx.sess),
         ) {
diff --git a/compiler/rustc_codegen_gcc/src/back/lto.rs b/compiler/rustc_codegen_gcc/src/back/lto.rs
index fcee6b6df62..9d8ce2383f2 100644
--- a/compiler/rustc_codegen_gcc/src/back/lto.rs
+++ b/compiler/rustc_codegen_gcc/src/back/lto.rs
@@ -204,7 +204,7 @@ fn fat_lto(
             let path = tmp_path.path().to_path_buf().join(&module.name);
             let path = path.to_str().expect("path");
             let context = &module.module_llvm.context;
-            let config = cgcx.config(module.kind);
+            let config = &cgcx.module_config;
             // NOTE: we need to set the optimization level here in order for LTO to do its job.
             context.set_optimization_level(to_gcc_opt_level(config.opt_level));
             context.add_command_line_option("-flto=auto");
diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs
index 5affb26483a..9f2d37d39d8 100644
--- a/compiler/rustc_codegen_llvm/src/attributes.rs
+++ b/compiler/rustc_codegen_llvm/src/attributes.rs
@@ -29,8 +29,18 @@ pub(crate) fn apply_to_callsite(callsite: &Value, idx: AttributePlace, attrs: &[
 }
 
 /// Get LLVM attribute for the provided inline heuristic.
-#[inline]
-fn inline_attr<'ll>(cx: &CodegenCx<'ll, '_>, inline: InlineAttr) -> Option<&'ll Attribute> {
+pub(crate) fn inline_attr<'ll, 'tcx>(
+    cx: &CodegenCx<'ll, 'tcx>,
+    instance: ty::Instance<'tcx>,
+) -> Option<&'ll Attribute> {
+    // `optnone` requires `noinline`
+    let codegen_fn_attrs = cx.tcx.codegen_fn_attrs(instance.def_id());
+    let inline = match (codegen_fn_attrs.inline, &codegen_fn_attrs.optimize) {
+        (_, OptimizeAttr::DoNotOptimize) => InlineAttr::Never,
+        (InlineAttr::None, _) if instance.def.requires_inline(cx.tcx) => InlineAttr::Hint,
+        (inline, _) => inline,
+    };
+
     if !cx.tcx.sess.opts.unstable_opts.inline_llvm {
         // disable LLVM inlining
         return Some(AttributeKind::NoInline.create_attr(cx.llcx));
@@ -346,14 +356,6 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
         OptimizeAttr::Speed => {}
     }
 
-    // `optnone` requires `noinline`
-    let inline = match (codegen_fn_attrs.inline, &codegen_fn_attrs.optimize) {
-        (_, OptimizeAttr::DoNotOptimize) => InlineAttr::Never,
-        (InlineAttr::None, _) if instance.def.requires_inline(cx.tcx) => InlineAttr::Hint,
-        (inline, _) => inline,
-    };
-    to_add.extend(inline_attr(cx, inline));
-
     if cx.sess().must_emit_unwind_tables() {
         to_add.push(uwtable_attr(cx.llcx, cx.sess().opts.unstable_opts.use_sync_unwind));
     }
@@ -488,6 +490,14 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
     let function_features =
         codegen_fn_attrs.target_features.iter().map(|f| f.name.as_str()).collect::<Vec<&str>>();
 
+    // Apply function attributes as per usual if there are no user defined
+    // target features otherwise this will get applied at the callsite.
+    if function_features.is_empty() {
+        if let Some(inline_attr) = inline_attr(cx, instance) {
+            to_add.push(inline_attr);
+        }
+    }
+
     let function_features = function_features
         .iter()
         // Convert to LLVMFeatures and filter out unavailable ones
@@ -517,6 +527,7 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
     let function_features = function_features.iter().map(|s| s.as_str());
     let target_features: String =
         global_features.chain(function_features).intersperse(",").collect();
+
     if !target_features.is_empty() {
         to_add.push(llvm::CreateAttrStringValue(cx.llcx, "target-features", &target_features));
     }
diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs
index fc38c4f3e51..326b876e7e6 100644
--- a/compiler/rustc_codegen_llvm/src/back/lto.rs
+++ b/compiler/rustc_codegen_llvm/src/back/lto.rs
@@ -11,7 +11,7 @@ use object::{Object, ObjectSection};
 use rustc_codegen_ssa::back::lto::{SerializedModule, ThinModule, ThinShared};
 use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput};
 use rustc_codegen_ssa::traits::*;
-use rustc_codegen_ssa::{ModuleCodegen, ModuleKind, looks_like_rust_object_file};
+use rustc_codegen_ssa::{ModuleCodegen, looks_like_rust_object_file};
 use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::memmap::Mmap;
 use rustc_errors::DiagCtxtHandle;
@@ -43,9 +43,7 @@ fn prepare_lto(
         .map(|symbol| CString::new(symbol.to_owned()).unwrap())
         .collect::<Vec<CString>>();
 
-    if cgcx.regular_module_config.instrument_coverage
-        || cgcx.regular_module_config.pgo_gen.enabled()
-    {
+    if cgcx.module_config.instrument_coverage || cgcx.module_config.pgo_gen.enabled() {
         // These are weak symbols that point to the profile version and the
         // profile name, which need to be treated as exported so LTO doesn't nix
         // them.
@@ -55,15 +53,15 @@ fn prepare_lto(
         symbols_below_threshold.extend(PROFILER_WEAK_SYMBOLS.iter().map(|&sym| sym.to_owned()));
     }
 
-    if cgcx.regular_module_config.sanitizer.contains(SanitizerSet::MEMORY) {
+    if cgcx.module_config.sanitizer.contains(SanitizerSet::MEMORY) {
         let mut msan_weak_symbols = Vec::new();
 
         // Similar to profiling, preserve weak msan symbol during LTO.
-        if cgcx.regular_module_config.sanitizer_recover.contains(SanitizerSet::MEMORY) {
+        if cgcx.module_config.sanitizer_recover.contains(SanitizerSet::MEMORY) {
             msan_weak_symbols.push(c"__msan_keep_going");
         }
 
-        if cgcx.regular_module_config.sanitizer_memory_track_origins != 0 {
+        if cgcx.module_config.sanitizer_memory_track_origins != 0 {
             msan_weak_symbols.push(c"__msan_track_origins");
         }
 
@@ -227,15 +225,9 @@ fn fat_lto(
     // All the other modules will be serialized and reparsed into the new
     // context, so this hopefully avoids serializing and parsing the largest
     // codegen unit.
-    //
-    // Additionally use a regular module as the base here to ensure that various
-    // file copy operations in the backend work correctly. The only other kind
-    // of module here should be an allocator one, and if your crate is smaller
-    // than the allocator module then the size doesn't really matter anyway.
     let costliest_module = in_memory
         .iter()
         .enumerate()
-        .filter(|&(_, module)| module.kind == ModuleKind::Regular)
         .map(|(i, module)| {
             let cost = unsafe { llvm::LLVMRustModuleCost(module.module_llvm.llmod()) };
             (cost, i)
@@ -583,7 +575,7 @@ pub(crate) fn run_pass_manager(
     thin: bool,
 ) {
     let _timer = cgcx.prof.generic_activity_with_arg("LLVM_lto_optimize", &*module.name);
-    let config = cgcx.config(module.kind);
+    let config = &cgcx.module_config;
 
     // Now we have one massive module inside of llmod. Time to run the
     // LTO-specific optimization passes that LLVM provides.
@@ -745,7 +737,7 @@ pub(crate) fn optimize_thin_module(
     let module_llvm = ModuleLlvm::parse(cgcx, module_name, thin_module.data(), dcx);
     let mut module = ModuleCodegen::new_regular(thin_module.name(), module_llvm);
     // Given that the newly created module lacks a thinlto buffer for embedding, we need to re-add it here.
-    if cgcx.config(ModuleKind::Regular).embed_bitcode() {
+    if cgcx.module_config.embed_bitcode() {
         module.thin_lto_buffer = Some(thin_module.data().to_vec());
     }
     {
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
index 37379586d58..7d0691366e6 100644
--- a/compiler/rustc_codegen_llvm/src/builder.rs
+++ b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -1392,7 +1392,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
     fn call(
         &mut self,
         llty: &'ll Type,
-        fn_attrs: Option<&CodegenFnAttrs>,
+        fn_call_attrs: Option<&CodegenFnAttrs>,
         fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
         llfn: &'ll Value,
         args: &[&'ll Value],
@@ -1409,10 +1409,10 @@ 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, instance, llfn);
+        self.cfi_type_test(fn_call_attrs, fn_abi, instance, llfn);
 
         // Emit KCFI operand bundle
-        let kcfi_bundle = self.kcfi_operand_bundle(fn_attrs, fn_abi, instance, llfn);
+        let kcfi_bundle = self.kcfi_operand_bundle(fn_call_attrs, fn_abi, instance, llfn);
         if let Some(kcfi_bundle) = kcfi_bundle.as_ref().map(|b| b.as_ref()) {
             bundles.push(kcfi_bundle);
         }
@@ -1429,6 +1429,29 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
                 c"".as_ptr(),
             )
         };
+
+        if let Some(instance) = instance {
+            // Attributes on the function definition being called
+            let fn_defn_attrs = self.cx.tcx.codegen_fn_attrs(instance.def_id());
+            if let Some(fn_call_attrs) = fn_call_attrs
+                && !fn_call_attrs.target_features.is_empty()
+                // If there is an inline attribute and a target feature that matches
+                // we will add the attribute to the callsite otherwise we'll omit
+                // this and not add the attribute to prevent soundness issues.
+                && let Some(inlining_rule) = attributes::inline_attr(&self.cx, instance)
+                && self.cx.tcx.is_target_feature_call_safe(
+                    &fn_call_attrs.target_features,
+                    &fn_defn_attrs.target_features,
+                )
+            {
+                attributes::apply_to_callsite(
+                    call,
+                    llvm::AttributePlace::Function,
+                    &[inlining_rule],
+                );
+            }
+        }
+
         if let Some(fn_abi) = fn_abi {
             fn_abi.apply_attrs_callsite(self, call);
         }
diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs
index 4fd6110ac4a..257c7b95666 100644
--- a/compiler/rustc_codegen_llvm/src/context.rs
+++ b/compiler/rustc_codegen_llvm/src/context.rs
@@ -217,6 +217,10 @@ pub(crate) unsafe fn create_module<'ll>(
             // LLVM 22.0 updated the default layout on avr: https://github.com/llvm/llvm-project/pull/153010
             target_data_layout = target_data_layout.replace("n8:16", "n8")
         }
+        if sess.target.arch == "nvptx64" {
+            // LLVM 22 updated the NVPTX layout to indicate 256-bit vector load/store: https://github.com/llvm/llvm-project/pull/155198
+            target_data_layout = target_data_layout.replace("-i256:256", "");
+        }
     }
 
     // Ensure the data-layout values hardcoded remain the defaults.
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index 49d3dedbeab..85f71f331a4 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -383,7 +383,9 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
             | sym::rotate_left
             | sym::rotate_right
             | sym::saturating_add
-            | sym::saturating_sub => {
+            | sym::saturating_sub
+            | sym::unchecked_funnel_shl
+            | sym::unchecked_funnel_shr => {
                 let ty = args[0].layout.ty;
                 if !ty.is_integral() {
                     tcx.dcx().emit_err(InvalidMonomorphization::BasicIntegerType {
@@ -424,18 +426,26 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     sym::bitreverse => {
                         self.call_intrinsic("llvm.bitreverse", &[llty], &[args[0].immediate()])
                     }
-                    sym::rotate_left | sym::rotate_right => {
-                        let is_left = name == sym::rotate_left;
-                        let val = args[0].immediate();
-                        let raw_shift = args[1].immediate();
-                        // rotate = funnel shift with first two args the same
+                    sym::rotate_left
+                    | sym::rotate_right
+                    | sym::unchecked_funnel_shl
+                    | sym::unchecked_funnel_shr => {
+                        let is_left = name == sym::rotate_left || name == sym::unchecked_funnel_shl;
+                        let lhs = args[0].immediate();
+                        let (rhs, raw_shift) =
+                            if name == sym::rotate_left || name == sym::rotate_right {
+                                // rotate = funnel shift with first two args the same
+                                (lhs, args[1].immediate())
+                            } else {
+                                (args[1].immediate(), args[2].immediate())
+                            };
                         let llvm_name = format!("llvm.fsh{}", if is_left { 'l' } else { 'r' });
 
                         // llvm expects shift to be the same type as the values, but rust
                         // always uses `u32`.
-                        let raw_shift = self.intcast(raw_shift, self.val_ty(val), false);
+                        let raw_shift = self.intcast(raw_shift, self.val_ty(lhs), false);
 
-                        self.call_intrinsic(llvm_name, &[llty], &[val, val, raw_shift])
+                        self.call_intrinsic(llvm_name, &[llty], &[lhs, rhs, raw_shift])
                     }
                     sym::saturating_add | sym::saturating_sub => {
                         let is_add = name == sym::saturating_add;
diff --git a/compiler/rustc_codegen_ssa/messages.ftl b/compiler/rustc_codegen_ssa/messages.ftl
index 44b9941691a..dd688b8b345 100644
--- a/compiler/rustc_codegen_ssa/messages.ftl
+++ b/compiler/rustc_codegen_ssa/messages.ftl
@@ -40,15 +40,15 @@ codegen_ssa_dynamic_linking_with_lto =
     .note = only 'staticlib', 'bin', and 'cdylib' outputs are supported with LTO
 
 codegen_ssa_error_calling_dlltool =
-    Error calling dlltool '{$dlltool_path}': {$error}
+    error calling dlltool '{$dlltool_path}': {$error}
 
 codegen_ssa_error_creating_import_library =
-    Error creating import library for {$lib_name}: {$error}
+    error creating import library for {$lib_name}: {$error}
 
 codegen_ssa_error_creating_remark_dir = failed to create remark directory: {$error}
 
 codegen_ssa_error_writing_def_file =
-    Error writing .DEF file: {$error}
+    error writing .DEF file: {$error}
 
 codegen_ssa_expected_name_value_pair = expected name value pair
 
@@ -264,9 +264,9 @@ codegen_ssa_shuffle_indices_evaluation = could not evaluate shuffle_indices at c
 
 codegen_ssa_specify_libraries_to_link = use the `-l` flag to specify native libraries to link
 
-codegen_ssa_static_library_native_artifacts = Link against the following native artifacts when linking against this static library. The order and any duplication can be significant on some platforms.
+codegen_ssa_static_library_native_artifacts = link against the following native artifacts when linking against this static library. The order and any duplication can be significant on some platforms.
 
-codegen_ssa_static_library_native_artifacts_to_file = Native artifacts to link against have been written to {$path}. The order and any duplication can be significant on some platforms.
+codegen_ssa_static_library_native_artifacts_to_file = native artifacts to link against have been written to {$path}. The order and any duplication can be significant on some platforms.
 
 codegen_ssa_stripping_debug_info_failed = stripping debug info with `{$util}` failed: {$status}
     .note = {$output}
@@ -364,13 +364,13 @@ codegen_ssa_unable_to_run = unable to run `{$util}`: {$error}
 
 codegen_ssa_unable_to_run_dsymutil = unable to run `dsymutil`: {$error}
 
-codegen_ssa_unable_to_write_debugger_visualizer = Unable to write debugger visualizer file `{$path}`: {$error}
+codegen_ssa_unable_to_write_debugger_visualizer = unable to write debugger visualizer file `{$path}`: {$error}
 
 codegen_ssa_unexpected_parameter_name = unexpected parameter name
     .label = expected `{$prefix_nops}` or `{$entry_nops}`
 
 codegen_ssa_unknown_archive_kind =
-    Don't know how to build archive of type: {$kind}
+    don't know how to build archive of type: {$kind}
 
 codegen_ssa_unknown_ctarget_feature =
     unknown and unstable feature specified for `-Ctarget-feature`: `{$feature}`
diff --git a/compiler/rustc_codegen_ssa/src/back/link.rs b/compiler/rustc_codegen_ssa/src/back/link.rs
index 19c919c0e4e..48b01ea2df1 100644
--- a/compiler/rustc_codegen_ssa/src/back/link.rs
+++ b/compiler/rustc_codegen_ssa/src/back/link.rs
@@ -58,6 +58,7 @@ use super::linker::{self, Linker};
 use super::metadata::{MetadataPosition, create_wrapper_file};
 use super::rpath::{self, RPathConfig};
 use super::{apple, versioned_llvm_target};
+use crate::base::needs_allocator_shim_for_linking;
 use crate::{
     CodegenResults, CompiledModule, CrateInfo, NativeLib, errors, looks_like_rust_object_file,
 };
@@ -2080,9 +2081,17 @@ fn add_local_crate_regular_objects(cmd: &mut dyn Linker, codegen_results: &Codeg
 }
 
 /// Add object files for allocator code linked once for the whole crate tree.
-fn add_local_crate_allocator_objects(cmd: &mut dyn Linker, codegen_results: &CodegenResults) {
-    if let Some(obj) = codegen_results.allocator_module.as_ref().and_then(|m| m.object.as_ref()) {
-        cmd.add_object(obj);
+fn add_local_crate_allocator_objects(
+    cmd: &mut dyn Linker,
+    codegen_results: &CodegenResults,
+    crate_type: CrateType,
+) {
+    if needs_allocator_shim_for_linking(&codegen_results.crate_info.dependency_formats, crate_type)
+    {
+        if let Some(obj) = codegen_results.allocator_module.as_ref().and_then(|m| m.object.as_ref())
+        {
+            cmd.add_object(obj);
+        }
     }
 }
 
@@ -2281,7 +2290,7 @@ fn linker_with_args(
         codegen_results,
         metadata,
     );
-    add_local_crate_allocator_objects(cmd, codegen_results);
+    add_local_crate_allocator_objects(cmd, codegen_results, crate_type);
 
     // Avoid linking to dynamic libraries unless they satisfy some undefined symbols
     // at the point at which they are specified on the command line.
diff --git a/compiler/rustc_codegen_ssa/src/back/linker.rs b/compiler/rustc_codegen_ssa/src/back/linker.rs
index df1e91b12f9..a2efd420a32 100644
--- a/compiler/rustc_codegen_ssa/src/back/linker.rs
+++ b/compiler/rustc_codegen_ssa/src/back/linker.rs
@@ -11,8 +11,9 @@ use rustc_metadata::{
 };
 use rustc_middle::bug;
 use rustc_middle::middle::dependency_format::Linkage;
-use rustc_middle::middle::exported_symbols;
-use rustc_middle::middle::exported_symbols::{ExportedSymbol, SymbolExportInfo, SymbolExportKind};
+use rustc_middle::middle::exported_symbols::{
+    self, ExportedSymbol, SymbolExportInfo, SymbolExportKind, SymbolExportLevel,
+};
 use rustc_middle::ty::TyCtxt;
 use rustc_session::Session;
 use rustc_session::config::{self, CrateType, DebugInfo, LinkerPluginLto, Lto, OptLevel, Strip};
@@ -22,6 +23,8 @@ use tracing::{debug, warn};
 
 use super::command::Command;
 use super::symbol_export;
+use crate::back::symbol_export::allocator_shim_symbols;
+use crate::base::needs_allocator_shim_for_linking;
 use crate::errors;
 
 #[cfg(test)]
@@ -1827,7 +1830,7 @@ fn exported_symbols_for_non_proc_macro(
     let export_threshold = symbol_export::crates_export_threshold(&[crate_type]);
     for_each_exported_symbols_include_dep(tcx, crate_type, |symbol, info, cnum| {
         // Do not export mangled symbols from cdylibs and don't attempt to export compiler-builtins
-        // from any cdylib. The latter doesn't work anyway as we use hidden visibility for
+        // from any dylib. The latter doesn't work anyway as we use hidden visibility for
         // compiler-builtins. Most linkers silently ignore it, but ld64 gives a warning.
         if info.level.is_below_threshold(export_threshold) && !tcx.is_compiler_builtins(cnum) {
             symbols.push((
@@ -1838,6 +1841,14 @@ fn exported_symbols_for_non_proc_macro(
         }
     });
 
+    // Mark allocator shim symbols as exported only if they were generated.
+    if export_threshold == SymbolExportLevel::Rust
+        && needs_allocator_shim_for_linking(tcx.dependency_formats(()), crate_type)
+        && tcx.allocator_kind(()).is_some()
+    {
+        symbols.extend(allocator_shim_symbols(tcx));
+    }
+
     symbols
 }
 
diff --git a/compiler/rustc_codegen_ssa/src/back/lto.rs b/compiler/rustc_codegen_ssa/src/back/lto.rs
index c95038375a1..f4a9037940a 100644
--- a/compiler/rustc_codegen_ssa/src/back/lto.rs
+++ b/compiler/rustc_codegen_ssa/src/back/lto.rs
@@ -1,6 +1,7 @@
 use std::ffi::CString;
 use std::sync::Arc;
 
+use rustc_ast::expand::allocator::AllocatorKind;
 use rustc_data_structures::memmap::Mmap;
 use rustc_hir::def_id::{CrateNum, LOCAL_CRATE};
 use rustc_middle::middle::exported_symbols::{ExportedSymbol, SymbolExportInfo, SymbolExportLevel};
@@ -8,8 +9,9 @@ use rustc_middle::ty::TyCtxt;
 use rustc_session::config::{CrateType, Lto};
 use tracing::info;
 
-use crate::back::symbol_export::{self, symbol_name_for_instance_in_crate};
+use crate::back::symbol_export::{self, allocator_shim_symbols, symbol_name_for_instance_in_crate};
 use crate::back::write::CodegenContext;
+use crate::base::allocator_kind_for_codegen;
 use crate::errors::{DynamicLinkingWithLTO, LtoDisallowed, LtoDylib, LtoProcMacro};
 use crate::traits::*;
 
@@ -94,6 +96,19 @@ pub(super) fn exported_symbols_for_lto(
             .filter_map(|&(s, info): &(ExportedSymbol<'_>, SymbolExportInfo)| {
                 if info.level.is_below_threshold(export_threshold) || info.used {
                     Some(symbol_name_for_instance_in_crate(tcx, s, cnum))
+                } else if export_threshold == SymbolExportLevel::C
+                    && info.rustc_std_internal_symbol
+                    && let Some(AllocatorKind::Default) = allocator_kind_for_codegen(tcx)
+                {
+                    // Export the __rdl_* exports for usage by the allocator shim when not using
+                    // #[global_allocator]. Most of the conditions above are only used to avoid
+                    // unnecessary expensive symbol_name_for_instance_in_crate calls.
+                    let sym = symbol_name_for_instance_in_crate(tcx, s, cnum);
+                    if sym.contains("__rdl_") || sym.contains("__rg_oom") {
+                        Some(sym)
+                    } else {
+                        None
+                    }
                 } else {
                     None
                 }
@@ -115,6 +130,11 @@ pub(super) fn exported_symbols_for_lto(
         }
     }
 
+    // Mark allocator shim symbols as exported only if they were generated.
+    if export_threshold == SymbolExportLevel::Rust && allocator_kind_for_codegen(tcx).is_some() {
+        symbols_below_threshold.extend(allocator_shim_symbols(tcx).map(|(name, _kind)| name));
+    }
+
     symbols_below_threshold
 }
 
diff --git a/compiler/rustc_codegen_ssa/src/back/symbol_export.rs b/compiler/rustc_codegen_ssa/src/back/symbol_export.rs
index d8a1480e911..b49e67217fb 100644
--- a/compiler/rustc_codegen_ssa/src/back/symbol_export.rs
+++ b/compiler/rustc_codegen_ssa/src/back/symbol_export.rs
@@ -18,7 +18,7 @@ use rustc_symbol_mangling::mangle_internal_symbol;
 use rustc_target::spec::TlsModel;
 use tracing::debug;
 
-use crate::base::allocator_kind_for_codegen;
+use crate::back::symbol_export;
 
 fn threshold(tcx: TyCtxt<'_>) -> SymbolExportLevel {
     crates_export_threshold(tcx.crate_types())
@@ -217,31 +217,6 @@ fn exported_non_generic_symbols_provider_local<'tcx>(
         ));
     }
 
-    // Mark allocator shim symbols as exported only if they were generated.
-    if allocator_kind_for_codegen(tcx).is_some() {
-        for symbol_name in ALLOCATOR_METHODS
-            .iter()
-            .map(|method| mangle_internal_symbol(tcx, global_fn_name(method.name).as_str()))
-            .chain([
-                mangle_internal_symbol(tcx, "__rust_alloc_error_handler"),
-                mangle_internal_symbol(tcx, OomStrategy::SYMBOL),
-                mangle_internal_symbol(tcx, NO_ALLOC_SHIM_IS_UNSTABLE),
-            ])
-        {
-            let exported_symbol = ExportedSymbol::NoDefId(SymbolName::new(tcx, &symbol_name));
-
-            symbols.push((
-                exported_symbol,
-                SymbolExportInfo {
-                    level: SymbolExportLevel::Rust,
-                    kind: SymbolExportKind::Text,
-                    used: false,
-                    rustc_std_internal_symbol: true,
-                },
-            ));
-        }
-    }
-
     // Sort so we get a stable incr. comp. hash.
     symbols.sort_by_cached_key(|s| s.0.symbol_name_for_local_instance(tcx));
 
@@ -516,6 +491,31 @@ pub(crate) fn provide(providers: &mut Providers) {
         upstream_monomorphizations_for_provider;
 }
 
+pub(crate) fn allocator_shim_symbols(
+    tcx: TyCtxt<'_>,
+) -> impl Iterator<Item = (String, SymbolExportKind)> {
+    ALLOCATOR_METHODS
+        .iter()
+        .map(move |method| mangle_internal_symbol(tcx, global_fn_name(method.name).as_str()))
+        .chain([
+            mangle_internal_symbol(tcx, "__rust_alloc_error_handler"),
+            mangle_internal_symbol(tcx, OomStrategy::SYMBOL),
+            mangle_internal_symbol(tcx, NO_ALLOC_SHIM_IS_UNSTABLE),
+        ])
+        .map(move |symbol_name| {
+            let exported_symbol = ExportedSymbol::NoDefId(SymbolName::new(tcx, &symbol_name));
+
+            (
+                symbol_export::exporting_symbol_name_for_instance_in_crate(
+                    tcx,
+                    exported_symbol,
+                    LOCAL_CRATE,
+                ),
+                SymbolExportKind::Text,
+            )
+        })
+}
+
 fn symbol_export_level(tcx: TyCtxt<'_>, sym_def_id: DefId) -> SymbolExportLevel {
     // We export anything that's not mangled at the "C" layer as it probably has
     // to do with ABI concerns. We do not, however, apply such treatment to
diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs
index 8586615f7c7..f637e7f58db 100644
--- a/compiler/rustc_codegen_ssa/src/back/write.rs
+++ b/compiler/rustc_codegen_ssa/src/back/write.rs
@@ -333,8 +333,7 @@ pub struct CodegenContext<B: WriteBackendMethods> {
     pub crate_types: Vec<CrateType>,
     pub output_filenames: Arc<OutputFilenames>,
     pub invocation_temp: Option<String>,
-    pub regular_module_config: Arc<ModuleConfig>,
-    pub allocator_module_config: Arc<ModuleConfig>,
+    pub module_config: Arc<ModuleConfig>,
     pub tm_factory: TargetMachineFactoryFn<B>,
     pub msvc_imps_needed: bool,
     pub is_pe_coff: bool,
@@ -372,13 +371,6 @@ impl<B: WriteBackendMethods> CodegenContext<B> {
     pub fn create_dcx(&self) -> DiagCtxt {
         DiagCtxt::new(Box::new(self.diag_emitter.clone()))
     }
-
-    pub fn config(&self, kind: ModuleKind) -> &ModuleConfig {
-        match kind {
-            ModuleKind::Regular => &self.regular_module_config,
-            ModuleKind::Allocator => &self.allocator_module_config,
-        }
-    }
 }
 
 fn generate_thin_lto_work<B: ExtraBackendMethods>(
@@ -442,6 +434,7 @@ pub(crate) fn start_async_codegen<B: ExtraBackendMethods>(
     backend: B,
     tcx: TyCtxt<'_>,
     target_cpu: String,
+    allocator_module: Option<ModuleCodegen<B::Module>>,
 ) -> OngoingCodegen<B> {
     let (coordinator_send, coordinator_receive) = channel();
 
@@ -465,6 +458,7 @@ pub(crate) fn start_async_codegen<B: ExtraBackendMethods>(
         coordinator_receive,
         Arc::new(regular_config),
         Arc::new(allocator_config),
+        allocator_module,
         coordinator_send.clone(),
     );
 
@@ -495,7 +489,7 @@ fn copy_all_cgu_workproducts_to_incr_comp_cache_dir(
 
     let _timer = sess.timer("copy_all_cgu_workproducts_to_incr_comp_cache_dir");
 
-    for module in compiled_modules.modules.iter().filter(|m| m.kind == ModuleKind::Regular) {
+    for module in &compiled_modules.modules {
         let mut files = Vec::new();
         if let Some(object_file_path) = &module.object {
             files.push((OutputType::Object.extension(), object_file_path.as_path()));
@@ -720,15 +714,6 @@ pub(crate) enum WorkItem<B: WriteBackendMethods> {
 }
 
 impl<B: WriteBackendMethods> WorkItem<B> {
-    fn module_kind(&self) -> ModuleKind {
-        match *self {
-            WorkItem::Optimize(ref m) => m.kind,
-            WorkItem::CopyPostLtoArtifacts(_) | WorkItem::FatLto { .. } | WorkItem::ThinLto(_) => {
-                ModuleKind::Regular
-            }
-        }
-    }
-
     /// Generate a short description of this work item suitable for use as a thread name.
     fn short_description(&self) -> String {
         // `pthread_setname()` on *nix ignores anything beyond the first 15
@@ -809,19 +794,12 @@ pub(crate) fn compute_per_cgu_lto_type(
     sess_lto: &Lto,
     opts: &config::Options,
     sess_crate_types: &[CrateType],
-    module_kind: ModuleKind,
 ) -> ComputedLtoType {
     // If the linker does LTO, we don't have to do it. Note that we
     // keep doing full LTO, if it is requested, as not to break the
     // assumption that the output will be a single module.
     let linker_does_lto = opts.cg.linker_plugin_lto.enabled();
 
-    // When we're automatically doing ThinLTO for multi-codegen-unit
-    // builds we don't actually want to LTO the allocator modules if
-    // it shows up. This is due to various linker shenanigans that
-    // we'll encounter later.
-    let is_allocator = module_kind == ModuleKind::Allocator;
-
     // We ignore a request for full crate graph LTO if the crate type
     // is only an rlib, as there is no full crate graph to process,
     // that'll happen later.
@@ -833,7 +811,7 @@ pub(crate) fn compute_per_cgu_lto_type(
     let is_rlib = matches!(sess_crate_types, [CrateType::Rlib]);
 
     match sess_lto {
-        Lto::ThinLocal if !linker_does_lto && !is_allocator => ComputedLtoType::Thin,
+        Lto::ThinLocal if !linker_does_lto => ComputedLtoType::Thin,
         Lto::Thin if !linker_does_lto && !is_rlib => ComputedLtoType::Thin,
         Lto::Fat if !is_rlib => ComputedLtoType::Fat,
         _ => ComputedLtoType::No,
@@ -843,23 +821,22 @@ pub(crate) fn compute_per_cgu_lto_type(
 fn execute_optimize_work_item<B: ExtraBackendMethods>(
     cgcx: &CodegenContext<B>,
     mut module: ModuleCodegen<B::Module>,
-    module_config: &ModuleConfig,
 ) -> WorkItemResult<B> {
     let dcx = cgcx.create_dcx();
     let dcx = dcx.handle();
 
-    B::optimize(cgcx, dcx, &mut module, module_config);
+    B::optimize(cgcx, dcx, &mut module, &cgcx.module_config);
 
     // After we've done the initial round of optimizations we need to
     // decide whether to synchronously codegen this module or ship it
     // back to the coordinator thread for further LTO processing (which
     // has to wait for all the initial modules to be optimized).
 
-    let lto_type = compute_per_cgu_lto_type(&cgcx.lto, &cgcx.opts, &cgcx.crate_types, module.kind);
+    let lto_type = compute_per_cgu_lto_type(&cgcx.lto, &cgcx.opts, &cgcx.crate_types);
 
     // If we're doing some form of incremental LTO then we need to be sure to
     // save our module to disk first.
-    let bitcode = if cgcx.config(module.kind).emit_pre_lto_bc {
+    let bitcode = if cgcx.module_config.emit_pre_lto_bc {
         let filename = pre_lto_bitcode_filename(&module.name);
         cgcx.incr_comp_session_dir.as_ref().map(|path| path.join(&filename))
     } else {
@@ -868,7 +845,7 @@ fn execute_optimize_work_item<B: ExtraBackendMethods>(
 
     match lto_type {
         ComputedLtoType::No => {
-            let module = B::codegen(cgcx, module, module_config);
+            let module = B::codegen(cgcx, module, &cgcx.module_config);
             WorkItemResult::Finished(module)
         }
         ComputedLtoType::Thin => {
@@ -899,7 +876,6 @@ fn execute_optimize_work_item<B: ExtraBackendMethods>(
 fn execute_copy_from_cache_work_item<B: ExtraBackendMethods>(
     cgcx: &CodegenContext<B>,
     module: CachedModuleCodegen,
-    module_config: &ModuleConfig,
 ) -> WorkItemResult<B> {
     let incr_comp_session_dir = cgcx.incr_comp_session_dir.as_ref().unwrap();
 
@@ -959,6 +935,7 @@ fn execute_copy_from_cache_work_item<B: ExtraBackendMethods>(
         }
     };
 
+    let module_config = &cgcx.module_config;
     let should_emit_obj = module_config.emit_obj != EmitObj::None;
     let assembly = load_from_incr_cache(module_config.emit_asm, OutputType::Assembly);
     let llvm_ir = load_from_incr_cache(module_config.emit_ir, OutputType::LlvmAssembly);
@@ -971,7 +948,6 @@ fn execute_copy_from_cache_work_item<B: ExtraBackendMethods>(
     WorkItemResult::Finished(CompiledModule {
         links_from_incr_cache,
         name: module.name,
-        kind: ModuleKind::Regular,
         object,
         dwarf_object,
         bytecode,
@@ -986,7 +962,6 @@ fn execute_fat_lto_work_item<B: ExtraBackendMethods>(
     each_linked_rlib_for_lto: &[PathBuf],
     mut needs_fat_lto: Vec<FatLtoInput<B>>,
     import_only_modules: Vec<(SerializedModule<B::ModuleBuffer>, WorkProduct)>,
-    module_config: &ModuleConfig,
 ) -> WorkItemResult<B> {
     for (module, wp) in import_only_modules {
         needs_fat_lto.push(FatLtoInput::Serialized { name: wp.cgu_name, buffer: module })
@@ -998,17 +973,16 @@ fn execute_fat_lto_work_item<B: ExtraBackendMethods>(
         each_linked_rlib_for_lto,
         needs_fat_lto,
     );
-    let module = B::codegen(cgcx, module, module_config);
+    let module = B::codegen(cgcx, module, &cgcx.module_config);
     WorkItemResult::Finished(module)
 }
 
 fn execute_thin_lto_work_item<B: ExtraBackendMethods>(
     cgcx: &CodegenContext<B>,
     module: lto::ThinModule<B>,
-    module_config: &ModuleConfig,
 ) -> WorkItemResult<B> {
     let module = B::optimize_thin(cgcx, module);
-    let module = B::codegen(cgcx, module, module_config);
+    let module = B::codegen(cgcx, module, &cgcx.module_config);
     WorkItemResult::Finished(module)
 }
 
@@ -1093,6 +1067,7 @@ fn start_executing_work<B: ExtraBackendMethods>(
     coordinator_receive: Receiver<Message<B>>,
     regular_config: Arc<ModuleConfig>,
     allocator_config: Arc<ModuleConfig>,
+    allocator_module: Option<ModuleCodegen<B::Module>>,
     tx_to_llvm_workers: Sender<Message<B>>,
 ) -> thread::JoinHandle<Result<CompiledModules, ()>> {
     let coordinator_send = tx_to_llvm_workers;
@@ -1157,8 +1132,7 @@ fn start_executing_work<B: ExtraBackendMethods>(
         expanded_args: tcx.sess.expanded_args.clone(),
         diag_emitter: shared_emitter.clone(),
         output_filenames: Arc::clone(tcx.output_filenames(())),
-        regular_module_config: regular_config,
-        allocator_module_config: allocator_config,
+        module_config: regular_config,
         tm_factory: backend.target_machine_factory(tcx.sess, ol, backend_features),
         msvc_imps_needed: msvc_imps_needed(tcx),
         is_pe_coff: tcx.sess.target.is_like_windows,
@@ -1173,6 +1147,11 @@ fn start_executing_work<B: ExtraBackendMethods>(
         invocation_temp: sess.invocation_temp.clone(),
     };
 
+    let compiled_allocator_module = allocator_module.map(|mut allocator_module| {
+        B::optimize(&cgcx, tcx.sess.dcx(), &mut allocator_module, &allocator_config);
+        B::codegen(&cgcx, allocator_module, &allocator_config)
+    });
+
     // This is the "main loop" of parallel work happening for parallel codegen.
     // It's here that we manage parallelism, schedule work, and work with
     // messages coming from clients.
@@ -1312,7 +1291,6 @@ fn start_executing_work<B: ExtraBackendMethods>(
         // This is where we collect codegen units that have gone all the way
         // through codegen and LLVM.
         let mut compiled_modules = vec![];
-        let mut compiled_allocator_module = None;
         let mut needs_fat_lto = Vec::new();
         let mut needs_thin_lto = Vec::new();
         let mut lto_import_only_modules = Vec::new();
@@ -1586,15 +1564,7 @@ fn start_executing_work<B: ExtraBackendMethods>(
 
                     match result {
                         Ok(WorkItemResult::Finished(compiled_module)) => {
-                            match compiled_module.kind {
-                                ModuleKind::Regular => {
-                                    compiled_modules.push(compiled_module);
-                                }
-                                ModuleKind::Allocator => {
-                                    assert!(compiled_allocator_module.is_none());
-                                    compiled_allocator_module = Some(compiled_module);
-                                }
-                            }
+                            compiled_modules.push(compiled_module);
                         }
                         Ok(WorkItemResult::NeedsFatLto(fat_lto_input)) => {
                             assert!(!started_lto);
@@ -1722,45 +1692,38 @@ fn spawn_work<'a, B: ExtraBackendMethods>(
     let cgcx = cgcx.clone();
 
     B::spawn_named_thread(cgcx.time_trace, work.short_description(), move || {
-        let result = std::panic::catch_unwind(AssertUnwindSafe(|| {
-            let module_config = cgcx.config(work.module_kind());
-
-            match work {
-                WorkItem::Optimize(m) => {
-                    let _timer =
-                        cgcx.prof.generic_activity_with_arg("codegen_module_optimize", &*m.name);
-                    execute_optimize_work_item(&cgcx, m, module_config)
-                }
-                WorkItem::CopyPostLtoArtifacts(m) => {
-                    let _timer = cgcx.prof.generic_activity_with_arg(
-                        "codegen_copy_artifacts_from_incr_cache",
-                        &*m.name,
-                    );
-                    execute_copy_from_cache_work_item(&cgcx, m, module_config)
-                }
-                WorkItem::FatLto {
-                    exported_symbols_for_lto,
-                    each_linked_rlib_for_lto,
+        let result = std::panic::catch_unwind(AssertUnwindSafe(|| match work {
+            WorkItem::Optimize(m) => {
+                let _timer =
+                    cgcx.prof.generic_activity_with_arg("codegen_module_optimize", &*m.name);
+                execute_optimize_work_item(&cgcx, m)
+            }
+            WorkItem::CopyPostLtoArtifacts(m) => {
+                let _timer = cgcx
+                    .prof
+                    .generic_activity_with_arg("codegen_copy_artifacts_from_incr_cache", &*m.name);
+                execute_copy_from_cache_work_item(&cgcx, m)
+            }
+            WorkItem::FatLto {
+                exported_symbols_for_lto,
+                each_linked_rlib_for_lto,
+                needs_fat_lto,
+                import_only_modules,
+            } => {
+                let _timer =
+                    cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", "everything");
+                execute_fat_lto_work_item(
+                    &cgcx,
+                    &exported_symbols_for_lto,
+                    &each_linked_rlib_for_lto,
                     needs_fat_lto,
                     import_only_modules,
-                } => {
-                    let _timer = cgcx
-                        .prof
-                        .generic_activity_with_arg("codegen_module_perform_lto", "everything");
-                    execute_fat_lto_work_item(
-                        &cgcx,
-                        &exported_symbols_for_lto,
-                        &each_linked_rlib_for_lto,
-                        needs_fat_lto,
-                        import_only_modules,
-                        module_config,
-                    )
-                }
-                WorkItem::ThinLto(m) => {
-                    let _timer =
-                        cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", m.name());
-                    execute_thin_lto_work_item(&cgcx, m, module_config)
-                }
+                )
+            }
+            WorkItem::ThinLto(m) => {
+                let _timer =
+                    cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", m.name());
+                execute_thin_lto_work_item(&cgcx, m)
             }
         }));
 
diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs
index 8abaf201aba..a9a2ae1b3db 100644
--- a/compiler/rustc_codegen_ssa/src/base.rs
+++ b/compiler/rustc_codegen_ssa/src/base.rs
@@ -17,6 +17,7 @@ use rustc_hir::lang_items::LangItem;
 use rustc_hir::{ItemId, Target};
 use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrs;
 use rustc_middle::middle::debugger_visualizer::{DebuggerVisualizerFile, DebuggerVisualizerType};
+use rustc_middle::middle::dependency_format::Dependencies;
 use rustc_middle::middle::exported_symbols::{self, SymbolExportKind};
 use rustc_middle::middle::lang_items;
 use rustc_middle::mir::BinOp;
@@ -45,9 +46,7 @@ use crate::meth::load_vtable;
 use crate::mir::operand::OperandValue;
 use crate::mir::place::PlaceRef;
 use crate::traits::*;
-use crate::{
-    CachedModuleCodegen, CodegenLintLevels, CrateInfo, ModuleCodegen, ModuleKind, errors, meth, mir,
-};
+use crate::{CachedModuleCodegen, CodegenLintLevels, CrateInfo, ModuleCodegen, errors, meth, mir};
 
 pub(crate) fn bin_op_to_icmp_predicate(op: BinOp, signed: bool) -> IntPredicate {
     match (op, signed) {
@@ -630,14 +629,30 @@ pub fn allocator_kind_for_codegen(tcx: TyCtxt<'_>) -> Option<AllocatorKind> {
     // If the crate doesn't have an `allocator_kind` set then there's definitely
     // no shim to generate. Otherwise we also check our dependency graph for all
     // our output crate types. If anything there looks like its a `Dynamic`
-    // linkage, then it's already got an allocator shim and we'll be using that
-    // one instead. If nothing exists then it's our job to generate the
-    // allocator!
-    let any_dynamic_crate = tcx.dependency_formats(()).iter().any(|(_, list)| {
+    // linkage for all crate types we may link as, then it's already got an
+    // allocator shim and we'll be using that one instead. If nothing exists
+    // then it's our job to generate the allocator! If crate types disagree
+    // about whether an allocator shim is necessary or not, we generate one
+    // and let needs_allocator_shim_for_linking decide at link time whether or
+    // not to use it for any particular linker invocation.
+    let all_crate_types_any_dynamic_crate = tcx.dependency_formats(()).iter().all(|(_, list)| {
         use rustc_middle::middle::dependency_format::Linkage;
         list.iter().any(|&linkage| linkage == Linkage::Dynamic)
     });
-    if any_dynamic_crate { None } else { tcx.allocator_kind(()) }
+    if all_crate_types_any_dynamic_crate { None } else { tcx.allocator_kind(()) }
+}
+
+/// Decide if this particular crate type needs an allocator shim linked in.
+/// This may return true even when allocator_kind_for_codegen returns false. In
+/// this case no allocator shim shall be linked.
+pub(crate) fn needs_allocator_shim_for_linking(
+    dependency_formats: &Dependencies,
+    crate_type: CrateType,
+) -> bool {
+    use rustc_middle::middle::dependency_format::Linkage;
+    let any_dynamic_crate =
+        dependency_formats[&crate_type].iter().any(|&linkage| linkage == Linkage::Dynamic);
+    !any_dynamic_crate
 }
 
 pub fn codegen_crate<B: ExtraBackendMethods>(
@@ -647,7 +662,7 @@ pub fn codegen_crate<B: ExtraBackendMethods>(
 ) -> OngoingCodegen<B> {
     // Skip crate items and just output metadata in -Z no-codegen mode.
     if tcx.sess.opts.unstable_opts.no_codegen || !tcx.sess.opts.output_types.should_codegen() {
-        let ongoing_codegen = start_async_codegen(backend, tcx, target_cpu);
+        let ongoing_codegen = start_async_codegen(backend, tcx, target_cpu, None);
 
         ongoing_codegen.codegen_finished(tcx);
 
@@ -678,7 +693,27 @@ pub fn codegen_crate<B: ExtraBackendMethods>(
         }
     }
 
-    let ongoing_codegen = start_async_codegen(backend.clone(), tcx, target_cpu);
+    // Codegen an allocator shim, if necessary.
+    let allocator_module = if let Some(kind) = allocator_kind_for_codegen(tcx) {
+        let llmod_id =
+            cgu_name_builder.build_cgu_name(LOCAL_CRATE, &["crate"], Some("allocator")).to_string();
+
+        tcx.sess.time("write_allocator_module", || {
+            let module = backend.codegen_allocator(
+                tcx,
+                &llmod_id,
+                kind,
+                // If allocator_kind is Some then alloc_error_handler_kind must
+                // also be Some.
+                tcx.alloc_error_handler_kind(()).unwrap(),
+            );
+            Some(ModuleCodegen::new_allocator(llmod_id, module))
+        })
+    } else {
+        None
+    };
+
+    let ongoing_codegen = start_async_codegen(backend.clone(), tcx, target_cpu, allocator_module);
 
     // For better throughput during parallel processing by LLVM, we used to sort
     // CGUs largest to smallest. This would lead to better thread utilization
@@ -795,35 +830,6 @@ pub fn codegen_crate<B: ExtraBackendMethods>(
         }
     }
 
-    // Codegen an allocator shim, if necessary.
-    // Do this last to ensure the LLVM_passes timer doesn't start while no CGUs have been codegened
-    // yet for the backend to optimize.
-    if let Some(kind) = allocator_kind_for_codegen(tcx) {
-        let llmod_id =
-            cgu_name_builder.build_cgu_name(LOCAL_CRATE, &["crate"], Some("allocator")).to_string();
-        let module_llvm = tcx.sess.time("write_allocator_module", || {
-            backend.codegen_allocator(
-                tcx,
-                &llmod_id,
-                kind,
-                // If allocator_kind is Some then alloc_error_handler_kind must
-                // also be Some.
-                tcx.alloc_error_handler_kind(()).unwrap(),
-            )
-        });
-
-        ongoing_codegen.wait_for_signal_to_codegen_item();
-        ongoing_codegen.check_for_errors(tcx.sess);
-
-        // These modules are generally cheap and won't throw off scheduling.
-        let cost = 0;
-        submit_codegened_module_to_llvm(
-            &ongoing_codegen.coordinator,
-            ModuleCodegen::new_allocator(llmod_id, module_llvm),
-            cost,
-        );
-    }
-
     ongoing_codegen.codegen_finished(tcx);
 
     // Since the main thread is sometimes blocked during codegen, we keep track
@@ -1118,12 +1124,7 @@ pub fn determine_cgu_reuse<'tcx>(tcx: TyCtxt<'tcx>, cgu: &CodegenUnit<'tcx>) ->
         // We can re-use either the pre- or the post-thinlto state. If no LTO is
         // being performed then we can use post-LTO artifacts, otherwise we must
         // reuse pre-LTO artifacts
-        match compute_per_cgu_lto_type(
-            &tcx.sess.lto(),
-            &tcx.sess.opts,
-            tcx.crate_types(),
-            ModuleKind::Regular,
-        ) {
+        match compute_per_cgu_lto_type(&tcx.sess.lto(), &tcx.sess.opts, tcx.crate_types()) {
             ComputedLtoType::No => CguReuse::PostLto,
             _ => CguReuse::PreLto,
         }
diff --git a/compiler/rustc_codegen_ssa/src/codegen_attrs.rs b/compiler/rustc_codegen_ssa/src/codegen_attrs.rs
index 961bb788149..008340e614d 100644
--- a/compiler/rustc_codegen_ssa/src/codegen_attrs.rs
+++ b/compiler/rustc_codegen_ssa/src/codegen_attrs.rs
@@ -428,9 +428,16 @@ fn check_result(
     // llvm/llvm-project#70563).
     if !codegen_fn_attrs.target_features.is_empty()
         && matches!(codegen_fn_attrs.inline, InlineAttr::Always)
+        && !tcx.features().target_feature_inline_always()
         && let Some(span) = interesting_spans.inline
     {
-        tcx.dcx().span_err(span, "cannot use `#[inline(always)]` with `#[target_feature]`");
+        feature_err(
+            tcx.sess,
+            sym::target_feature_inline_always,
+            span,
+            "cannot use `#[inline(always)]` with `#[target_feature]`",
+        )
+        .emit();
     }
 
     // warn that inline has no effect when no_sanitize is present
diff --git a/compiler/rustc_codegen_ssa/src/lib.rs b/compiler/rustc_codegen_ssa/src/lib.rs
index fe0500a5d4c..5b90ffaa056 100644
--- a/compiler/rustc_codegen_ssa/src/lib.rs
+++ b/compiler/rustc_codegen_ssa/src/lib.rs
@@ -120,7 +120,6 @@ impl<M> ModuleCodegen<M> {
 
         CompiledModule {
             name: self.name.clone(),
-            kind: self.kind,
             object,
             dwarf_object,
             bytecode,
@@ -134,7 +133,6 @@ impl<M> ModuleCodegen<M> {
 #[derive(Debug, Encodable, Decodable)]
 pub struct CompiledModule {
     pub name: String,
-    pub kind: ModuleKind,
     pub object: Option<PathBuf>,
     pub dwarf_object: Option<PathBuf>,
     pub bytecode: Option<PathBuf>,
diff --git a/compiler/rustc_const_eval/messages.ftl b/compiler/rustc_const_eval/messages.ftl
index 60518dafbf2..700d7c26752 100644
--- a/compiler/rustc_const_eval/messages.ftl
+++ b/compiler/rustc_const_eval/messages.ftl
@@ -457,7 +457,7 @@ const_eval_validation_failure =
     it is undefined behavior to use this value
 
 const_eval_validation_failure_note =
-    The rules on what exactly is undefined behavior aren't clear, so this check might be overzealous. Please open an issue on the rustc repository if you believe it should not be considered undefined behavior.
+    the rules on what exactly is undefined behavior aren't clear, so this check might be overzealous. Please open an issue on the rustc repository if you believe it should not be considered undefined behavior.
 
 const_eval_validation_front_matter_invalid_value = constructing invalid value
 const_eval_validation_front_matter_invalid_value_with_path = constructing invalid value at {$path}
diff --git a/compiler/rustc_driver_impl/messages.ftl b/compiler/rustc_driver_impl/messages.ftl
index 2c6a0291ac2..b62cdc35f51 100644
--- a/compiler/rustc_driver_impl/messages.ftl
+++ b/compiler/rustc_driver_impl/messages.ftl
@@ -14,7 +14,7 @@ driver_impl_ice_version = rustc {$version} running on {$triple}
 
 driver_impl_rlink_corrupt_file = corrupt metadata encountered in `{$file}`
 
-driver_impl_rlink_empty_version_number = The input does not contain version number
+driver_impl_rlink_empty_version_number = the input does not contain version number
 
 driver_impl_rlink_encoding_version_mismatch = .rlink file was produced with encoding version `{$version_array}`, but the current version is `{$rlink_version}`
 
@@ -24,6 +24,6 @@ driver_impl_rlink_rustc_version_mismatch = .rlink file was produced by rustc ver
 
 driver_impl_rlink_unable_to_read = failed to read rlink file: `{$err}`
 
-driver_impl_rlink_wrong_file_type = The input does not look like a .rlink file
+driver_impl_rlink_wrong_file_type = the input does not look like a .rlink file
 
 driver_impl_unstable_feature_usage = cannot dump feature usage metrics: {$error}
diff --git a/compiler/rustc_errors/src/lib.rs b/compiler/rustc_errors/src/lib.rs
index 71fc54f0d33..a56e0f3fae1 100644
--- a/compiler/rustc_errors/src/lib.rs
+++ b/compiler/rustc_errors/src/lib.rs
@@ -1160,7 +1160,7 @@ impl<'a> DiagCtxtHandle<'a> {
         // - It's only produce with JSON output.
         // - It's not emitted the usual way, via `emit_diagnostic`.
         // - The `$message_type` field is "unused_externs" rather than the usual
-        //   "diagnosic".
+        //   "diagnostic".
         //
         // We count it as a lint error because it has a lint level. The value
         // of `loud` (which comes from "unused-externs" or
diff --git a/compiler/rustc_expand/src/mbe/macro_parser.rs b/compiler/rustc_expand/src/mbe/macro_parser.rs
index 0324057e331..ab8e059b7b7 100644
--- a/compiler/rustc_expand/src/mbe/macro_parser.rs
+++ b/compiler/rustc_expand/src/mbe/macro_parser.rs
@@ -77,7 +77,7 @@ use std::rc::Rc;
 
 pub(crate) use NamedMatch::*;
 pub(crate) use ParseResult::*;
-use rustc_ast::token::{self, DocComment, NonterminalKind, Token};
+use rustc_ast::token::{self, DocComment, NonterminalKind, Token, TokenKind};
 use rustc_data_structures::fx::FxHashMap;
 use rustc_errors::ErrorGuaranteed;
 use rustc_lint_defs::pluralize;
@@ -397,7 +397,23 @@ fn token_name_eq(t1: &Token, t2: &Token) -> bool {
     {
         ident1.name == ident2.name && is_raw1 == is_raw2
     } else {
-        t1.kind == t2.kind
+        // Note: we SHOULD NOT use `t1.kind == t2.kind` here, and we should instead compare the
+        // tokens using the special comparison logic below.
+        // It makes sure that variants containing `InvisibleOrigin` will
+        // never compare equal to one another.
+        //
+        // When we had AST-based nonterminals we couldn't compare them, and the
+        // old `Nonterminal` type had an `eq` that always returned false,
+        // resulting in this restriction:
+        // <https://doc.rust-lang.org/nightly/reference/macros-by-example.html#forwarding-a-matched-fragment>
+        // This comparison logic emulates that behaviour. We could consider lifting this
+        // restriction now but there are still cases involving invisible
+        // delimiters that make it harder than it first appears.
+        match (t1.kind, t2.kind) {
+            (TokenKind::OpenInvisible(_) | TokenKind::CloseInvisible(_), _)
+            | (_, TokenKind::OpenInvisible(_) | TokenKind::CloseInvisible(_)) => false,
+            (a, b) => a == b,
+        }
     }
 }
 
diff --git a/compiler/rustc_feature/src/unstable.rs b/compiler/rustc_feature/src/unstable.rs
index 573e9aab055..4f35bf63a1a 100644
--- a/compiler/rustc_feature/src/unstable.rs
+++ b/compiler/rustc_feature/src/unstable.rs
@@ -642,6 +642,8 @@ declare_features! (
     (unstable, super_let, "1.88.0", Some(139076)),
     /// Allows subtrait items to shadow supertrait items.
     (unstable, supertrait_item_shadowing, "1.86.0", Some(89151)),
+    /// Allows the use of target_feature when a function is marked inline(always).
+    (unstable, target_feature_inline_always, "CURRENT_RUSTC_VERSION", Some(145574)),
     /// Allows using `#[thread_local]` on `static` items.
     (unstable, thread_local, "1.0.0", Some(29594)),
     /// Allows defining `trait X = A + B;` alias items.
diff --git a/compiler/rustc_hir_analysis/src/check/intrinsic.rs b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
index cfc6bc2f3a0..aa2d27ab809 100644
--- a/compiler/rustc_hir_analysis/src/check/intrinsic.rs
+++ b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
@@ -449,6 +449,9 @@ pub(crate) fn check_intrinsic_type(
         }
         sym::unchecked_shl | sym::unchecked_shr => (2, 0, vec![param(0), param(1)], param(0)),
         sym::rotate_left | sym::rotate_right => (1, 0, vec![param(0), tcx.types.u32], param(0)),
+        sym::unchecked_funnel_shl | sym::unchecked_funnel_shr => {
+            (1, 0, vec![param(0), param(0), tcx.types.u32], param(0))
+        }
         sym::unchecked_add | sym::unchecked_sub | sym::unchecked_mul => {
             (1, 0, vec![param(0), param(0)], param(0))
         }
diff --git a/compiler/rustc_hir_analysis/src/check/region.rs b/compiler/rustc_hir_analysis/src/check/region.rs
index f5770b7312d..2ba7ed46f92 100644
--- a/compiler/rustc_hir_analysis/src/check/region.rs
+++ b/compiler/rustc_hir_analysis/src/check/region.rs
@@ -490,12 +490,8 @@ fn resolve_local<'tcx>(
             //
             // Iterate up to the enclosing destruction scope to find the same scope that will also
             // be used for the result of the block itself.
-            while let Some(s) = visitor.cx.var_parent {
-                let parent = visitor.scope_tree.parent_map.get(&s).cloned();
-                if let Some(Scope { data: ScopeData::Destruction, .. }) = parent {
-                    break;
-                }
-                visitor.cx.var_parent = parent;
+            if let Some(inner_scope) = visitor.cx.var_parent {
+                (visitor.cx.var_parent, _) = visitor.scope_tree.default_temporary_scope(inner_scope)
             }
         }
     }
diff --git a/compiler/rustc_interface/src/interface.rs b/compiler/rustc_interface/src/interface.rs
index 8f131f45bbd..4c820b8877b 100644
--- a/compiler/rustc_interface/src/interface.rs
+++ b/compiler/rustc_interface/src/interface.rs
@@ -13,7 +13,7 @@ use rustc_lint::LintStore;
 use rustc_middle::ty;
 use rustc_middle::ty::CurrentGcx;
 use rustc_middle::util::Providers;
-use rustc_parse::new_parser_from_source_str;
+use rustc_parse::new_parser_from_simple_source_str;
 use rustc_parse::parser::attr::AllowLeadingUnsafe;
 use rustc_query_impl::QueryCtxt;
 use rustc_query_system::query::print_query_stack;
@@ -68,7 +68,7 @@ pub(crate) fn parse_cfg(dcx: DiagCtxtHandle<'_>, cfgs: Vec<String>) -> Cfg {
                 };
             }
 
-            match new_parser_from_source_str(&psess, filename, s.to_string()) {
+            match new_parser_from_simple_source_str(&psess, filename, s.to_string()) {
                 Ok(mut parser) => match parser.parse_meta_item(AllowLeadingUnsafe::No) {
                     Ok(meta_item) if parser.token == token::Eof => {
                         if meta_item.path.segments.len() != 1 {
@@ -166,7 +166,7 @@ pub(crate) fn parse_check_cfg(dcx: DiagCtxtHandle<'_>, specs: Vec<String>) -> Ch
             error!("expected `cfg(name, values(\"value1\", \"value2\", ... \"valueN\"))`")
         };
 
-        let mut parser = match new_parser_from_source_str(&psess, filename, s.to_string()) {
+        let mut parser = match new_parser_from_simple_source_str(&psess, filename, s.to_string()) {
             Ok(parser) => parser,
             Err(errs) => {
                 errs.into_iter().for_each(|err| err.cancel());
diff --git a/compiler/rustc_lint_defs/src/builtin.rs b/compiler/rustc_lint_defs/src/builtin.rs
index ce92263babd..b79075ac09b 100644
--- a/compiler/rustc_lint_defs/src/builtin.rs
+++ b/compiler/rustc_lint_defs/src/builtin.rs
@@ -5141,3 +5141,57 @@ declare_lint! {
     "detects tail calls of functions marked with `#[track_caller]`",
     @feature_gate = explicit_tail_calls;
 }
+declare_lint! {
+    /// The `inline_always_mismatching_target_features` lint will trigger when a
+    /// function with the `#[inline(always)]` and `#[target_feature(enable = "...")]`
+    /// attributes is called and cannot be inlined due to missing target features in the caller.
+    ///
+    /// ### Example
+    ///
+    /// ```rust,ignore (fails on x86_64)
+    /// #[inline(always)]
+    /// #[target_feature(enable = "fp16")]
+    /// unsafe fn callee() {
+    ///     // operations using fp16 types
+    /// }
+    ///
+    /// // Caller does not enable the required target feature
+    /// fn caller() {
+    ///     unsafe { callee(); }
+    /// }
+    ///
+    /// fn main() {
+    ///     caller();
+    /// }
+    /// ```
+    ///
+    /// This will produce:
+    ///
+    /// ```text
+    /// warning: call to `#[inline(always)]`-annotated `callee` requires the same target features. Function will not have `alwaysinline` attribute applied
+    ///   --> $DIR/builtin.rs:5192:14
+    ///    |
+    /// 10 |     unsafe { callee(); }
+    ///    |              ^^^^^^^^
+    ///    |
+    /// note: `fp16` target feature enabled in `callee` here but missing from `caller`
+    ///   --> $DIR/builtin.rs:5185:1
+    ///    |
+    /// 3  | #[target_feature(enable = "fp16")]
+    ///    | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+    /// 4  | unsafe fn callee() {
+    ///    | ------------------
+    ///    = note: `#[warn(inline_always_mismatching_target_features)]` on by default
+    /// warning: 1 warning emitted
+    /// ```
+    ///
+    /// ### Explanation
+    ///
+    /// Inlining a function with a target feature attribute into a caller that
+    /// lacks the corresponding target feature can lead to unsound behavior.
+    /// LLVM may select the wrong instructions or registers, or reorder
+    /// operations, potentially resulting in runtime errors.
+    pub INLINE_ALWAYS_MISMATCHING_TARGET_FEATURES,
+    Warn,
+    r#"detects when a function annotated with `#[inline(always)]` and `#[target_feature(enable = "..")]` is inlined into a caller without the required target feature"#,
+}
diff --git a/compiler/rustc_metadata/src/creader.rs b/compiler/rustc_metadata/src/creader.rs
index 5d776ea581d..9e23da88f5e 100644
--- a/compiler/rustc_metadata/src/creader.rs
+++ b/compiler/rustc_metadata/src/creader.rs
@@ -412,7 +412,7 @@ impl CStore {
             match (&left_name_val, &right_name_val) {
                 (Some(l), Some(r)) => match l.1.opt.cmp(&r.1.opt) {
                     cmp::Ordering::Equal => {
-                        if l.0.tech_value != r.0.tech_value {
+                        if !l.1.consistent(&tcx.sess.opts, Some(&r.1)) {
                             report_diff(
                                 &l.0.prefix,
                                 &l.0.name,
@@ -424,20 +424,28 @@ impl CStore {
                         right_name_val = None;
                     }
                     cmp::Ordering::Greater => {
-                        report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name));
+                        if !r.1.consistent(&tcx.sess.opts, None) {
+                            report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name));
+                        }
                         right_name_val = None;
                     }
                     cmp::Ordering::Less => {
-                        report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None);
+                        if !l.1.consistent(&tcx.sess.opts, None) {
+                            report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None);
+                        }
                         left_name_val = None;
                     }
                 },
                 (Some(l), None) => {
-                    report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None);
+                    if !l.1.consistent(&tcx.sess.opts, None) {
+                        report_diff(&l.0.prefix, &l.0.name, Some(&l.1.value_name), None);
+                    }
                     left_name_val = None;
                 }
                 (None, Some(r)) => {
-                    report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name));
+                    if !r.1.consistent(&tcx.sess.opts, None) {
+                        report_diff(&r.0.prefix, &r.0.name, None, Some(&r.1.value_name));
+                    }
                     right_name_val = None;
                 }
                 (None, None) => break,
diff --git a/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs b/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs
index 866736f74a0..8b4503073b0 100644
--- a/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs
+++ b/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs
@@ -82,7 +82,7 @@ pub enum TargetFeatureKind {
     Forced,
 }
 
-#[derive(Copy, Clone, Debug, TyEncodable, TyDecodable, HashStable)]
+#[derive(Copy, Clone, Debug, Eq, PartialEq, TyEncodable, TyDecodable, HashStable)]
 pub struct TargetFeature {
     /// The name of the target feature (e.g. "avx")
     pub name: Symbol,
diff --git a/compiler/rustc_middle/src/middle/region.rs b/compiler/rustc_middle/src/middle/region.rs
index 857d041224f..5367e5edd49 100644
--- a/compiler/rustc_middle/src/middle/region.rs
+++ b/compiler/rustc_middle/src/middle/region.rs
@@ -299,4 +299,43 @@ impl ScopeTree {
 
         true
     }
+
+    /// Returns the scope of non-lifetime-extended temporaries within a given scope, as well as
+    /// whether we've recorded a potential backwards-incompatible change to lint on.
+    /// Returns `None` when no enclosing temporary scope is found, such as for static items.
+    pub fn default_temporary_scope(&self, inner: Scope) -> (Option<Scope>, Option<Scope>) {
+        let mut id = inner;
+        let mut backwards_incompatible = None;
+
+        while let Some(&p) = self.parent_map.get(&id) {
+            match p.data {
+                ScopeData::Destruction => {
+                    debug!("temporary_scope({inner:?}) = {id:?} [enclosing]");
+                    return (Some(id), backwards_incompatible);
+                }
+                ScopeData::IfThenRescope | ScopeData::MatchGuard => {
+                    debug!("temporary_scope({inner:?}) = {p:?} [enclosing]");
+                    return (Some(p), backwards_incompatible);
+                }
+                ScopeData::Node
+                | ScopeData::CallSite
+                | ScopeData::Arguments
+                | ScopeData::IfThen
+                | ScopeData::Remainder(_) => {
+                    // If we haven't already passed through a backwards-incompatible node,
+                    // then check if we are passing through one now and record it if so.
+                    // This is for now only working for cases where a temporary lifetime is
+                    // *shortened*.
+                    if backwards_incompatible.is_none() {
+                        backwards_incompatible =
+                            self.backwards_incompatible_scope.get(&p.local_id).copied();
+                    }
+                    id = p
+                }
+            }
+        }
+
+        debug!("temporary_scope({inner:?}) = None");
+        (None, backwards_incompatible)
+    }
 }
diff --git a/compiler/rustc_middle/src/ty/context.rs b/compiler/rustc_middle/src/ty/context.rs
index 72ab6ac612c..97c8a0336c2 100644
--- a/compiler/rustc_middle/src/ty/context.rs
+++ b/compiler/rustc_middle/src/ty/context.rs
@@ -738,8 +738,8 @@ impl<'tcx> Interner for TyCtxt<'tcx> {
         )
     }
 
-    type ProbeRef = &'tcx inspect::Probe<TyCtxt<'tcx>>;
-    fn mk_probe_ref(self, probe: inspect::Probe<Self>) -> &'tcx inspect::Probe<TyCtxt<'tcx>> {
+    type Probe = &'tcx inspect::Probe<TyCtxt<'tcx>>;
+    fn mk_probe(self, probe: inspect::Probe<Self>) -> &'tcx inspect::Probe<TyCtxt<'tcx>> {
         self.arena.alloc(probe)
     }
     fn evaluate_root_goal_for_proof_tree_raw(
diff --git a/compiler/rustc_middle/src/ty/rvalue_scopes.rs b/compiler/rustc_middle/src/ty/rvalue_scopes.rs
index 7dfe2d28051..8b92e48ed1a 100644
--- a/compiler/rustc_middle/src/ty/rvalue_scopes.rs
+++ b/compiler/rustc_middle/src/ty/rvalue_scopes.rs
@@ -35,41 +35,8 @@ impl RvalueScopes {
         // if there's one. Static items, for instance, won't
         // have an enclosing scope, hence no scope will be
         // returned.
-        let mut id = Scope { local_id: expr_id, data: ScopeData::Node };
-        let mut backwards_incompatible = None;
-
-        while let Some(&p) = region_scope_tree.parent_map.get(&id) {
-            match p.data {
-                ScopeData::Destruction => {
-                    debug!("temporary_scope({expr_id:?}) = {id:?} [enclosing]");
-                    return (Some(id), backwards_incompatible);
-                }
-                ScopeData::IfThenRescope | ScopeData::MatchGuard => {
-                    debug!("temporary_scope({expr_id:?}) = {p:?} [enclosing]");
-                    return (Some(p), backwards_incompatible);
-                }
-                ScopeData::Node
-                | ScopeData::CallSite
-                | ScopeData::Arguments
-                | ScopeData::IfThen
-                | ScopeData::Remainder(_) => {
-                    // If we haven't already passed through a backwards-incompatible node,
-                    // then check if we are passing through one now and record it if so.
-                    // This is for now only working for cases where a temporary lifetime is
-                    // *shortened*.
-                    if backwards_incompatible.is_none() {
-                        backwards_incompatible = region_scope_tree
-                            .backwards_incompatible_scope
-                            .get(&p.local_id)
-                            .copied();
-                    }
-                    id = p
-                }
-            }
-        }
-
-        debug!("temporary_scope({expr_id:?}) = None");
-        (None, backwards_incompatible)
+        region_scope_tree
+            .default_temporary_scope(Scope { local_id: expr_id, data: ScopeData::Node })
     }
 
     /// Make an association between a sub-expression and an extended lifetime
diff --git a/compiler/rustc_mir_transform/src/check_inline_always_target_features.rs b/compiler/rustc_mir_transform/src/check_inline_always_target_features.rs
new file mode 100644
index 00000000000..abad28f0a8f
--- /dev/null
+++ b/compiler/rustc_mir_transform/src/check_inline_always_target_features.rs
@@ -0,0 +1,88 @@
+use rustc_hir::attrs::InlineAttr;
+use rustc_middle::middle::codegen_fn_attrs::{TargetFeature, TargetFeatureKind};
+use rustc_middle::mir::{Body, TerminatorKind};
+use rustc_middle::ty::{self, TyCtxt};
+
+use crate::pass_manager::MirLint;
+
+pub(super) struct CheckInlineAlwaysTargetFeature;
+
+impl<'tcx> MirLint<'tcx> for CheckInlineAlwaysTargetFeature {
+    fn run_lint(&self, tcx: TyCtxt<'tcx>, body: &Body<'tcx>) {
+        check_inline_always_target_features(tcx, body)
+    }
+}
+
+/// `#[target_feature]`-annotated functions can be marked `#[inline]` and will only be inlined if
+/// the target features match (as well as all of the other inlining heuristics). `#[inline(always)]`
+/// will always inline regardless of matching target features, which can result in errors from LLVM.
+/// However, it is desirable to be able to always annotate certain functions (e.g. SIMD intrinsics)
+/// as `#[inline(always)]` but check the target features match in Rust to avoid the LLVM errors.
+///
+/// We check the caller and callee target features to ensure that this can
+/// be done or emit a lint.
+#[inline]
+fn check_inline_always_target_features<'tcx>(tcx: TyCtxt<'tcx>, body: &Body<'tcx>) {
+    let caller_def_id = body.source.def_id().expect_local();
+    if !tcx.def_kind(caller_def_id).has_codegen_attrs() {
+        return;
+    }
+
+    let caller_codegen_fn_attrs = tcx.codegen_fn_attrs(caller_def_id);
+
+    for bb in body.basic_blocks.iter() {
+        let terminator = bb.terminator();
+        match &terminator.kind {
+            TerminatorKind::Call { func, .. } | TerminatorKind::TailCall { func, .. } => {
+                let fn_ty = func.ty(body, tcx);
+                let ty::FnDef(callee_def_id, _) = *fn_ty.kind() else {
+                    continue;
+                };
+
+                if !tcx.def_kind(callee_def_id).has_codegen_attrs() {
+                    continue;
+                }
+                let callee_codegen_fn_attrs = tcx.codegen_fn_attrs(callee_def_id);
+                if callee_codegen_fn_attrs.inline != InlineAttr::Always
+                    || callee_codegen_fn_attrs.target_features.is_empty()
+                {
+                    continue;
+                }
+
+                // Scan the users defined target features and ensure they
+                // match the caller.
+                if tcx.is_target_feature_call_safe(
+                    &callee_codegen_fn_attrs.target_features,
+                    &caller_codegen_fn_attrs
+                        .target_features
+                        .iter()
+                        .cloned()
+                        .chain(tcx.sess.target_features.iter().map(|feat| TargetFeature {
+                            name: *feat,
+                            kind: TargetFeatureKind::Implied,
+                        }))
+                        .collect::<Vec<_>>(),
+                ) {
+                    continue;
+                }
+
+                let callee_only: Vec<_> = callee_codegen_fn_attrs
+                    .target_features
+                    .iter()
+                    .filter(|it| !caller_codegen_fn_attrs.target_features.contains(it))
+                    .filter(|it| !matches!(it.kind, TargetFeatureKind::Implied))
+                    .map(|it| it.name.as_str())
+                    .collect();
+
+                crate::errors::emit_inline_always_target_feature_diagnostic(
+                    tcx,
+                    terminator.source_info.span,
+                    callee_def_id,
+                    caller_def_id.into(),
+                    &callee_only,
+                );
+            }
+            _ => (),
+        }
+    }
+}
diff --git a/compiler/rustc_mir_transform/src/errors.rs b/compiler/rustc_mir_transform/src/errors.rs
index ad9635aae33..775f5f9a7cf 100644
--- a/compiler/rustc_mir_transform/src/errors.rs
+++ b/compiler/rustc_mir_transform/src/errors.rs
@@ -2,6 +2,7 @@ use rustc_errors::codes::*;
 use rustc_errors::{Diag, LintDiagnostic};
 use rustc_macros::{Diagnostic, LintDiagnostic, Subdiagnostic};
 use rustc_middle::mir::AssertKind;
+use rustc_middle::query::Key;
 use rustc_middle::ty::TyCtxt;
 use rustc_session::lint::{self, Lint};
 use rustc_span::def_id::DefId;
@@ -9,6 +10,46 @@ use rustc_span::{Ident, Span, Symbol};
 
 use crate::fluent_generated as fluent;
 
+/// Emit diagnostic for calls to `#[inline(always)]`-annotated functions with a
+/// `#[target_feature]` attribute where the caller enables a different set of target features.
+pub(crate) fn emit_inline_always_target_feature_diagnostic<'a, 'tcx>(
+    tcx: TyCtxt<'tcx>,
+    call_span: Span,
+    callee_def_id: DefId,
+    caller_def_id: DefId,
+    callee_only: &[&'a str],
+) {
+    let callee = tcx.def_path_str(callee_def_id);
+    let caller = tcx.def_path_str(caller_def_id);
+
+    tcx.node_span_lint(
+        lint::builtin::INLINE_ALWAYS_MISMATCHING_TARGET_FEATURES,
+        tcx.local_def_id_to_hir_id(caller_def_id.as_local().unwrap()),
+        call_span,
+        |lint| {
+            lint.primary_message(format!(
+                "call to `#[inline(always)]`-annotated `{callee}` \
+                requires the same target features to be inlined"
+            ));
+            lint.note("function will not be inlined");
+
+            lint.note(format!(
+                "the following target features are on `{callee}` but missing from `{caller}`: {}",
+                callee_only.join(", ")
+            ));
+            lint.span_note(callee_def_id.default_span(tcx), format!("`{callee}` is defined here"));
+
+            let feats = callee_only.join(",");
+            lint.span_suggestion(
+                tcx.def_span(caller_def_id).shrink_to_lo(),
+                format!("add `#[target_feature]` attribute to `{caller}`"),
+                format!("#[target_feature(enable = \"{feats}\")]\n"),
+                lint::Applicability::MaybeIncorrect,
+            );
+        },
+    );
+}
+
 #[derive(LintDiagnostic)]
 #[diag(mir_transform_unconditional_recursion)]
 #[help]
diff --git a/compiler/rustc_mir_transform/src/lib.rs b/compiler/rustc_mir_transform/src/lib.rs
index 8f319e64916..1663dfa744f 100644
--- a/compiler/rustc_mir_transform/src/lib.rs
+++ b/compiler/rustc_mir_transform/src/lib.rs
@@ -117,6 +117,7 @@ declare_passes! {
     mod add_subtyping_projections : Subtyper;
     mod check_inline : CheckForceInline;
     mod check_call_recursion : CheckCallRecursion, CheckDropRecursion;
+    mod check_inline_always_target_features: CheckInlineAlwaysTargetFeature;
     mod check_alignment : CheckAlignment;
     mod check_enums : CheckEnums;
     mod check_const_item_mutation : CheckConstItemMutation;
@@ -384,6 +385,9 @@ fn mir_built(tcx: TyCtxt<'_>, def: LocalDefId) -> &Steal<Body<'_>> {
             // MIR-level lints.
             &Lint(check_inline::CheckForceInline),
             &Lint(check_call_recursion::CheckCallRecursion),
+            // Check callee's target features match callers target features when
+            // using `#[inline(always)]`
+            &Lint(check_inline_always_target_features::CheckInlineAlwaysTargetFeature),
             &Lint(check_packed_ref::CheckPackedRef),
             &Lint(check_const_item_mutation::CheckConstItemMutation),
             &Lint(function_item_references::FunctionItemReferences),
diff --git a/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs b/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs
index 31106a74527..443aebbdb4d 100644
--- a/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs
+++ b/compiler/rustc_next_trait_solver/src/solve/eval_ctxt/mod.rs
@@ -1275,7 +1275,7 @@ pub fn evaluate_root_goal_for_proof_tree_raw_provider<
 >(
     cx: I,
     canonical_goal: CanonicalInput<I>,
-) -> (QueryResult<I>, I::ProbeRef) {
+) -> (QueryResult<I>, I::Probe) {
     let mut inspect = inspect::ProofTreeBuilder::new();
     let canonical_result = SearchGraph::<D>::evaluate_root_goal_for_proof_tree(
         cx,
@@ -1284,7 +1284,7 @@ pub fn evaluate_root_goal_for_proof_tree_raw_provider<
         &mut inspect,
     );
     let final_revision = inspect.unwrap();
-    (canonical_result, cx.mk_probe_ref(final_revision))
+    (canonical_result, cx.mk_probe(final_revision))
 }
 
 /// Evaluate a goal to build a proof tree.
diff --git a/compiler/rustc_parse/messages.ftl b/compiler/rustc_parse/messages.ftl
index 4ca2f57bd87..77dd313d9b8 100644
--- a/compiler/rustc_parse/messages.ftl
+++ b/compiler/rustc_parse/messages.ftl
@@ -58,7 +58,7 @@ parse_async_use_order_incorrect = the order of `use` and `async` is incorrect
 parse_at_dot_dot_in_struct_pattern = `@ ..` is not supported in struct patterns
     .suggestion = bind to each field separately or, if you don't need them, just remove `{$ident} @`
 
-parse_at_in_struct_pattern = Unexpected `@` in struct pattern
+parse_at_in_struct_pattern = unexpected `@` in struct pattern
     .note = struct patterns use `field: pattern` syntax to bind to fields
     .help = consider replacing `new_name @ field_name` with `field_name: new_name` if that is what you intended
 
diff --git a/compiler/rustc_parse/src/lib.rs b/compiler/rustc_parse/src/lib.rs
index 197333d942d..b790966acfd 100644
--- a/compiler/rustc_parse/src/lib.rs
+++ b/compiler/rustc_parse/src/lib.rs
@@ -62,7 +62,20 @@ pub fn new_parser_from_source_str(
     source: String,
 ) -> Result<Parser<'_>, Vec<Diag<'_>>> {
     let source_file = psess.source_map().new_source_file(name, source);
-    new_parser_from_source_file(psess, source_file)
+    new_parser_from_source_file(psess, source_file, FrontmatterAllowed::Yes)
+}
+
+/// Creates a new parser from a simple (no frontmatter) source string.
+///
+/// On failure, the errors must be consumed via `unwrap_or_emit_fatal`, `emit`, `cancel`,
+/// etc., otherwise a panic will occur when they are dropped.
+pub fn new_parser_from_simple_source_str(
+    psess: &ParseSess,
+    name: FileName,
+    source: String,
+) -> Result<Parser<'_>, Vec<Diag<'_>>> {
+    let source_file = psess.source_map().new_source_file(name, source);
+    new_parser_from_source_file(psess, source_file, FrontmatterAllowed::No)
 }
 
 /// Creates a new parser from a filename. On failure, the errors must be consumed via
@@ -96,7 +109,7 @@ pub fn new_parser_from_file<'a>(
         }
         err.emit();
     });
-    new_parser_from_source_file(psess, source_file)
+    new_parser_from_source_file(psess, source_file, FrontmatterAllowed::Yes)
 }
 
 pub fn utf8_error<E: EmissionGuarantee>(
@@ -147,9 +160,10 @@ pub fn utf8_error<E: EmissionGuarantee>(
 fn new_parser_from_source_file(
     psess: &ParseSess,
     source_file: Arc<SourceFile>,
+    frontmatter_allowed: FrontmatterAllowed,
 ) -> Result<Parser<'_>, Vec<Diag<'_>>> {
     let end_pos = source_file.end_position();
-    let stream = source_file_to_stream(psess, source_file, None, FrontmatterAllowed::Yes)?;
+    let stream = source_file_to_stream(psess, source_file, None, frontmatter_allowed)?;
     let mut parser = Parser::new(psess, stream, None);
     if parser.token == token::Eof {
         parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt(), None);
diff --git a/compiler/rustc_passes/messages.ftl b/compiler/rustc_passes/messages.ftl
index 498d9a405fa..afd08319738 100644
--- a/compiler/rustc_passes/messages.ftl
+++ b/compiler/rustc_passes/messages.ftl
@@ -687,6 +687,7 @@ passes_unused_var_maybe_capture_ref = unused variable: `{$name}`
 
 passes_unused_var_remove_field = unused variable: `{$name}`
 passes_unused_var_remove_field_suggestion = try removing the field
+passes_unused_var_typo = you might have meant to pattern match on the similarly named {$kind} `{$item_name}`
 
 passes_unused_variable_args_in_macro = `{$name}` is captured in macro and introduced a unused variable
 
diff --git a/compiler/rustc_passes/src/errors.rs b/compiler/rustc_passes/src/errors.rs
index 680e2a26d84..23dcabef1a1 100644
--- a/compiler/rustc_passes/src/errors.rs
+++ b/compiler/rustc_passes/src/errors.rs
@@ -1367,6 +1367,22 @@ pub(crate) struct UnusedVarRemoveFieldSugg {
 #[note]
 pub(crate) struct UnusedVarAssignedOnly {
     pub name: String,
+    #[subdiagnostic]
+    pub typo: Option<PatternTypo>,
+}
+
+#[derive(Subdiagnostic)]
+#[multipart_suggestion(
+    passes_unused_var_typo,
+    style = "verbose",
+    applicability = "machine-applicable"
+)]
+pub(crate) struct PatternTypo {
+    #[suggestion_part(code = "{code}")]
+    pub span: Span,
+    pub code: String,
+    pub item_name: String,
+    pub kind: String,
 }
 
 #[derive(LintDiagnostic)]
@@ -1434,6 +1450,8 @@ pub(crate) struct UnusedVariableTryPrefix {
     #[subdiagnostic]
     pub sugg: UnusedVariableSugg,
     pub name: String,
+    #[subdiagnostic]
+    pub typo: Option<PatternTypo>,
 }
 
 #[derive(Subdiagnostic)]
diff --git a/compiler/rustc_passes/src/liveness.rs b/compiler/rustc_passes/src/liveness.rs
index ae16a51bc69..1b2ffb5b3db 100644
--- a/compiler/rustc_passes/src/liveness.rs
+++ b/compiler/rustc_passes/src/liveness.rs
@@ -95,8 +95,10 @@ use rustc_hir::{Expr, HirId, HirIdMap, HirIdSet, find_attr};
 use rustc_index::IndexVec;
 use rustc_middle::query::Providers;
 use rustc_middle::span_bug;
+use rustc_middle::ty::print::with_no_trimmed_paths;
 use rustc_middle::ty::{self, RootVariableMinCaptureList, Ty, TyCtxt};
 use rustc_session::lint;
+use rustc_span::edit_distance::find_best_match_for_name;
 use rustc_span::{BytePos, Span, Symbol};
 use tracing::{debug, instrument};
 
@@ -1688,6 +1690,51 @@ impl<'tcx> Liveness<'_, 'tcx> {
             let is_assigned =
                 if ln == self.exit_ln { false } else { self.assigned_on_exit(ln, var) };
 
+            let mut typo = None;
+            for (hir_id, _, span) in &hir_ids_and_spans {
+                let ty = self.typeck_results.node_type(*hir_id);
+                if let ty::Adt(adt, _) = ty.peel_refs().kind() {
+                    let name = Symbol::intern(&name);
+                    let adt_def = self.ir.tcx.adt_def(adt.did());
+                    let variant_names: Vec<_> = adt_def
+                        .variants()
+                        .iter()
+                        .filter(|v| matches!(v.ctor, Some((CtorKind::Const, _))))
+                        .map(|v| v.name)
+                        .collect();
+                    if let Some(name) = find_best_match_for_name(&variant_names, name, None)
+                        && let Some(variant) = adt_def.variants().iter().find(|v| {
+                            v.name == name && matches!(v.ctor, Some((CtorKind::Const, _)))
+                        })
+                    {
+                        typo = Some(errors::PatternTypo {
+                            span: *span,
+                            code: with_no_trimmed_paths!(self.ir.tcx.def_path_str(variant.def_id)),
+                            kind: self.ir.tcx.def_descr(variant.def_id).to_string(),
+                            item_name: variant.name.to_string(),
+                        });
+                    }
+                }
+            }
+            if typo.is_none() {
+                for (hir_id, _, span) in &hir_ids_and_spans {
+                    let ty = self.typeck_results.node_type(*hir_id);
+                    // Look for consts of the same type with similar names as well, not just unit
+                    // structs and variants.
+                    for def_id in self.ir.tcx.hir_body_owners() {
+                        if let DefKind::Const = self.ir.tcx.def_kind(def_id)
+                            && self.ir.tcx.type_of(def_id).instantiate_identity() == ty
+                        {
+                            typo = Some(errors::PatternTypo {
+                                span: *span,
+                                code: with_no_trimmed_paths!(self.ir.tcx.def_path_str(def_id)),
+                                kind: "constant".to_string(),
+                                item_name: self.ir.tcx.item_name(def_id).to_string(),
+                            });
+                        }
+                    }
+                }
+            }
             if is_assigned {
                 self.ir.tcx.emit_node_span_lint(
                     lint::builtin::UNUSED_VARIABLES,
@@ -1696,7 +1743,7 @@ impl<'tcx> Liveness<'_, 'tcx> {
                         .into_iter()
                         .map(|(_, _, ident_span)| ident_span)
                         .collect::<Vec<_>>(),
-                    errors::UnusedVarAssignedOnly { name },
+                    errors::UnusedVarAssignedOnly { name, typo },
                 )
             } else if can_remove {
                 let spans = hir_ids_and_spans
@@ -1788,6 +1835,7 @@ impl<'tcx> Liveness<'_, 'tcx> {
                             name,
                             sugg,
                             string_interp: suggestions,
+                            typo,
                         },
                     );
                 }
diff --git a/compiler/rustc_query_system/messages.ftl b/compiler/rustc_query_system/messages.ftl
index f48dc60afa0..f686608034c 100644
--- a/compiler/rustc_query_system/messages.ftl
+++ b/compiler/rustc_query_system/messages.ftl
@@ -18,8 +18,8 @@ query_system_cycle_usage = cycle used when {$usage}
 query_system_increment_compilation = internal compiler error: encountered incremental compilation error with {$dep_node}
     .help = This is a known issue with the compiler. Run {$run_cmd} to allow your project to compile
 
-query_system_increment_compilation_note1 = Please follow the instructions below to create a bug report with the provided information
-query_system_increment_compilation_note2 = See <https://github.com/rust-lang/rust/issues/84970> for more information
+query_system_increment_compilation_note1 = please follow the instructions below to create a bug report with the provided information
+query_system_increment_compilation_note2 = see <https://github.com/rust-lang/rust/issues/84970> for more information
 
 query_system_overflow_note = query depth increased by {$depth} when {$desc}
 
diff --git a/compiler/rustc_resolve/messages.ftl b/compiler/rustc_resolve/messages.ftl
index 47280a93677..0e566e20a12 100644
--- a/compiler/rustc_resolve/messages.ftl
+++ b/compiler/rustc_resolve/messages.ftl
@@ -470,6 +470,8 @@ resolve_variable_bound_with_different_mode =
     .label = bound in different ways
     .first_binding_span = first binding
 
+resolve_variable_is_a_typo = you might have meant to use the similarly named previously used binding `{$typo}`
+
 resolve_variable_is_not_bound_in_all_patterns =
     variable `{$name}` is not bound in all patterns
 
diff --git a/compiler/rustc_resolve/src/diagnostics.rs b/compiler/rustc_resolve/src/diagnostics.rs
index 5dfc4292a38..abbacc70b3e 100644
--- a/compiler/rustc_resolve/src/diagnostics.rs
+++ b/compiler/rustc_resolve/src/diagnostics.rs
@@ -661,8 +661,12 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> {
             ResolutionError::VariableNotBoundInPattern(binding_error, parent_scope) => {
                 let BindingError { name, target, origin, could_be_path } = binding_error;
 
-                let target_sp = target.iter().copied().collect::<Vec<_>>();
-                let origin_sp = origin.iter().copied().collect::<Vec<_>>();
+                let mut target_sp = target.iter().map(|pat| pat.span).collect::<Vec<_>>();
+                target_sp.sort();
+                target_sp.dedup();
+                let mut origin_sp = origin.iter().map(|(span, _)| *span).collect::<Vec<_>>();
+                origin_sp.sort();
+                origin_sp.dedup();
 
                 let msp = MultiSpan::from_spans(target_sp.clone());
                 let mut err = self
@@ -671,8 +675,35 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> {
                 for sp in target_sp {
                     err.subdiagnostic(errors::PatternDoesntBindName { span: sp, name });
                 }
-                for sp in origin_sp {
-                    err.subdiagnostic(errors::VariableNotInAllPatterns { span: sp });
+                for sp in &origin_sp {
+                    err.subdiagnostic(errors::VariableNotInAllPatterns { span: *sp });
+                }
+                let mut suggested_typo = false;
+                if !target.iter().all(|pat| matches!(pat.kind, ast::PatKind::Ident(..)))
+                    && !origin.iter().all(|(_, pat)| matches!(pat.kind, ast::PatKind::Ident(..)))
+                {
+                    // The check above is so that when we encounter `match foo { (a | b) => {} }`,
+                    // we don't suggest `(a | a) => {}`, which would never be what the user wants.
+                    let mut target_visitor = BindingVisitor::default();
+                    for pat in &target {
+                        target_visitor.visit_pat(pat);
+                    }
+                    target_visitor.identifiers.sort();
+                    target_visitor.identifiers.dedup();
+                    let mut origin_visitor = BindingVisitor::default();
+                    for (_, pat) in &origin {
+                        origin_visitor.visit_pat(pat);
+                    }
+                    origin_visitor.identifiers.sort();
+                    origin_visitor.identifiers.dedup();
+                    // Find if the binding could have been a typo
+                    if let Some(typo) =
+                        find_best_match_for_name(&target_visitor.identifiers, name.name, None)
+                        && !origin_visitor.identifiers.contains(&typo)
+                    {
+                        err.subdiagnostic(errors::PatternBindingTypo { spans: origin_sp, typo });
+                        suggested_typo = true;
+                    }
                 }
                 if could_be_path {
                     let import_suggestions = self.lookup_import_candidates(
@@ -693,10 +724,86 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> {
                         },
                     );
 
-                    if import_suggestions.is_empty() {
+                    if import_suggestions.is_empty() && !suggested_typo {
+                        let kinds = [
+                            DefKind::Ctor(CtorOf::Variant, CtorKind::Const),
+                            DefKind::Ctor(CtorOf::Struct, CtorKind::Const),
+                            DefKind::Const,
+                            DefKind::AssocConst,
+                        ];
+                        let mut local_names = vec![];
+                        self.add_module_candidates(
+                            parent_scope.module,
+                            &mut local_names,
+                            &|res| matches!(res, Res::Def(_, _)),
+                            None,
+                        );
+                        let local_names: FxHashSet<_> = local_names
+                            .into_iter()
+                            .filter_map(|s| match s.res {
+                                Res::Def(_, def_id) => Some(def_id),
+                                _ => None,
+                            })
+                            .collect();
+
+                        let mut local_suggestions = vec![];
+                        let mut suggestions = vec![];
+                        for kind in kinds {
+                            if let Some(suggestion) = self.early_lookup_typo_candidate(
+                                ScopeSet::All(Namespace::ValueNS),
+                                &parent_scope,
+                                name,
+                                &|res: Res| match res {
+                                    Res::Def(k, _) => k == kind,
+                                    _ => false,
+                                },
+                            ) && let Res::Def(kind, mut def_id) = suggestion.res
+                            {
+                                if let DefKind::Ctor(_, _) = kind {
+                                    def_id = self.tcx.parent(def_id);
+                                }
+                                let kind = kind.descr(def_id);
+                                if local_names.contains(&def_id) {
+                                    // The item is available in the current scope. Very likely to
+                                    // be a typo. Don't use the full path.
+                                    local_suggestions.push((
+                                        suggestion.candidate,
+                                        suggestion.candidate.to_string(),
+                                        kind,
+                                    ));
+                                } else {
+                                    suggestions.push((
+                                        suggestion.candidate,
+                                        self.def_path_str(def_id),
+                                        kind,
+                                    ));
+                                }
+                            }
+                        }
+                        let suggestions = if !local_suggestions.is_empty() {
+                            // There is at least one item available in the current scope that is a
+                            // likely typo. We only show those.
+                            local_suggestions
+                        } else {
+                            suggestions
+                        };
+                        for (name, sugg, kind) in suggestions {
+                            err.span_suggestion_verbose(
+                                span,
+                                format!(
+                                    "you might have meant to use the similarly named {kind} `{name}`",
+                                ),
+                                sugg,
+                                Applicability::MaybeIncorrect,
+                            );
+                            suggested_typo = true;
+                        }
+                    }
+                    if import_suggestions.is_empty() && !suggested_typo {
                         let help_msg = format!(
-                            "if you meant to match on a variant or a `const` item, consider \
-                             making the path in the pattern qualified: `path::to::ModOrType::{name}`",
+                            "if you meant to match on a unit struct, unit variant or a `const` \
+                             item, consider making the path in the pattern qualified: \
+                             `path::to::ModOrType::{name}`",
                         );
                         err.span_help(span, help_msg);
                     }
@@ -1016,6 +1123,39 @@ impl<'ra, 'tcx> Resolver<'ra, 'tcx> {
         .emit()
     }
 
+    fn def_path_str(&self, mut def_id: DefId) -> String {
+        // We can't use `def_path_str` in resolve.
+        let mut path = vec![def_id];
+        while let Some(parent) = self.tcx.opt_parent(def_id) {
+            def_id = parent;
+            path.push(def_id);
+            if def_id.is_top_level_module() {
+                break;
+            }
+        }
+        // We will only suggest importing directly if it is accessible through that path.
+        path.into_iter()
+            .rev()
+            .map(|def_id| {
+                self.tcx
+                    .opt_item_name(def_id)
+                    .map(|name| {
+                        match (
+                            def_id.is_top_level_module(),
+                            def_id.is_local(),
+                            self.tcx.sess.edition(),
+                        ) {
+                            (true, true, Edition::Edition2015) => String::new(),
+                            (true, true, _) => kw::Crate.to_string(),
+                            (true, false, _) | (false, _, _) => name.to_string(),
+                        }
+                    })
+                    .unwrap_or_else(|| "_".to_string())
+            })
+            .collect::<Vec<String>>()
+            .join("::")
+    }
+
     pub(crate) fn add_scope_set_candidates(
         &mut self,
         suggestions: &mut Vec<TypoSuggestion>,
@@ -3396,7 +3536,7 @@ impl UsePlacementFinder {
     }
 }
 
-impl<'tcx> visit::Visitor<'tcx> for UsePlacementFinder {
+impl<'tcx> Visitor<'tcx> for UsePlacementFinder {
     fn visit_crate(&mut self, c: &Crate) {
         if self.target_module == CRATE_NODE_ID {
             let inject = c.spans.inject_use_span;
@@ -3424,6 +3564,22 @@ impl<'tcx> visit::Visitor<'tcx> for UsePlacementFinder {
     }
 }
 
+#[derive(Default)]
+struct BindingVisitor {
+    identifiers: Vec<Symbol>,
+    spans: FxHashMap<Symbol, Vec<Span>>,
+}
+
+impl<'tcx> Visitor<'tcx> for BindingVisitor {
+    fn visit_pat(&mut self, pat: &ast::Pat) {
+        if let ast::PatKind::Ident(_, ident, _) = pat.kind {
+            self.identifiers.push(ident.name);
+            self.spans.entry(ident.name).or_default().push(ident.span);
+        }
+        visit::walk_pat(self, pat);
+    }
+}
+
 fn search_for_any_use_in_items(items: &[Box<ast::Item>]) -> Option<Span> {
     for item in items {
         if let ItemKind::Use(..) = item.kind
diff --git a/compiler/rustc_resolve/src/errors.rs b/compiler/rustc_resolve/src/errors.rs
index 63d6fa23a14..72be94e58db 100644
--- a/compiler/rustc_resolve/src/errors.rs
+++ b/compiler/rustc_resolve/src/errors.rs
@@ -986,6 +986,18 @@ pub(crate) struct VariableNotInAllPatterns {
     pub(crate) span: Span,
 }
 
+#[derive(Subdiagnostic)]
+#[multipart_suggestion(
+    resolve_variable_is_a_typo,
+    applicability = "maybe-incorrect",
+    style = "verbose"
+)]
+pub(crate) struct PatternBindingTypo {
+    #[suggestion_part(code = "{typo}")]
+    pub(crate) spans: Vec<Span>,
+    pub(crate) typo: Symbol,
+}
+
 #[derive(Diagnostic)]
 #[diag(resolve_name_defined_multiple_time)]
 #[note]
diff --git a/compiler/rustc_resolve/src/late.rs b/compiler/rustc_resolve/src/late.rs
index 6a12fafcb1e..521fef2b9d4 100644
--- a/compiler/rustc_resolve/src/late.rs
+++ b/compiler/rustc_resolve/src/late.rs
@@ -8,7 +8,6 @@
 
 use std::assert_matches::debug_assert_matches;
 use std::borrow::Cow;
-use std::collections::BTreeSet;
 use std::collections::hash_map::Entry;
 use std::mem::{replace, swap, take};
 
@@ -3682,31 +3681,30 @@ impl<'a, 'ast, 'ra, 'tcx> LateResolutionVisitor<'a, 'ast, 'ra, 'tcx> {
         // 2) Record any missing bindings or binding mode inconsistencies.
         for (map_outer, pat_outer) in not_never_pats.iter() {
             // Check against all arms except for the same pattern which is always self-consistent.
-            let inners = not_never_pats
-                .iter()
-                .filter(|(_, pat)| pat.id != pat_outer.id)
-                .flat_map(|(map, _)| map);
-
-            for (&name, binding_inner) in inners {
-                match map_outer.get(&name) {
-                    None => {
-                        // The inner binding is missing in the outer.
-                        let binding_error =
-                            missing_vars.entry(name).or_insert_with(|| BindingError {
-                                name,
-                                origin: BTreeSet::new(),
-                                target: BTreeSet::new(),
-                                could_be_path: name.as_str().starts_with(char::is_uppercase),
-                            });
-                        binding_error.origin.insert(binding_inner.span);
-                        binding_error.target.insert(pat_outer.span);
-                    }
-                    Some(binding_outer) => {
-                        if binding_outer.annotation != binding_inner.annotation {
-                            // The binding modes in the outer and inner bindings differ.
-                            inconsistent_vars
-                                .entry(name)
-                                .or_insert((binding_inner.span, binding_outer.span));
+            let inners = not_never_pats.iter().filter(|(_, pat)| pat.id != pat_outer.id);
+
+            for (map, pat) in inners {
+                for (&name, binding_inner) in map {
+                    match map_outer.get(&name) {
+                        None => {
+                            // The inner binding is missing in the outer.
+                            let binding_error =
+                                missing_vars.entry(name).or_insert_with(|| BindingError {
+                                    name,
+                                    origin: Default::default(),
+                                    target: Default::default(),
+                                    could_be_path: name.as_str().starts_with(char::is_uppercase),
+                                });
+                            binding_error.origin.push((binding_inner.span, (***pat).clone()));
+                            binding_error.target.push((***pat_outer).clone());
+                        }
+                        Some(binding_outer) => {
+                            if binding_outer.annotation != binding_inner.annotation {
+                                // The binding modes in the outer and inner bindings differ.
+                                inconsistent_vars
+                                    .entry(name)
+                                    .or_insert((binding_inner.span, binding_outer.span));
+                            }
                         }
                     }
                 }
@@ -3719,7 +3717,7 @@ impl<'a, 'ast, 'ra, 'tcx> LateResolutionVisitor<'a, 'ast, 'ra, 'tcx> {
                 v.could_be_path = false;
             }
             self.report_error(
-                *v.origin.iter().next().unwrap(),
+                v.origin.iter().next().unwrap().0,
                 ResolutionError::VariableNotBoundInPattern(v, self.parent_scope),
             );
         }
diff --git a/compiler/rustc_resolve/src/lib.rs b/compiler/rustc_resolve/src/lib.rs
index 9674c0356c2..0dea6ae3327 100644
--- a/compiler/rustc_resolve/src/lib.rs
+++ b/compiler/rustc_resolve/src/lib.rs
@@ -230,8 +230,8 @@ enum Used {
 #[derive(Debug)]
 struct BindingError {
     name: Ident,
-    origin: BTreeSet<Span>,
-    target: BTreeSet<Span>,
+    origin: Vec<(Span, ast::Pat)>,
+    target: Vec<ast::Pat>,
     could_be_path: bool,
 }
 
diff --git a/compiler/rustc_session/src/options.rs b/compiler/rustc_session/src/options.rs
index 6d5be2d92cd..69facde6936 100644
--- a/compiler/rustc_session/src/options.rs
+++ b/compiler/rustc_session/src/options.rs
@@ -83,10 +83,77 @@ pub struct TargetModifier {
     pub value_name: String,
 }
 
+mod target_modifier_consistency_check {
+    use super::*;
+    pub(super) fn sanitizer(l: &TargetModifier, r: Option<&TargetModifier>) -> bool {
+        let mut lparsed: SanitizerSet = Default::default();
+        let lval = if l.value_name.is_empty() { None } else { Some(l.value_name.as_str()) };
+        parse::parse_sanitizers(&mut lparsed, lval);
+
+        let mut rparsed: SanitizerSet = Default::default();
+        let rval = r.filter(|v| !v.value_name.is_empty()).map(|v| v.value_name.as_str());
+        parse::parse_sanitizers(&mut rparsed, rval);
+
+        // Some sanitizers need to be target modifiers, and some do not.
+        // For now, we should mark all sanitizers as target modifiers except for these:
+        // AddressSanitizer, LeakSanitizer
+        let tmod_sanitizers = SanitizerSet::MEMORY
+            | SanitizerSet::THREAD
+            | SanitizerSet::HWADDRESS
+            | SanitizerSet::CFI
+            | SanitizerSet::MEMTAG
+            | SanitizerSet::SHADOWCALLSTACK
+            | SanitizerSet::KCFI
+            | SanitizerSet::KERNELADDRESS
+            | SanitizerSet::SAFESTACK
+            | SanitizerSet::DATAFLOW;
+
+        lparsed & tmod_sanitizers == rparsed & tmod_sanitizers
+    }
+    pub(super) fn sanitizer_cfi_normalize_integers(
+        opts: &Options,
+        l: &TargetModifier,
+        r: Option<&TargetModifier>,
+    ) -> bool {
+        // For kCFI, the helper flag -Zsanitizer-cfi-normalize-integers should also be a target modifier
+        if opts.unstable_opts.sanitizer.contains(SanitizerSet::KCFI) {
+            if let Some(r) = r {
+                return l.extend().tech_value == r.extend().tech_value;
+            } else {
+                return false;
+            }
+        }
+        true
+    }
+}
+
 impl TargetModifier {
     pub fn extend(&self) -> ExtendedTargetModifierInfo {
         self.opt.reparse(&self.value_name)
     }
+    // Custom consistency check for target modifiers (or default `l.tech_value == r.tech_value`)
+    // When other is None, consistency with default value is checked
+    pub fn consistent(&self, opts: &Options, other: Option<&TargetModifier>) -> bool {
+        assert!(other.is_none() || self.opt == other.unwrap().opt);
+        match self.opt {
+            OptionsTargetModifiers::UnstableOptions(unstable) => match unstable {
+                UnstableOptionsTargetModifiers::sanitizer => {
+                    return target_modifier_consistency_check::sanitizer(self, other);
+                }
+                UnstableOptionsTargetModifiers::sanitizer_cfi_normalize_integers => {
+                    return target_modifier_consistency_check::sanitizer_cfi_normalize_integers(
+                        opts, self, other,
+                    );
+                }
+                _ => {}
+            },
+            _ => {}
+        };
+        match other {
+            Some(other) => self.extend().tech_value == other.extend().tech_value,
+            None => false,
+        }
+    }
 }
 
 fn tmod_push_impl(
@@ -2504,13 +2571,13 @@ written to standard error output)"),
     retpoline_external_thunk: bool = (false, parse_bool, [TRACKED TARGET_MODIFIER],
         "enables retpoline-external-thunk, retpoline-indirect-branches and retpoline-indirect-calls \
         target features (default: no)"),
-    sanitizer: SanitizerSet = (SanitizerSet::empty(), parse_sanitizers, [TRACKED],
+    sanitizer: SanitizerSet = (SanitizerSet::empty(), parse_sanitizers, [TRACKED TARGET_MODIFIER],
         "use a sanitizer"),
     sanitizer_cfi_canonical_jump_tables: Option<bool> = (Some(true), parse_opt_bool, [TRACKED],
         "enable canonical jump tables (default: yes)"),
     sanitizer_cfi_generalize_pointers: Option<bool> = (None, parse_opt_bool, [TRACKED],
         "enable generalizing pointer types (default: no)"),
-    sanitizer_cfi_normalize_integers: Option<bool> = (None, parse_opt_bool, [TRACKED],
+    sanitizer_cfi_normalize_integers: Option<bool> = (None, parse_opt_bool, [TRACKED TARGET_MODIFIER],
         "enable normalizing integer types (default: no)"),
     sanitizer_dataflow_abilist: Vec<String> = (Vec::new(), parse_comma_list, [TRACKED],
         "additional ABI list files that control how shadow parameters are passed (comma separated)"),
diff --git a/compiler/rustc_span/src/analyze_source_file.rs b/compiler/rustc_span/src/analyze_source_file.rs
index c32593a6d95..bb2cda77dff 100644
--- a/compiler/rustc_span/src/analyze_source_file.rs
+++ b/compiler/rustc_span/src/analyze_source_file.rs
@@ -81,8 +81,8 @@ cfg_select! {
                 // use `loadu`, which supports unaligned loading.
                 let chunk = unsafe { _mm_loadu_si128(chunk.as_ptr() as *const __m128i) };
 
-                // For character in the chunk, see if its byte value is < 0, which
-                // indicates that it's part of a UTF-8 char.
+                // For each character in the chunk, see if its byte value is < 0,
+                // which indicates that it's part of a UTF-8 char.
                 let multibyte_test = _mm_cmplt_epi8(chunk, _mm_set1_epi8(0));
                 // Create a bit mask from the comparison results.
                 let multibyte_mask = _mm_movemask_epi8(multibyte_test);
@@ -132,8 +132,111 @@ cfg_select! {
             }
         }
     }
+    target_arch = "loongarch64" => {
+        fn analyze_source_file_dispatch(
+            src: &str,
+            lines: &mut Vec<RelativeBytePos>,
+            multi_byte_chars: &mut Vec<MultiByteChar>,
+        ) {
+            use std::arch::is_loongarch_feature_detected;
+
+            if is_loongarch_feature_detected!("lsx") {
+                unsafe {
+                    analyze_source_file_lsx(src, lines, multi_byte_chars);
+                }
+            } else {
+                analyze_source_file_generic(
+                    src,
+                    src.len(),
+                    RelativeBytePos::from_u32(0),
+                    lines,
+                    multi_byte_chars,
+                );
+            }
+        }
+
+        /// Checks 16 byte chunks of text at a time. If the chunk contains
+        /// something other than printable ASCII characters and newlines, the
+        /// function falls back to the generic implementation. Otherwise it uses
+        /// LSX intrinsics to quickly find all newlines.
+        #[target_feature(enable = "lsx")]
+        unsafe fn analyze_source_file_lsx(
+            src: &str,
+            lines: &mut Vec<RelativeBytePos>,
+            multi_byte_chars: &mut Vec<MultiByteChar>,
+        ) {
+            use std::arch::loongarch64::*;
+
+            const CHUNK_SIZE: usize = 16;
+
+            let (chunks, tail) = src.as_bytes().as_chunks::<CHUNK_SIZE>();
+
+            // This variable keeps track of where we should start decoding a
+            // chunk. If a multi-byte character spans across chunk boundaries,
+            // we need to skip that part in the next chunk because we already
+            // handled it.
+            let mut intra_chunk_offset = 0;
+
+            for (chunk_index, chunk) in chunks.iter().enumerate() {
+                // All LSX memory instructions support unaligned access, so using
+                // vld is fine.
+                let chunk = unsafe { lsx_vld::<0>(chunk.as_ptr() as *const i8) };
+
+                // For each character in the chunk, see if its byte value is < 0,
+                // which indicates that it's part of a UTF-8 char.
+                let multibyte_mask = lsx_vmskltz_b(chunk);
+                // Create a bit mask from the comparison results.
+                let multibyte_mask = lsx_vpickve2gr_w::<0>(multibyte_mask);
+
+                // If the bit mask is all zero, we only have ASCII chars here:
+                if multibyte_mask == 0 {
+                    assert!(intra_chunk_offset == 0);
+
+                    // Check for newlines in the chunk
+                    let newlines_test = lsx_vseqi_b::<{b'\n' as i32}>(chunk);
+                    let newlines_mask = lsx_vmskltz_b(newlines_test);
+                    let mut newlines_mask = lsx_vpickve2gr_w::<0>(newlines_mask);
+
+                    let output_offset = RelativeBytePos::from_usize(chunk_index * CHUNK_SIZE + 1);
+
+                    while newlines_mask != 0 {
+                        let index = newlines_mask.trailing_zeros();
+
+                        lines.push(RelativeBytePos(index) + output_offset);
+
+                        // Clear the bit, so we can find the next one.
+                        newlines_mask &= newlines_mask - 1;
+                    }
+                } else {
+                    // The slow path.
+                    // There are multibyte chars in here, fallback to generic decoding.
+                    let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
+                    intra_chunk_offset = analyze_source_file_generic(
+                        &src[scan_start..],
+                        CHUNK_SIZE - intra_chunk_offset,
+                        RelativeBytePos::from_usize(scan_start),
+                        lines,
+                        multi_byte_chars,
+                    );
+                }
+            }
+
+            // There might still be a tail left to analyze
+            let tail_start = src.len() - tail.len() + intra_chunk_offset;
+            if tail_start < src.len() {
+                analyze_source_file_generic(
+                    &src[tail_start..],
+                    src.len() - tail_start,
+                    RelativeBytePos::from_usize(tail_start),
+                    lines,
+                    multi_byte_chars,
+                );
+            }
+        }
+    }
     _ => {
-        // The target (or compiler version) does not support SSE2 ...
+        // The target (or compiler version) does not support vector instructions
+        // our specialized implementations need (x86 SSE2, loongarch64 LSX)...
         fn analyze_source_file_dispatch(
             src: &str,
             lines: &mut Vec<RelativeBytePos>,
diff --git a/compiler/rustc_span/src/lib.rs b/compiler/rustc_span/src/lib.rs
index ae6755f0764..8907c5e4c4a 100644
--- a/compiler/rustc_span/src/lib.rs
+++ b/compiler/rustc_span/src/lib.rs
@@ -18,6 +18,7 @@
 // tidy-alphabetical-start
 #![allow(internal_features)]
 #![cfg_attr(bootstrap, feature(round_char_boundary))]
+#![cfg_attr(target_arch = "loongarch64", feature(stdarch_loongarch))]
 #![doc(html_root_url = "https://doc.rust-lang.org/nightly/nightly-rustc/")]
 #![doc(rust_logo)]
 #![feature(array_windows)]
diff --git a/compiler/rustc_span/src/symbol.rs b/compiler/rustc_span/src/symbol.rs
index 77260d07c99..e5108d8b7e9 100644
--- a/compiler/rustc_span/src/symbol.rs
+++ b/compiler/rustc_span/src/symbol.rs
@@ -2162,6 +2162,7 @@ symbols! {
         target_family,
         target_feature,
         target_feature_11,
+        target_feature_inline_always,
         target_has_atomic,
         target_has_atomic_equal_alignment,
         target_has_atomic_load_store,
@@ -2280,6 +2281,8 @@ symbols! {
         unboxed_closures,
         unchecked_add,
         unchecked_div,
+        unchecked_funnel_shl,
+        unchecked_funnel_shr,
         unchecked_mul,
         unchecked_rem,
         unchecked_shl,
diff --git a/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs b/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs
index f1b6fa123de..cd55576ef81 100644
--- a/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs
+++ b/compiler/rustc_target/src/spec/targets/aarch64_pc_windows_msvc.rs
@@ -15,7 +15,7 @@ pub(crate) fn target() -> Target {
         llvm_target: "aarch64-pc-windows-msvc".into(),
         metadata: TargetMetadata {
             description: Some("ARM64 Windows MSVC".into()),
-            tier: Some(2),
+            tier: Some(1),
             host_tools: Some(true),
             std: Some(true),
         },
diff --git a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs
index 598f0f19f0d..cada0dd640a 100644
--- a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs
+++ b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs
@@ -6,7 +6,7 @@ use crate::spec::{
 pub(crate) fn target() -> Target {
     Target {
         arch: "nvptx64".into(),
-        data_layout: "e-p6:32:32-i64:64-i128:128-v16:16-v32:32-n16:32:64".into(),
+        data_layout: "e-p6:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64".into(),
         llvm_target: "nvptx64-nvidia-cuda".into(),
         metadata: TargetMetadata {
             description: Some("--emit=asm generates PTX code that runs on NVIDIA GPUs".into()),
diff --git a/compiler/rustc_type_ir/src/interner.rs b/compiler/rustc_type_ir/src/interner.rs
index f5448baf8c0..3355c0c2104 100644
--- a/compiler/rustc_type_ir/src/interner.rs
+++ b/compiler/rustc_type_ir/src/interner.rs
@@ -1,3 +1,4 @@
+use std::borrow::Borrow;
 use std::fmt::Debug;
 use std::hash::Hash;
 use std::ops::Deref;
@@ -383,12 +384,12 @@ pub trait Interner:
         defining_anchor: Self::LocalDefId,
     ) -> Self::LocalDefIds;
 
-    type ProbeRef: Copy + Debug + Hash + Eq + Deref<Target = inspect::Probe<Self>>;
-    fn mk_probe_ref(self, probe: inspect::Probe<Self>) -> Self::ProbeRef;
+    type Probe: Debug + Hash + Eq + Borrow<inspect::Probe<Self>>;
+    fn mk_probe(self, probe: inspect::Probe<Self>) -> Self::Probe;
     fn evaluate_root_goal_for_proof_tree_raw(
         self,
         canonical_goal: CanonicalInput<Self>,
-    ) -> (QueryResult<Self>, Self::ProbeRef);
+    ) -> (QueryResult<Self>, Self::Probe);
 }
 
 /// Imagine you have a function `F: FnOnce(&[T]) -> R`, plus an iterator `iter`
diff --git a/compiler/rustc_type_ir/src/solve/inspect.rs b/compiler/rustc_type_ir/src/solve/inspect.rs
index 5452ac44158..3af2f8dbaf2 100644
--- a/compiler/rustc_type_ir/src/solve/inspect.rs
+++ b/compiler/rustc_type_ir/src/solve/inspect.rs
@@ -49,7 +49,7 @@ pub type CanonicalState<I, T> = Canonical<I, State<I, T>>;
 pub struct GoalEvaluation<I: Interner> {
     pub uncanonicalized_goal: Goal<I, I::Predicate>,
     pub orig_values: Vec<I::GenericArg>,
-    pub final_revision: I::ProbeRef,
+    pub final_revision: I::Probe,
     pub result: QueryResult<I>,
 }