diff --git a/compiler/rustc_codegen_llvm/src/abi.rs b/compiler/rustc_codegen_llvm/src/abi.rs
index dca9c1f..07adfff 100644
--- a/compiler/rustc_codegen_llvm/src/abi.rs
+++ b/compiler/rustc_codegen_llvm/src/abi.rs
@@ -1,7 +1,6 @@
 use crate::builder::Builder;
 use crate::context::CodegenCx;
 use crate::llvm::{self, AttributePlace};
-use crate::llvm_util;
 use crate::type_::Type;
 use crate::type_of::LayoutLlvmExt;
 use crate::value::Value;
@@ -53,15 +52,10 @@
 }
 
 fn should_use_mutable_noalias(cx: &CodegenCx<'_, '_>) -> bool {
-    // LLVM prior to version 12 has known miscompiles in the presence of
-    // noalias attributes (see #54878). Only enable mutable noalias by
-    // default for versions we believe to be safe.
-    cx.tcx
-        .sess
-        .opts
-        .debugging_opts
-        .mutable_noalias
-        .unwrap_or_else(|| llvm_util::get_version() >= (12, 0, 0))
+    // LLVM prior to version 12 had known miscompiles in the presence of
+    // noalias attributes (see #54878), but we don't support earlier
+    // versions at all anymore. We now enable mutable noalias by default.
+    cx.tcx.sess.opts.debugging_opts.mutable_noalias.unwrap_or(true)
 }
 
 impl ArgAttributesExt for ArgAttributes {
@@ -613,7 +607,7 @@
         fn_abi.apply_attrs_callsite(self, callsite)
     }
 
-    fn get_param(&self, index: usize) -> Self::Value {
+    fn get_param(&mut self, index: usize) -> Self::Value {
         llvm::get_param(self.llfn(), index as c_uint)
     }
 }
diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs
index 341a888..02096f4 100644
--- a/compiler/rustc_codegen_llvm/src/asm.rs
+++ b/compiler/rustc_codegen_llvm/src/asm.rs
@@ -13,7 +13,7 @@
 use rustc_data_structures::fx::FxHashMap;
 use rustc_hir as hir;
 use rustc_middle::ty::layout::TyAndLayout;
-use rustc_middle::{bug, span_bug};
+use rustc_middle::{bug, span_bug, ty::Instance};
 use rustc_span::{Pos, Span, Symbol};
 use rustc_target::abi::*;
 use rustc_target::asm::*;
@@ -120,6 +120,7 @@
         operands: &[InlineAsmOperandRef<'tcx, Self>],
         options: InlineAsmOptions,
         line_spans: &[Span],
+        instance: Instance<'_>,
     ) {
         let asm_arch = self.tcx.sess.asm_arch.unwrap();
 
@@ -135,7 +136,10 @@
                     let is_target_supported = |reg_class: InlineAsmRegClass| {
                         for &(_, feature) in reg_class.supported_types(asm_arch) {
                             if let Some(feature) = feature {
-                                if self.tcx.sess.target_features.contains(&Symbol::intern(feature))
+                                let codegen_fn_attrs = self.tcx.codegen_fn_attrs(instance.def_id());
+                                let feature_name = Symbol::intern(feature);
+                                if self.tcx.sess.target_features.contains(&feature_name)
+                                    || codegen_fn_attrs.target_features.contains(&feature_name)
                                 {
                                     return true;
                                 }
@@ -316,7 +320,7 @@
                 InlineAsmArch::Mips | InlineAsmArch::Mips64 => {}
                 InlineAsmArch::S390x => {}
                 InlineAsmArch::SpirV => {}
-                InlineAsmArch::Wasm32 => {}
+                InlineAsmArch::Wasm32 | InlineAsmArch::Wasm64 => {}
                 InlineAsmArch::Bpf => {}
             }
         }
diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs
index 659cf9e..8e6329a 100644
--- a/compiler/rustc_codegen_llvm/src/attributes.rs
+++ b/compiler/rustc_codegen_llvm/src/attributes.rs
@@ -12,7 +12,7 @@
 use rustc_session::config::OptLevel;
 use rustc_session::Session;
 use rustc_target::spec::abi::Abi;
-use rustc_target::spec::{FramePointer, SanitizerSet, StackProbeType};
+use rustc_target::spec::{FramePointer, SanitizerSet, StackProbeType, StackProtector};
 
 use crate::attributes;
 use crate::llvm::AttributePlace::Function;
@@ -161,6 +161,17 @@
     }
 }
 
+fn set_stackprotector(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
+    let sspattr = match cx.sess().stack_protector() {
+        StackProtector::None => return,
+        StackProtector::All => Attribute::StackProtectReq,
+        StackProtector::Strong => Attribute::StackProtectStrong,
+        StackProtector::Basic => Attribute::StackProtect,
+    };
+
+    sspattr.apply_llfn(Function, llfn)
+}
+
 pub fn apply_target_cpu_attr(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
     let target_cpu = SmallCStr::new(llvm_util::target_cpu(cx.tcx.sess));
     llvm::AddFunctionAttrStringValue(
@@ -271,6 +282,7 @@
     set_frame_pointer_type(cx, llfn);
     set_instrument_function(cx, llfn);
     set_probestack(cx, llfn);
+    set_stackprotector(cx, llfn);
 
     if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::COLD) {
         Attribute::Cold.apply_llfn(Function, llfn);
diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs
index b6d682f..6d2ad70 100644
--- a/compiler/rustc_codegen_llvm/src/back/write.rs
+++ b/compiler/rustc_codegen_llvm/src/back/write.rs
@@ -17,6 +17,7 @@
 };
 use rustc_codegen_ssa::traits::*;
 use rustc_codegen_ssa::{CompiledModule, ModuleCodegen};
+use rustc_data_structures::profiling::SelfProfilerRef;
 use rustc_data_structures::small_c_str::SmallCStr;
 use rustc_errors::{FatalError, Handler, Level};
 use rustc_fs_util::{link_or_copy, path_to_c_string};
@@ -53,6 +54,7 @@
     output: &Path,
     dwo_output: Option<&Path>,
     file_type: llvm::FileType,
+    self_profiler_ref: &SelfProfilerRef,
 ) -> Result<(), FatalError> {
     unsafe {
         let output_c = path_to_c_string(output);
@@ -76,6 +78,19 @@
                 file_type,
             )
         };
+
+        // Record artifact sizes for self-profiling
+        if result == llvm::LLVMRustResult::Success {
+            let artifact_kind = match file_type {
+                llvm::FileType::ObjectFile => "object_file",
+                llvm::FileType::AssemblyFile => "assembly_file",
+            };
+            record_artifact_size(self_profiler_ref, artifact_kind, output);
+            if let Some(dwo_file) = dwo_output {
+                record_artifact_size(self_profiler_ref, "dwo_file", dwo_file);
+            }
+        }
+
         result.into_result().map_err(|()| {
             let msg = format!("could not write output to {}", output.display());
             llvm_err(handler, &msg)
@@ -161,6 +176,7 @@
     let ffunction_sections =
         sess.opts.debugging_opts.function_sections.unwrap_or(sess.target.function_sections);
     let fdata_sections = ffunction_sections;
+    let funique_section_names = !sess.opts.debugging_opts.no_unique_section_names;
 
     let code_model = to_llvm_code_model(sess.code_model());
 
@@ -205,6 +221,7 @@
                 use_softfp,
                 ffunction_sections,
                 fdata_sections,
+                funique_section_names,
                 trap_unreachable,
                 singlethread,
                 asm_comments,
@@ -284,7 +301,7 @@
         cookie = 0;
     }
     let level = match level {
-        llvm::DiagnosticLevel::Error => Level::Error,
+        llvm::DiagnosticLevel::Error => Level::Error { lint: false },
         llvm::DiagnosticLevel::Warning => Level::Warning,
         llvm::DiagnosticLevel::Note | llvm::DiagnosticLevel::Remark => Level::Note,
     };
@@ -746,6 +763,14 @@
             let thin = ThinBuffer::new(llmod);
             let data = thin.data();
 
+            if let Some(bitcode_filename) = bc_out.file_name() {
+                cgcx.prof.artifact_size(
+                    "llvm_bitcode",
+                    bitcode_filename.to_string_lossy(),
+                    data.len() as u64,
+                );
+            }
+
             if config.emit_bc || config.emit_obj == EmitObj::Bitcode {
                 let _timer = cgcx.prof.generic_activity_with_arg(
                     "LLVM_module_codegen_emit_bitcode",
@@ -806,6 +831,11 @@
             }
 
             let result = llvm::LLVMRustPrintModule(llmod, out_c.as_ptr(), demangle_callback);
+
+            if result == llvm::LLVMRustResult::Success {
+                record_artifact_size(&cgcx.prof, "llvm_ir", &out);
+            }
+
             result.into_result().map_err(|()| {
                 let msg = format!("failed to write LLVM IR to {}", out.display());
                 llvm_err(diag_handler, &msg)
@@ -836,6 +866,7 @@
                     &path,
                     None,
                     llvm::FileType::AssemblyFile,
+                    &cgcx.prof,
                 )
             })?;
         }
@@ -869,6 +900,7 @@
                         &obj_out,
                         dwo_out,
                         llvm::FileType::ObjectFile,
+                        &cgcx.prof,
                     )
                 })?;
             }
@@ -1125,3 +1157,19 @@
         symbol_name.starts_with(b"__llvm_profile_")
     }
 }
+
+fn record_artifact_size(
+    self_profiler_ref: &SelfProfilerRef,
+    artifact_kind: &'static str,
+    path: &Path,
+) {
+    // Don't stat the file if we are not going to record its size.
+    if !self_profiler_ref.enabled() {
+        return;
+    }
+
+    if let Some(artifact_name) = path.file_name() {
+        let file_size = std::fs::metadata(path).map(|m| m.len()).unwrap_or(0);
+        self_profiler_ref.artifact_size(artifact_kind, artifact_name.to_string_lossy(), file_size);
+    }
+}
diff --git a/compiler/rustc_codegen_llvm/src/base.rs b/compiler/rustc_codegen_llvm/src/base.rs
index 3026c2f..8766cae 100644
--- a/compiler/rustc_codegen_llvm/src/base.rs
+++ b/compiler/rustc_codegen_llvm/src/base.rs
@@ -113,8 +113,13 @@
     let start_time = Instant::now();
 
     let dep_node = tcx.codegen_unit(cgu_name).codegen_dep_node(tcx);
-    let (module, _) =
-        tcx.dep_graph.with_task(dep_node, tcx, cgu_name, module_codegen, dep_graph::hash_result);
+    let (module, _) = tcx.dep_graph.with_task(
+        dep_node,
+        tcx,
+        cgu_name,
+        module_codegen,
+        Some(dep_graph::hash_result),
+    );
     let time_to_codegen = start_time.elapsed();
 
     // We assume that the cost to run LLVM on a CGU is proportional to
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
index d5deacf..ff88302 100644
--- a/compiler/rustc_codegen_llvm/src/builder.rs
+++ b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -604,6 +604,32 @@
         }
     }
 
+    fn type_metadata(&mut self, function: &'ll Value, typeid: String) {
+        let typeid_metadata = self.typeid_metadata(typeid);
+        let v = [self.const_usize(0), typeid_metadata];
+        unsafe {
+            llvm::LLVMGlobalSetMetadata(
+                function,
+                llvm::MD_type as c_uint,
+                llvm::LLVMValueAsMetadata(llvm::LLVMMDNodeInContext(
+                    self.cx.llcx,
+                    v.as_ptr(),
+                    v.len() as c_uint,
+                )),
+            )
+        }
+    }
+
+    fn typeid_metadata(&mut self, typeid: String) -> Self::Value {
+        unsafe {
+            llvm::LLVMMDStringInContext(
+                self.cx.llcx,
+                typeid.as_ptr() as *const c_char,
+                typeid.as_bytes().len() as c_uint,
+            )
+        }
+    }
+
     fn store(&mut self, val: &'ll Value, ptr: &'ll Value, align: Align) -> &'ll Value {
         self.store_with_flags(val, ptr, align, MemFlags::empty())
     }
@@ -705,7 +731,7 @@
     }
 
     fn fptoui_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> Option<&'ll Value> {
-        if llvm_util::get_version() >= (12, 0, 0) && !self.fptoint_sat_broken_in_llvm() {
+        if !self.fptoint_sat_broken_in_llvm() {
             let src_ty = self.cx.val_ty(val);
             let float_width = self.cx.float_width(src_ty);
             let int_width = self.cx.int_width(dest_ty);
@@ -717,7 +743,7 @@
     }
 
     fn fptosi_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> Option<&'ll Value> {
-        if llvm_util::get_version() >= (12, 0, 0) && !self.fptoint_sat_broken_in_llvm() {
+        if !self.fptoint_sat_broken_in_llvm() {
             let src_ty = self.cx.val_ty(val);
             let float_width = self.cx.float_width(src_ty);
             let int_width = self.cx.int_width(dest_ty);
@@ -743,7 +769,7 @@
         // we like. To ensure that LLVM picks the right instruction we choose
         // the raw wasm intrinsic functions which avoid LLVM inserting all the
         // other control flow automatically.
-        if self.sess().target.arch == "wasm32" {
+        if self.sess().target.is_like_wasm {
             let src_ty = self.cx.val_ty(val);
             if self.cx.type_kind(src_ty) != TypeKind::Vector {
                 let float_width = self.cx.float_width(src_ty);
@@ -765,7 +791,7 @@
 
     fn fptosi(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value {
         // see `fptoui` above for why wasm is different here
-        if self.sess().target.arch == "wasm32" {
+        if self.sess().target.is_like_wasm {
             let src_ty = self.cx.val_ty(val);
             if self.cx.type_kind(src_ty) != TypeKind::Vector {
                 let float_width = self.cx.float_width(src_ty);
diff --git a/compiler/rustc_codegen_llvm/src/consts.rs b/compiler/rustc_codegen_llvm/src/consts.rs
index 1afa6f0..b154ced 100644
--- a/compiler/rustc_codegen_llvm/src/consts.rs
+++ b/compiler/rustc_codegen_llvm/src/consts.rs
@@ -490,7 +490,7 @@
 
             // Wasm statics with custom link sections get special treatment as they
             // go into custom sections of the wasm executable.
-            if self.tcx.sess.opts.target_triple.triple().starts_with("wasm32") {
+            if self.tcx.sess.target.is_like_wasm {
                 if let Some(section) = attrs.link_section {
                     let section = llvm::LLVMMDStringInContext(
                         self.llcx,
diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs
index 257a0ac..613a8df 100644
--- a/compiler/rustc_codegen_llvm/src/context.rs
+++ b/compiler/rustc_codegen_llvm/src/context.rs
@@ -134,9 +134,6 @@
     let llmod = llvm::LLVMModuleCreateWithNameInContext(mod_name.as_ptr(), llcx);
 
     let mut target_data_layout = sess.target.data_layout.clone();
-    if llvm_util::get_version() < (12, 0, 0) && sess.target.arch == "powerpc64" {
-        target_data_layout = target_data_layout.replace("-v256:256:256-v512:512:512", "");
-    }
     if llvm_util::get_version() < (13, 0, 0) {
         if sess.target.arch == "powerpc64" {
             target_data_layout = target_data_layout.replace("-S128", "");
@@ -221,6 +218,15 @@
         llvm::LLVMRustAddModuleFlag(llmod, avoid_plt, 1);
     }
 
+    if sess.is_sanitizer_cfi_enabled() {
+        // FIXME(rcvalle): Add support for non canonical jump tables.
+        let canonical_jump_tables = "CFI Canonical Jump Tables\0".as_ptr().cast();
+        // FIXME(rcvalle): Add it with Override behavior flag--LLVMRustAddModuleFlag adds it with
+        // Warning behavior flag. Add support for specifying the behavior flag to
+        // LLVMRustAddModuleFlag.
+        llvm::LLVMRustAddModuleFlag(llmod, canonical_jump_tables, 1);
+    }
+
     // Control Flow Guard is currently only supported by the MSVC linker on Windows.
     if sess.target.is_like_msvc {
         match sess.opts.cg.control_flow_guard {
@@ -591,7 +597,6 @@
         ifn!("llvm.trap", fn() -> void);
         ifn!("llvm.debugtrap", fn() -> void);
         ifn!("llvm.frameaddress", fn(t_i32) -> i8p);
-        ifn!("llvm.sideeffect", fn() -> void);
 
         ifn!("llvm.powi.f32", fn(t_f32, t_i32) -> t_f32);
         ifn!("llvm.powi.f64", fn(t_f64, t_i32) -> t_f64);
@@ -779,6 +784,8 @@
             ifn!("llvm.instrprof.increment", fn(i8p, t_i64, t_i32, t_i32) -> void);
         }
 
+        ifn!("llvm.type.test", fn(i8p, self.type_metadata()) -> i1);
+
         if self.sess().opts.debuginfo != DebugInfo::None {
             ifn!("llvm.dbg.declare", fn(self.type_metadata(), self.type_metadata()) -> void);
             ifn!("llvm.dbg.value", fn(self.type_metadata(), t_i64, self.type_metadata()) -> void);
diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs b/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs
index 1f1bd73..2a6bf7d 100644
--- a/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs
+++ b/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs
@@ -322,7 +322,7 @@
         type_names::push_item_name(self.tcx(), def_id, false, &mut name);
 
         // Find the enclosing function, in case this is a closure.
-        let enclosing_fn_def_id = self.tcx().closure_base_def_id(def_id);
+        let enclosing_fn_def_id = self.tcx().typeck_root_def_id(def_id);
 
         // Get_template_parameters() will append a `<...>` clause to the function
         // name if necessary.
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index c43141c..a7e34b0 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -19,7 +19,7 @@
 use rustc_middle::ty::{self, Ty};
 use rustc_middle::{bug, span_bug};
 use rustc_span::{sym, symbol::kw, Span, Symbol};
-use rustc_target::abi::{self, HasDataLayout, Primitive};
+use rustc_target::abi::{self, Align, HasDataLayout, Primitive};
 use rustc_target::spec::{HasTargetSpec, PanicStrategy};
 
 use std::cmp::Ordering;
@@ -392,13 +392,12 @@
         self.call_intrinsic("llvm.expect.i1", &[cond, self.const_bool(expected)])
     }
 
-    fn sideeffect(&mut self) {
-        // This kind of check would make a ton of sense in the caller, but currently the only
-        // caller of this function is in `rustc_codegen_ssa`, which is agnostic to whether LLVM
-        // codegen backend being used, and so is unable to check the LLVM version.
-        if unsafe { llvm::LLVMRustVersionMajor() } < 12 {
-            self.call_intrinsic("llvm.sideeffect", &[]);
-        }
+    fn type_test(&mut self, pointer: Self::Value, typeid: Self::Value) -> Self::Value {
+        // Test the called operand using llvm.type.test intrinsic. The LowerTypeTests link-time
+        // optimization pass replaces calls to this intrinsic with code to test type membership.
+        let i8p_ty = self.type_i8p();
+        let bitcast = self.bitcast(pointer, i8p_ty);
+        self.call_intrinsic("llvm.type.test", &[bitcast, typeid])
     }
 
     fn va_start(&mut self, va_list: &'ll Value) -> &'ll Value {
@@ -849,28 +848,39 @@
     let arg_tys = sig.inputs();
 
     if name == sym::simd_select_bitmask {
-        let in_ty = arg_tys[0];
-        let m_len = match in_ty.kind() {
-            // Note that this `.unwrap()` crashes for isize/usize, that's sort
-            // of intentional as there's not currently a use case for that.
-            ty::Int(i) => i.bit_width().unwrap(),
-            ty::Uint(i) => i.bit_width().unwrap(),
-            _ => return_error!("`{}` is not an integral type", in_ty),
-        };
         require_simd!(arg_tys[1], "argument");
-        let (v_len, _) = arg_tys[1].simd_size_and_type(bx.tcx());
-        require!(
-            // Allow masks for vectors with fewer than 8 elements to be
-            // represented with a u8 or i8.
-            m_len == v_len || (m_len == 8 && v_len < 8),
-            "mismatched lengths: mask length `{}` != other vector length `{}`",
-            m_len,
-            v_len
-        );
+        let (len, _) = arg_tys[1].simd_size_and_type(bx.tcx());
+
+        let expected_int_bits = (len.max(8) - 1).next_power_of_two();
+        let expected_bytes = len / 8 + ((len % 8 > 0) as u64);
+
+        let mask_ty = arg_tys[0];
+        let mask = match mask_ty.kind() {
+            ty::Int(i) if i.bit_width() == Some(expected_int_bits) => args[0].immediate(),
+            ty::Uint(i) if i.bit_width() == Some(expected_int_bits) => args[0].immediate(),
+            ty::Array(elem, len)
+                if matches!(elem.kind(), ty::Uint(ty::UintTy::U8))
+                    && len.try_eval_usize(bx.tcx, ty::ParamEnv::reveal_all())
+                        == Some(expected_bytes) =>
+            {
+                let place = PlaceRef::alloca(bx, args[0].layout);
+                args[0].val.store(bx, place);
+                let int_ty = bx.type_ix(expected_bytes * 8);
+                let ptr = bx.pointercast(place.llval, bx.cx.type_ptr_to(int_ty));
+                bx.load(int_ty, ptr, Align::ONE)
+            }
+            _ => return_error!(
+                "invalid bitmask `{}`, expected `u{}` or `[u8; {}]`",
+                mask_ty,
+                expected_int_bits,
+                expected_bytes
+            ),
+        };
+
         let i1 = bx.type_i1();
-        let im = bx.type_ix(v_len);
-        let i1xn = bx.type_vector(i1, v_len);
-        let m_im = bx.trunc(args[0].immediate(), im);
+        let im = bx.type_ix(len);
+        let i1xn = bx.type_vector(i1, len);
+        let m_im = bx.trunc(mask, im);
         let m_i1s = bx.bitcast(m_im, i1xn);
         return Ok(bx.select(m_i1s, args[1].immediate(), args[2].immediate()));
     }
@@ -1048,16 +1058,16 @@
 
     if name == sym::simd_bitmask {
         // The `fn simd_bitmask(vector) -> unsigned integer` intrinsic takes a
-        // vector mask and returns an unsigned integer containing the most
-        // significant bit (MSB) of each lane.
-
-        // If the vector has less than 8 lanes, a u8 is returned with zeroed
-        // trailing bits.
+        // vector mask and returns the most significant bit (MSB) of each lane in the form
+        // of either:
+        // * an unsigned integer
+        // * an array of `u8`
+        // If the vector has less than 8 lanes, a u8 is returned with zeroed trailing bits.
+        //
+        // The bit order of the result depends on the byte endianness, LSB-first for little
+        // endian and MSB-first for big endian.
         let expected_int_bits = in_len.max(8);
-        match ret_ty.kind() {
-            ty::Uint(i) if i.bit_width() == Some(expected_int_bits) => (),
-            _ => return_error!("bitmask `{}`, expected `u{}`", ret_ty, expected_int_bits),
-        }
+        let expected_bytes = expected_int_bits / 8 + ((expected_int_bits % 8 > 0) as u64);
 
         // Integer vector <i{in_bitwidth} x in_len>:
         let (i_xn, in_elem_bitwidth) = match in_elem.kind() {
@@ -1087,8 +1097,34 @@
         let i1xn = bx.trunc(i_xn_msb, bx.type_vector(bx.type_i1(), in_len));
         // Bitcast <i1 x N> to iN:
         let i_ = bx.bitcast(i1xn, bx.type_ix(in_len));
-        // Zero-extend iN to the bitmask type:
-        return Ok(bx.zext(i_, bx.type_ix(expected_int_bits)));
+
+        match ret_ty.kind() {
+            ty::Uint(i) if i.bit_width() == Some(expected_int_bits) => {
+                // Zero-extend iN to the bitmask type:
+                return Ok(bx.zext(i_, bx.type_ix(expected_int_bits)));
+            }
+            ty::Array(elem, len)
+                if matches!(elem.kind(), ty::Uint(ty::UintTy::U8))
+                    && len.try_eval_usize(bx.tcx, ty::ParamEnv::reveal_all())
+                        == Some(expected_bytes) =>
+            {
+                // Zero-extend iN to the array lengh:
+                let ze = bx.zext(i_, bx.type_ix(expected_bytes * 8));
+
+                // Convert the integer to a byte array
+                let ptr = bx.alloca(bx.type_ix(expected_bytes * 8), Align::ONE);
+                bx.store(ze, ptr, Align::ONE);
+                let array_ty = bx.type_array(bx.type_i8(), expected_bytes);
+                let ptr = bx.pointercast(ptr, bx.cx.type_ptr_to(array_ty));
+                return Ok(bx.load(array_ty, ptr, Align::ONE));
+            }
+            _ => return_error!(
+                "cannot return `{}`, expected `u{}` or `[u8; {}]`",
+                ret_ty,
+                expected_int_bits,
+                expected_bytes
+            ),
+        }
     }
 
     fn simd_simple_float_intrinsic(
diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs
index 8f4d79e..c66d7d8 100644
--- a/compiler/rustc_codegen_llvm/src/lib.rs
+++ b/compiler/rustc_codegen_llvm/src/lib.rs
@@ -76,6 +76,27 @@
 #[derive(Clone)]
 pub struct LlvmCodegenBackend(());
 
+struct TimeTraceProfiler {
+    enabled: bool,
+}
+
+impl TimeTraceProfiler {
+    fn new(enabled: bool) -> Self {
+        if enabled {
+            unsafe { llvm::LLVMTimeTraceProfilerInitialize() }
+        }
+        TimeTraceProfiler { enabled }
+    }
+}
+
+impl Drop for TimeTraceProfiler {
+    fn drop(&mut self) {
+        if self.enabled {
+            unsafe { llvm::LLVMTimeTraceProfilerFinishThread() }
+        }
+    }
+}
+
 impl ExtraBackendMethods for LlvmCodegenBackend {
     fn new_metadata(&self, tcx: TyCtxt<'_>, mod_name: &str) -> ModuleLlvm {
         ModuleLlvm::new_metadata(tcx, mod_name)
@@ -119,6 +140,34 @@
     fn tune_cpu<'b>(&self, sess: &'b Session) -> Option<&'b str> {
         llvm_util::tune_cpu(sess)
     }
+
+    fn spawn_thread<F, T>(time_trace: bool, f: F) -> std::thread::JoinHandle<T>
+    where
+        F: FnOnce() -> T,
+        F: Send + 'static,
+        T: Send + 'static,
+    {
+        std::thread::spawn(move || {
+            let _profiler = TimeTraceProfiler::new(time_trace);
+            f()
+        })
+    }
+
+    fn spawn_named_thread<F, T>(
+        time_trace: bool,
+        name: String,
+        f: F,
+    ) -> std::io::Result<std::thread::JoinHandle<T>>
+    where
+        F: FnOnce() -> T,
+        F: Send + 'static,
+        T: Send + 'static,
+    {
+        std::thread::Builder::new().name(name).spawn(move || {
+            let _profiler = TimeTraceProfiler::new(time_trace);
+            f()
+        })
+    }
 }
 
 impl WriteBackendMethods for LlvmCodegenBackend {
@@ -239,6 +288,31 @@
                 }
                 println!();
             }
+            PrintRequest::StackProtectorStrategies => {
+                println!(
+                    r#"Available stack protector strategies:
+    all
+        Generate stack canaries in all functions.
+
+    strong
+        Generate stack canaries in a function if it either:
+        - has a local variable of `[T; N]` type, regardless of `T` and `N`
+        - takes the address of a local variable.
+
+          (Note that a local variable being borrowed is not equivalent to its
+          address being taken: e.g. some borrows may be removed by optimization,
+          while by-value argument passing may be implemented with reference to a
+          local stack variable in the ABI.)
+
+    basic
+        Generate stack canaries in functions with:
+        - local variables of `[T; N]` type, where `T` is byte-sized and `N` > 8.
+
+    none
+        Do not generate stack canaries.
+"#
+                );
+            }
             req => llvm_util::print(req, sess),
         }
     }
diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
index 4c9ae4f..1d255c0 100644
--- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
@@ -166,6 +166,9 @@
     InaccessibleMemOnly = 27,
     SanitizeHWAddress = 28,
     WillReturn = 29,
+    StackProtectReq = 30,
+    StackProtectStrong = 31,
+    StackProtect = 32,
 }
 
 /// LLVMIntPredicate
@@ -416,6 +419,7 @@
     MD_nontemporal = 9,
     MD_mem_parallel_loop_access = 10,
     MD_nonnull = 11,
+    MD_type = 19,
 }
 
 /// LLVMRustAsmDialect
@@ -1002,6 +1006,8 @@
     pub fn LLVMSetValueName2(Val: &Value, Name: *const c_char, NameLen: size_t);
     pub fn LLVMReplaceAllUsesWith(OldVal: &'a Value, NewVal: &'a Value);
     pub fn LLVMSetMetadata(Val: &'a Value, KindID: c_uint, Node: &'a Value);
+    pub fn LLVMGlobalSetMetadata(Val: &'a Value, KindID: c_uint, Metadata: &'a Metadata);
+    pub fn LLVMValueAsMetadata(Node: &'a Value) -> &Metadata;
 
     // Operations on constants of any type
     pub fn LLVMConstNull(Ty: &Type) -> &Value;
@@ -1734,6 +1740,8 @@
 
     pub fn LLVMTimeTraceProfilerInitialize();
 
+    pub fn LLVMTimeTraceProfilerFinishThread();
+
     pub fn LLVMTimeTraceProfilerFinish(FileName: *const c_char);
 
     pub fn LLVMAddAnalysisPasses(T: &'a TargetMachine, PM: &PassManager<'a>);
@@ -1770,7 +1778,7 @@
 
     pub fn LLVMDisposeMessage(message: *mut c_char);
 
-    pub fn LLVMStartMultithreaded() -> Bool;
+    pub fn LLVMIsMultithreaded() -> Bool;
 
     /// Returns a string describing the last error caused by an LLVMRust* call.
     pub fn LLVMRustGetLastError() -> *const c_char;
@@ -2187,6 +2195,7 @@
         UseSoftFP: bool,
         FunctionSections: bool,
         DataSections: bool,
+        UniqueSectionNames: bool,
         TrapUnreachable: bool,
         Singlethread: bool,
         AsmComments: bool,
diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs
index c2136f1..3393c9b 100644
--- a/compiler/rustc_codegen_llvm/src/llvm_util.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs
@@ -17,35 +17,25 @@
 use std::ptr;
 use std::slice;
 use std::str;
-use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Once;
 
-static POISONED: AtomicBool = AtomicBool::new(false);
 static INIT: Once = Once::new();
 
 pub(crate) fn init(sess: &Session) {
     unsafe {
         // Before we touch LLVM, make sure that multithreading is enabled.
+        if llvm::LLVMIsMultithreaded() != 1 {
+            bug!("LLVM compiled without support for threads");
+        }
         INIT.call_once(|| {
-            if llvm::LLVMStartMultithreaded() != 1 {
-                // use an extra bool to make sure that all future usage of LLVM
-                // cannot proceed despite the Once not running more than once.
-                POISONED.store(true, Ordering::SeqCst);
-            }
-
             configure_llvm(sess);
         });
-
-        if POISONED.load(Ordering::SeqCst) {
-            bug!("couldn't enable multi-threaded LLVM");
-        }
     }
 }
 
 fn require_inited() {
-    INIT.call_once(|| bug!("llvm is not initialized"));
-    if POISONED.load(Ordering::SeqCst) {
-        bug!("couldn't enable multi-threaded LLVM");
+    if !INIT.is_completed() {
+        bug!("LLVM is not initialized");
     }
 }
 
@@ -85,7 +75,9 @@
         if sess.print_llvm_passes() {
             add("-debug-pass=Structure", false);
         }
-        if !sess.opts.debugging_opts.no_generate_arange_section {
+        if sess.target.generate_arange_section
+            && !sess.opts.debugging_opts.no_generate_arange_section
+        {
             add("-generate-arange-section", false);
         }
 
@@ -95,8 +87,7 @@
         // Ref:
         // - https://github.com/rust-lang/rust/issues/85351
         // - https://reviews.llvm.org/D103167
-        let llvm_version = llvm_util::get_version();
-        if llvm_version >= (11, 0, 0) && llvm_version < (13, 0, 0) {
+        if llvm_util::get_version() < (13, 0, 0) {
             add("-enable-machine-outliner=never", false);
         }
 
@@ -124,11 +115,6 @@
     }
 
     if sess.opts.debugging_opts.llvm_time_trace {
-        // time-trace is not thread safe and running it in parallel will cause seg faults.
-        if !sess.opts.debugging_opts.no_parallel_llvm {
-            bug!("`-Z llvm-time-trace` requires `-Z no-parallel-llvm")
-        }
-
         llvm::LLVMTimeTraceProfilerInitialize();
     }
 
@@ -191,6 +177,7 @@
         ("aarch64", "dpb2") => vec!["ccdp"],
         ("aarch64", "frintts") => vec!["fptoint"],
         ("aarch64", "fcma") => vec!["complxnum"],
+        ("aarch64", "pmuv3") => vec!["perfmon"],
         (_, s) => vec![s],
     }
 }
@@ -416,14 +403,6 @@
     // -Ctarget-features
     features.extend(sess.opts.cg.target_feature.split(',').flat_map(&filter));
 
-    // FIXME: Move outline-atomics to target definition when earliest supported LLVM is 12.
-    if get_version() >= (12, 0, 0)
-        && sess.target.llvm_target.contains("aarch64-unknown-linux")
-        && sess.target.llvm_target != "aarch64-unknown-linux-musl"
-    {
-        features.push("+outline-atomics".to_string());
-    }
-
     features
 }