Merge ref '82310651b93a' from rust-lang/rust

Pull recent changes from https://github.com/rust-lang/rust via Josh.

Upstream ref: 82310651b93a594a3fd69015e1562186a080d94c
Filtered ref: e13c0be8f13737c64082b89ce834546079767ac4

This merge was created using https://github.com/rust-lang/josh-sync.

25 files changed, 955 insertions, 1635 deletions

diff --git a/compiler/rustc_codegen_llvm/src/abi.rs b/compiler/rustc_codegen_llvm/src/abi.rs
index aba63d75f1d..009e7e2487b 100644
--- a/compiler/rustc_codegen_llvm/src/abi.rs
+++ b/compiler/rustc_codegen_llvm/src/abi.rs
@@ -146,7 +146,7 @@ impl LlvmType for CastTarget {
                 "total size {:?} cannot be divided into units of zero size",
                 self.rest.total
             );
-            if self.rest.total.bytes() % self.rest.unit.size.bytes() != 0 {
+            if !self.rest.total.bytes().is_multiple_of(self.rest.unit.size.bytes()) {
                 assert_eq!(self.rest.unit.kind, RegKind::Integer, "only int regs can be split");
             }
             self.rest.total.bytes().div_ceil(self.rest.unit.size.bytes())
@@ -229,7 +229,7 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> {
                 let llscratch = bx.alloca(scratch_size, scratch_align);
                 bx.lifetime_start(llscratch, scratch_size);
                 // ...store the value...
-                bx.store(val, llscratch, scratch_align);
+                rustc_codegen_ssa::mir::store_cast(bx, cast, val, llscratch, scratch_align);
                 // ... and then memcpy it to the intended destination.
                 bx.memcpy(
                     dst.val.llval,
diff --git a/compiler/rustc_codegen_llvm/src/allocator.rs b/compiler/rustc_codegen_llvm/src/allocator.rs
index 4a78e694979..2b5090ed6db 100644
--- a/compiler/rustc_codegen_llvm/src/allocator.rs
+++ b/compiler/rustc_codegen_llvm/src/allocator.rs
@@ -11,7 +11,7 @@ use rustc_symbol_mangling::mangle_internal_symbol;
 
 use crate::builder::SBuilder;
 use crate::declare::declare_simple_fn;
-use crate::llvm::{self, False, True, Type};
+use crate::llvm::{self, False, True, Type, Value};
 use crate::{SimpleCx, attributes, debuginfo};
 
 pub(crate) unsafe fn codegen(
@@ -57,7 +57,7 @@ pub(crate) unsafe fn codegen(
             let from_name = mangle_internal_symbol(tcx, &global_fn_name(method.name));
             let to_name = mangle_internal_symbol(tcx, &default_fn_name(method.name));
 
-            create_wrapper_function(tcx, &cx, &from_name, &to_name, &args, output, false);
+            create_wrapper_function(tcx, &cx, &from_name, Some(&to_name), &args, output, false);
         }
     }
 
@@ -66,26 +66,32 @@ pub(crate) unsafe fn codegen(
         tcx,
         &cx,
         &mangle_internal_symbol(tcx, "__rust_alloc_error_handler"),
-        &mangle_internal_symbol(tcx, alloc_error_handler_name(alloc_error_handler_kind)),
+        Some(&mangle_internal_symbol(tcx, alloc_error_handler_name(alloc_error_handler_kind))),
         &[usize, usize], // size, align
         None,
         true,
     );
 
     unsafe {
-        // __rust_alloc_error_handler_should_panic
-        let name = mangle_internal_symbol(tcx, OomStrategy::SYMBOL);
-        let ll_g = cx.declare_global(&name, i8);
-        llvm::set_visibility(ll_g, llvm::Visibility::from_generic(tcx.sess.default_visibility()));
-        let val = tcx.sess.opts.unstable_opts.oom.should_panic();
-        let llval = llvm::LLVMConstInt(i8, val as u64, False);
-        llvm::set_initializer(ll_g, llval);
-
-        let name = mangle_internal_symbol(tcx, NO_ALLOC_SHIM_IS_UNSTABLE);
-        let ll_g = cx.declare_global(&name, i8);
-        llvm::set_visibility(ll_g, llvm::Visibility::from_generic(tcx.sess.default_visibility()));
-        let llval = llvm::LLVMConstInt(i8, 0, False);
-        llvm::set_initializer(ll_g, llval);
+        // __rust_alloc_error_handler_should_panic_v2
+        create_const_value_function(
+            tcx,
+            &cx,
+            &mangle_internal_symbol(tcx, OomStrategy::SYMBOL),
+            &i8,
+            &llvm::LLVMConstInt(i8, tcx.sess.opts.unstable_opts.oom.should_panic() as u64, False),
+        );
+
+        // __rust_no_alloc_shim_is_unstable_v2
+        create_wrapper_function(
+            tcx,
+            &cx,
+            &mangle_internal_symbol(tcx, NO_ALLOC_SHIM_IS_UNSTABLE),
+            None,
+            &[],
+            None,
+            false,
+        );
     }
 
     if tcx.sess.opts.debuginfo != DebugInfo::None {
@@ -95,11 +101,39 @@ pub(crate) unsafe fn codegen(
     }
 }
 
+fn create_const_value_function(
+    tcx: TyCtxt<'_>,
+    cx: &SimpleCx<'_>,
+    name: &str,
+    output: &Type,
+    value: &Value,
+) {
+    let ty = cx.type_func(&[], output);
+    let llfn = declare_simple_fn(
+        &cx,
+        name,
+        llvm::CallConv::CCallConv,
+        llvm::UnnamedAddr::Global,
+        llvm::Visibility::from_generic(tcx.sess.default_visibility()),
+        ty,
+    );
+
+    attributes::apply_to_llfn(
+        llfn,
+        llvm::AttributePlace::Function,
+        &[llvm::AttributeKind::AlwaysInline.create_attr(cx.llcx)],
+    );
+
+    let llbb = unsafe { llvm::LLVMAppendBasicBlockInContext(cx.llcx, llfn, c"entry".as_ptr()) };
+    let mut bx = SBuilder::build(&cx, llbb);
+    bx.ret(value);
+}
+
 fn create_wrapper_function(
     tcx: TyCtxt<'_>,
     cx: &SimpleCx<'_>,
     from_name: &str,
-    to_name: &str,
+    to_name: Option<&str>,
     args: &[&Type],
     output: Option<&Type>,
     no_return: bool,
@@ -128,33 +162,38 @@ fn create_wrapper_function(
         attributes::apply_to_llfn(llfn, llvm::AttributePlace::Function, &[uwtable]);
     }
 
-    let callee = declare_simple_fn(
-        &cx,
-        to_name,
-        llvm::CallConv::CCallConv,
-        llvm::UnnamedAddr::Global,
-        llvm::Visibility::Hidden,
-        ty,
-    );
-    if let Some(no_return) = no_return {
-        // -> ! DIFlagNoReturn
-        attributes::apply_to_llfn(callee, llvm::AttributePlace::Function, &[no_return]);
-    }
-    llvm::set_visibility(callee, llvm::Visibility::Hidden);
-
     let llbb = unsafe { llvm::LLVMAppendBasicBlockInContext(cx.llcx, llfn, c"entry".as_ptr()) };
-
     let mut bx = SBuilder::build(&cx, llbb);
-    let args = args
-        .iter()
-        .enumerate()
-        .map(|(i, _)| llvm::get_param(llfn, i as c_uint))
-        .collect::<Vec<_>>();
-    let ret = bx.call(ty, callee, &args, None);
-    llvm::LLVMSetTailCall(ret, True);
-    if output.is_some() {
-        bx.ret(ret);
+
+    if let Some(to_name) = to_name {
+        let callee = declare_simple_fn(
+            &cx,
+            to_name,
+            llvm::CallConv::CCallConv,
+            llvm::UnnamedAddr::Global,
+            llvm::Visibility::Hidden,
+            ty,
+        );
+        if let Some(no_return) = no_return {
+            // -> ! DIFlagNoReturn
+            attributes::apply_to_llfn(callee, llvm::AttributePlace::Function, &[no_return]);
+        }
+        llvm::set_visibility(callee, llvm::Visibility::Hidden);
+
+        let args = args
+            .iter()
+            .enumerate()
+            .map(|(i, _)| llvm::get_param(llfn, i as c_uint))
+            .collect::<Vec<_>>();
+        let ret = bx.call(ty, callee, &args, None);
+        llvm::LLVMSetTailCall(ret, True);
+        if output.is_some() {
+            bx.ret(ret);
+        } else {
+            bx.ret_void()
+        }
     } else {
+        assert!(output.is_none());
         bx.ret_void()
     }
 }
diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs
index 4185aef8b31..a643a91141e 100644
--- a/compiler/rustc_codegen_llvm/src/asm.rs
+++ b/compiler/rustc_codegen_llvm/src/asm.rs
@@ -384,15 +384,19 @@ impl<'tcx> AsmCodegenMethods<'tcx> for CodegenCx<'_, 'tcx> {
     ) {
         let asm_arch = self.tcx.sess.asm_arch.unwrap();
 
-        // Default to Intel syntax on x86
-        let intel_syntax = matches!(asm_arch, InlineAsmArch::X86 | InlineAsmArch::X86_64)
-            && !options.contains(InlineAsmOptions::ATT_SYNTAX);
-
         // Build the template string
         let mut template_str = String::new();
-        if intel_syntax {
-            template_str.push_str(".intel_syntax\n");
+
+        // On X86 platforms there are two assembly syntaxes. Rust uses intel by default,
+        // but AT&T can be specified explicitly.
+        if matches!(asm_arch, InlineAsmArch::X86 | InlineAsmArch::X86_64) {
+            if options.contains(InlineAsmOptions::ATT_SYNTAX) {
+                template_str.push_str(".att_syntax\n")
+            } else {
+                template_str.push_str(".intel_syntax\n")
+            }
         }
+
         for piece in template {
             match *piece {
                 InlineAsmTemplatePiece::String(ref s) => template_str.push_str(s),
@@ -431,7 +435,11 @@ impl<'tcx> AsmCodegenMethods<'tcx> for CodegenCx<'_, 'tcx> {
                 }
             }
         }
-        if intel_syntax {
+
+        // Just to play it safe, if intel was used, reset the assembly syntax to att.
+        if matches!(asm_arch, InlineAsmArch::X86 | InlineAsmArch::X86_64)
+            && !options.contains(InlineAsmOptions::ATT_SYNTAX)
+        {
             template_str.push_str("\n.att_syntax\n");
         }
 
@@ -1021,6 +1029,15 @@ fn llvm_fixup_input<'ll, 'tcx>(
         ) if element.primitive() == Primitive::Float(Float::F16) => {
             bx.bitcast(value, bx.type_vector(bx.type_i16(), count))
         }
+        (LoongArch(LoongArchInlineAsmRegClass::freg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F16) =>
+        {
+            // Smaller floats are always "NaN-boxed" inside larger floats on LoongArch.
+            let value = bx.bitcast(value, bx.type_i16());
+            let value = bx.zext(value, bx.type_i32());
+            let value = bx.or(value, bx.const_u32(0xFFFF_0000));
+            bx.bitcast(value, bx.type_f32())
+        }
         (Mips(MipsInlineAsmRegClass::reg), BackendRepr::Scalar(s)) => {
             match s.primitive() {
                 // MIPS only supports register-length arithmetics.
@@ -1178,6 +1195,13 @@ fn llvm_fixup_output<'ll, 'tcx>(
         ) if element.primitive() == Primitive::Float(Float::F16) => {
             bx.bitcast(value, bx.type_vector(bx.type_f16(), count))
         }
+        (LoongArch(LoongArchInlineAsmRegClass::freg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F16) =>
+        {
+            let value = bx.bitcast(value, bx.type_i32());
+            let value = bx.trunc(value, bx.type_i16());
+            bx.bitcast(value, bx.type_f16())
+        }
         (Mips(MipsInlineAsmRegClass::reg), BackendRepr::Scalar(s)) => {
             match s.primitive() {
                 // MIPS only supports register-length arithmetics.
@@ -1318,6 +1342,11 @@ fn llvm_fixup_output_type<'ll, 'tcx>(
         ) if element.primitive() == Primitive::Float(Float::F16) => {
             cx.type_vector(cx.type_i16(), count)
         }
+        (LoongArch(LoongArchInlineAsmRegClass::freg), BackendRepr::Scalar(s))
+            if s.primitive() == Primitive::Float(Float::F16) =>
+        {
+            cx.type_f32()
+        }
         (Mips(MipsInlineAsmRegClass::reg), BackendRepr::Scalar(s)) => {
             match s.primitive() {
                 // MIPS only supports register-length arithmetics.
diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs
index 27fd09745ff..c32f11b27f3 100644
--- a/compiler/rustc_codegen_llvm/src/attributes.rs
+++ b/compiler/rustc_codegen_llvm/src/attributes.rs
@@ -344,7 +344,7 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
     llfn: &'ll Value,
     instance: ty::Instance<'tcx>,
 ) {
-    let codegen_fn_attrs = cx.tcx.codegen_fn_attrs(instance.def_id());
+    let codegen_fn_attrs = cx.tcx.codegen_instance_attrs(instance.def);
 
     let mut to_add = SmallVec::<[_; 16]>::new();
 
@@ -370,22 +370,6 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
     };
     to_add.extend(inline_attr(cx, inline));
 
-    // The `uwtable` attribute according to LLVM is:
-    //
-    //     This attribute indicates that the ABI being targeted requires that an
-    //     unwind table entry be produced for this function even if we can show
-    //     that no exceptions passes by it. This is normally the case for the
-    //     ELF x86-64 abi, but it can be disabled for some compilation units.
-    //
-    // Typically when we're compiling with `-C panic=abort` (which implies this
-    // `no_landing_pads` check) we don't need `uwtable` because we can't
-    // generate any exceptions! On Windows, however, exceptions include other
-    // events such as illegal instructions, segfaults, etc. This means that on
-    // Windows we end up still needing the `uwtable` attribute even if the `-C
-    // panic=abort` flag is passed.
-    //
-    // You can also find more info on why Windows always requires uwtables here:
-    //      https://bugzilla.mozilla.org/show_bug.cgi?id=1302078
     if cx.sess().must_emit_unwind_tables() {
         to_add.push(uwtable_attr(cx.llcx, cx.sess().opts.unstable_opts.use_sync_unwind));
     }
@@ -491,11 +475,7 @@ pub(crate) fn llfn_attrs_from_instance<'ll, 'tcx>(
         let allocated_pointer = AttributeKind::AllocatedPointer.create_attr(cx.llcx);
         attributes::apply_to_llfn(llfn, AttributePlace::Argument(0), &[allocated_pointer]);
     }
-    // function alignment can be set globally with the `-Zmin-function-alignment=<n>` flag;
-    // the alignment from a `#[repr(align(<n>))]` is used if it specifies a higher alignment.
-    if let Some(align) =
-        Ord::max(cx.tcx.sess.opts.unstable_opts.min_function_alignment, codegen_fn_attrs.alignment)
-    {
+    if let Some(align) = codegen_fn_attrs.alignment {
         llvm::set_alignment(llfn, align);
     }
     if let Some(backchain) = backchain_attr(cx) {
diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs
index ee46b49a094..655e1c95373 100644
--- a/compiler/rustc_codegen_llvm/src/back/lto.rs
+++ b/compiler/rustc_codegen_llvm/src/back/lto.rs
@@ -7,7 +7,7 @@ use std::sync::Arc;
 use std::{io, iter, slice};
 
 use object::read::archive::ArchiveFile;
-use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule, ThinShared};
+use rustc_codegen_ssa::back::lto::{SerializedModule, ThinModule, ThinShared};
 use rustc_codegen_ssa::back::symbol_export;
 use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput};
 use rustc_codegen_ssa::traits::*;
@@ -201,7 +201,7 @@ pub(crate) fn run_fat(
     cgcx: &CodegenContext<LlvmCodegenBackend>,
     modules: Vec<FatLtoInput<LlvmCodegenBackend>>,
     cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
-) -> Result<LtoModuleCodegen<LlvmCodegenBackend>, FatalError> {
+) -> Result<ModuleCodegen<ModuleLlvm>, FatalError> {
     let dcx = cgcx.create_dcx();
     let dcx = dcx.handle();
     let (symbols_below_threshold, upstream_modules) = prepare_lto(cgcx, dcx)?;
@@ -217,7 +217,7 @@ pub(crate) fn run_thin(
     cgcx: &CodegenContext<LlvmCodegenBackend>,
     modules: Vec<(String, ThinBuffer)>,
     cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
-) -> Result<(Vec<LtoModuleCodegen<LlvmCodegenBackend>>, Vec<WorkProduct>), FatalError> {
+) -> Result<(Vec<ThinModule<LlvmCodegenBackend>>, Vec<WorkProduct>), FatalError> {
     let dcx = cgcx.create_dcx();
     let dcx = dcx.handle();
     let (symbols_below_threshold, upstream_modules) = prepare_lto(cgcx, dcx)?;
@@ -248,7 +248,7 @@ fn fat_lto(
     cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
     mut serialized_modules: Vec<(SerializedModule<ModuleBuffer>, CString)>,
     symbols_below_threshold: &[*const libc::c_char],
-) -> Result<LtoModuleCodegen<LlvmCodegenBackend>, FatalError> {
+) -> Result<ModuleCodegen<ModuleLlvm>, FatalError> {
     let _timer = cgcx.prof.generic_activity("LLVM_fat_lto_build_monolithic_module");
     info!("going for a fat lto");
 
@@ -366,7 +366,7 @@ fn fat_lto(
         save_temp_bitcode(cgcx, &module, "lto.after-restriction");
     }
 
-    Ok(LtoModuleCodegen::Fat(module))
+    Ok(module)
 }
 
 pub(crate) struct Linker<'a>(&'a mut llvm::Linker<'a>);
@@ -436,7 +436,7 @@ fn thin_lto(
     serialized_modules: Vec<(SerializedModule<ModuleBuffer>, CString)>,
     cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
     symbols_below_threshold: &[*const libc::c_char],
-) -> Result<(Vec<LtoModuleCodegen<LlvmCodegenBackend>>, Vec<WorkProduct>), FatalError> {
+) -> Result<(Vec<ThinModule<LlvmCodegenBackend>>, Vec<WorkProduct>), FatalError> {
     let _timer = cgcx.prof.generic_activity("LLVM_thin_lto_global_analysis");
     unsafe {
         info!("going for that thin, thin LTO");
@@ -568,10 +568,7 @@ fn thin_lto(
             }
 
             info!(" - {}: re-compiled", module_name);
-            opt_jobs.push(LtoModuleCodegen::Thin(ThinModule {
-                shared: Arc::clone(&shared),
-                idx: module_index,
-            }));
+            opt_jobs.push(ThinModule { shared: Arc::clone(&shared), idx: module_index });
         }
 
         // Save the current ThinLTO import information for the next compilation
@@ -587,7 +584,7 @@ fn thin_lto(
 }
 
 fn enable_autodiff_settings(ad: &[config::AutoDiff]) {
-    for &val in ad {
+    for val in ad {
         // We intentionally don't use a wildcard, to not forget handling anything new.
         match val {
             config::AutoDiff::PrintPerf => {
@@ -599,6 +596,10 @@ fn enable_autodiff_settings(ad: &[config::AutoDiff]) {
             config::AutoDiff::PrintTA => {
                 llvm::set_print_type(true);
             }
+            config::AutoDiff::PrintTAFn(fun) => {
+                llvm::set_print_type(true); // Enable general type printing
+                llvm::set_print_type_fun(&fun); // Set specific function to analyze
+            }
             config::AutoDiff::Inline => {
                 llvm::set_inline(true);
             }
@@ -676,7 +677,7 @@ pub(crate) fn run_pass_manager(
             if attributes::has_string_attr(function, enzyme_marker) {
                 // Sanity check: Ensure 'noinline' is present before replacing it.
                 assert!(
-                    !attributes::has_attr(function, Function, llvm::AttributeKind::NoInline),
+                    attributes::has_attr(function, Function, llvm::AttributeKind::NoInline),
                     "Expected __enzyme function to have 'noinline' before adding 'alwaysinline'"
                 );
 
diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs
index bde6a9cf4bc..68279008c03 100644
--- a/compiler/rustc_codegen_llvm/src/back/write.rs
+++ b/compiler/rustc_codegen_llvm/src/back/write.rs
@@ -817,10 +817,12 @@ pub(crate) fn link(
 
 pub(crate) fn codegen(
     cgcx: &CodegenContext<LlvmCodegenBackend>,
-    dcx: DiagCtxtHandle<'_>,
     module: ModuleCodegen<ModuleLlvm>,
     config: &ModuleConfig,
 ) -> Result<CompiledModule, FatalError> {
+    let dcx = cgcx.create_dcx();
+    let dcx = dcx.handle();
+
     let _timer = cgcx.prof.generic_activity_with_arg("LLVM_module_codegen", &*module.name);
     {
         let llmod = module.module_llvm.llmod();
@@ -879,9 +881,7 @@ pub(crate) fn codegen(
                     .generic_activity_with_arg("LLVM_module_codegen_embed_bitcode", &*module.name);
                 let thin_bc =
                     module.thin_lto_buffer.as_deref().expect("cannot find embedded bitcode");
-                unsafe {
-                    embed_bitcode(cgcx, llcx, llmod, &config.bc_cmdline, &thin_bc);
-                }
+                embed_bitcode(cgcx, llcx, llmod, &config.bc_cmdline, &thin_bc);
             }
         }
 
@@ -945,7 +945,7 @@ pub(crate) fn codegen(
             // binaries. So we must clone the module to produce the asm output
             // if we are also producing object code.
             let llmod = if let EmitObj::ObjectCode(_) = config.emit_obj {
-                unsafe { llvm::LLVMCloneModule(llmod) }
+                llvm::LLVMCloneModule(llmod)
             } else {
                 llmod
             };
@@ -1073,7 +1073,7 @@ pub(crate) fn bitcode_section_name(cgcx: &CodegenContext<LlvmCodegenBackend>) ->
 }
 
 /// Embed the bitcode of an LLVM module for LTO in the LLVM module itself.
-unsafe fn embed_bitcode(
+fn embed_bitcode(
     cgcx: &CodegenContext<LlvmCodegenBackend>,
     llcx: &llvm::Context,
     llmod: &llvm::Module,
@@ -1115,43 +1115,40 @@ unsafe fn embed_bitcode(
     // Unfortunately, LLVM provides no way to set custom section flags. For ELF
     // and COFF we emit the sections using module level inline assembly for that
     // reason (see issue #90326 for historical background).
-    unsafe {
-        if cgcx.target_is_like_darwin
-            || cgcx.target_is_like_aix
-            || cgcx.target_arch == "wasm32"
-            || cgcx.target_arch == "wasm64"
-        {
-            // We don't need custom section flags, create LLVM globals.
-            let llconst = common::bytes_in_context(llcx, bitcode);
-            let llglobal =
-                llvm::add_global(llmod, common::val_ty(llconst), c"rustc.embedded.module");
-            llvm::set_initializer(llglobal, llconst);
-
-            llvm::set_section(llglobal, bitcode_section_name(cgcx));
-            llvm::set_linkage(llglobal, llvm::Linkage::PrivateLinkage);
-            llvm::LLVMSetGlobalConstant(llglobal, llvm::True);
-
-            let llconst = common::bytes_in_context(llcx, cmdline.as_bytes());
-            let llglobal =
-                llvm::add_global(llmod, common::val_ty(llconst), c"rustc.embedded.cmdline");
-            llvm::set_initializer(llglobal, llconst);
-            let section = if cgcx.target_is_like_darwin {
-                c"__LLVM,__cmdline"
-            } else if cgcx.target_is_like_aix {
-                c".info"
-            } else {
-                c".llvmcmd"
-            };
-            llvm::set_section(llglobal, section);
-            llvm::set_linkage(llglobal, llvm::Linkage::PrivateLinkage);
+
+    if cgcx.target_is_like_darwin
+        || cgcx.target_is_like_aix
+        || cgcx.target_arch == "wasm32"
+        || cgcx.target_arch == "wasm64"
+    {
+        // We don't need custom section flags, create LLVM globals.
+        let llconst = common::bytes_in_context(llcx, bitcode);
+        let llglobal = llvm::add_global(llmod, common::val_ty(llconst), c"rustc.embedded.module");
+        llvm::set_initializer(llglobal, llconst);
+
+        llvm::set_section(llglobal, bitcode_section_name(cgcx));
+        llvm::set_linkage(llglobal, llvm::Linkage::PrivateLinkage);
+        llvm::LLVMSetGlobalConstant(llglobal, llvm::True);
+
+        let llconst = common::bytes_in_context(llcx, cmdline.as_bytes());
+        let llglobal = llvm::add_global(llmod, common::val_ty(llconst), c"rustc.embedded.cmdline");
+        llvm::set_initializer(llglobal, llconst);
+        let section = if cgcx.target_is_like_darwin {
+            c"__LLVM,__cmdline"
+        } else if cgcx.target_is_like_aix {
+            c".info"
         } else {
-            // We need custom section flags, so emit module-level inline assembly.
-            let section_flags = if cgcx.is_pe_coff { "n" } else { "e" };
-            let asm = create_section_with_flags_asm(".llvmbc", section_flags, bitcode);
-            llvm::append_module_inline_asm(llmod, &asm);
-            let asm = create_section_with_flags_asm(".llvmcmd", section_flags, cmdline.as_bytes());
-            llvm::append_module_inline_asm(llmod, &asm);
-        }
+            c".llvmcmd"
+        };
+        llvm::set_section(llglobal, section);
+        llvm::set_linkage(llglobal, llvm::Linkage::PrivateLinkage);
+    } else {
+        // We need custom section flags, so emit module-level inline assembly.
+        let section_flags = if cgcx.is_pe_coff { "n" } else { "e" };
+        let asm = create_section_with_flags_asm(".llvmbc", section_flags, bitcode);
+        llvm::append_module_inline_asm(llmod, &asm);
+        let asm = create_section_with_flags_asm(".llvmcmd", section_flags, cmdline.as_bytes());
+        llvm::append_module_inline_asm(llmod, &asm);
     }
 }
 
@@ -1182,7 +1179,7 @@ fn create_msvc_imps(
         .filter_map(|val| {
             // Exclude some symbols that we know are not Rust symbols.
             let name = llvm::get_value_name(val);
-            if ignored(name) { None } else { Some((val, name)) }
+            if ignored(&name) { None } else { Some((val, name)) }
         })
         .map(move |(val, name)| {
             let mut imp_name = prefix.as_bytes().to_vec();
diff --git a/compiler/rustc_codegen_llvm/src/builder.rs b/compiler/rustc_codegen_llvm/src/builder.rs
index ec006b59192..514923ad6f3 100644
--- a/compiler/rustc_codegen_llvm/src/builder.rs
+++ b/compiler/rustc_codegen_llvm/src/builder.rs
@@ -14,7 +14,6 @@ use rustc_codegen_ssa::mir::place::PlaceRef;
 use rustc_codegen_ssa::traits::*;
 use rustc_data_structures::small_c_str::SmallCStr;
 use rustc_hir::def_id::DefId;
-use rustc_middle::bug;
 use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrs;
 use rustc_middle::ty::layout::{
     FnAbiError, FnAbiOfHelpers, FnAbiRequest, HasTypingEnv, LayoutError, LayoutOfHelpers,
@@ -303,10 +302,7 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
             return;
         }
 
-        let id_str = "branch_weights";
-        let id = unsafe {
-            llvm::LLVMMDStringInContext2(self.cx.llcx, id_str.as_ptr().cast(), id_str.len())
-        };
+        let id = self.cx.create_metadata(b"branch_weights");
 
         // For switch instructions with 2 targets, the `llvm.expect` intrinsic is used.
         // This function handles switch instructions with more than 2 targets and it needs to
@@ -484,73 +480,31 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
     fn checked_binop(
         &mut self,
         oop: OverflowOp,
-        ty: Ty<'_>,
+        ty: Ty<'tcx>,
         lhs: Self::Value,
         rhs: Self::Value,
     ) -> (Self::Value, Self::Value) {
-        use rustc_middle::ty::IntTy::*;
-        use rustc_middle::ty::UintTy::*;
-        use rustc_middle::ty::{Int, Uint};
-
-        let new_kind = match ty.kind() {
-            Int(t @ Isize) => Int(t.normalize(self.tcx.sess.target.pointer_width)),
-            Uint(t @ Usize) => Uint(t.normalize(self.tcx.sess.target.pointer_width)),
-            t @ (Uint(_) | Int(_)) => *t,
-            _ => panic!("tried to get overflow intrinsic for op applied to non-int type"),
+        let (size, signed) = ty.int_size_and_signed(self.tcx);
+        let width = size.bits();
+
+        if oop == OverflowOp::Sub && !signed {
+            // Emit sub and icmp instead of llvm.usub.with.overflow. LLVM considers these
+            // to be the canonical form. It will attempt to reform llvm.usub.with.overflow
+            // in the backend if profitable.
+            let sub = self.sub(lhs, rhs);
+            let cmp = self.icmp(IntPredicate::IntULT, lhs, rhs);
+            return (sub, cmp);
+        }
+
+        let oop_str = match oop {
+            OverflowOp::Add => "add",
+            OverflowOp::Sub => "sub",
+            OverflowOp::Mul => "mul",
         };
 
-        let name = match oop {
-            OverflowOp::Add => match new_kind {
-                Int(I8) => "llvm.sadd.with.overflow.i8",
-                Int(I16) => "llvm.sadd.with.overflow.i16",
-                Int(I32) => "llvm.sadd.with.overflow.i32",
-                Int(I64) => "llvm.sadd.with.overflow.i64",
-                Int(I128) => "llvm.sadd.with.overflow.i128",
-
-                Uint(U8) => "llvm.uadd.with.overflow.i8",
-                Uint(U16) => "llvm.uadd.with.overflow.i16",
-                Uint(U32) => "llvm.uadd.with.overflow.i32",
-                Uint(U64) => "llvm.uadd.with.overflow.i64",
-                Uint(U128) => "llvm.uadd.with.overflow.i128",
-
-                _ => unreachable!(),
-            },
-            OverflowOp::Sub => match new_kind {
-                Int(I8) => "llvm.ssub.with.overflow.i8",
-                Int(I16) => "llvm.ssub.with.overflow.i16",
-                Int(I32) => "llvm.ssub.with.overflow.i32",
-                Int(I64) => "llvm.ssub.with.overflow.i64",
-                Int(I128) => "llvm.ssub.with.overflow.i128",
-
-                Uint(_) => {
-                    // Emit sub and icmp instead of llvm.usub.with.overflow. LLVM considers these
-                    // to be the canonical form. It will attempt to reform llvm.usub.with.overflow
-                    // in the backend if profitable.
-                    let sub = self.sub(lhs, rhs);
-                    let cmp = self.icmp(IntPredicate::IntULT, lhs, rhs);
-                    return (sub, cmp);
-                }
-
-                _ => unreachable!(),
-            },
-            OverflowOp::Mul => match new_kind {
-                Int(I8) => "llvm.smul.with.overflow.i8",
-                Int(I16) => "llvm.smul.with.overflow.i16",
-                Int(I32) => "llvm.smul.with.overflow.i32",
-                Int(I64) => "llvm.smul.with.overflow.i64",
-                Int(I128) => "llvm.smul.with.overflow.i128",
-
-                Uint(U8) => "llvm.umul.with.overflow.i8",
-                Uint(U16) => "llvm.umul.with.overflow.i16",
-                Uint(U32) => "llvm.umul.with.overflow.i32",
-                Uint(U64) => "llvm.umul.with.overflow.i64",
-                Uint(U128) => "llvm.umul.with.overflow.i128",
-
-                _ => unreachable!(),
-            },
-        };
+        let name = format!("llvm.{}{oop_str}.with.overflow", if signed { 's' } else { 'u' });
 
-        let res = self.call_intrinsic(name, &[lhs, rhs]);
+        let res = self.call_intrinsic(name, &[self.type_ix(width)], &[lhs, rhs]);
         (self.extract_value(res, 0), self.extract_value(res, 1))
     }
 
@@ -581,16 +535,6 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         }
     }
 
-    fn dynamic_alloca(&mut self, size: &'ll Value, align: Align) -> &'ll Value {
-        unsafe {
-            let alloca =
-                llvm::LLVMBuildArrayAlloca(self.llbuilder, self.cx().type_i8(), size, UNNAMED);
-            llvm::LLVMSetAlignment(alloca, align.bytes() as c_uint);
-            // Cast to default addrspace if necessary
-            llvm::LLVMBuildPointerCast(self.llbuilder, alloca, self.cx().type_ptr(), UNNAMED)
-        }
-    }
-
     fn load(&mut self, ty: &'ll Type, ptr: &'ll Value, align: Align) -> &'ll Value {
         unsafe {
             let load = llvm::LLVMBuildLoad2(self.llbuilder, ty, ptr, UNNAMED);
@@ -690,17 +634,16 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         } else if place.layout.is_llvm_immediate() {
             let mut const_llval = None;
             let llty = place.layout.llvm_type(self);
-            unsafe {
-                if let Some(global) = llvm::LLVMIsAGlobalVariable(place.val.llval) {
-                    if llvm::LLVMIsGlobalConstant(global) == llvm::True {
-                        if let Some(init) = llvm::LLVMGetInitializer(global) {
-                            if self.val_ty(init) == llty {
-                                const_llval = Some(init);
-                            }
+            if let Some(global) = llvm::LLVMIsAGlobalVariable(place.val.llval) {
+                if llvm::LLVMIsGlobalConstant(global) == llvm::True {
+                    if let Some(init) = llvm::LLVMGetInitializer(global) {
+                        if self.val_ty(init) == llty {
+                            const_llval = Some(init);
                         }
                     }
                 }
             }
+
             let llval = const_llval.unwrap_or_else(|| {
                 let load = self.load(llty, place.val.llval, place.val.align);
                 if let abi::BackendRepr::Scalar(scalar) = place.layout.backend_repr {
@@ -954,11 +897,11 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
     }
 
     fn fptoui_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value {
-        self.fptoint_sat(false, val, dest_ty)
+        self.call_intrinsic("llvm.fptoui.sat", &[dest_ty, self.val_ty(val)], &[val])
     }
 
     fn fptosi_sat(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value {
-        self.fptoint_sat(true, val, dest_ty)
+        self.call_intrinsic("llvm.fptosi.sat", &[dest_ty, self.val_ty(val)], &[val])
     }
 
     fn fptoui(&mut self, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value {
@@ -981,15 +924,12 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
             if self.cx.type_kind(src_ty) != TypeKind::Vector {
                 let float_width = self.cx.float_width(src_ty);
                 let int_width = self.cx.int_width(dest_ty);
-                let name = match (int_width, float_width) {
-                    (32, 32) => Some("llvm.wasm.trunc.unsigned.i32.f32"),
-                    (32, 64) => Some("llvm.wasm.trunc.unsigned.i32.f64"),
-                    (64, 32) => Some("llvm.wasm.trunc.unsigned.i64.f32"),
-                    (64, 64) => Some("llvm.wasm.trunc.unsigned.i64.f64"),
-                    _ => None,
-                };
-                if let Some(name) = name {
-                    return self.call_intrinsic(name, &[val]);
+                if matches!((int_width, float_width), (32 | 64, 32 | 64)) {
+                    return self.call_intrinsic(
+                        "llvm.wasm.trunc.unsigned",
+                        &[dest_ty, src_ty],
+                        &[val],
+                    );
                 }
             }
         }
@@ -1003,15 +943,12 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
             if self.cx.type_kind(src_ty) != TypeKind::Vector {
                 let float_width = self.cx.float_width(src_ty);
                 let int_width = self.cx.int_width(dest_ty);
-                let name = match (int_width, float_width) {
-                    (32, 32) => Some("llvm.wasm.trunc.signed.i32.f32"),
-                    (32, 64) => Some("llvm.wasm.trunc.signed.i32.f64"),
-                    (64, 32) => Some("llvm.wasm.trunc.signed.i64.f32"),
-                    (64, 64) => Some("llvm.wasm.trunc.signed.i64.f64"),
-                    _ => None,
-                };
-                if let Some(name) = name {
-                    return self.call_intrinsic(name, &[val]);
+                if matches!((int_width, float_width), (32 | 64, 32 | 64)) {
+                    return self.call_intrinsic(
+                        "llvm.wasm.trunc.signed",
+                        &[dest_ty, src_ty],
+                        &[val],
+                    );
                 }
             }
         }
@@ -1084,22 +1021,10 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
             return None;
         }
 
-        let name = match (ty.is_signed(), ty.primitive_size(self.tcx).bits()) {
-            (true, 8) => "llvm.scmp.i8.i8",
-            (true, 16) => "llvm.scmp.i8.i16",
-            (true, 32) => "llvm.scmp.i8.i32",
-            (true, 64) => "llvm.scmp.i8.i64",
-            (true, 128) => "llvm.scmp.i8.i128",
-
-            (false, 8) => "llvm.ucmp.i8.i8",
-            (false, 16) => "llvm.ucmp.i8.i16",
-            (false, 32) => "llvm.ucmp.i8.i32",
-            (false, 64) => "llvm.ucmp.i8.i64",
-            (false, 128) => "llvm.ucmp.i8.i128",
+        let size = ty.primitive_size(self.tcx);
+        let name = if ty.is_signed() { "llvm.scmp" } else { "llvm.ucmp" };
 
-            _ => bug!("three-way compare unsupported for type {ty:?}"),
-        };
-        Some(self.call_intrinsic(name, &[lhs, rhs]))
+        Some(self.call_intrinsic(name, &[self.type_i8(), self.type_ix(size.bits())], &[lhs, rhs]))
     }
 
     /* Miscellaneous instructions */
@@ -1227,11 +1152,10 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
         (self.extract_value(landing_pad, 0), self.extract_value(landing_pad, 1))
     }
 
-    fn filter_landing_pad(&mut self, pers_fn: &'ll Value) -> (&'ll Value, &'ll Value) {
+    fn filter_landing_pad(&mut self, pers_fn: &'ll Value) {
         let ty = self.type_struct(&[self.type_ptr(), self.type_i32()], false);
         let landing_pad = self.landing_pad(ty, pers_fn, 1);
         self.add_clause(landing_pad, self.const_array(self.type_ptr(), &[]));
-        (self.extract_value(landing_pad, 0), self.extract_value(landing_pad, 1))
     }
 
     fn resume(&mut self, exn0: &'ll Value, exn1: &'ll Value) {
@@ -1385,11 +1309,11 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
     }
 
     fn lifetime_start(&mut self, ptr: &'ll Value, size: Size) {
-        self.call_lifetime_intrinsic("llvm.lifetime.start.p0i8", ptr, size);
+        self.call_lifetime_intrinsic("llvm.lifetime.start", ptr, size);
     }
 
     fn lifetime_end(&mut self, ptr: &'ll Value, size: Size) {
-        self.call_lifetime_intrinsic("llvm.lifetime.end.p0i8", ptr, size);
+        self.call_lifetime_intrinsic("llvm.lifetime.end", ptr, size);
     }
 
     fn call(
@@ -1454,7 +1378,8 @@ impl<'ll> StaticBuilderMethods for Builder<'_, 'll, '_> {
         // Forward to the `get_static` method of `CodegenCx`
         let global = self.cx().get_static(def_id);
         if self.cx().tcx.is_thread_local_static(def_id) {
-            let pointer = self.call_intrinsic("llvm.threadlocal.address", &[global]);
+            let pointer =
+                self.call_intrinsic("llvm.threadlocal.address", &[self.val_ty(global)], &[global]);
             // Cast to default address space if globals are in a different addrspace
             self.pointercast(pointer, self.type_ptr())
         } else {
@@ -1649,12 +1574,17 @@ impl<'a, 'll, CX: Borrow<SCx<'ll>>> GenericBuilder<'a, 'll, CX> {
 }
 
 impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
-    pub(crate) fn call_intrinsic(&mut self, intrinsic: &str, args: &[&'ll Value]) -> &'ll Value {
-        let (ty, f) = self.cx.get_intrinsic(intrinsic);
+    pub(crate) fn call_intrinsic(
+        &mut self,
+        base_name: impl Into<Cow<'static, str>>,
+        type_params: &[&'ll Type],
+        args: &[&'ll Value],
+    ) -> &'ll Value {
+        let (ty, f) = self.cx.get_intrinsic(base_name.into(), type_params);
         self.call(ty, None, None, f, args, None, None)
     }
 
-    fn call_lifetime_intrinsic(&mut self, intrinsic: &str, ptr: &'ll Value, size: Size) {
+    fn call_lifetime_intrinsic(&mut self, intrinsic: &'static str, ptr: &'ll Value, size: Size) {
         let size = size.bytes();
         if size == 0 {
             return;
@@ -1664,7 +1594,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
             return;
         }
 
-        self.call_intrinsic(intrinsic, &[self.cx.const_u64(size), ptr]);
+        self.call_intrinsic(intrinsic, &[self.val_ty(ptr)], &[self.cx.const_u64(size), ptr]);
     }
 }
 impl<'a, 'll, CX: Borrow<SCx<'ll>>> GenericBuilder<'a, 'll, CX> {
@@ -1689,31 +1619,6 @@ impl<'a, 'll, CX: Borrow<SCx<'ll>>> GenericBuilder<'a, 'll, CX> {
     }
 }
 impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
-    fn fptoint_sat(&mut self, signed: bool, val: &'ll Value, dest_ty: &'ll Type) -> &'ll Value {
-        let src_ty = self.cx.val_ty(val);
-        let (float_ty, int_ty, vector_length) = if self.cx.type_kind(src_ty) == TypeKind::Vector {
-            assert_eq!(self.cx.vector_length(src_ty), self.cx.vector_length(dest_ty));
-            (
-                self.cx.element_type(src_ty),
-                self.cx.element_type(dest_ty),
-                Some(self.cx.vector_length(src_ty)),
-            )
-        } else {
-            (src_ty, dest_ty, None)
-        };
-        let float_width = self.cx.float_width(float_ty);
-        let int_width = self.cx.int_width(int_ty);
-
-        let instr = if signed { "fptosi" } else { "fptoui" };
-        let name = if let Some(vector_length) = vector_length {
-            format!("llvm.{instr}.sat.v{vector_length}i{int_width}.v{vector_length}f{float_width}")
-        } else {
-            format!("llvm.{instr}.sat.i{int_width}.f{float_width}")
-        };
-        let f = self.declare_cfn(&name, llvm::UnnamedAddr::No, self.type_func(&[src_ty], dest_ty));
-        self.call(self.type_func(&[src_ty], dest_ty), None, None, f, &[val], None, None)
-    }
-
     pub(crate) fn landing_pad(
         &mut self,
         ty: &'ll Type,
@@ -1812,14 +1717,14 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
             } else {
                 cfi::typeid_for_fnabi(self.tcx, fn_abi, options)
             };
-            let typeid_metadata = self.cx.typeid_metadata(typeid).unwrap();
+            let typeid_metadata = self.cx.create_metadata(typeid.as_bytes());
             let dbg_loc = self.get_dbg_loc();
 
             // Test whether the function pointer is associated with the type identifier using the
             // llvm.type.test intrinsic. The LowerTypeTests link-time optimization pass replaces
             // calls to this intrinsic with code to test type membership.
             let typeid = self.get_metadata_value(typeid_metadata);
-            let cond = self.call_intrinsic("llvm.type.test", &[llfn, typeid]);
+            let cond = self.call_intrinsic("llvm.type.test", &[], &[llfn, typeid]);
             let bb_pass = self.append_sibling_block("type_test.pass");
             let bb_fail = self.append_sibling_block("type_test.fail");
             self.cond_br(cond, bb_pass, bb_fail);
@@ -1887,7 +1792,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         num_counters: &'ll Value,
         index: &'ll Value,
     ) {
-        self.call_intrinsic("llvm.instrprof.increment", &[fn_name, hash, num_counters, index]);
+        self.call_intrinsic("llvm.instrprof.increment", &[], &[fn_name, hash, num_counters, index]);
     }
 
     /// Emits a call to `llvm.instrprof.mcdc.parameters`.
@@ -1906,7 +1811,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         hash: &'ll Value,
         bitmap_bits: &'ll Value,
     ) {
-        self.call_intrinsic("llvm.instrprof.mcdc.parameters", &[fn_name, hash, bitmap_bits]);
+        self.call_intrinsic("llvm.instrprof.mcdc.parameters", &[], &[fn_name, hash, bitmap_bits]);
     }
 
     #[instrument(level = "debug", skip(self))]
@@ -1918,7 +1823,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
         mcdc_temp: &'ll Value,
     ) {
         let args = &[fn_name, hash, bitmap_index, mcdc_temp];
-        self.call_intrinsic("llvm.instrprof.mcdc.tvbitmap.update", args);
+        self.call_intrinsic("llvm.instrprof.mcdc.tvbitmap.update", &[], args);
     }
 
     #[instrument(level = "debug", skip(self))]
diff --git a/compiler/rustc_codegen_llvm/src/builder/autodiff.rs b/compiler/rustc_codegen_llvm/src/builder/autodiff.rs
index c5c13ac097a..829b3c513c2 100644
--- a/compiler/rustc_codegen_llvm/src/builder/autodiff.rs
+++ b/compiler/rustc_codegen_llvm/src/builder/autodiff.rs
@@ -2,7 +2,6 @@ use std::ptr;
 
 use rustc_ast::expand::autodiff_attrs::{AutoDiffAttrs, AutoDiffItem, DiffActivity, DiffMode};
 use rustc_codegen_ssa::ModuleCodegen;
-use rustc_codegen_ssa::back::write::ModuleConfig;
 use rustc_codegen_ssa::common::TypeKind;
 use rustc_codegen_ssa::traits::BaseTypeCodegenMethods;
 use rustc_errors::FatalError;
@@ -76,12 +75,12 @@ fn match_args_from_caller_to_enzyme<'ll>(
         outer_pos = 1;
     }
 
-    let enzyme_const = cx.create_metadata("enzyme_const".to_string()).unwrap();
-    let enzyme_out = cx.create_metadata("enzyme_out".to_string()).unwrap();
-    let enzyme_dup = cx.create_metadata("enzyme_dup".to_string()).unwrap();
-    let enzyme_dupv = cx.create_metadata("enzyme_dupv".to_string()).unwrap();
-    let enzyme_dupnoneed = cx.create_metadata("enzyme_dupnoneed".to_string()).unwrap();
-    let enzyme_dupnoneedv = cx.create_metadata("enzyme_dupnoneedv".to_string()).unwrap();
+    let enzyme_const = cx.create_metadata(b"enzyme_const");
+    let enzyme_out = cx.create_metadata(b"enzyme_out");
+    let enzyme_dup = cx.create_metadata(b"enzyme_dup");
+    let enzyme_dupv = cx.create_metadata(b"enzyme_dupv");
+    let enzyme_dupnoneed = cx.create_metadata(b"enzyme_dupnoneed");
+    let enzyme_dupnoneedv = cx.create_metadata(b"enzyme_dupnoneedv");
 
     while activity_pos < inputs.len() {
         let diff_activity = inputs[activity_pos as usize];
@@ -114,7 +113,7 @@ fn match_args_from_caller_to_enzyme<'ll>(
             let mul = unsafe {
                 llvm::LLVMBuildMul(
                     builder.llbuilder,
-                    cx.get_const_i64(elem_bytes_size),
+                    cx.get_const_int(cx.type_i64(), elem_bytes_size),
                     next_outer_arg,
                     UNNAMED,
                 )
@@ -306,7 +305,7 @@ fn generate_enzyme_call<'ll>(
     // add outer_fn name to ad_name to make it unique, in case users apply autodiff to multiple
     // functions. Unwrap will only panic, if LLVM gave us an invalid string.
     let name = llvm::get_value_name(outer_fn);
-    let outer_fn_name = std::str::from_utf8(name).unwrap();
+    let outer_fn_name = std::str::from_utf8(&name).unwrap();
     ad_name.push_str(outer_fn_name);
 
     // Let us assume the user wrote the following function square:
@@ -378,14 +377,14 @@ fn generate_enzyme_call<'ll>(
         let mut args = Vec::with_capacity(num_args as usize + 1);
         args.push(fn_to_diff);
 
-        let enzyme_primal_ret = cx.create_metadata("enzyme_primal_return".to_string()).unwrap();
+        let enzyme_primal_ret = cx.create_metadata(b"enzyme_primal_return");
         if matches!(attrs.ret_activity, DiffActivity::Dual | DiffActivity::Active) {
             args.push(cx.get_metadata_value(enzyme_primal_ret));
         }
         if attrs.width > 1 {
-            let enzyme_width = cx.create_metadata("enzyme_width".to_string()).unwrap();
+            let enzyme_width = cx.create_metadata(b"enzyme_width");
             args.push(cx.get_metadata_value(enzyme_width));
-            args.push(cx.get_const_i64(attrs.width as u64));
+            args.push(cx.get_const_int(cx.type_i64(), attrs.width as u64));
         }
 
         let has_sret = has_sret(outer_fn);
@@ -461,7 +460,6 @@ pub(crate) fn differentiate<'ll>(
     module: &'ll ModuleCodegen<ModuleLlvm>,
     cgcx: &CodegenContext<LlvmCodegenBackend>,
     diff_items: Vec<AutoDiffItem>,
-    _config: &ModuleConfig,
 ) -> Result<(), FatalError> {
     for item in &diff_items {
         trace!("{}", item);
diff --git a/compiler/rustc_codegen_llvm/src/callee.rs b/compiler/rustc_codegen_llvm/src/callee.rs
index 6d68eca60af..5a3dd90ab24 100644
--- a/compiler/rustc_codegen_llvm/src/callee.rs
+++ b/compiler/rustc_codegen_llvm/src/callee.rs
@@ -102,7 +102,7 @@ pub(crate) fn get_fn<'ll, 'tcx>(cx: &CodegenCx<'ll, 'tcx>, instance: Instance<'t
         let is_hidden = if is_generic {
             // This is a monomorphization of a generic function.
             if !(cx.tcx.sess.opts.share_generics()
-                || tcx.codegen_fn_attrs(instance_def_id).inline
+                || tcx.codegen_instance_attrs(instance.def).inline
                     == rustc_attr_data_structures::InlineAttr::Never)
             {
                 // When not sharing generics, all instances are in the same
diff --git a/compiler/rustc_codegen_llvm/src/common.rs b/compiler/rustc_codegen_llvm/src/common.rs
index 3cfa96393e9..f9ab96b5789 100644
--- a/compiler/rustc_codegen_llvm/src/common.rs
+++ b/compiler/rustc_codegen_llvm/src/common.rs
@@ -3,9 +3,8 @@
 use std::borrow::Borrow;
 
 use libc::{c_char, c_uint};
-use rustc_abi as abi;
-use rustc_abi::HasDataLayout;
 use rustc_abi::Primitive::Pointer;
+use rustc_abi::{self as abi, HasDataLayout as _};
 use rustc_ast::Mutability;
 use rustc_codegen_ssa::common::TypeKind;
 use rustc_codegen_ssa::traits::*;
@@ -99,14 +98,14 @@ impl<'ll, CX: Borrow<SCx<'ll>>> BackendTypes for GenericCx<'ll, CX> {
     type DIVariable = &'ll llvm::debuginfo::DIVariable;
 }
 
-impl<'ll> CodegenCx<'ll, '_> {
+impl<'ll, CX: Borrow<SCx<'ll>>> GenericCx<'ll, CX> {
     pub(crate) fn const_array(&self, ty: &'ll Type, elts: &[&'ll Value]) -> &'ll Value {
         let len = u64::try_from(elts.len()).expect("LLVMConstArray2 elements len overflow");
         unsafe { llvm::LLVMConstArray2(ty, elts.as_ptr(), len) }
     }
 
     pub(crate) fn const_bytes(&self, bytes: &[u8]) -> &'ll Value {
-        bytes_in_context(self.llcx, bytes)
+        bytes_in_context(self.llcx(), bytes)
     }
 
     pub(crate) fn const_get_elt(&self, v: &'ll Value, idx: u64) -> &'ll Value {
@@ -175,7 +174,7 @@ impl<'ll, 'tcx> ConstCodegenMethods for CodegenCx<'ll, 'tcx> {
     }
 
     fn const_usize(&self, i: u64) -> &'ll Value {
-        let bit_size = self.data_layout().pointer_size.bits();
+        let bit_size = self.data_layout().pointer_size().bits();
         if bit_size < 64 {
             // make sure it doesn't overflow
             assert!(i < (1 << bit_size));
@@ -216,10 +215,10 @@ impl<'ll, 'tcx> ConstCodegenMethods for CodegenCx<'ll, 'tcx> {
                 bug!("symbol `{}` is already defined", sym);
             });
             llvm::set_initializer(g, sc);
-            unsafe {
-                llvm::LLVMSetGlobalConstant(g, True);
-                llvm::LLVMSetUnnamedAddress(g, llvm::UnnamedAddr::Global);
-            }
+
+            llvm::set_global_constant(g, true);
+            llvm::set_unnamed_address(g, llvm::UnnamedAddr::Global);
+
             llvm::set_linkage(g, llvm::Linkage::InternalLinkage);
             // Cast to default address space if globals are in a different addrspace
             let g = self.const_pointercast(g, self.type_ptr());
@@ -268,7 +267,7 @@ impl<'ll, 'tcx> ConstCodegenMethods for CodegenCx<'ll, 'tcx> {
                 }
             }
             Scalar::Ptr(ptr, _size) => {
-                let (prov, offset) = ptr.into_parts();
+                let (prov, offset) = ptr.prov_and_relative_offset();
                 let global_alloc = self.tcx.global_alloc(prov.alloc_id());
                 let base_addr = match global_alloc {
                     GlobalAlloc::Memory(alloc) => {
@@ -284,7 +283,8 @@ impl<'ll, 'tcx> ConstCodegenMethods for CodegenCx<'ll, 'tcx> {
                                 self.const_bitcast(llval, llty)
                             };
                         } else {
-                            let init = const_alloc_to_llvm(self, alloc, /*static*/ false);
+                            let init =
+                                const_alloc_to_llvm(self, alloc.inner(), /*static*/ false);
                             let alloc = alloc.inner();
                             let value = match alloc.mutability {
                                 Mutability::Mut => self.static_addr_of_mut(init, alloc.align, None),
@@ -316,15 +316,19 @@ impl<'ll, 'tcx> ConstCodegenMethods for CodegenCx<'ll, 'tcx> {
                                 }),
                             )))
                             .unwrap_memory();
-                        let init = const_alloc_to_llvm(self, alloc, /*static*/ false);
-                        let value = self.static_addr_of_impl(init, alloc.inner().align, None);
-                        value
+                        let init = const_alloc_to_llvm(self, alloc.inner(), /*static*/ false);
+                        self.static_addr_of_impl(init, alloc.inner().align, None)
                     }
                     GlobalAlloc::Static(def_id) => {
                         assert!(self.tcx.is_static(def_id));
                         assert!(!self.tcx.is_thread_local_static(def_id));
                         self.get_static(def_id)
                     }
+                    GlobalAlloc::TypeId { .. } => {
+                        // Drop the provenance, the offset contains the bytes of the hash
+                        let llval = self.const_usize(offset.bytes());
+                        return unsafe { llvm::LLVMConstIntToPtr(llval, llty) };
+                    }
                 };
                 let base_addr_space = global_alloc.address_space(self);
                 let llval = unsafe {
@@ -346,7 +350,7 @@ impl<'ll, 'tcx> ConstCodegenMethods for CodegenCx<'ll, 'tcx> {
     }
 
     fn const_data_from_alloc(&self, alloc: ConstAllocation<'_>) -> Self::Value {
-        const_alloc_to_llvm(self, alloc, /*static*/ false)
+        const_alloc_to_llvm(self, alloc.inner(), /*static*/ false)
     }
 
     fn const_ptr_byte_offset(&self, base_addr: Self::Value, offset: abi::Size) -> Self::Value {
diff --git a/compiler/rustc_codegen_llvm/src/consts.rs b/compiler/rustc_codegen_llvm/src/consts.rs
index a4492d76c3c..0b96b63bc85 100644
--- a/compiler/rustc_codegen_llvm/src/consts.rs
+++ b/compiler/rustc_codegen_llvm/src/consts.rs
@@ -17,20 +17,18 @@ use rustc_middle::ty::{self, Instance};
 use rustc_middle::{bug, span_bug};
 use tracing::{debug, instrument, trace};
 
-use crate::common::{AsCCharPtr, CodegenCx};
+use crate::common::CodegenCx;
 use crate::errors::SymbolAlreadyDefined;
-use crate::llvm::{self, True};
 use crate::type_::Type;
 use crate::type_of::LayoutLlvmExt;
 use crate::value::Value;
-use crate::{base, debuginfo};
+use crate::{base, debuginfo, llvm};
 
 pub(crate) fn const_alloc_to_llvm<'ll>(
     cx: &CodegenCx<'ll, '_>,
-    alloc: ConstAllocation<'_>,
+    alloc: &Allocation,
     is_static: bool,
 ) -> &'ll Value {
-    let alloc = alloc.inner();
     // We expect that callers of const_alloc_to_llvm will instead directly codegen a pointer or
     // integer for any &ZST where the ZST is a constant (i.e. not a static). We should never be
     // producing empty LLVM allocations as they're just adding noise to binaries and forcing less
@@ -43,7 +41,8 @@ pub(crate) fn const_alloc_to_llvm<'ll>(
     }
     let mut llvals = Vec::with_capacity(alloc.provenance().ptrs().len() + 1);
     let dl = cx.data_layout();
-    let pointer_size = dl.pointer_size.bytes() as usize;
+    let pointer_size = dl.pointer_size();
+    let pointer_size_bytes = pointer_size.bytes() as usize;
 
     // Note: this function may call `inspect_with_uninit_and_ptr_outside_interpreter`, so `range`
     // must be within the bounds of `alloc` and not contain or overlap a pointer provenance.
@@ -100,7 +99,9 @@ pub(crate) fn const_alloc_to_llvm<'ll>(
             // This `inspect` is okay since it is within the bounds of the allocation, it doesn't
             // affect interpreter execution (we inspect the result after interpreter execution),
             // and we properly interpret the provenance as a relocation pointer offset.
-            alloc.inspect_with_uninit_and_ptr_outside_interpreter(offset..(offset + pointer_size)),
+            alloc.inspect_with_uninit_and_ptr_outside_interpreter(
+                offset..(offset + pointer_size_bytes),
+            ),
         )
         .expect("const_alloc_to_llvm: could not read relocation pointer")
             as u64;
@@ -111,11 +112,11 @@ pub(crate) fn const_alloc_to_llvm<'ll>(
             InterpScalar::from_pointer(Pointer::new(prov, Size::from_bytes(ptr_offset)), &cx.tcx),
             Scalar::Initialized {
                 value: Primitive::Pointer(address_space),
-                valid_range: WrappingRange::full(dl.pointer_size),
+                valid_range: WrappingRange::full(pointer_size),
             },
             cx.type_ptr_ext(address_space),
         ));
-        next_offset = offset + pointer_size;
+        next_offset = offset + pointer_size_bytes;
     }
     if alloc.len() >= next_offset {
         let range = next_offset..alloc.len();
@@ -138,7 +139,7 @@ fn codegen_static_initializer<'ll, 'tcx>(
     def_id: DefId,
 ) -> Result<(&'ll Value, ConstAllocation<'tcx>), ErrorHandled> {
     let alloc = cx.tcx.eval_static_initializer(def_id)?;
-    Ok((const_alloc_to_llvm(cx, alloc, /*static*/ true), alloc))
+    Ok((const_alloc_to_llvm(cx, alloc.inner(), /*static*/ true), alloc))
 }
 
 fn set_global_alignment<'ll>(cx: &CodegenCx<'ll, '_>, gv: &'ll Value, mut align: Align) {
@@ -245,7 +246,7 @@ impl<'ll> CodegenCx<'ll, '_> {
         };
         llvm::set_initializer(gv, cv);
         set_global_alignment(self, gv, align);
-        llvm::SetUnnamedAddress(gv, llvm::UnnamedAddr::Global);
+        llvm::set_unnamed_address(gv, llvm::UnnamedAddr::Global);
         gv
     }
 
@@ -270,9 +271,8 @@ impl<'ll> CodegenCx<'ll, '_> {
             return gv;
         }
         let gv = self.static_addr_of_mut(cv, align, kind);
-        unsafe {
-            llvm::LLVMSetGlobalConstant(gv, True);
-        }
+        llvm::set_global_constant(gv, true);
+
         self.const_globals.borrow_mut().insert(cv, gv);
         gv
     }
@@ -396,149 +396,140 @@ impl<'ll> CodegenCx<'ll, '_> {
     }
 
     fn codegen_static_item(&mut self, def_id: DefId) {
-        unsafe {
-            assert!(
-                llvm::LLVMGetInitializer(
-                    self.instances.borrow().get(&Instance::mono(self.tcx, def_id)).unwrap()
-                )
-                .is_none()
-            );
-            let attrs = self.tcx.codegen_fn_attrs(def_id);
+        assert!(
+            llvm::LLVMGetInitializer(
+                self.instances.borrow().get(&Instance::mono(self.tcx, def_id)).unwrap()
+            )
+            .is_none()
+        );
+        let attrs = self.tcx.codegen_fn_attrs(def_id);
 
-            let Ok((v, alloc)) = codegen_static_initializer(self, def_id) else {
-                // Error has already been reported
-                return;
-            };
-            let alloc = alloc.inner();
+        let Ok((v, alloc)) = codegen_static_initializer(self, def_id) else {
+            // Error has already been reported
+            return;
+        };
+        let alloc = alloc.inner();
 
-            let val_llty = self.val_ty(v);
+        let val_llty = self.val_ty(v);
 
-            let g = self.get_static_inner(def_id, val_llty);
-            let llty = self.get_type_of_global(g);
+        let g = self.get_static_inner(def_id, val_llty);
+        let llty = self.get_type_of_global(g);
 
-            let g = if val_llty == llty {
-                g
-            } else {
-                // codegen_static_initializer creates the global value just from the
-                // `Allocation` data by generating one big struct value that is just
-                // all the bytes and pointers after each other. This will almost never
-                // match the type that the static was declared with. Unfortunately
-                // we can't just LLVMConstBitCast our way out of it because that has very
-                // specific rules on what can be cast. So instead of adding a new way to
-                // generate static initializers that match the static's type, we picked
-                // the easier option and retroactively change the type of the static item itself.
-                let name = llvm::get_value_name(g).to_vec();
-                llvm::set_value_name(g, b"");
-
-                let linkage = llvm::get_linkage(g);
-                let visibility = llvm::get_visibility(g);
-
-                let new_g = llvm::LLVMRustGetOrInsertGlobal(
-                    self.llmod,
-                    name.as_c_char_ptr(),
-                    name.len(),
-                    val_llty,
-                );
-
-                llvm::set_linkage(new_g, linkage);
-                llvm::set_visibility(new_g, visibility);
-
-                // The old global has had its name removed but is returned by
-                // get_static since it is in the instance cache. Provide an
-                // alternative lookup that points to the new global so that
-                // global_asm! can compute the correct mangled symbol name
-                // for the global.
-                self.renamed_statics.borrow_mut().insert(def_id, new_g);
-
-                // To avoid breaking any invariants, we leave around the old
-                // global for the moment; we'll replace all references to it
-                // with the new global later. (See base::codegen_backend.)
-                self.statics_to_rauw.borrow_mut().push((g, new_g));
-                new_g
-            };
-            set_global_alignment(self, g, alloc.align);
-            llvm::set_initializer(g, v);
-
-            self.assume_dso_local(g, true);
-
-            // Forward the allocation's mutability (picked by the const interner) to LLVM.
-            if alloc.mutability.is_not() {
-                llvm::LLVMSetGlobalConstant(g, llvm::True);
-            }
+        let g = if val_llty == llty {
+            g
+        } else {
+            // codegen_static_initializer creates the global value just from the
+            // `Allocation` data by generating one big struct value that is just
+            // all the bytes and pointers after each other. This will almost never
+            // match the type that the static was declared with. Unfortunately
+            // we can't just LLVMConstBitCast our way out of it because that has very
+            // specific rules on what can be cast. So instead of adding a new way to
+            // generate static initializers that match the static's type, we picked
+            // the easier option and retroactively change the type of the static item itself.
+            let name = String::from_utf8(llvm::get_value_name(g))
+                .expect("we declare our statics with a utf8-valid name");
+            llvm::set_value_name(g, b"");
+
+            let linkage = llvm::get_linkage(g);
+            let visibility = llvm::get_visibility(g);
+
+            let new_g = self.declare_global(&name, val_llty);
+
+            llvm::set_linkage(new_g, linkage);
+            llvm::set_visibility(new_g, visibility);
+
+            // The old global has had its name removed but is returned by
+            // get_static since it is in the instance cache. Provide an
+            // alternative lookup that points to the new global so that
+            // global_asm! can compute the correct mangled symbol name
+            // for the global.
+            self.renamed_statics.borrow_mut().insert(def_id, new_g);
+
+            // To avoid breaking any invariants, we leave around the old
+            // global for the moment; we'll replace all references to it
+            // with the new global later. (See base::codegen_backend.)
+            self.statics_to_rauw.borrow_mut().push((g, new_g));
+            new_g
+        };
+        set_global_alignment(self, g, alloc.align);
+        llvm::set_initializer(g, v);
 
-            debuginfo::build_global_var_di_node(self, def_id, g);
+        self.assume_dso_local(g, true);
 
-            if attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL) {
-                llvm::set_thread_local_mode(g, self.tls_model);
-            }
+        // Forward the allocation's mutability (picked by the const interner) to LLVM.
+        if alloc.mutability.is_not() {
+            llvm::set_global_constant(g, true);
+        }
 
-            // Wasm statics with custom link sections get special treatment as they
-            // go into custom sections of the wasm executable. The exception to this
-            // is the `.init_array` section which are treated specially by the wasm linker.
-            if self.tcx.sess.target.is_like_wasm
-                && attrs
-                    .link_section
-                    .map(|link_section| !link_section.as_str().starts_with(".init_array"))
-                    .unwrap_or(true)
-            {
-                if let Some(section) = attrs.link_section {
-                    let section = llvm::LLVMMDStringInContext2(
-                        self.llcx,
-                        section.as_str().as_c_char_ptr(),
-                        section.as_str().len(),
-                    );
-                    assert!(alloc.provenance().ptrs().is_empty());
-
-                    // The `inspect` method is okay here because we checked for provenance, and
-                    // because we are doing this access to inspect the final interpreter state (not
-                    // as part of the interpreter execution).
-                    let bytes =
-                        alloc.inspect_with_uninit_and_ptr_outside_interpreter(0..alloc.len());
-                    let alloc =
-                        llvm::LLVMMDStringInContext2(self.llcx, bytes.as_c_char_ptr(), bytes.len());
-                    let data = [section, alloc];
-                    let meta = llvm::LLVMMDNodeInContext2(self.llcx, data.as_ptr(), data.len());
-                    let val = self.get_metadata_value(meta);
+        debuginfo::build_global_var_di_node(self, def_id, g);
+
+        if attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL) {
+            llvm::set_thread_local_mode(g, self.tls_model);
+        }
+
+        // Wasm statics with custom link sections get special treatment as they
+        // go into custom sections of the wasm executable. The exception to this
+        // is the `.init_array` section which are treated specially by the wasm linker.
+        if self.tcx.sess.target.is_like_wasm
+            && attrs
+                .link_section
+                .map(|link_section| !link_section.as_str().starts_with(".init_array"))
+                .unwrap_or(true)
+        {
+            if let Some(section) = attrs.link_section {
+                let section = self.create_metadata(section.as_str().as_bytes());
+                assert!(alloc.provenance().ptrs().is_empty());
+
+                // The `inspect` method is okay here because we checked for provenance, and
+                // because we are doing this access to inspect the final interpreter state (not
+                // as part of the interpreter execution).
+                let bytes = alloc.inspect_with_uninit_and_ptr_outside_interpreter(0..alloc.len());
+                let alloc = self.create_metadata(bytes);
+                let data = [section, alloc];
+                let meta =
+                    unsafe { llvm::LLVMMDNodeInContext2(self.llcx, data.as_ptr(), data.len()) };
+                let val = self.get_metadata_value(meta);
+                unsafe {
                     llvm::LLVMAddNamedMetadataOperand(
                         self.llmod,
                         c"wasm.custom_sections".as_ptr(),
                         val,
-                    );
-                }
-            } else {
-                base::set_link_section(g, attrs);
+                    )
+                };
             }
+        } else {
+            base::set_link_section(g, attrs);
+        }
 
-            base::set_variable_sanitizer_attrs(g, attrs);
-
-            if attrs.flags.contains(CodegenFnAttrFlags::USED_COMPILER) {
-                // `USED` and `USED_LINKER` can't be used together.
-                assert!(!attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER));
-
-                // The semantics of #[used] in Rust only require the symbol to make it into the
-                // object file. It is explicitly allowed for the linker to strip the symbol if it
-                // is dead, which means we are allowed to use `llvm.compiler.used` instead of
-                // `llvm.used` here.
-                //
-                // Additionally, https://reviews.llvm.org/D97448 in LLVM 13 started emitting unique
-                // sections with SHF_GNU_RETAIN flag for llvm.used symbols, which may trigger bugs
-                // in the handling of `.init_array` (the static constructor list) in versions of
-                // the gold linker (prior to the one released with binutils 2.36).
-                //
-                // That said, we only ever emit these when `#[used(compiler)]` is explicitly
-                // requested. This is to avoid similar breakage on other targets, in particular
-                // MachO targets have *their* static constructor lists broken if `llvm.compiler.used`
-                // is emitted rather than `llvm.used`. However, that check happens when assigning
-                // the `CodegenFnAttrFlags` in the `codegen_fn_attrs` query, so we don't need to
-                // take care of it here.
-                self.add_compiler_used_global(g);
-            }
-            if attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER) {
-                // `USED` and `USED_LINKER` can't be used together.
-                assert!(!attrs.flags.contains(CodegenFnAttrFlags::USED_COMPILER));
+        base::set_variable_sanitizer_attrs(g, attrs);
+
+        if attrs.flags.contains(CodegenFnAttrFlags::USED_COMPILER) {
+            // `USED` and `USED_LINKER` can't be used together.
+            assert!(!attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER));
+
+            // The semantics of #[used] in Rust only require the symbol to make it into the
+            // object file. It is explicitly allowed for the linker to strip the symbol if it
+            // is dead, which means we are allowed to use `llvm.compiler.used` instead of
+            // `llvm.used` here.
+            //
+            // Additionally, https://reviews.llvm.org/D97448 in LLVM 13 started emitting unique
+            // sections with SHF_GNU_RETAIN flag for llvm.used symbols, which may trigger bugs
+            // in the handling of `.init_array` (the static constructor list) in versions of
+            // the gold linker (prior to the one released with binutils 2.36).
+            //
+            // That said, we only ever emit these when `#[used(compiler)]` is explicitly
+            // requested. This is to avoid similar breakage on other targets, in particular
+            // MachO targets have *their* static constructor lists broken if `llvm.compiler.used`
+            // is emitted rather than `llvm.used`. However, that check happens when assigning
+            // the `CodegenFnAttrFlags` in the `codegen_fn_attrs` query, so we don't need to
+            // take care of it here.
+            self.add_compiler_used_global(g);
+        }
+        if attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER) {
+            // `USED` and `USED_LINKER` can't be used together.
+            assert!(!attrs.flags.contains(CodegenFnAttrFlags::USED_COMPILER));
 
-                self.add_used_global(g);
-            }
+            self.add_used_global(g);
         }
     }
 
diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs
index 8d6e1d8941b..6a23becaa96 100644
--- a/compiler/rustc_codegen_llvm/src/context.rs
+++ b/compiler/rustc_codegen_llvm/src/context.rs
@@ -1,4 +1,4 @@
-use std::borrow::Borrow;
+use std::borrow::{Borrow, Cow};
 use std::cell::{Cell, RefCell};
 use std::ffi::{CStr, c_char, c_uint};
 use std::marker::PhantomData;
@@ -34,7 +34,6 @@ use smallvec::SmallVec;
 
 use crate::back::write::to_llvm_code_model;
 use crate::callee::get_fn;
-use crate::common::AsCCharPtr;
 use crate::debuginfo::metadata::apply_vcall_visibility_metadata;
 use crate::llvm::Metadata;
 use crate::type_::Type;
@@ -137,7 +136,8 @@ pub(crate) struct FullCx<'ll, 'tcx> {
     eh_catch_typeinfo: Cell<Option<&'ll Value>>,
     pub rust_try_fn: Cell<Option<(&'ll Type, &'ll Value)>>,
 
-    intrinsics: RefCell<FxHashMap<&'static str, (&'ll Type, &'ll Value)>>,
+    intrinsics:
+        RefCell<FxHashMap<(Cow<'static, str>, SmallVec<[&'ll Type; 2]>), (&'ll Type, &'ll Value)>>,
 
     /// A counter that is used for generating local symbol names
     local_gen_sym_counter: Cell<usize>,
@@ -168,6 +168,8 @@ pub(crate) unsafe fn create_module<'ll>(
     let mod_name = SmallCStr::new(mod_name);
     let llmod = unsafe { llvm::LLVMModuleCreateWithNameInContext(mod_name.as_ptr(), llcx) };
 
+    let cx = SimpleCx::new(llmod, llcx, tcx.data_layout.pointer_size());
+
     let mut target_data_layout = sess.target.data_layout.to_string();
     let llvm_version = llvm_util::get_version();
 
@@ -472,18 +474,14 @@ pub(crate) unsafe fn create_module<'ll>(
     #[allow(clippy::option_env_unwrap)]
     let rustc_producer =
         format!("rustc version {}", option_env!("CFG_VERSION").expect("CFG_VERSION"));
-    let name_metadata = unsafe {
-        llvm::LLVMMDStringInContext2(
-            llcx,
-            rustc_producer.as_c_char_ptr(),
-            rustc_producer.as_bytes().len(),
-        )
-    };
+
+    let name_metadata = cx.create_metadata(rustc_producer.as_bytes());
+
     unsafe {
         llvm::LLVMAddNamedMetadataOperand(
             llmod,
             c"llvm.ident".as_ptr(),
-            &llvm::LLVMMetadataAsValue(llcx, llvm::LLVMMDNodeInContext2(llcx, &name_metadata, 1)),
+            &cx.get_metadata_value(llvm::LLVMMDNodeInContext2(llcx, &name_metadata, 1)),
         );
     }
 
@@ -604,7 +602,7 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
         GenericCx(
             FullCx {
                 tcx,
-                scx: SimpleCx::new(llmod, llcx, tcx.data_layout.pointer_size),
+                scx: SimpleCx::new(llmod, llcx, tcx.data_layout.pointer_size()),
                 use_dll_storage_attrs,
                 tls_model,
                 codegen_unit,
@@ -678,11 +676,8 @@ impl<'ll, CX: Borrow<SCx<'ll>>> GenericCx<'ll, CX> {
         llvm::LLVMMetadataAsValue(self.llcx(), metadata)
     }
 
-    // FIXME(autodiff): We should split `ConstCodegenMethods` to pull the reusable parts
-    // onto a trait that is also implemented for GenericCx.
-    pub(crate) fn get_const_i64(&self, n: u64) -> &'ll Value {
-        let ty = unsafe { llvm::LLVMInt64TypeInContext(self.llcx()) };
-        unsafe { llvm::LLVMConstInt(ty, n, llvm::False) }
+    pub(crate) fn get_const_int(&self, ty: &'ll Type, val: u64) -> &'ll Value {
+        unsafe { llvm::LLVMConstInt(ty, val, llvm::False) }
     }
 
     pub(crate) fn get_function(&self, name: &str) -> Option<&'ll Value> {
@@ -700,10 +695,10 @@ impl<'ll, CX: Borrow<SCx<'ll>>> GenericCx<'ll, CX> {
         }
     }
 
-    pub(crate) fn create_metadata(&self, name: String) -> Option<&'ll Metadata> {
-        Some(unsafe {
+    pub(crate) fn create_metadata(&self, name: &[u8]) -> &'ll Metadata {
+        unsafe {
             llvm::LLVMMDStringInContext2(self.llcx(), name.as_ptr() as *const c_char, name.len())
-        })
+        }
     }
 
     pub(crate) fn get_functions(&self) -> Vec<&'ll Value> {
@@ -842,410 +837,41 @@ impl<'ll, 'tcx> MiscCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
 }
 
 impl<'ll> CodegenCx<'ll, '_> {
-    pub(crate) fn get_intrinsic(&self, key: &str) -> (&'ll Type, &'ll Value) {
-        if let Some(v) = self.intrinsics.borrow().get(key).cloned() {
-            return v;
-        }
-
-        self.declare_intrinsic(key).unwrap_or_else(|| bug!("unknown intrinsic '{}'", key))
-    }
-
-    fn insert_intrinsic(
+    pub(crate) fn get_intrinsic(
         &self,
-        name: &'static str,
-        args: Option<&[&'ll llvm::Type]>,
-        ret: &'ll llvm::Type,
-    ) -> (&'ll llvm::Type, &'ll llvm::Value) {
-        let fn_ty = if let Some(args) = args {
-            self.type_func(args, ret)
-        } else {
-            self.type_variadic_func(&[], ret)
-        };
-        let f = self.declare_cfn(name, llvm::UnnamedAddr::No, fn_ty);
-        self.intrinsics.borrow_mut().insert(name, (fn_ty, f));
-        (fn_ty, f)
-    }
-
-    fn declare_intrinsic(&self, key: &str) -> Option<(&'ll Type, &'ll Value)> {
-        macro_rules! ifn {
-            ($name:expr, fn() -> $ret:expr) => (
-                if key == $name {
-                    return Some(self.insert_intrinsic($name, Some(&[]), $ret));
-                }
-            );
-            ($name:expr, fn(...) -> $ret:expr) => (
-                if key == $name {
-                    return Some(self.insert_intrinsic($name, None, $ret));
-                }
-            );
-            ($name:expr, fn($($arg:expr),*) -> $ret:expr) => (
-                if key == $name {
-                    return Some(self.insert_intrinsic($name, Some(&[$($arg),*]), $ret));
-                }
-            );
-        }
-        macro_rules! mk_struct {
-            ($($field_ty:expr),*) => (self.type_struct( &[$($field_ty),*], false))
-        }
-
-        let ptr = self.type_ptr();
-        let void = self.type_void();
-        let i1 = self.type_i1();
-        let t_i8 = self.type_i8();
-        let t_i16 = self.type_i16();
-        let t_i32 = self.type_i32();
-        let t_i64 = self.type_i64();
-        let t_i128 = self.type_i128();
-        let t_isize = self.type_isize();
-        let t_f16 = self.type_f16();
-        let t_f32 = self.type_f32();
-        let t_f64 = self.type_f64();
-        let t_f128 = self.type_f128();
-        let t_metadata = self.type_metadata();
-        let t_token = self.type_token();
-
-        ifn!("llvm.wasm.get.exception", fn(t_token) -> ptr);
-        ifn!("llvm.wasm.get.ehselector", fn(t_token) -> t_i32);
-
-        ifn!("llvm.wasm.trunc.unsigned.i32.f32", fn(t_f32) -> t_i32);
-        ifn!("llvm.wasm.trunc.unsigned.i32.f64", fn(t_f64) -> t_i32);
-        ifn!("llvm.wasm.trunc.unsigned.i64.f32", fn(t_f32) -> t_i64);
-        ifn!("llvm.wasm.trunc.unsigned.i64.f64", fn(t_f64) -> t_i64);
-        ifn!("llvm.wasm.trunc.signed.i32.f32", fn(t_f32) -> t_i32);
-        ifn!("llvm.wasm.trunc.signed.i32.f64", fn(t_f64) -> t_i32);
-        ifn!("llvm.wasm.trunc.signed.i64.f32", fn(t_f32) -> t_i64);
-        ifn!("llvm.wasm.trunc.signed.i64.f64", fn(t_f64) -> t_i64);
-
-        ifn!("llvm.fptosi.sat.i8.f32", fn(t_f32) -> t_i8);
-        ifn!("llvm.fptosi.sat.i16.f32", fn(t_f32) -> t_i16);
-        ifn!("llvm.fptosi.sat.i32.f32", fn(t_f32) -> t_i32);
-        ifn!("llvm.fptosi.sat.i64.f32", fn(t_f32) -> t_i64);
-        ifn!("llvm.fptosi.sat.i128.f32", fn(t_f32) -> t_i128);
-        ifn!("llvm.fptosi.sat.i8.f64", fn(t_f64) -> t_i8);
-        ifn!("llvm.fptosi.sat.i16.f64", fn(t_f64) -> t_i16);
-        ifn!("llvm.fptosi.sat.i32.f64", fn(t_f64) -> t_i32);
-        ifn!("llvm.fptosi.sat.i64.f64", fn(t_f64) -> t_i64);
-        ifn!("llvm.fptosi.sat.i128.f64", fn(t_f64) -> t_i128);
-
-        ifn!("llvm.fptoui.sat.i8.f32", fn(t_f32) -> t_i8);
-        ifn!("llvm.fptoui.sat.i16.f32", fn(t_f32) -> t_i16);
-        ifn!("llvm.fptoui.sat.i32.f32", fn(t_f32) -> t_i32);
-        ifn!("llvm.fptoui.sat.i64.f32", fn(t_f32) -> t_i64);
-        ifn!("llvm.fptoui.sat.i128.f32", fn(t_f32) -> t_i128);
-        ifn!("llvm.fptoui.sat.i8.f64", fn(t_f64) -> t_i8);
-        ifn!("llvm.fptoui.sat.i16.f64", fn(t_f64) -> t_i16);
-        ifn!("llvm.fptoui.sat.i32.f64", fn(t_f64) -> t_i32);
-        ifn!("llvm.fptoui.sat.i64.f64", fn(t_f64) -> t_i64);
-        ifn!("llvm.fptoui.sat.i128.f64", fn(t_f64) -> t_i128);
-
-        ifn!("llvm.trap", fn() -> void);
-        ifn!("llvm.debugtrap", fn() -> void);
-        ifn!("llvm.frameaddress", fn(t_i32) -> ptr);
-
-        ifn!("llvm.powi.f16.i32", fn(t_f16, t_i32) -> t_f16);
-        ifn!("llvm.powi.f32.i32", fn(t_f32, t_i32) -> t_f32);
-        ifn!("llvm.powi.f64.i32", fn(t_f64, t_i32) -> t_f64);
-        ifn!("llvm.powi.f128.i32", fn(t_f128, t_i32) -> t_f128);
-
-        ifn!("llvm.pow.f16", fn(t_f16, t_f16) -> t_f16);
-        ifn!("llvm.pow.f32", fn(t_f32, t_f32) -> t_f32);
-        ifn!("llvm.pow.f64", fn(t_f64, t_f64) -> t_f64);
-        ifn!("llvm.pow.f128", fn(t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.sqrt.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.sqrt.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.sqrt.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.sqrt.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.sin.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.sin.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.sin.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.sin.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.cos.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.cos.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.cos.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.cos.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.exp.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.exp.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.exp.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.exp.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.exp2.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.exp2.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.exp2.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.exp2.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.log.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.log.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.log.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.log.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.log10.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.log10.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.log10.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.log10.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.log2.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.log2.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.log2.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.log2.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.fma.f16", fn(t_f16, t_f16, t_f16) -> t_f16);
-        ifn!("llvm.fma.f32", fn(t_f32, t_f32, t_f32) -> t_f32);
-        ifn!("llvm.fma.f64", fn(t_f64, t_f64, t_f64) -> t_f64);
-        ifn!("llvm.fma.f128", fn(t_f128, t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.fmuladd.f16", fn(t_f16, t_f16, t_f16) -> t_f16);
-        ifn!("llvm.fmuladd.f32", fn(t_f32, t_f32, t_f32) -> t_f32);
-        ifn!("llvm.fmuladd.f64", fn(t_f64, t_f64, t_f64) -> t_f64);
-        ifn!("llvm.fmuladd.f128", fn(t_f128, t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.fabs.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.fabs.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.fabs.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.fabs.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.minnum.f16", fn(t_f16, t_f16) -> t_f16);
-        ifn!("llvm.minnum.f32", fn(t_f32, t_f32) -> t_f32);
-        ifn!("llvm.minnum.f64", fn(t_f64, t_f64) -> t_f64);
-        ifn!("llvm.minnum.f128", fn(t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.minimum.f16", fn(t_f16, t_f16) -> t_f16);
-        ifn!("llvm.minimum.f32", fn(t_f32, t_f32) -> t_f32);
-        ifn!("llvm.minimum.f64", fn(t_f64, t_f64) -> t_f64);
-        // There are issues on x86_64 and aarch64 with the f128 variant.
-        //  - https://github.com/llvm/llvm-project/issues/139380
-        //  - https://github.com/llvm/llvm-project/issues/139381
-        // ifn!("llvm.minimum.f128", fn(t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.maxnum.f16", fn(t_f16, t_f16) -> t_f16);
-        ifn!("llvm.maxnum.f32", fn(t_f32, t_f32) -> t_f32);
-        ifn!("llvm.maxnum.f64", fn(t_f64, t_f64) -> t_f64);
-        ifn!("llvm.maxnum.f128", fn(t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.maximum.f16", fn(t_f16, t_f16) -> t_f16);
-        ifn!("llvm.maximum.f32", fn(t_f32, t_f32) -> t_f32);
-        ifn!("llvm.maximum.f64", fn(t_f64, t_f64) -> t_f64);
-        // There are issues on x86_64 and aarch64 with the f128 variant.
-        //  - https://github.com/llvm/llvm-project/issues/139380
-        //  - https://github.com/llvm/llvm-project/issues/139381
-        // ifn!("llvm.maximum.f128", fn(t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.floor.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.floor.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.floor.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.floor.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.ceil.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.ceil.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.ceil.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.ceil.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.trunc.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.trunc.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.trunc.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.trunc.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.copysign.f16", fn(t_f16, t_f16) -> t_f16);
-        ifn!("llvm.copysign.f32", fn(t_f32, t_f32) -> t_f32);
-        ifn!("llvm.copysign.f64", fn(t_f64, t_f64) -> t_f64);
-        ifn!("llvm.copysign.f128", fn(t_f128, t_f128) -> t_f128);
-
-        ifn!("llvm.round.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.round.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.round.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.round.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.roundeven.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.roundeven.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.roundeven.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.roundeven.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.rint.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.rint.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.rint.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.rint.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.nearbyint.f16", fn(t_f16) -> t_f16);
-        ifn!("llvm.nearbyint.f32", fn(t_f32) -> t_f32);
-        ifn!("llvm.nearbyint.f64", fn(t_f64) -> t_f64);
-        ifn!("llvm.nearbyint.f128", fn(t_f128) -> t_f128);
-
-        ifn!("llvm.ctpop.i8", fn(t_i8) -> t_i8);
-        ifn!("llvm.ctpop.i16", fn(t_i16) -> t_i16);
-        ifn!("llvm.ctpop.i32", fn(t_i32) -> t_i32);
-        ifn!("llvm.ctpop.i64", fn(t_i64) -> t_i64);
-        ifn!("llvm.ctpop.i128", fn(t_i128) -> t_i128);
-
-        ifn!("llvm.ctlz.i8", fn(t_i8, i1) -> t_i8);
-        ifn!("llvm.ctlz.i16", fn(t_i16, i1) -> t_i16);
-        ifn!("llvm.ctlz.i32", fn(t_i32, i1) -> t_i32);
-        ifn!("llvm.ctlz.i64", fn(t_i64, i1) -> t_i64);
-        ifn!("llvm.ctlz.i128", fn(t_i128, i1) -> t_i128);
-
-        ifn!("llvm.cttz.i8", fn(t_i8, i1) -> t_i8);
-        ifn!("llvm.cttz.i16", fn(t_i16, i1) -> t_i16);
-        ifn!("llvm.cttz.i32", fn(t_i32, i1) -> t_i32);
-        ifn!("llvm.cttz.i64", fn(t_i64, i1) -> t_i64);
-        ifn!("llvm.cttz.i128", fn(t_i128, i1) -> t_i128);
-
-        ifn!("llvm.bswap.i16", fn(t_i16) -> t_i16);
-        ifn!("llvm.bswap.i32", fn(t_i32) -> t_i32);
-        ifn!("llvm.bswap.i64", fn(t_i64) -> t_i64);
-        ifn!("llvm.bswap.i128", fn(t_i128) -> t_i128);
-
-        ifn!("llvm.bitreverse.i8", fn(t_i8) -> t_i8);
-        ifn!("llvm.bitreverse.i16", fn(t_i16) -> t_i16);
-        ifn!("llvm.bitreverse.i32", fn(t_i32) -> t_i32);
-        ifn!("llvm.bitreverse.i64", fn(t_i64) -> t_i64);
-        ifn!("llvm.bitreverse.i128", fn(t_i128) -> t_i128);
-
-        ifn!("llvm.fshl.i8", fn(t_i8, t_i8, t_i8) -> t_i8);
-        ifn!("llvm.fshl.i16", fn(t_i16, t_i16, t_i16) -> t_i16);
-        ifn!("llvm.fshl.i32", fn(t_i32, t_i32, t_i32) -> t_i32);
-        ifn!("llvm.fshl.i64", fn(t_i64, t_i64, t_i64) -> t_i64);
-        ifn!("llvm.fshl.i128", fn(t_i128, t_i128, t_i128) -> t_i128);
-
-        ifn!("llvm.fshr.i8", fn(t_i8, t_i8, t_i8) -> t_i8);
-        ifn!("llvm.fshr.i16", fn(t_i16, t_i16, t_i16) -> t_i16);
-        ifn!("llvm.fshr.i32", fn(t_i32, t_i32, t_i32) -> t_i32);
-        ifn!("llvm.fshr.i64", fn(t_i64, t_i64, t_i64) -> t_i64);
-        ifn!("llvm.fshr.i128", fn(t_i128, t_i128, t_i128) -> t_i128);
-
-        ifn!("llvm.sadd.with.overflow.i8", fn(t_i8, t_i8) -> mk_struct! {t_i8, i1});
-        ifn!("llvm.sadd.with.overflow.i16", fn(t_i16, t_i16) -> mk_struct! {t_i16, i1});
-        ifn!("llvm.sadd.with.overflow.i32", fn(t_i32, t_i32) -> mk_struct! {t_i32, i1});
-        ifn!("llvm.sadd.with.overflow.i64", fn(t_i64, t_i64) -> mk_struct! {t_i64, i1});
-        ifn!("llvm.sadd.with.overflow.i128", fn(t_i128, t_i128) -> mk_struct! {t_i128, i1});
-
-        ifn!("llvm.uadd.with.overflow.i8", fn(t_i8, t_i8) -> mk_struct! {t_i8, i1});
-        ifn!("llvm.uadd.with.overflow.i16", fn(t_i16, t_i16) -> mk_struct! {t_i16, i1});
-        ifn!("llvm.uadd.with.overflow.i32", fn(t_i32, t_i32) -> mk_struct! {t_i32, i1});
-        ifn!("llvm.uadd.with.overflow.i64", fn(t_i64, t_i64) -> mk_struct! {t_i64, i1});
-        ifn!("llvm.uadd.with.overflow.i128", fn(t_i128, t_i128) -> mk_struct! {t_i128, i1});
-
-        ifn!("llvm.ssub.with.overflow.i8", fn(t_i8, t_i8) -> mk_struct! {t_i8, i1});
-        ifn!("llvm.ssub.with.overflow.i16", fn(t_i16, t_i16) -> mk_struct! {t_i16, i1});
-        ifn!("llvm.ssub.with.overflow.i32", fn(t_i32, t_i32) -> mk_struct! {t_i32, i1});
-        ifn!("llvm.ssub.with.overflow.i64", fn(t_i64, t_i64) -> mk_struct! {t_i64, i1});
-        ifn!("llvm.ssub.with.overflow.i128", fn(t_i128, t_i128) -> mk_struct! {t_i128, i1});
-
-        ifn!("llvm.usub.with.overflow.i8", fn(t_i8, t_i8) -> mk_struct! {t_i8, i1});
-        ifn!("llvm.usub.with.overflow.i16", fn(t_i16, t_i16) -> mk_struct! {t_i16, i1});
-        ifn!("llvm.usub.with.overflow.i32", fn(t_i32, t_i32) -> mk_struct! {t_i32, i1});
-        ifn!("llvm.usub.with.overflow.i64", fn(t_i64, t_i64) -> mk_struct! {t_i64, i1});
-        ifn!("llvm.usub.with.overflow.i128", fn(t_i128, t_i128) -> mk_struct! {t_i128, i1});
-
-        ifn!("llvm.smul.with.overflow.i8", fn(t_i8, t_i8) -> mk_struct! {t_i8, i1});
-        ifn!("llvm.smul.with.overflow.i16", fn(t_i16, t_i16) -> mk_struct! {t_i16, i1});
-        ifn!("llvm.smul.with.overflow.i32", fn(t_i32, t_i32) -> mk_struct! {t_i32, i1});
-        ifn!("llvm.smul.with.overflow.i64", fn(t_i64, t_i64) -> mk_struct! {t_i64, i1});
-        ifn!("llvm.smul.with.overflow.i128", fn(t_i128, t_i128) -> mk_struct! {t_i128, i1});
-
-        ifn!("llvm.umul.with.overflow.i8", fn(t_i8, t_i8) -> mk_struct! {t_i8, i1});
-        ifn!("llvm.umul.with.overflow.i16", fn(t_i16, t_i16) -> mk_struct! {t_i16, i1});
-        ifn!("llvm.umul.with.overflow.i32", fn(t_i32, t_i32) -> mk_struct! {t_i32, i1});
-        ifn!("llvm.umul.with.overflow.i64", fn(t_i64, t_i64) -> mk_struct! {t_i64, i1});
-        ifn!("llvm.umul.with.overflow.i128", fn(t_i128, t_i128) -> mk_struct! {t_i128, i1});
-
-        ifn!("llvm.sadd.sat.i8", fn(t_i8, t_i8) -> t_i8);
-        ifn!("llvm.sadd.sat.i16", fn(t_i16, t_i16) -> t_i16);
-        ifn!("llvm.sadd.sat.i32", fn(t_i32, t_i32) -> t_i32);
-        ifn!("llvm.sadd.sat.i64", fn(t_i64, t_i64) -> t_i64);
-        ifn!("llvm.sadd.sat.i128", fn(t_i128, t_i128) -> t_i128);
-
-        ifn!("llvm.uadd.sat.i8", fn(t_i8, t_i8) -> t_i8);
-        ifn!("llvm.uadd.sat.i16", fn(t_i16, t_i16) -> t_i16);
-        ifn!("llvm.uadd.sat.i32", fn(t_i32, t_i32) -> t_i32);
-        ifn!("llvm.uadd.sat.i64", fn(t_i64, t_i64) -> t_i64);
-        ifn!("llvm.uadd.sat.i128", fn(t_i128, t_i128) -> t_i128);
-
-        ifn!("llvm.ssub.sat.i8", fn(t_i8, t_i8) -> t_i8);
-        ifn!("llvm.ssub.sat.i16", fn(t_i16, t_i16) -> t_i16);
-        ifn!("llvm.ssub.sat.i32", fn(t_i32, t_i32) -> t_i32);
-        ifn!("llvm.ssub.sat.i64", fn(t_i64, t_i64) -> t_i64);
-        ifn!("llvm.ssub.sat.i128", fn(t_i128, t_i128) -> t_i128);
-
-        ifn!("llvm.usub.sat.i8", fn(t_i8, t_i8) -> t_i8);
-        ifn!("llvm.usub.sat.i16", fn(t_i16, t_i16) -> t_i16);
-        ifn!("llvm.usub.sat.i32", fn(t_i32, t_i32) -> t_i32);
-        ifn!("llvm.usub.sat.i64", fn(t_i64, t_i64) -> t_i64);
-        ifn!("llvm.usub.sat.i128", fn(t_i128, t_i128) -> t_i128);
-
-        ifn!("llvm.scmp.i8.i8", fn(t_i8, t_i8) -> t_i8);
-        ifn!("llvm.scmp.i8.i16", fn(t_i16, t_i16) -> t_i8);
-        ifn!("llvm.scmp.i8.i32", fn(t_i32, t_i32) -> t_i8);
-        ifn!("llvm.scmp.i8.i64", fn(t_i64, t_i64) -> t_i8);
-        ifn!("llvm.scmp.i8.i128", fn(t_i128, t_i128) -> t_i8);
-
-        ifn!("llvm.ucmp.i8.i8", fn(t_i8, t_i8) -> t_i8);
-        ifn!("llvm.ucmp.i8.i16", fn(t_i16, t_i16) -> t_i8);
-        ifn!("llvm.ucmp.i8.i32", fn(t_i32, t_i32) -> t_i8);
-        ifn!("llvm.ucmp.i8.i64", fn(t_i64, t_i64) -> t_i8);
-        ifn!("llvm.ucmp.i8.i128", fn(t_i128, t_i128) -> t_i8);
-
-        ifn!("llvm.lifetime.start.p0i8", fn(t_i64, ptr) -> void);
-        ifn!("llvm.lifetime.end.p0i8", fn(t_i64, ptr) -> void);
-
-        // FIXME: This is an infinitesimally small portion of the types you can
-        // pass to this intrinsic, if we can ever lazily register intrinsics we
-        // should register these when they're used, that way any type can be
-        // passed.
-        ifn!("llvm.is.constant.i1", fn(i1) -> i1);
-        ifn!("llvm.is.constant.i8", fn(t_i8) -> i1);
-        ifn!("llvm.is.constant.i16", fn(t_i16) -> i1);
-        ifn!("llvm.is.constant.i32", fn(t_i32) -> i1);
-        ifn!("llvm.is.constant.i64", fn(t_i64) -> i1);
-        ifn!("llvm.is.constant.i128", fn(t_i128) -> i1);
-        ifn!("llvm.is.constant.isize", fn(t_isize) -> i1);
-        ifn!("llvm.is.constant.f16", fn(t_f16) -> i1);
-        ifn!("llvm.is.constant.f32", fn(t_f32) -> i1);
-        ifn!("llvm.is.constant.f64", fn(t_f64) -> i1);
-        ifn!("llvm.is.constant.f128", fn(t_f128) -> i1);
-        ifn!("llvm.is.constant.ptr", fn(ptr) -> i1);
-
-        ifn!("llvm.expect.i1", fn(i1, i1) -> i1);
-        ifn!("llvm.eh.typeid.for", fn(ptr) -> t_i32);
-        ifn!("llvm.localescape", fn(...) -> void);
-        ifn!("llvm.localrecover", fn(ptr, ptr, t_i32) -> ptr);
-        ifn!("llvm.x86.seh.recoverfp", fn(ptr, ptr) -> ptr);
-
-        ifn!("llvm.assume", fn(i1) -> void);
-        ifn!("llvm.prefetch", fn(ptr, t_i32, t_i32, t_i32) -> void);
-
+        base_name: Cow<'static, str>,
+        type_params: &[&'ll Type],
+    ) -> (&'ll Type, &'ll Value) {
+        *self
+            .intrinsics
+            .borrow_mut()
+            .entry((base_name, SmallVec::from_slice(type_params)))
+            .or_insert_with_key(|(base_name, type_params)| {
+                self.declare_intrinsic(base_name, type_params)
+            })
+    }
+
+    fn declare_intrinsic(
+        &self,
+        base_name: &str,
+        type_params: &[&'ll Type],
+    ) -> (&'ll Type, &'ll Value) {
         // This isn't an "LLVM intrinsic", but LLVM's optimization passes
         // recognize it like one (including turning it into `bcmp` sometimes)
         // and we use it to implement intrinsics like `raw_eq` and `compare_bytes`
-        match self.sess().target.arch.as_ref() {
-            "avr" | "msp430" => ifn!("memcmp", fn(ptr, ptr, t_isize) -> t_i16),
-            _ => ifn!("memcmp", fn(ptr, ptr, t_isize) -> t_i32),
-        }
-
-        // variadic intrinsics
-        ifn!("llvm.va_start", fn(ptr) -> void);
-        ifn!("llvm.va_end", fn(ptr) -> void);
-        ifn!("llvm.va_copy", fn(ptr, ptr) -> void);
-
-        if self.sess().instrument_coverage() {
-            ifn!("llvm.instrprof.increment", fn(ptr, t_i64, t_i32, t_i32) -> void);
-            ifn!("llvm.instrprof.mcdc.parameters", fn(ptr, t_i64, t_i32) -> void);
-            ifn!("llvm.instrprof.mcdc.tvbitmap.update", fn(ptr, t_i64, t_i32, ptr) -> void);
-        }
-
-        ifn!("llvm.type.test", fn(ptr, t_metadata) -> i1);
-        ifn!("llvm.type.checked.load", fn(ptr, t_i32, t_metadata) -> mk_struct! {ptr, i1});
+        if base_name == "memcmp" {
+            let fn_ty = self
+                .type_func(&[self.type_ptr(), self.type_ptr(), self.type_isize()], self.type_int());
+            let f = self.declare_cfn("memcmp", llvm::UnnamedAddr::No, fn_ty);
 
-        if self.sess().opts.debuginfo != DebugInfo::None {
-            ifn!("llvm.dbg.declare", fn(t_metadata, t_metadata) -> void);
-            ifn!("llvm.dbg.value", fn(t_metadata, t_i64, t_metadata) -> void);
+            return (fn_ty, f);
         }
 
-        ifn!("llvm.ptrmask", fn(ptr, t_isize) -> ptr);
-        ifn!("llvm.threadlocal.address", fn(ptr) -> ptr);
+        let intrinsic = llvm::Intrinsic::lookup(base_name.as_bytes())
+            .unwrap_or_else(|| bug!("Unknown intrinsic: `{base_name}`"));
+        let f = intrinsic.get_declaration(self.llmod, &type_params);
 
-        None
+        (self.get_type_of_global(f), f)
     }
 
     pub(crate) fn eh_catch_typeinfo(&self) -> &'ll Value {
diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs b/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs
index 8f0948b8183..61555ac2f6f 100644
--- a/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs
+++ b/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs
@@ -1,13 +1,12 @@
 // .debug_gdb_scripts binary section.
 
-use rustc_ast::attr;
+use rustc_attr_data_structures::{AttributeKind, find_attr};
 use rustc_codegen_ssa::base::collect_debugger_visualizers_transitive;
 use rustc_codegen_ssa::traits::*;
 use rustc_hir::def_id::LOCAL_CRATE;
 use rustc_middle::bug;
 use rustc_middle::middle::debugger_visualizer::DebuggerVisualizerType;
 use rustc_session::config::{CrateType, DebugInfo};
-use rustc_span::sym;
 
 use crate::builder::Builder;
 use crate::common::CodegenCx;
@@ -75,7 +74,7 @@ pub(crate) fn get_or_insert_gdb_debug_scripts_section_global<'ll>(
             llvm::set_section(section_var, c".debug_gdb_scripts");
             llvm::set_initializer(section_var, cx.const_bytes(section_contents));
             llvm::LLVMSetGlobalConstant(section_var, llvm::True);
-            llvm::LLVMSetUnnamedAddress(section_var, llvm::UnnamedAddr::Global);
+            llvm::set_unnamed_address(section_var, llvm::UnnamedAddr::Global);
             llvm::set_linkage(section_var, llvm::Linkage::LinkOnceODRLinkage);
             // This should make sure that the whole section is not larger than
             // the string it contains. Otherwise we get a warning from GDB.
@@ -87,7 +86,7 @@ pub(crate) fn get_or_insert_gdb_debug_scripts_section_global<'ll>(
 
 pub(crate) fn needs_gdb_debug_scripts_section(cx: &CodegenCx<'_, '_>) -> bool {
     let omit_gdb_pretty_printer_section =
-        attr::contains_name(cx.tcx.hir_krate_attrs(), sym::omit_gdb_pretty_printer_section);
+        find_attr!(cx.tcx.hir_krate_attrs(), AttributeKind::OmitGdbPrettyPrinterSection);
 
     // To ensure the section `__rustc_debug_gdb_scripts_section__` will not create
     // ODR violations at link time, this section will not be emitted for rlibs since
diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs b/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs
index 7f3e486ca31..0e9dbfba658 100644
--- a/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs
+++ b/compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs
@@ -5,7 +5,7 @@ use std::path::{Path, PathBuf};
 use std::sync::Arc;
 use std::{iter, ptr};
 
-use libc::{c_char, c_longlong, c_uint};
+use libc::{c_longlong, c_uint};
 use rustc_abi::{Align, Size};
 use rustc_codegen_ssa::debuginfo::type_names::{VTableNameKind, cpp_like_debuginfo};
 use rustc_codegen_ssa::traits::*;
@@ -159,13 +159,15 @@ fn build_pointer_or_reference_di_node<'ll, 'tcx>(
     return_if_di_node_created_in_meantime!(cx, unique_type_id);
 
     let data_layout = &cx.tcx.data_layout;
+    let pointer_size = data_layout.pointer_size();
+    let pointer_align = data_layout.pointer_align();
     let ptr_type_debuginfo_name = compute_debuginfo_type_name(cx.tcx, ptr_type, true);
 
     match wide_pointer_kind(cx, pointee_type) {
         None => {
             // This is a thin pointer. Create a regular pointer type and give it the correct name.
             assert_eq!(
-                (data_layout.pointer_size, data_layout.pointer_align.abi),
+                (pointer_size, pointer_align.abi),
                 cx.size_and_align_of(ptr_type),
                 "ptr_type={ptr_type}, pointee_type={pointee_type}",
             );
@@ -174,8 +176,8 @@ fn build_pointer_or_reference_di_node<'ll, 'tcx>(
                 llvm::LLVMRustDIBuilderCreatePointerType(
                     DIB(cx),
                     pointee_type_di_node,
-                    data_layout.pointer_size.bits(),
-                    data_layout.pointer_align.abi.bits() as u32,
+                    pointer_size.bits(),
+                    pointer_align.abi.bits() as u32,
                     0, // Ignore DWARF address space.
                     ptr_type_debuginfo_name.as_c_char_ptr(),
                     ptr_type_debuginfo_name.len(),
@@ -319,7 +321,9 @@ fn build_subroutine_type_di_node<'ll, 'tcx>(
     let name = compute_debuginfo_type_name(cx.tcx, fn_ty, false);
     let (size, align) = match fn_ty.kind() {
         ty::FnDef(..) => (Size::ZERO, Align::ONE),
-        ty::FnPtr(..) => (cx.tcx.data_layout.pointer_size, cx.tcx.data_layout.pointer_align.abi),
+        ty::FnPtr(..) => {
+            (cx.tcx.data_layout.pointer_size(), cx.tcx.data_layout.pointer_align().abi)
+        }
         _ => unreachable!(),
     };
     let di_node = unsafe {
@@ -504,7 +508,7 @@ fn recursion_marker_type_di_node<'ll, 'tcx>(cx: &CodegenCx<'ll, 'tcx>) -> &'ll D
         create_basic_type(
             cx,
             "<recur_type>",
-            cx.tcx.data_layout.pointer_size,
+            cx.tcx.data_layout.pointer_size(),
             dwarf_const::DW_ATE_unsigned,
         )
     })
@@ -1578,13 +1582,9 @@ pub(crate) fn apply_vcall_visibility_metadata<'ll, 'tcx>(
     };
 
     let trait_ref_typeid = typeid_for_trait_ref(cx.tcx, trait_ref);
+    let typeid = cx.create_metadata(trait_ref_typeid.as_bytes());
 
     unsafe {
-        let typeid = llvm::LLVMMDStringInContext2(
-            cx.llcx,
-            trait_ref_typeid.as_ptr() as *const c_char,
-            trait_ref_typeid.as_bytes().len(),
-        );
         let v = [llvm::LLVMValueAsMetadata(cx.const_usize(0)), typeid];
         llvm::LLVMRustGlobalAddMetadata(
             vtable,
@@ -1626,7 +1626,7 @@ pub(crate) fn create_vtable_di_node<'ll, 'tcx>(
     // When full debuginfo is enabled, we want to try and prevent vtables from being
     // merged. Otherwise debuggers will have a hard time mapping from dyn pointer
     // to concrete type.
-    llvm::SetUnnamedAddress(vtable, llvm::UnnamedAddr::No);
+    llvm::set_unnamed_address(vtable, llvm::UnnamedAddr::No);
 
     let vtable_name =
         compute_debuginfo_vtable_name(cx.tcx, ty, poly_trait_ref, VTableNameKind::GlobalVariable);
diff --git a/compiler/rustc_codegen_llvm/src/declare.rs b/compiler/rustc_codegen_llvm/src/declare.rs
index 2419ec1f888..eb75716d768 100644
--- a/compiler/rustc_codegen_llvm/src/declare.rs
+++ b/compiler/rustc_codegen_llvm/src/declare.rs
@@ -49,7 +49,7 @@ pub(crate) fn declare_simple_fn<'ll>(
     };
 
     llvm::SetFunctionCallConv(llfn, callconv);
-    llvm::SetUnnamedAddress(llfn, unnamed);
+    llvm::set_unnamed_address(llfn, unnamed);
     llvm::set_visibility(llfn, visibility);
 
     llfn
@@ -176,7 +176,7 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
                 {
                     let typeid = cfi::typeid_for_instance(self.tcx, instance, options);
                     if typeids.insert(typeid.clone()) {
-                        self.add_type_metadata(llfn, typeid);
+                        self.add_type_metadata(llfn, typeid.as_bytes());
                     }
                 }
             } else {
@@ -189,7 +189,7 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
                 .map(cfi::TypeIdOptions::from_iter)
                 {
                     let typeid = cfi::typeid_for_fnabi(self.tcx, fn_abi, options);
-                    self.add_type_metadata(llfn, typeid);
+                    self.add_type_metadata(llfn, typeid.as_bytes());
                 }
             }
         }
diff --git a/compiler/rustc_codegen_llvm/src/errors.rs b/compiler/rustc_codegen_llvm/src/errors.rs
index 8bc74fbec7e..31d49e86319 100644
--- a/compiler/rustc_codegen_llvm/src/errors.rs
+++ b/compiler/rustc_codegen_llvm/src/errors.rs
@@ -3,36 +3,12 @@ use std::path::Path;
 
 use rustc_data_structures::small_c_str::SmallCStr;
 use rustc_errors::{Diag, DiagCtxtHandle, Diagnostic, EmissionGuarantee, Level};
-use rustc_macros::{Diagnostic, Subdiagnostic};
+use rustc_macros::Diagnostic;
 use rustc_span::Span;
 
 use crate::fluent_generated as fluent;
 
 #[derive(Diagnostic)]
-#[diag(codegen_llvm_unknown_ctarget_feature_prefix)]
-#[note]
-pub(crate) struct UnknownCTargetFeaturePrefix<'a> {
-    pub feature: &'a str,
-}
-
-#[derive(Diagnostic)]
-#[diag(codegen_llvm_unknown_ctarget_feature)]
-#[note]
-pub(crate) struct UnknownCTargetFeature<'a> {
-    pub feature: &'a str,
-    #[subdiagnostic]
-    pub rust_feature: PossibleFeature<'a>,
-}
-
-#[derive(Subdiagnostic)]
-pub(crate) enum PossibleFeature<'a> {
-    #[help(codegen_llvm_possible_feature)]
-    Some { rust_feature: &'a str },
-    #[help(codegen_llvm_consider_filing_feature_request)]
-    None,
-}
-
-#[derive(Diagnostic)]
 #[diag(codegen_llvm_symbol_already_defined)]
 pub(crate) struct SymbolAlreadyDefined<'a> {
     #[primary_span]
@@ -62,10 +38,6 @@ impl<G: EmissionGuarantee> Diagnostic<'_, G> for ParseTargetMachineConfig<'_> {
 }
 
 #[derive(Diagnostic)]
-#[diag(codegen_llvm_autodiff_without_lto)]
-pub(crate) struct AutoDiffWithoutLTO;
-
-#[derive(Diagnostic)]
 #[diag(codegen_llvm_autodiff_without_enable)]
 pub(crate) struct AutoDiffWithoutEnable;
 
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index 10697b9a71f..fcc0d378f06 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -15,7 +15,7 @@ use rustc_middle::ty::{self, GenericArgsRef, Ty};
 use rustc_middle::{bug, span_bug};
 use rustc_span::{Span, Symbol, sym};
 use rustc_symbol_mangling::mangle_internal_symbol;
-use rustc_target::spec::{HasTargetSpec, PanicStrategy};
+use rustc_target::spec::PanicStrategy;
 use tracing::debug;
 
 use crate::abi::FnAbiLlvmExt;
@@ -27,137 +27,142 @@ use crate::type_of::LayoutLlvmExt;
 use crate::va_arg::emit_va_arg;
 use crate::value::Value;
 
-fn get_simple_intrinsic<'ll>(
-    cx: &CodegenCx<'ll, '_>,
+fn call_simple_intrinsic<'ll, 'tcx>(
+    bx: &mut Builder<'_, 'll, 'tcx>,
     name: Symbol,
-) -> Option<(&'ll Type, &'ll Value)> {
-    let llvm_name = match name {
-        sym::sqrtf16 => "llvm.sqrt.f16",
-        sym::sqrtf32 => "llvm.sqrt.f32",
-        sym::sqrtf64 => "llvm.sqrt.f64",
-        sym::sqrtf128 => "llvm.sqrt.f128",
-
-        sym::powif16 => "llvm.powi.f16.i32",
-        sym::powif32 => "llvm.powi.f32.i32",
-        sym::powif64 => "llvm.powi.f64.i32",
-        sym::powif128 => "llvm.powi.f128.i32",
-
-        sym::sinf16 => "llvm.sin.f16",
-        sym::sinf32 => "llvm.sin.f32",
-        sym::sinf64 => "llvm.sin.f64",
-        sym::sinf128 => "llvm.sin.f128",
-
-        sym::cosf16 => "llvm.cos.f16",
-        sym::cosf32 => "llvm.cos.f32",
-        sym::cosf64 => "llvm.cos.f64",
-        sym::cosf128 => "llvm.cos.f128",
-
-        sym::powf16 => "llvm.pow.f16",
-        sym::powf32 => "llvm.pow.f32",
-        sym::powf64 => "llvm.pow.f64",
-        sym::powf128 => "llvm.pow.f128",
-
-        sym::expf16 => "llvm.exp.f16",
-        sym::expf32 => "llvm.exp.f32",
-        sym::expf64 => "llvm.exp.f64",
-        sym::expf128 => "llvm.exp.f128",
-
-        sym::exp2f16 => "llvm.exp2.f16",
-        sym::exp2f32 => "llvm.exp2.f32",
-        sym::exp2f64 => "llvm.exp2.f64",
-        sym::exp2f128 => "llvm.exp2.f128",
-
-        sym::logf16 => "llvm.log.f16",
-        sym::logf32 => "llvm.log.f32",
-        sym::logf64 => "llvm.log.f64",
-        sym::logf128 => "llvm.log.f128",
-
-        sym::log10f16 => "llvm.log10.f16",
-        sym::log10f32 => "llvm.log10.f32",
-        sym::log10f64 => "llvm.log10.f64",
-        sym::log10f128 => "llvm.log10.f128",
-
-        sym::log2f16 => "llvm.log2.f16",
-        sym::log2f32 => "llvm.log2.f32",
-        sym::log2f64 => "llvm.log2.f64",
-        sym::log2f128 => "llvm.log2.f128",
-
-        sym::fmaf16 => "llvm.fma.f16",
-        sym::fmaf32 => "llvm.fma.f32",
-        sym::fmaf64 => "llvm.fma.f64",
-        sym::fmaf128 => "llvm.fma.f128",
-
-        sym::fmuladdf16 => "llvm.fmuladd.f16",
-        sym::fmuladdf32 => "llvm.fmuladd.f32",
-        sym::fmuladdf64 => "llvm.fmuladd.f64",
-        sym::fmuladdf128 => "llvm.fmuladd.f128",
-
-        sym::fabsf16 => "llvm.fabs.f16",
-        sym::fabsf32 => "llvm.fabs.f32",
-        sym::fabsf64 => "llvm.fabs.f64",
-        sym::fabsf128 => "llvm.fabs.f128",
-
-        sym::minnumf16 => "llvm.minnum.f16",
-        sym::minnumf32 => "llvm.minnum.f32",
-        sym::minnumf64 => "llvm.minnum.f64",
-        sym::minnumf128 => "llvm.minnum.f128",
-
-        sym::minimumf16 => "llvm.minimum.f16",
-        sym::minimumf32 => "llvm.minimum.f32",
-        sym::minimumf64 => "llvm.minimum.f64",
-        // There are issues on x86_64 and aarch64 with the f128 variant,
-        // let's instead use the instrinsic fallback body.
-        // sym::minimumf128 => "llvm.minimum.f128",
-        sym::maxnumf16 => "llvm.maxnum.f16",
-        sym::maxnumf32 => "llvm.maxnum.f32",
-        sym::maxnumf64 => "llvm.maxnum.f64",
-        sym::maxnumf128 => "llvm.maxnum.f128",
-
-        sym::maximumf16 => "llvm.maximum.f16",
-        sym::maximumf32 => "llvm.maximum.f32",
-        sym::maximumf64 => "llvm.maximum.f64",
-        // There are issues on x86_64 and aarch64 with the f128 variant,
-        // let's instead use the instrinsic fallback body.
-        // sym::maximumf128 => "llvm.maximum.f128",
-        sym::copysignf16 => "llvm.copysign.f16",
-        sym::copysignf32 => "llvm.copysign.f32",
-        sym::copysignf64 => "llvm.copysign.f64",
-        sym::copysignf128 => "llvm.copysign.f128",
-
-        sym::floorf16 => "llvm.floor.f16",
-        sym::floorf32 => "llvm.floor.f32",
-        sym::floorf64 => "llvm.floor.f64",
-        sym::floorf128 => "llvm.floor.f128",
-
-        sym::ceilf16 => "llvm.ceil.f16",
-        sym::ceilf32 => "llvm.ceil.f32",
-        sym::ceilf64 => "llvm.ceil.f64",
-        sym::ceilf128 => "llvm.ceil.f128",
-
-        sym::truncf16 => "llvm.trunc.f16",
-        sym::truncf32 => "llvm.trunc.f32",
-        sym::truncf64 => "llvm.trunc.f64",
-        sym::truncf128 => "llvm.trunc.f128",
+    args: &[OperandRef<'tcx, &'ll Value>],
+) -> Option<&'ll Value> {
+    let (base_name, type_params): (&'static str, &[&'ll Type]) = match name {
+        sym::sqrtf16 => ("llvm.sqrt", &[bx.type_f16()]),
+        sym::sqrtf32 => ("llvm.sqrt", &[bx.type_f32()]),
+        sym::sqrtf64 => ("llvm.sqrt", &[bx.type_f64()]),
+        sym::sqrtf128 => ("llvm.sqrt", &[bx.type_f128()]),
+
+        sym::powif16 => ("llvm.powi", &[bx.type_f16(), bx.type_i32()]),
+        sym::powif32 => ("llvm.powi", &[bx.type_f32(), bx.type_i32()]),
+        sym::powif64 => ("llvm.powi", &[bx.type_f64(), bx.type_i32()]),
+        sym::powif128 => ("llvm.powi", &[bx.type_f128(), bx.type_i32()]),
+
+        sym::sinf16 => ("llvm.sin", &[bx.type_f16()]),
+        sym::sinf32 => ("llvm.sin", &[bx.type_f32()]),
+        sym::sinf64 => ("llvm.sin", &[bx.type_f64()]),
+        sym::sinf128 => ("llvm.sin", &[bx.type_f128()]),
+
+        sym::cosf16 => ("llvm.cos", &[bx.type_f16()]),
+        sym::cosf32 => ("llvm.cos", &[bx.type_f32()]),
+        sym::cosf64 => ("llvm.cos", &[bx.type_f64()]),
+        sym::cosf128 => ("llvm.cos", &[bx.type_f128()]),
+
+        sym::powf16 => ("llvm.pow", &[bx.type_f16()]),
+        sym::powf32 => ("llvm.pow", &[bx.type_f32()]),
+        sym::powf64 => ("llvm.pow", &[bx.type_f64()]),
+        sym::powf128 => ("llvm.pow", &[bx.type_f128()]),
+
+        sym::expf16 => ("llvm.exp", &[bx.type_f16()]),
+        sym::expf32 => ("llvm.exp", &[bx.type_f32()]),
+        sym::expf64 => ("llvm.exp", &[bx.type_f64()]),
+        sym::expf128 => ("llvm.exp", &[bx.type_f128()]),
+
+        sym::exp2f16 => ("llvm.exp2", &[bx.type_f16()]),
+        sym::exp2f32 => ("llvm.exp2", &[bx.type_f32()]),
+        sym::exp2f64 => ("llvm.exp2", &[bx.type_f64()]),
+        sym::exp2f128 => ("llvm.exp2", &[bx.type_f128()]),
+
+        sym::logf16 => ("llvm.log", &[bx.type_f16()]),
+        sym::logf32 => ("llvm.log", &[bx.type_f32()]),
+        sym::logf64 => ("llvm.log", &[bx.type_f64()]),
+        sym::logf128 => ("llvm.log", &[bx.type_f128()]),
+
+        sym::log10f16 => ("llvm.log10", &[bx.type_f16()]),
+        sym::log10f32 => ("llvm.log10", &[bx.type_f32()]),
+        sym::log10f64 => ("llvm.log10", &[bx.type_f64()]),
+        sym::log10f128 => ("llvm.log10", &[bx.type_f128()]),
+
+        sym::log2f16 => ("llvm.log2", &[bx.type_f16()]),
+        sym::log2f32 => ("llvm.log2", &[bx.type_f32()]),
+        sym::log2f64 => ("llvm.log2", &[bx.type_f64()]),
+        sym::log2f128 => ("llvm.log2", &[bx.type_f128()]),
+
+        sym::fmaf16 => ("llvm.fma", &[bx.type_f16()]),
+        sym::fmaf32 => ("llvm.fma", &[bx.type_f32()]),
+        sym::fmaf64 => ("llvm.fma", &[bx.type_f64()]),
+        sym::fmaf128 => ("llvm.fma", &[bx.type_f128()]),
+
+        sym::fmuladdf16 => ("llvm.fmuladd", &[bx.type_f16()]),
+        sym::fmuladdf32 => ("llvm.fmuladd", &[bx.type_f32()]),
+        sym::fmuladdf64 => ("llvm.fmuladd", &[bx.type_f64()]),
+        sym::fmuladdf128 => ("llvm.fmuladd", &[bx.type_f128()]),
+
+        sym::fabsf16 => ("llvm.fabs", &[bx.type_f16()]),
+        sym::fabsf32 => ("llvm.fabs", &[bx.type_f32()]),
+        sym::fabsf64 => ("llvm.fabs", &[bx.type_f64()]),
+        sym::fabsf128 => ("llvm.fabs", &[bx.type_f128()]),
+
+        sym::minnumf16 => ("llvm.minnum", &[bx.type_f16()]),
+        sym::minnumf32 => ("llvm.minnum", &[bx.type_f32()]),
+        sym::minnumf64 => ("llvm.minnum", &[bx.type_f64()]),
+        sym::minnumf128 => ("llvm.minnum", &[bx.type_f128()]),
+
+        // FIXME: LLVM currently mis-compile those intrinsics, re-enable them
+        // when llvm/llvm-project#{139380,139381,140445} are fixed.
+        //sym::minimumf16 => ("llvm.minimum", &[bx.type_f16()]),
+        //sym::minimumf32 => ("llvm.minimum", &[bx.type_f32()]),
+        //sym::minimumf64 => ("llvm.minimum", &[bx.type_f64()]),
+        //sym::minimumf128 => ("llvm.minimum", &[cx.type_f128()]),
+        //
+        sym::maxnumf16 => ("llvm.maxnum", &[bx.type_f16()]),
+        sym::maxnumf32 => ("llvm.maxnum", &[bx.type_f32()]),
+        sym::maxnumf64 => ("llvm.maxnum", &[bx.type_f64()]),
+        sym::maxnumf128 => ("llvm.maxnum", &[bx.type_f128()]),
+
+        // FIXME: LLVM currently mis-compile those intrinsics, re-enable them
+        // when llvm/llvm-project#{139380,139381,140445} are fixed.
+        //sym::maximumf16 => ("llvm.maximum", &[bx.type_f16()]),
+        //sym::maximumf32 => ("llvm.maximum", &[bx.type_f32()]),
+        //sym::maximumf64 => ("llvm.maximum", &[bx.type_f64()]),
+        //sym::maximumf128 => ("llvm.maximum", &[cx.type_f128()]),
+        //
+        sym::copysignf16 => ("llvm.copysign", &[bx.type_f16()]),
+        sym::copysignf32 => ("llvm.copysign", &[bx.type_f32()]),
+        sym::copysignf64 => ("llvm.copysign", &[bx.type_f64()]),
+        sym::copysignf128 => ("llvm.copysign", &[bx.type_f128()]),
+
+        sym::floorf16 => ("llvm.floor", &[bx.type_f16()]),
+        sym::floorf32 => ("llvm.floor", &[bx.type_f32()]),
+        sym::floorf64 => ("llvm.floor", &[bx.type_f64()]),
+        sym::floorf128 => ("llvm.floor", &[bx.type_f128()]),
+
+        sym::ceilf16 => ("llvm.ceil", &[bx.type_f16()]),
+        sym::ceilf32 => ("llvm.ceil", &[bx.type_f32()]),
+        sym::ceilf64 => ("llvm.ceil", &[bx.type_f64()]),
+        sym::ceilf128 => ("llvm.ceil", &[bx.type_f128()]),
+
+        sym::truncf16 => ("llvm.trunc", &[bx.type_f16()]),
+        sym::truncf32 => ("llvm.trunc", &[bx.type_f32()]),
+        sym::truncf64 => ("llvm.trunc", &[bx.type_f64()]),
+        sym::truncf128 => ("llvm.trunc", &[bx.type_f128()]),
 
         // We could use any of `rint`, `nearbyint`, or `roundeven`
         // for this -- they are all identical in semantics when
         // assuming the default FP environment.
         // `rint` is what we used for $forever.
-        sym::round_ties_even_f16 => "llvm.rint.f16",
-        sym::round_ties_even_f32 => "llvm.rint.f32",
-        sym::round_ties_even_f64 => "llvm.rint.f64",
-        sym::round_ties_even_f128 => "llvm.rint.f128",
+        sym::round_ties_even_f16 => ("llvm.rint", &[bx.type_f16()]),
+        sym::round_ties_even_f32 => ("llvm.rint", &[bx.type_f32()]),
+        sym::round_ties_even_f64 => ("llvm.rint", &[bx.type_f64()]),
+        sym::round_ties_even_f128 => ("llvm.rint", &[bx.type_f128()]),
 
-        sym::roundf16 => "llvm.round.f16",
-        sym::roundf32 => "llvm.round.f32",
-        sym::roundf64 => "llvm.round.f64",
-        sym::roundf128 => "llvm.round.f128",
-
-        sym::ptr_mask => "llvm.ptrmask",
+        sym::roundf16 => ("llvm.round", &[bx.type_f16()]),
+        sym::roundf32 => ("llvm.round", &[bx.type_f32()]),
+        sym::roundf64 => ("llvm.round", &[bx.type_f64()]),
+        sym::roundf128 => ("llvm.round", &[bx.type_f128()]),
 
         _ => return None,
     };
-    Some(cx.get_intrinsic(llvm_name))
+    Some(bx.call_intrinsic(
+        base_name,
+        type_params,
+        &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
+    ))
 }
 
 impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
@@ -173,36 +178,24 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
         let name = tcx.item_name(instance.def_id());
         let fn_args = instance.args;
 
-        let simple = get_simple_intrinsic(self, name);
+        let simple = call_simple_intrinsic(self, name, args);
         let llval = match name {
-            _ if simple.is_some() => {
-                let (simple_ty, simple_fn) = simple.unwrap();
-                self.call(
-                    simple_ty,
-                    None,
-                    None,
-                    simple_fn,
-                    &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
-                    None,
-                    Some(instance),
+            _ if simple.is_some() => simple.unwrap(),
+            sym::ptr_mask => {
+                let ptr = args[0].immediate();
+                self.call_intrinsic(
+                    "llvm.ptrmask",
+                    &[self.val_ty(ptr), self.type_isize()],
+                    &[ptr, args[1].immediate()],
                 )
             }
             sym::is_val_statically_known => {
-                let intrinsic_type = args[0].layout.immediate_llvm_type(self.cx);
-                let kind = self.type_kind(intrinsic_type);
-                let intrinsic_name = match kind {
-                    TypeKind::Pointer | TypeKind::Integer => {
-                        Some(format!("llvm.is.constant.{intrinsic_type:?}"))
-                    }
-                    // LLVM float types' intrinsic names differ from their type names.
-                    TypeKind::Half => Some(format!("llvm.is.constant.f16")),
-                    TypeKind::Float => Some(format!("llvm.is.constant.f32")),
-                    TypeKind::Double => Some(format!("llvm.is.constant.f64")),
-                    TypeKind::FP128 => Some(format!("llvm.is.constant.f128")),
-                    _ => None,
-                };
-                if let Some(intrinsic_name) = intrinsic_name {
-                    self.call_intrinsic(&intrinsic_name, &[args[0].immediate()])
+                if let OperandValue::Immediate(imm) = args[0].val {
+                    self.call_intrinsic(
+                        "llvm.is.constant",
+                        &[args[0].layout.immediate_llvm_type(self.cx)],
+                        &[imm],
+                    )
                 } else {
                     self.const_bool(false)
                 }
@@ -246,9 +239,14 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                 );
                 return Ok(());
             }
-            sym::breakpoint => self.call_intrinsic("llvm.debugtrap", &[]),
+            sym::breakpoint => self.call_intrinsic("llvm.debugtrap", &[], &[]),
             sym::va_copy => {
-                self.call_intrinsic("llvm.va_copy", &[args[0].immediate(), args[1].immediate()])
+                let dest = args[0].immediate();
+                self.call_intrinsic(
+                    "llvm.va_copy",
+                    &[self.val_ty(dest)],
+                    &[dest, args[1].immediate()],
+                )
             }
             sym::va_arg => {
                 match result.layout.backend_repr {
@@ -322,14 +320,11 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     sym::prefetch_write_instruction => (1, 0),
                     _ => bug!(),
                 };
+                let ptr = args[0].immediate();
                 self.call_intrinsic(
                     "llvm.prefetch",
-                    &[
-                        args[0].immediate(),
-                        self.const_i32(rw),
-                        args[1].immediate(),
-                        self.const_i32(cache_type),
-                    ],
+                    &[self.val_ty(ptr)],
+                    &[ptr, self.const_i32(rw), args[1].immediate(), self.const_i32(cache_type)],
                 )
             }
             sym::carrying_mul_add => {
@@ -385,11 +380,13 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                 }
                 let (size, signed) = ty.int_size_and_signed(self.tcx);
                 let width = size.bits();
+                let llty = self.type_ix(width);
                 match name {
                     sym::ctlz | sym::cttz => {
                         let y = self.const_bool(false);
                         let ret = self.call_intrinsic(
-                            &format!("llvm.{name}.i{width}"),
+                            format!("llvm.{name}"),
+                            &[llty],
                             &[args[0].immediate(), y],
                         );
 
@@ -397,62 +394,54 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     }
                     sym::ctlz_nonzero => {
                         let y = self.const_bool(true);
-                        let llvm_name = &format!("llvm.ctlz.i{width}");
-                        let ret = self.call_intrinsic(llvm_name, &[args[0].immediate(), y]);
+                        let ret =
+                            self.call_intrinsic("llvm.ctlz", &[llty], &[args[0].immediate(), y]);
                         self.intcast(ret, result.layout.llvm_type(self), false)
                     }
                     sym::cttz_nonzero => {
                         let y = self.const_bool(true);
-                        let llvm_name = &format!("llvm.cttz.i{width}");
-                        let ret = self.call_intrinsic(llvm_name, &[args[0].immediate(), y]);
+                        let ret =
+                            self.call_intrinsic("llvm.cttz", &[llty], &[args[0].immediate(), y]);
                         self.intcast(ret, result.layout.llvm_type(self), false)
                     }
                     sym::ctpop => {
-                        let ret = self.call_intrinsic(
-                            &format!("llvm.ctpop.i{width}"),
-                            &[args[0].immediate()],
-                        );
+                        let ret =
+                            self.call_intrinsic("llvm.ctpop", &[llty], &[args[0].immediate()]);
                         self.intcast(ret, result.layout.llvm_type(self), false)
                     }
                     sym::bswap => {
                         if width == 8 {
                             args[0].immediate() // byte swap a u8/i8 is just a no-op
                         } else {
-                            self.call_intrinsic(
-                                &format!("llvm.bswap.i{width}"),
-                                &[args[0].immediate()],
-                            )
+                            self.call_intrinsic("llvm.bswap", &[llty], &[args[0].immediate()])
                         }
                     }
-                    sym::bitreverse => self.call_intrinsic(
-                        &format!("llvm.bitreverse.i{width}"),
-                        &[args[0].immediate()],
-                    ),
+                    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
-                        let llvm_name =
-                            &format!("llvm.fsh{}.i{}", if is_left { 'l' } else { 'r' }, width);
+                        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);
 
-                        self.call_intrinsic(llvm_name, &[val, val, raw_shift])
+                        self.call_intrinsic(llvm_name, &[llty], &[val, val, raw_shift])
                     }
                     sym::saturating_add | sym::saturating_sub => {
                         let is_add = name == sym::saturating_add;
                         let lhs = args[0].immediate();
                         let rhs = args[1].immediate();
-                        let llvm_name = &format!(
-                            "llvm.{}{}.sat.i{}",
+                        let llvm_name = format!(
+                            "llvm.{}{}.sat",
                             if signed { 's' } else { 'u' },
                             if is_add { "add" } else { "sub" },
-                            width
                         );
-                        self.call_intrinsic(llvm_name, &[lhs, rhs])
+                        self.call_intrinsic(llvm_name, &[llty], &[lhs, rhs])
                     }
                     _ => bug!(),
                 }
@@ -469,7 +458,7 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                         // For rusty ABIs, small aggregates are actually passed
                         // as `RegKind::Integer` (see `FnAbi::adjust_for_abi`),
                         // so we re-use that same threshold here.
-                        layout.size() <= self.data_layout().pointer_size * 2
+                        layout.size() <= self.data_layout().pointer_size() * 2
                     }
                 };
 
@@ -484,11 +473,8 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     self.icmp(IntPredicate::IntEQ, a_val, b_val)
                 } else {
                     let n = self.const_usize(layout.size().bytes());
-                    let cmp = self.call_intrinsic("memcmp", &[a, b, n]);
-                    match self.cx.sess().target.arch.as_ref() {
-                        "avr" | "msp430" => self.icmp(IntPredicate::IntEQ, cmp, self.const_i16(0)),
-                        _ => self.icmp(IntPredicate::IntEQ, cmp, self.const_i32(0)),
-                    }
+                    let cmp = self.call_intrinsic("memcmp", &[], &[a, b, n]);
+                    self.icmp(IntPredicate::IntEQ, cmp, self.const_int(self.type_int(), 0))
                 }
             }
 
@@ -496,6 +482,7 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                 // Here we assume that the `memcmp` provided by the target is a NOP for size 0.
                 let cmp = self.call_intrinsic(
                     "memcmp",
+                    &[],
                     &[args[0].immediate(), args[1].immediate(), args[2].immediate()],
                 );
                 // Some targets have `memcmp` returning `i16`, but the intrinsic is always `i32`.
@@ -619,18 +606,22 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
     }
 
     fn abort(&mut self) {
-        self.call_intrinsic("llvm.trap", &[]);
+        self.call_intrinsic("llvm.trap", &[], &[]);
     }
 
     fn assume(&mut self, val: Self::Value) {
         if self.cx.sess().opts.optimize != rustc_session::config::OptLevel::No {
-            self.call_intrinsic("llvm.assume", &[val]);
+            self.call_intrinsic("llvm.assume", &[], &[val]);
         }
     }
 
     fn expect(&mut self, cond: Self::Value, expected: bool) -> Self::Value {
         if self.cx.sess().opts.optimize != rustc_session::config::OptLevel::No {
-            self.call_intrinsic("llvm.expect.i1", &[cond, self.const_bool(expected)])
+            self.call_intrinsic(
+                "llvm.expect",
+                &[self.type_i1()],
+                &[cond, self.const_bool(expected)],
+            )
         } else {
             cond
         }
@@ -644,17 +635,20 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
     ) -> Self::Value {
         let typeid = self.get_metadata_value(typeid);
         let vtable_byte_offset = self.const_i32(vtable_byte_offset as i32);
-        let type_checked_load =
-            self.call_intrinsic("llvm.type.checked.load", &[llvtable, vtable_byte_offset, typeid]);
+        let type_checked_load = self.call_intrinsic(
+            "llvm.type.checked.load",
+            &[],
+            &[llvtable, vtable_byte_offset, typeid],
+        );
         self.extract_value(type_checked_load, 0)
     }
 
     fn va_start(&mut self, va_list: &'ll Value) -> &'ll Value {
-        self.call_intrinsic("llvm.va_start", &[va_list])
+        self.call_intrinsic("llvm.va_start", &[self.val_ty(va_list)], &[va_list])
     }
 
     fn va_end(&mut self, va_list: &'ll Value) -> &'ll Value {
-        self.call_intrinsic("llvm.va_end", &[va_list])
+        self.call_intrinsic("llvm.va_end", &[self.val_ty(va_list)], &[va_list])
     }
 }
 
@@ -764,8 +758,8 @@ fn codegen_msvc_try<'ll, 'tcx>(
         //      }
         //
         // More information can be found in libstd's seh.rs implementation.
-        let ptr_size = bx.tcx().data_layout.pointer_size;
-        let ptr_align = bx.tcx().data_layout.pointer_align.abi;
+        let ptr_size = bx.tcx().data_layout.pointer_size();
+        let ptr_align = bx.tcx().data_layout.pointer_align().abi;
         let slot = bx.alloca(ptr_size, ptr_align);
         let try_func_ty = bx.type_func(&[bx.type_ptr()], bx.type_void());
         bx.invoke(try_func_ty, None, None, try_func, &[data], normal, catchswitch, None, None);
@@ -893,8 +887,8 @@ fn codegen_wasm_try<'ll, 'tcx>(
         let null = bx.const_null(bx.type_ptr());
         let funclet = bx.catch_pad(cs, &[null]);
 
-        let ptr = bx.call_intrinsic("llvm.wasm.get.exception", &[funclet.cleanuppad()]);
-        let _sel = bx.call_intrinsic("llvm.wasm.get.ehselector", &[funclet.cleanuppad()]);
+        let ptr = bx.call_intrinsic("llvm.wasm.get.exception", &[], &[funclet.cleanuppad()]);
+        let _sel = bx.call_intrinsic("llvm.wasm.get.ehselector", &[], &[funclet.cleanuppad()]);
 
         let catch_ty = bx.type_func(&[bx.type_ptr(), bx.type_ptr()], bx.type_void());
         bx.call(catch_ty, None, None, catch_func, &[data, ptr], Some(&funclet), None);
@@ -1031,14 +1025,14 @@ fn codegen_emcc_try<'ll, 'tcx>(
         let selector = bx.extract_value(vals, 1);
 
         // Check if the typeid we got is the one for a Rust panic.
-        let rust_typeid = bx.call_intrinsic("llvm.eh.typeid.for", &[tydesc]);
+        let rust_typeid = bx.call_intrinsic("llvm.eh.typeid.for", &[bx.val_ty(tydesc)], &[tydesc]);
         let is_rust_panic = bx.icmp(IntPredicate::IntEQ, selector, rust_typeid);
         let is_rust_panic = bx.zext(is_rust_panic, bx.type_bool());
 
         // We need to pass two values to catch_func (ptr and is_rust_panic), so
         // create an alloca and pass a pointer to that.
-        let ptr_size = bx.tcx().data_layout.pointer_size;
-        let ptr_align = bx.tcx().data_layout.pointer_align.abi;
+        let ptr_size = bx.tcx().data_layout.pointer_size();
+        let ptr_align = bx.tcx().data_layout.pointer_align().abi;
         let i8_align = bx.tcx().data_layout.i8_align.abi;
         // Required in order for there to be no padding between the fields.
         assert!(i8_align <= ptr_align);
@@ -1164,9 +1158,11 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
     macro_rules! require_int_or_uint_ty {
         ($ty: expr, $diag: expr) => {
             match $ty {
-                ty::Int(i) => i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size.bits()),
+                ty::Int(i) => {
+                    i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size().bits())
+                }
                 ty::Uint(i) => {
-                    i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size.bits())
+                    i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size().bits())
                 }
                 _ => {
                     return_error!($diag);
@@ -1522,56 +1518,38 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             }};
         }
 
-        let (elem_ty_str, elem_ty) = if let ty::Float(f) = in_elem.kind() {
-            let elem_ty = bx.cx.type_float_from_ty(*f);
-            match f.bit_width() {
-                16 => ("f16", elem_ty),
-                32 => ("f32", elem_ty),
-                64 => ("f64", elem_ty),
-                128 => ("f128", elem_ty),
-                _ => return_error!(InvalidMonomorphization::FloatingPointVector {
-                    span,
-                    name,
-                    f_ty: *f,
-                    in_ty,
-                }),
-            }
+        let elem_ty = if let ty::Float(f) = in_elem.kind() {
+            bx.cx.type_float_from_ty(*f)
         } else {
             return_error!(InvalidMonomorphization::FloatingPointType { span, name, in_ty });
         };
 
         let vec_ty = bx.type_vector(elem_ty, in_len);
 
-        let (intr_name, fn_ty) = match name {
-            sym::simd_ceil => ("ceil", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_fabs => ("fabs", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_fcos => ("cos", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_fexp2 => ("exp2", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_fexp => ("exp", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_flog10 => ("log10", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_flog2 => ("log2", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_flog => ("log", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_floor => ("floor", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_fma => ("fma", bx.type_func(&[vec_ty, vec_ty, vec_ty], vec_ty)),
-            sym::simd_relaxed_fma => ("fmuladd", bx.type_func(&[vec_ty, vec_ty, vec_ty], vec_ty)),
-            sym::simd_fsin => ("sin", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_fsqrt => ("sqrt", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_round => ("round", bx.type_func(&[vec_ty], vec_ty)),
-            sym::simd_trunc => ("trunc", bx.type_func(&[vec_ty], vec_ty)),
+        let intr_name = match name {
+            sym::simd_ceil => "llvm.ceil",
+            sym::simd_fabs => "llvm.fabs",
+            sym::simd_fcos => "llvm.cos",
+            sym::simd_fexp2 => "llvm.exp2",
+            sym::simd_fexp => "llvm.exp",
+            sym::simd_flog10 => "llvm.log10",
+            sym::simd_flog2 => "llvm.log2",
+            sym::simd_flog => "llvm.log",
+            sym::simd_floor => "llvm.floor",
+            sym::simd_fma => "llvm.fma",
+            sym::simd_relaxed_fma => "llvm.fmuladd",
+            sym::simd_fsin => "llvm.sin",
+            sym::simd_fsqrt => "llvm.sqrt",
+            sym::simd_round => "llvm.round",
+            sym::simd_round_ties_even => "llvm.rint",
+            sym::simd_trunc => "llvm.trunc",
             _ => return_error!(InvalidMonomorphization::UnrecognizedIntrinsic { span, name }),
         };
-        let llvm_name = &format!("llvm.{intr_name}.v{in_len}{elem_ty_str}");
-        let f = bx.declare_cfn(llvm_name, llvm::UnnamedAddr::No, fn_ty);
-        let c = bx.call(
-            fn_ty,
-            None,
-            None,
-            f,
+        Ok(bx.call_intrinsic(
+            intr_name,
+            &[vec_ty],
             &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
-            None,
-            None,
-        );
-        Ok(c)
+        ))
     }
 
     if std::matches!(
@@ -1590,34 +1568,12 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             | sym::simd_fsqrt
             | sym::simd_relaxed_fma
             | sym::simd_round
+            | sym::simd_round_ties_even
             | sym::simd_trunc
     ) {
         return simd_simple_float_intrinsic(name, in_elem, in_ty, in_len, bx, span, args);
     }
 
-    // FIXME: use:
-    //  https://github.com/llvm-mirror/llvm/blob/master/include/llvm/IR/Function.h#L182
-    //  https://github.com/llvm-mirror/llvm/blob/master/include/llvm/IR/Intrinsics.h#L81
-    fn llvm_vector_str(bx: &Builder<'_, '_, '_>, elem_ty: Ty<'_>, vec_len: u64) -> String {
-        match *elem_ty.kind() {
-            ty::Int(v) => format!(
-                "v{}i{}",
-                vec_len,
-                // Normalize to prevent crash if v: IntTy::Isize
-                v.normalize(bx.target_spec().pointer_width).bit_width().unwrap()
-            ),
-            ty::Uint(v) => format!(
-                "v{}i{}",
-                vec_len,
-                // Normalize to prevent crash if v: UIntTy::Usize
-                v.normalize(bx.target_spec().pointer_width).bit_width().unwrap()
-            ),
-            ty::Float(v) => format!("v{}f{}", vec_len, v.bit_width()),
-            ty::RawPtr(_, _) => format!("v{}p0", vec_len),
-            _ => unreachable!(),
-        }
-    }
-
     fn llvm_vector_ty<'ll>(cx: &CodegenCx<'ll, '_>, elem_ty: Ty<'_>, vec_len: u64) -> &'ll Type {
         let elem_ty = match *elem_ty.kind() {
             ty::Int(v) => cx.type_int_from_ty(v),
@@ -1698,38 +1654,22 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         );
 
         // Alignment of T, must be a constant integer value:
-        let alignment_ty = bx.type_i32();
         let alignment = bx.const_i32(bx.align_of(in_elem).bytes() as i32);
 
         // Truncate the mask vector to a vector of i1s:
         let mask = vector_mask_to_bitmask(bx, args[2].immediate(), mask_elem_bitwidth, in_len);
-        let mask_ty = bx.type_vector(bx.type_i1(), in_len);
 
         // Type of the vector of pointers:
         let llvm_pointer_vec_ty = llvm_vector_ty(bx, element_ty1, in_len);
-        let llvm_pointer_vec_str = llvm_vector_str(bx, element_ty1, in_len);
 
         // Type of the vector of elements:
         let llvm_elem_vec_ty = llvm_vector_ty(bx, element_ty0, in_len);
-        let llvm_elem_vec_str = llvm_vector_str(bx, element_ty0, in_len);
 
-        let llvm_intrinsic =
-            format!("llvm.masked.gather.{llvm_elem_vec_str}.{llvm_pointer_vec_str}");
-        let fn_ty = bx.type_func(
-            &[llvm_pointer_vec_ty, alignment_ty, mask_ty, llvm_elem_vec_ty],
-            llvm_elem_vec_ty,
-        );
-        let f = bx.declare_cfn(&llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-        let v = bx.call(
-            fn_ty,
-            None,
-            None,
-            f,
+        return Ok(bx.call_intrinsic(
+            "llvm.masked.gather",
+            &[llvm_elem_vec_ty, llvm_pointer_vec_ty],
             &[args[1].immediate(), alignment, mask, args[0].immediate()],
-            None,
-            None,
-        );
-        return Ok(v);
+        ));
     }
 
     if name == sym::simd_masked_load {
@@ -1795,32 +1735,20 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         );
 
         let mask = vector_mask_to_bitmask(bx, args[0].immediate(), m_elem_bitwidth, mask_len);
-        let mask_ty = bx.type_vector(bx.type_i1(), mask_len);
 
         // Alignment of T, must be a constant integer value:
-        let alignment_ty = bx.type_i32();
         let alignment = bx.const_i32(bx.align_of(values_elem).bytes() as i32);
 
         let llvm_pointer = bx.type_ptr();
 
         // Type of the vector of elements:
         let llvm_elem_vec_ty = llvm_vector_ty(bx, values_elem, values_len);
-        let llvm_elem_vec_str = llvm_vector_str(bx, values_elem, values_len);
-
-        let llvm_intrinsic = format!("llvm.masked.load.{llvm_elem_vec_str}.p0");
-        let fn_ty = bx
-            .type_func(&[llvm_pointer, alignment_ty, mask_ty, llvm_elem_vec_ty], llvm_elem_vec_ty);
-        let f = bx.declare_cfn(&llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-        let v = bx.call(
-            fn_ty,
-            None,
-            None,
-            f,
+
+        return Ok(bx.call_intrinsic(
+            "llvm.masked.load",
+            &[llvm_elem_vec_ty, llvm_pointer],
             &[args[1].immediate(), alignment, mask, args[2].immediate()],
-            None,
-            None,
-        );
-        return Ok(v);
+        ));
     }
 
     if name == sym::simd_masked_store {
@@ -1880,33 +1808,20 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         );
 
         let mask = vector_mask_to_bitmask(bx, args[0].immediate(), m_elem_bitwidth, mask_len);
-        let mask_ty = bx.type_vector(bx.type_i1(), mask_len);
 
         // Alignment of T, must be a constant integer value:
-        let alignment_ty = bx.type_i32();
         let alignment = bx.const_i32(bx.align_of(values_elem).bytes() as i32);
 
-        let ret_t = bx.type_void();
-
         let llvm_pointer = bx.type_ptr();
 
         // Type of the vector of elements:
         let llvm_elem_vec_ty = llvm_vector_ty(bx, values_elem, values_len);
-        let llvm_elem_vec_str = llvm_vector_str(bx, values_elem, values_len);
-
-        let llvm_intrinsic = format!("llvm.masked.store.{llvm_elem_vec_str}.p0");
-        let fn_ty = bx.type_func(&[llvm_elem_vec_ty, llvm_pointer, alignment_ty, mask_ty], ret_t);
-        let f = bx.declare_cfn(&llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-        let v = bx.call(
-            fn_ty,
-            None,
-            None,
-            f,
+
+        return Ok(bx.call_intrinsic(
+            "llvm.masked.store",
+            &[llvm_elem_vec_ty, llvm_pointer],
             &[args[2].immediate(), args[1].immediate(), alignment, mask],
-            None,
-            None,
-        );
-        return Ok(v);
+        ));
     }
 
     if name == sym::simd_scatter {
@@ -1971,38 +1886,22 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         );
 
         // Alignment of T, must be a constant integer value:
-        let alignment_ty = bx.type_i32();
         let alignment = bx.const_i32(bx.align_of(in_elem).bytes() as i32);
 
         // Truncate the mask vector to a vector of i1s:
         let mask = vector_mask_to_bitmask(bx, args[2].immediate(), mask_elem_bitwidth, in_len);
-        let mask_ty = bx.type_vector(bx.type_i1(), in_len);
-
-        let ret_t = bx.type_void();
 
         // Type of the vector of pointers:
         let llvm_pointer_vec_ty = llvm_vector_ty(bx, element_ty1, in_len);
-        let llvm_pointer_vec_str = llvm_vector_str(bx, element_ty1, in_len);
 
         // Type of the vector of elements:
         let llvm_elem_vec_ty = llvm_vector_ty(bx, element_ty0, in_len);
-        let llvm_elem_vec_str = llvm_vector_str(bx, element_ty0, in_len);
-
-        let llvm_intrinsic =
-            format!("llvm.masked.scatter.{llvm_elem_vec_str}.{llvm_pointer_vec_str}");
-        let fn_ty =
-            bx.type_func(&[llvm_elem_vec_ty, llvm_pointer_vec_ty, alignment_ty, mask_ty], ret_t);
-        let f = bx.declare_cfn(&llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-        let v = bx.call(
-            fn_ty,
-            None,
-            None,
-            f,
+
+        return Ok(bx.call_intrinsic(
+            "llvm.masked.scatter",
+            &[llvm_elem_vec_ty, llvm_pointer_vec_ty],
             &[args[0].immediate(), args[1].immediate(), alignment, mask],
-            None,
-            None,
-        );
-        return Ok(v);
+        ));
     }
 
     macro_rules! arith_red {
@@ -2117,10 +2016,10 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
                 } else {
                     let bitwidth = match in_elem.kind() {
                         ty::Int(i) => {
-                            i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size.bits())
+                            i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size().bits())
                         }
                         ty::Uint(i) => {
-                            i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size.bits())
+                            i.bit_width().unwrap_or_else(|| bx.data_layout().pointer_size().bits())
                         }
                         _ => return_error!(InvalidMonomorphization::UnsupportedSymbol {
                             span,
@@ -2416,7 +2315,13 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
     // Unary integer intrinsics
     if matches!(
         name,
-        sym::simd_bswap | sym::simd_bitreverse | sym::simd_ctlz | sym::simd_ctpop | sym::simd_cttz
+        sym::simd_bswap
+            | sym::simd_bitreverse
+            | sym::simd_ctlz
+            | sym::simd_ctpop
+            | sym::simd_cttz
+            | sym::simd_funnel_shl
+            | sym::simd_funnel_shr
     ) {
         let vec_ty = bx.cx.type_vector(
             match *in_elem.kind() {
@@ -2431,41 +2336,39 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
             },
             in_len as u64,
         );
-        let intrinsic_name = match name {
-            sym::simd_bswap => "bswap",
-            sym::simd_bitreverse => "bitreverse",
-            sym::simd_ctlz => "ctlz",
-            sym::simd_ctpop => "ctpop",
-            sym::simd_cttz => "cttz",
+        let llvm_intrinsic = match name {
+            sym::simd_bswap => "llvm.bswap",
+            sym::simd_bitreverse => "llvm.bitreverse",
+            sym::simd_ctlz => "llvm.ctlz",
+            sym::simd_ctpop => "llvm.ctpop",
+            sym::simd_cttz => "llvm.cttz",
+            sym::simd_funnel_shl => "llvm.fshl",
+            sym::simd_funnel_shr => "llvm.fshr",
             _ => unreachable!(),
         };
         let int_size = in_elem.int_size_and_signed(bx.tcx()).0.bits();
-        let llvm_intrinsic = &format!("llvm.{}.v{}i{}", intrinsic_name, in_len, int_size,);
 
         return match name {
             // byte swap is no-op for i8/u8
             sym::simd_bswap if int_size == 8 => Ok(args[0].immediate()),
             sym::simd_ctlz | sym::simd_cttz => {
                 // for the (int, i1 immediate) pair, the second arg adds `(0, true) => poison`
-                let fn_ty = bx.type_func(&[vec_ty, bx.type_i1()], vec_ty);
                 let dont_poison_on_zero = bx.const_int(bx.type_i1(), 0);
-                let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-                Ok(bx.call(
-                    fn_ty,
-                    None,
-                    None,
-                    f,
+                Ok(bx.call_intrinsic(
+                    llvm_intrinsic,
+                    &[vec_ty],
                     &[args[0].immediate(), dont_poison_on_zero],
-                    None,
-                    None,
                 ))
             }
             sym::simd_bswap | sym::simd_bitreverse | sym::simd_ctpop => {
                 // simple unary argument cases
-                let fn_ty = bx.type_func(&[vec_ty], vec_ty);
-                let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-                Ok(bx.call(fn_ty, None, None, f, &[args[0].immediate()], None, None))
+                Ok(bx.call_intrinsic(llvm_intrinsic, &[vec_ty], &[args[0].immediate()]))
             }
+            sym::simd_funnel_shl | sym::simd_funnel_shr => Ok(bx.call_intrinsic(
+                llvm_intrinsic,
+                &[vec_ty],
+                &[args[0].immediate(), args[1].immediate(), args[2].immediate()],
+            )),
             _ => unreachable!(),
         };
     }
@@ -2495,10 +2398,9 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         let lhs = args[0].immediate();
         let rhs = args[1].immediate();
         let is_add = name == sym::simd_saturating_add;
-        let ptr_bits = bx.tcx().data_layout.pointer_size.bits() as _;
-        let (signed, elem_width, elem_ty) = match *in_elem.kind() {
-            ty::Int(i) => (true, i.bit_width().unwrap_or(ptr_bits), bx.cx.type_int_from_ty(i)),
-            ty::Uint(i) => (false, i.bit_width().unwrap_or(ptr_bits), bx.cx.type_uint_from_ty(i)),
+        let (signed, elem_ty) = match *in_elem.kind() {
+            ty::Int(i) => (true, bx.cx.type_int_from_ty(i)),
+            ty::Uint(i) => (false, bx.cx.type_uint_from_ty(i)),
             _ => {
                 return_error!(InvalidMonomorphization::ExpectedVectorElementType {
                     span,
@@ -2508,19 +2410,14 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
                 });
             }
         };
-        let llvm_intrinsic = &format!(
-            "llvm.{}{}.sat.v{}i{}",
+        let llvm_intrinsic = format!(
+            "llvm.{}{}.sat",
             if signed { 's' } else { 'u' },
             if is_add { "add" } else { "sub" },
-            in_len,
-            elem_width
         );
         let vec_ty = bx.cx.type_vector(elem_ty, in_len as u64);
 
-        let fn_ty = bx.type_func(&[vec_ty, vec_ty], vec_ty);
-        let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
-        let v = bx.call(fn_ty, None, None, f, &[lhs, rhs], None, None);
-        return Ok(v);
+        return Ok(bx.call_intrinsic(llvm_intrinsic, &[vec_ty], &[lhs, rhs]));
     }
 
     span_bug!(span, "unknown SIMD intrinsic");
diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs
index 6890923a594..6db4e122ad6 100644
--- a/compiler/rustc_codegen_llvm/src/lib.rs
+++ b/compiler/rustc_codegen_llvm/src/lib.rs
@@ -26,11 +26,11 @@ use std::mem::ManuallyDrop;
 use back::owned_target_machine::OwnedTargetMachine;
 use back::write::{create_informational_target_machine, create_target_machine};
 use context::SimpleCx;
-use errors::{AutoDiffWithoutLTO, ParseTargetMachineConfig};
+use errors::ParseTargetMachineConfig;
 use llvm_util::target_config;
 use rustc_ast::expand::allocator::AllocatorKind;
 use rustc_ast::expand::autodiff_attrs::AutoDiffItem;
-use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule};
+use rustc_codegen_ssa::back::lto::{SerializedModule, ThinModule};
 use rustc_codegen_ssa::back::write::{
     CodegenContext, FatLtoInput, ModuleConfig, TargetMachineFactoryConfig, TargetMachineFactoryFn,
 };
@@ -43,7 +43,7 @@ use rustc_middle::dep_graph::{WorkProduct, WorkProductId};
 use rustc_middle::ty::TyCtxt;
 use rustc_middle::util::Providers;
 use rustc_session::Session;
-use rustc_session::config::{Lto, OptLevel, OutputFilenames, PrintKind, PrintRequest};
+use rustc_session::config::{OptLevel, OutputFilenames, PrintKind, PrintRequest};
 use rustc_span::Symbol;
 
 mod back {
@@ -113,7 +113,7 @@ impl ExtraBackendMethods for LlvmCodegenBackend {
     ) -> ModuleLlvm {
         let module_llvm = ModuleLlvm::new_metadata(tcx, module_name);
         let cx =
-            SimpleCx::new(module_llvm.llmod(), &module_llvm.llcx, tcx.data_layout.pointer_size);
+            SimpleCx::new(module_llvm.llmod(), &module_llvm.llcx, tcx.data_layout.pointer_size());
         unsafe {
             allocator::codegen(tcx, cx, module_name, kind, alloc_error_handler_kind);
         }
@@ -174,18 +174,29 @@ impl WriteBackendMethods for LlvmCodegenBackend {
     ) -> Result<ModuleCodegen<Self::Module>, FatalError> {
         back::write::link(cgcx, dcx, modules)
     }
-    fn run_fat_lto(
+    fn run_and_optimize_fat_lto(
         cgcx: &CodegenContext<Self>,
         modules: Vec<FatLtoInput<Self>>,
         cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>,
-    ) -> Result<LtoModuleCodegen<Self>, FatalError> {
-        back::lto::run_fat(cgcx, modules, cached_modules)
+        diff_fncs: Vec<AutoDiffItem>,
+    ) -> Result<ModuleCodegen<Self::Module>, FatalError> {
+        let mut module = back::lto::run_fat(cgcx, modules, cached_modules)?;
+
+        if !diff_fncs.is_empty() {
+            builder::autodiff::differentiate(&module, cgcx, diff_fncs)?;
+        }
+
+        let dcx = cgcx.create_dcx();
+        let dcx = dcx.handle();
+        back::lto::run_pass_manager(cgcx, dcx, &mut module, false)?;
+
+        Ok(module)
     }
     fn run_thin_lto(
         cgcx: &CodegenContext<Self>,
         modules: Vec<(String, Self::ThinBuffer)>,
         cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>,
-    ) -> Result<(Vec<LtoModuleCodegen<Self>>, Vec<WorkProduct>), FatalError> {
+    ) -> Result<(Vec<ThinModule<Self>>, Vec<WorkProduct>), FatalError> {
         back::lto::run_thin(cgcx, modules, cached_modules)
     }
     fn optimize(
@@ -196,14 +207,6 @@ impl WriteBackendMethods for LlvmCodegenBackend {
     ) -> Result<(), FatalError> {
         back::write::optimize(cgcx, dcx, module, config)
     }
-    fn optimize_fat(
-        cgcx: &CodegenContext<Self>,
-        module: &mut ModuleCodegen<Self::Module>,
-    ) -> Result<(), FatalError> {
-        let dcx = cgcx.create_dcx();
-        let dcx = dcx.handle();
-        back::lto::run_pass_manager(cgcx, dcx, module, false)
-    }
     fn optimize_thin(
         cgcx: &CodegenContext<Self>,
         thin: ThinModule<Self>,
@@ -212,11 +215,10 @@ impl WriteBackendMethods for LlvmCodegenBackend {
     }
     fn codegen(
         cgcx: &CodegenContext<Self>,
-        dcx: DiagCtxtHandle<'_>,
         module: ModuleCodegen<Self::Module>,
         config: &ModuleConfig,
     ) -> Result<CompiledModule, FatalError> {
-        back::write::codegen(cgcx, dcx, module, config)
+        back::write::codegen(cgcx, module, config)
     }
     fn prepare_thin(
         module: ModuleCodegen<Self::Module>,
@@ -227,19 +229,6 @@ impl WriteBackendMethods for LlvmCodegenBackend {
     fn serialize_module(module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer) {
         (module.name, back::lto::ModuleBuffer::new(module.module_llvm.llmod()))
     }
-    /// Generate autodiff rules
-    fn autodiff(
-        cgcx: &CodegenContext<Self>,
-        module: &ModuleCodegen<Self::Module>,
-        diff_fncs: Vec<AutoDiffItem>,
-        config: &ModuleConfig,
-    ) -> Result<(), FatalError> {
-        if cgcx.lto != Lto::Fat {
-            let dcx = cgcx.create_dcx();
-            return Err(dcx.handle().emit_almost_fatal(AutoDiffWithoutLTO));
-        }
-        builder::autodiff::differentiate(module, cgcx, diff_fncs, config)
-    }
 }
 
 impl LlvmCodegenBackend {
@@ -340,18 +329,11 @@ impl CodegenBackend for LlvmCodegenBackend {
         target_config(sess)
     }
 
-    fn codegen_crate<'tcx>(
-        &self,
-        tcx: TyCtxt<'tcx>,
-        metadata: EncodedMetadata,
-        need_metadata_module: bool,
-    ) -> Box<dyn Any> {
+    fn codegen_crate<'tcx>(&self, tcx: TyCtxt<'tcx>) -> Box<dyn Any> {
         Box::new(rustc_codegen_ssa::base::codegen_crate(
             LlvmCodegenBackend(()),
             tcx,
             crate::llvm_util::target_cpu(tcx.sess).to_string(),
-            metadata,
-            need_metadata_module,
         ))
     }
 
@@ -376,14 +358,20 @@ impl CodegenBackend for LlvmCodegenBackend {
         (codegen_results, work_products)
     }
 
-    fn link(&self, sess: &Session, codegen_results: CodegenResults, outputs: &OutputFilenames) {
+    fn link(
+        &self,
+        sess: &Session,
+        codegen_results: CodegenResults,
+        metadata: EncodedMetadata,
+        outputs: &OutputFilenames,
+    ) {
         use rustc_codegen_ssa::back::link::link_binary;
 
         use crate::back::archive::LlvmArchiveBuilderBuilder;
 
         // Run the linker on any artifacts that resulted from the LLVM run.
         // This should produce either a finished executable or library.
-        link_binary(sess, &LlvmArchiveBuilderBuilder, codegen_results, outputs);
+        link_binary(sess, &LlvmArchiveBuilderBuilder, codegen_results, metadata, outputs);
     }
 }
 
diff --git a/compiler/rustc_codegen_llvm/src/llvm/enzyme_ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/enzyme_ffi.rs
index 2ad39fc8538..c696b8d8ff2 100644
--- a/compiler/rustc_codegen_llvm/src/llvm/enzyme_ffi.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm/enzyme_ffi.rs
@@ -1,4 +1,3 @@
-#![allow(non_camel_case_types)]
 #![expect(dead_code)]
 
 use libc::{c_char, c_uint};
@@ -40,7 +39,7 @@ unsafe extern "C" {
     pub(crate) fn LLVMDumpValue(V: &Value);
     pub(crate) fn LLVMGetFunctionCallConv(F: &Value) -> c_uint;
     pub(crate) fn LLVMGetReturnType(T: &Type) -> &Type;
-    pub(crate) fn LLVMGetParams(Fnc: &Value, parms: *mut &Value);
+    pub(crate) fn LLVMGetParams(Fnc: &Value, params: *mut &Value);
     pub(crate) fn LLVMGetNamedFunction(M: &Module, Name: *const c_char) -> Option<&Value>;
 }
 
@@ -57,14 +56,19 @@ pub(crate) use self::Enzyme_AD::*;
 
 #[cfg(llvm_enzyme)]
 pub(crate) mod Enzyme_AD {
+    use std::ffi::{CString, c_char};
+
     use libc::c_void;
+
     unsafe extern "C" {
         pub(crate) fn EnzymeSetCLBool(arg1: *mut ::std::os::raw::c_void, arg2: u8);
+        pub(crate) fn EnzymeSetCLString(arg1: *mut ::std::os::raw::c_void, arg2: *const c_char);
     }
     unsafe extern "C" {
         static mut EnzymePrintPerf: c_void;
         static mut EnzymePrintActivity: c_void;
         static mut EnzymePrintType: c_void;
+        static mut EnzymeFunctionToAnalyze: c_void;
         static mut EnzymePrint: c_void;
         static mut EnzymeStrictAliasing: c_void;
         static mut looseTypeAnalysis: c_void;
@@ -86,6 +90,15 @@ pub(crate) mod Enzyme_AD {
             EnzymeSetCLBool(std::ptr::addr_of_mut!(EnzymePrintType), print as u8);
         }
     }
+    pub(crate) fn set_print_type_fun(fun_name: &str) {
+        let c_fun_name = CString::new(fun_name).unwrap();
+        unsafe {
+            EnzymeSetCLString(
+                std::ptr::addr_of_mut!(EnzymeFunctionToAnalyze),
+                c_fun_name.as_ptr() as *const c_char,
+            );
+        }
+    }
     pub(crate) fn set_print(print: bool) {
         unsafe {
             EnzymeSetCLBool(std::ptr::addr_of_mut!(EnzymePrint), print as u8);
@@ -132,6 +145,9 @@ pub(crate) mod Fallback_AD {
     pub(crate) fn set_print_type(print: bool) {
         unimplemented!()
     }
+    pub(crate) fn set_print_type_fun(fun_name: &str) {
+        unimplemented!()
+    }
     pub(crate) fn set_print(print: bool) {
         unimplemented!()
     }
diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
index e27fbf94f34..0b1e632cbc4 100644
--- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
@@ -15,6 +15,7 @@
 
 use std::fmt::Debug;
 use std::marker::PhantomData;
+use std::num::NonZero;
 use std::ptr;
 
 use bitflags::bitflags;
@@ -1008,7 +1009,7 @@ unsafe extern "C" {
         ModuleID: *const c_char,
         C: &Context,
     ) -> &Module;
-    pub(crate) fn LLVMCloneModule(M: &Module) -> &Module;
+    pub(crate) safe fn LLVMCloneModule(M: &Module) -> &Module;
 
     /// Data layout. See Module::getDataLayout.
     pub(crate) fn LLVMGetDataLayoutStr(M: &Module) -> *const c_char;
@@ -1077,8 +1078,6 @@ unsafe extern "C" {
 
     // Operations on other types
     pub(crate) fn LLVMVoidTypeInContext(C: &Context) -> &Type;
-    pub(crate) fn LLVMTokenTypeInContext(C: &Context) -> &Type;
-    pub(crate) fn LLVMMetadataTypeInContext(C: &Context) -> &Type;
 
     // Operations on all values
     pub(crate) fn LLVMTypeOf(Val: &Value) -> &Type;
@@ -1169,18 +1168,18 @@ unsafe extern "C" {
     pub(crate) fn LLVMGlobalGetValueType(Global: &Value) -> &Type;
 
     // Operations on global variables
-    pub(crate) fn LLVMIsAGlobalVariable(GlobalVar: &Value) -> Option<&Value>;
+    pub(crate) safe fn LLVMIsAGlobalVariable(GlobalVar: &Value) -> Option<&Value>;
     pub(crate) fn LLVMAddGlobal<'a>(M: &'a Module, Ty: &'a Type, Name: *const c_char) -> &'a Value;
     pub(crate) fn LLVMGetNamedGlobal(M: &Module, Name: *const c_char) -> Option<&Value>;
     pub(crate) fn LLVMGetFirstGlobal(M: &Module) -> Option<&Value>;
     pub(crate) fn LLVMGetNextGlobal(GlobalVar: &Value) -> Option<&Value>;
     pub(crate) fn LLVMDeleteGlobal(GlobalVar: &Value);
-    pub(crate) fn LLVMGetInitializer(GlobalVar: &Value) -> Option<&Value>;
+    pub(crate) safe fn LLVMGetInitializer(GlobalVar: &Value) -> Option<&Value>;
     pub(crate) fn LLVMSetInitializer<'a>(GlobalVar: &'a Value, ConstantVal: &'a Value);
-    pub(crate) fn LLVMIsThreadLocal(GlobalVar: &Value) -> Bool;
+    pub(crate) safe fn LLVMIsThreadLocal(GlobalVar: &Value) -> Bool;
     pub(crate) fn LLVMSetThreadLocalMode(GlobalVar: &Value, Mode: ThreadLocalMode);
-    pub(crate) fn LLVMIsGlobalConstant(GlobalVar: &Value) -> Bool;
-    pub(crate) fn LLVMSetGlobalConstant(GlobalVar: &Value, IsConstant: Bool);
+    pub(crate) safe fn LLVMIsGlobalConstant(GlobalVar: &Value) -> Bool;
+    pub(crate) safe fn LLVMSetGlobalConstant(GlobalVar: &Value, IsConstant: Bool);
     pub(crate) safe fn LLVMSetTailCall(CallInst: &Value, IsTailCall: Bool);
 
     // Operations on attributes
@@ -1195,6 +1194,15 @@ unsafe extern "C" {
     // Operations on functions
     pub(crate) fn LLVMSetFunctionCallConv(Fn: &Value, CC: c_uint);
 
+    // Operations about llvm intrinsics
+    pub(crate) fn LLVMLookupIntrinsicID(Name: *const c_char, NameLen: size_t) -> c_uint;
+    pub(crate) fn LLVMGetIntrinsicDeclaration<'a>(
+        Mod: &'a Module,
+        ID: NonZero<c_uint>,
+        ParamTypes: *const &'a Type,
+        ParamCount: size_t,
+    ) -> &'a Value;
+
     // Operations on parameters
     pub(crate) fn LLVMIsAArgument(Val: &Value) -> Option<&Value>;
     pub(crate) safe fn LLVMCountParams(Fn: &Value) -> c_uint;
@@ -1484,12 +1492,6 @@ unsafe extern "C" {
         Ty: &'a Type,
         Name: *const c_char,
     ) -> &'a Value;
-    pub(crate) fn LLVMBuildArrayAlloca<'a>(
-        B: &Builder<'a>,
-        Ty: &'a Type,
-        Val: &'a Value,
-        Name: *const c_char,
-    ) -> &'a Value;
     pub(crate) fn LLVMBuildLoad2<'a>(
         B: &Builder<'a>,
         Ty: &'a Type,
@@ -1716,7 +1718,7 @@ unsafe extern "C" {
 
     pub(crate) safe fn LLVMMetadataAsValue<'a>(C: &'a Context, MD: &'a Metadata) -> &'a Value;
 
-    pub(crate) fn LLVMSetUnnamedAddress(Global: &Value, UnnamedAddr: UnnamedAddr);
+    pub(crate) safe fn LLVMSetUnnamedAddress(Global: &Value, UnnamedAddr: UnnamedAddr);
 
     pub(crate) fn LLVMIsAConstantInt(value_ref: &Value) -> Option<&ConstantInt>;
 
@@ -1972,12 +1974,12 @@ unsafe extern "C" {
     pub(crate) fn LLVMRustBuildMinNum<'a>(
         B: &Builder<'a>,
         LHS: &'a Value,
-        LHS: &'a Value,
+        RHS: &'a Value,
     ) -> &'a Value;
     pub(crate) fn LLVMRustBuildMaxNum<'a>(
         B: &Builder<'a>,
         LHS: &'a Value,
-        LHS: &'a Value,
+        RHS: &'a Value,
     ) -> &'a Value;
 
     // Atomic Operations
diff --git a/compiler/rustc_codegen_llvm/src/llvm/mod.rs b/compiler/rustc_codegen_llvm/src/llvm/mod.rs
index ed23f911930..154ba4fd690 100644
--- a/compiler/rustc_codegen_llvm/src/llvm/mod.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm/mod.rs
@@ -1,6 +1,7 @@
 #![allow(non_snake_case)]
 
 use std::ffi::{CStr, CString};
+use std::num::NonZero;
 use std::ptr;
 use std::str::FromStr;
 use std::string::FromUtf8Error;
@@ -210,16 +211,14 @@ pub(crate) fn SetFunctionCallConv(fn_: &Value, cc: CallConv) {
 // function.
 // For more details on COMDAT sections see e.g., https://www.airs.com/blog/archives/52
 pub(crate) fn SetUniqueComdat(llmod: &Module, val: &Value) {
-    let name_buf = get_value_name(val).to_vec();
+    let name_buf = get_value_name(val);
     let name =
         CString::from_vec_with_nul(name_buf).or_else(|buf| CString::new(buf.into_bytes())).unwrap();
     set_comdat(llmod, val, &name);
 }
 
-pub(crate) fn SetUnnamedAddress(global: &Value, unnamed: UnnamedAddr) {
-    unsafe {
-        LLVMSetUnnamedAddress(global, unnamed);
-    }
+pub(crate) fn set_unnamed_address(global: &Value, unnamed: UnnamedAddr) {
+    LLVMSetUnnamedAddress(global, unnamed);
 }
 
 pub(crate) fn set_thread_local_mode(global: &Value, mode: ThreadLocalMode) {
@@ -259,9 +258,7 @@ pub(crate) fn set_initializer(llglobal: &Value, constant_val: &Value) {
 }
 
 pub(crate) fn set_global_constant(llglobal: &Value, is_constant: bool) {
-    unsafe {
-        LLVMSetGlobalConstant(llglobal, if is_constant { ffi::True } else { ffi::False });
-    }
+    LLVMSetGlobalConstant(llglobal, if is_constant { ffi::True } else { ffi::False });
 }
 
 pub(crate) fn get_linkage(llglobal: &Value) -> Linkage {
@@ -318,12 +315,36 @@ pub(crate) fn get_param(llfn: &Value, index: c_uint) -> &Value {
     }
 }
 
-/// Safe wrapper for `LLVMGetValueName2` into a byte slice
-pub(crate) fn get_value_name(value: &Value) -> &[u8] {
+/// Safe wrapper for `LLVMGetValueName2`
+/// Needs to allocate the value, because `set_value_name` will invalidate
+/// the pointer.
+pub(crate) fn get_value_name(value: &Value) -> Vec<u8> {
     unsafe {
         let mut len = 0;
         let data = LLVMGetValueName2(value, &mut len);
-        std::slice::from_raw_parts(data.cast(), len)
+        std::slice::from_raw_parts(data.cast(), len).to_vec()
+    }
+}
+
+#[derive(Debug, Copy, Clone)]
+pub(crate) struct Intrinsic {
+    id: NonZero<c_uint>,
+}
+
+impl Intrinsic {
+    pub(crate) fn lookup(name: &[u8]) -> Option<Self> {
+        let id = unsafe { LLVMLookupIntrinsicID(name.as_c_char_ptr(), name.len()) };
+        NonZero::new(id).map(|id| Self { id })
+    }
+
+    pub(crate) fn get_declaration<'ll>(
+        self,
+        llmod: &'ll Module,
+        type_params: &[&'ll Type],
+    ) -> &'ll Value {
+        unsafe {
+            LLVMGetIntrinsicDeclaration(llmod, self.id, type_params.as_ptr(), type_params.len())
+        }
     }
 }
 
diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs
index 0e77bc43df8..0fb987bdf82 100644
--- a/compiler/rustc_codegen_llvm/src/llvm_util.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs
@@ -6,27 +6,20 @@ use std::sync::Once;
 use std::{ptr, slice, str};
 
 use libc::c_int;
-use rustc_codegen_ssa::TargetConfig;
 use rustc_codegen_ssa::base::wants_wasm_eh;
-use rustc_codegen_ssa::codegen_attrs::check_tied_features;
-use rustc_data_structures::fx::{FxHashMap, FxHashSet};
+use rustc_codegen_ssa::target_features::cfg_target_feature;
+use rustc_codegen_ssa::{TargetConfig, target_features};
+use rustc_data_structures::fx::FxHashSet;
 use rustc_data_structures::small_c_str::SmallCStr;
-use rustc_data_structures::unord::UnordSet;
 use rustc_fs_util::path_to_c_string;
 use rustc_middle::bug;
 use rustc_session::Session;
 use rustc_session::config::{PrintKind, PrintRequest};
-use rustc_session::features::{StabilityExt, retpoline_features_by_flags};
-use rustc_span::Symbol;
 use rustc_target::spec::{MergeFunctions, PanicStrategy, SmallDataThresholdSupport};
-use rustc_target::target_features::{RUSTC_SPECIAL_FEATURES, RUSTC_SPECIFIC_FEATURES};
 use smallvec::{SmallVec, smallvec};
 
 use crate::back::write::create_informational_target_machine;
-use crate::errors::{
-    FixedX18InvalidArch, PossibleFeature, UnknownCTargetFeature, UnknownCTargetFeaturePrefix,
-};
-use crate::llvm;
+use crate::{errors, llvm};
 
 static INIT: Once = Once::new();
 
@@ -195,15 +188,6 @@ impl<'a> LLVMFeature<'a> {
     ) -> Self {
         Self { llvm_feature_name, dependencies }
     }
-
-    fn contains(&'a self, feat: &str) -> bool {
-        self.iter().any(|dep| dep == feat)
-    }
-
-    fn iter(&'a self) -> impl Iterator<Item = &'a str> {
-        let dependencies = self.dependencies.iter().map(|feat| feat.as_str());
-        std::iter::once(self.llvm_feature_name).chain(dependencies)
-    }
 }
 
 impl<'a> IntoIterator for LLVMFeature<'a> {
@@ -216,18 +200,22 @@ impl<'a> IntoIterator for LLVMFeature<'a> {
     }
 }
 
-// WARNING: the features after applying `to_llvm_features` must be known
-// to LLVM or the feature detection code will walk past the end of the feature
-// array, leading to crashes.
-//
-// To find a list of LLVM's names, see llvm-project/llvm/lib/Target/{ARCH}/*.td
-// where `{ARCH}` is the architecture name. Look for instances of `SubtargetFeature`.
-//
-// Check the current rustc fork of LLVM in the repo at https://github.com/rust-lang/llvm-project/.
-// The commit in use can be found via the `llvm-project` submodule in
-// https://github.com/rust-lang/rust/tree/master/src Though note that Rust can also be build with
-// an external precompiled version of LLVM which might lead to failures if the oldest tested /
-// supported LLVM version doesn't yet support the relevant intrinsics.
+/// Convert a Rust feature name to an LLVM feature name. Returning `None` means the
+/// feature should be skipped, usually because it is not supported by the current
+/// LLVM version.
+///
+/// WARNING: the features after applying `to_llvm_features` must be known
+/// to LLVM or the feature detection code will walk past the end of the feature
+/// array, leading to crashes.
+///
+/// To find a list of LLVM's names, see llvm-project/llvm/lib/Target/{ARCH}/*.td
+/// where `{ARCH}` is the architecture name. Look for instances of `SubtargetFeature`.
+///
+/// Check the current rustc fork of LLVM in the repo at
+/// <https://github.com/rust-lang/llvm-project/>. The commit in use can be found via the
+/// `llvm-project` submodule in <https://github.com/rust-lang/rust/tree/master/src> Though note that
+/// Rust can also be build with an external precompiled version of LLVM which might lead to failures
+/// if the oldest tested / supported LLVM version doesn't yet support the relevant intrinsics.
 pub(crate) fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> Option<LLVMFeature<'a>> {
     let arch = if sess.target.arch == "x86_64" {
         "x86"
@@ -343,98 +331,25 @@ pub(crate) fn target_config(sess: &Session) -> TargetConfig {
     // the target CPU, that is still expanded to target features (with all their implied features)
     // by LLVM.
     let target_machine = create_informational_target_machine(sess, true);
-    // Compute which of the known target features are enabled in the 'base' target machine. We only
-    // consider "supported" features; "forbidden" features are not reflected in `cfg` as of now.
-    let mut features: FxHashSet<Symbol> = sess
-        .target
-        .rust_target_features()
-        .iter()
-        .filter(|(feature, _, _)| {
-            // skip checking special features, as LLVM may not understand them
-            if RUSTC_SPECIAL_FEATURES.contains(feature) {
-                return true;
-            }
-            if let Some(feat) = to_llvm_features(sess, feature) {
-                for llvm_feature in feat {
-                    let cstr = SmallCStr::new(llvm_feature);
-                    // `LLVMRustHasFeature` is moderately expensive. On targets with many
-                    // features (e.g. x86) these calls take a non-trivial fraction of runtime
-                    // when compiling very small programs.
-                    if !unsafe { llvm::LLVMRustHasFeature(target_machine.raw(), cstr.as_ptr()) } {
-                        return false;
-                    }
+
+    let (unstable_target_features, target_features) = cfg_target_feature(sess, |feature| {
+        if let Some(feat) = to_llvm_features(sess, feature) {
+            // All the LLVM features this expands to must be enabled.
+            for llvm_feature in feat {
+                let cstr = SmallCStr::new(llvm_feature);
+                // `LLVMRustHasFeature` is moderately expensive. On targets with many
+                // features (e.g. x86) these calls take a non-trivial fraction of runtime
+                // when compiling very small programs.
+                if !unsafe { llvm::LLVMRustHasFeature(target_machine.raw(), cstr.as_ptr()) } {
+                    return false;
                 }
-                true
-            } else {
-                false
             }
-        })
-        .map(|(feature, _, _)| Symbol::intern(feature))
-        .collect();
-
-    // Add enabled and remove disabled features.
-    for (enabled, feature) in
-        sess.opts.cg.target_feature.split(',').filter_map(|s| match s.chars().next() {
-            Some('+') => Some((true, Symbol::intern(&s[1..]))),
-            Some('-') => Some((false, Symbol::intern(&s[1..]))),
-            _ => None,
-        })
-    {
-        if enabled {
-            // Also add all transitively implied features.
-
-            // We don't care about the order in `features` since the only thing we use it for is the
-            // `features.contains` below.
-            #[allow(rustc::potential_query_instability)]
-            features.extend(
-                sess.target
-                    .implied_target_features(feature.as_str())
-                    .iter()
-                    .map(|s| Symbol::intern(s)),
-            );
+            true
         } else {
-            // Remove transitively reverse-implied features.
-
-            // We don't care about the order in `features` since the only thing we use it for is the
-            // `features.contains` below.
-            #[allow(rustc::potential_query_instability)]
-            features.retain(|f| {
-                if sess.target.implied_target_features(f.as_str()).contains(&feature.as_str()) {
-                    // If `f` if implies `feature`, then `!feature` implies `!f`, so we have to
-                    // remove `f`. (This is the standard logical contraposition principle.)
-                    false
-                } else {
-                    // We can keep `f`.
-                    true
-                }
-            });
+            false
         }
-    }
+    });
 
-    // Filter enabled features based on feature gates.
-    let f = |allow_unstable| {
-        sess.target
-            .rust_target_features()
-            .iter()
-            .filter_map(|(feature, gate, _)| {
-                // The `allow_unstable` set is used by rustc internally to determined which target
-                // features are truly available, so we want to return even perma-unstable
-                // "forbidden" features.
-                if allow_unstable
-                    || (gate.in_cfg()
-                        && (sess.is_nightly_build() || gate.requires_nightly().is_none()))
-                {
-                    Some(Symbol::intern(feature))
-                } else {
-                    None
-                }
-            })
-            .filter(|feature| features.contains(&feature))
-            .collect()
-    };
-
-    let target_features = f(false);
-    let unstable_target_features = f(true);
     let mut cfg = TargetConfig {
         target_features,
         unstable_target_features,
@@ -455,10 +370,18 @@ fn update_target_reliable_float_cfg(sess: &Session, cfg: &mut TargetConfig) {
     let target_env = sess.target.options.env.as_ref();
     let target_abi = sess.target.options.abi.as_ref();
     let target_pointer_width = sess.target.pointer_width;
+    let version = get_version();
 
     cfg.has_reliable_f16 = match (target_arch, target_os) {
         // Selection failure <https://github.com/llvm/llvm-project/issues/50374>
         ("s390x", _) => false,
+        // LLVM crash without neon <https://github.com/llvm/llvm-project/issues/129394> (now fixed)
+        ("aarch64", _)
+            if !cfg.target_features.iter().any(|f| f.as_str() == "neon")
+                && version < (20, 1, 1) =>
+        {
+            false
+        }
         // Unsupported <https://github.com/llvm/llvm-project/issues/94434>
         ("arm64ec", _) => false,
         // MinGW ABI bugs <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115054>
@@ -707,10 +630,18 @@ pub(crate) fn target_cpu(sess: &Session) -> &str {
     handle_native(cpu_name)
 }
 
-fn llvm_features_by_flags(sess: &Session) -> Vec<&str> {
-    let mut features: Vec<&str> = Vec::new();
-    retpoline_features_by_flags(sess, &mut features);
-    features
+/// The target features for compiler flags other than `-Ctarget-features`.
+fn llvm_features_by_flags(sess: &Session, features: &mut Vec<String>) {
+    target_features::retpoline_features_by_flags(sess, features);
+
+    // -Zfixed-x18
+    if sess.opts.unstable_opts.fixed_x18 {
+        if sess.target.arch != "aarch64" {
+            sess.dcx().emit_fatal(errors::FixedX18InvalidArch { arch: &sess.target.arch });
+        } else {
+            features.push("+reserve-x18".into());
+        }
+    }
 }
 
 /// The list of LLVM features computed from CLI flags (`-Ctarget-cpu`, `-Ctarget-feature`,
@@ -777,6 +708,8 @@ pub(crate) fn global_llvm_features(
             .split(',')
             .filter(|v| !v.is_empty())
             // Drop +v8plus feature introduced in LLVM 20.
+            // (Hard-coded target features do not go through `to_llvm_feature` since they already
+            // are LLVM feature names, hence we need a special case here.)
             .filter(|v| *v != "+v8plus" || get_version() >= (20, 0, 0))
             .map(String::from),
     );
@@ -787,86 +720,23 @@ pub(crate) fn global_llvm_features(
 
     // -Ctarget-features
     if !only_base_features {
-        let known_features = sess.target.rust_target_features();
-        // Will only be filled when `diagnostics` is set!
-        let mut featsmap = FxHashMap::default();
-
-        // Compute implied features
-        let mut all_rust_features = vec![];
-        for feature in sess.opts.cg.target_feature.split(',').chain(llvm_features_by_flags(sess)) {
-            if let Some(feature) = feature.strip_prefix('+') {
-                all_rust_features.extend(
-                    UnordSet::from(sess.target.implied_target_features(feature))
-                        .to_sorted_stable_ord()
-                        .iter()
-                        .map(|&&s| (true, s)),
-                )
-            } else if let Some(feature) = feature.strip_prefix('-') {
-                // FIXME: Why do we not remove implied features on "-" here?
-                // We do the equivalent above in `target_config`.
-                // See <https://github.com/rust-lang/rust/issues/134792>.
-                all_rust_features.push((false, feature));
-            } else if !feature.is_empty() {
-                if diagnostics {
-                    sess.dcx().emit_warn(UnknownCTargetFeaturePrefix { feature });
-                }
-            }
-        }
-        // Remove features that are meant for rustc, not LLVM.
-        all_rust_features.retain(|(_, feature)| {
-            // Retain if it is not a rustc feature
-            !RUSTC_SPECIFIC_FEATURES.contains(feature)
-        });
-
-        // Check feature validity.
-        if diagnostics {
-            for &(enable, feature) in &all_rust_features {
-                let feature_state = known_features.iter().find(|&&(v, _, _)| v == feature);
-                match feature_state {
-                    None => {
-                        let rust_feature =
-                            known_features.iter().find_map(|&(rust_feature, _, _)| {
-                                let llvm_features = to_llvm_features(sess, rust_feature)?;
-                                if llvm_features.contains(feature)
-                                    && !llvm_features.contains(rust_feature)
-                                {
-                                    Some(rust_feature)
-                                } else {
-                                    None
-                                }
-                            });
-                        let unknown_feature = if let Some(rust_feature) = rust_feature {
-                            UnknownCTargetFeature {
-                                feature,
-                                rust_feature: PossibleFeature::Some { rust_feature },
-                            }
-                        } else {
-                            UnknownCTargetFeature { feature, rust_feature: PossibleFeature::None }
-                        };
-                        sess.dcx().emit_warn(unknown_feature);
-                    }
-                    Some((_, stability, _)) => {
-                        stability.verify_feature_enabled_by_flag(sess, enable, feature);
-                    }
-                }
-
-                // FIXME(nagisa): figure out how to not allocate a full hashset here.
-                featsmap.insert(feature, enable);
-            }
-        }
-
-        // Translate this into LLVM features.
-        let feats = all_rust_features
-            .iter()
-            .filter_map(|&(enable, feature)| {
+        target_features::flag_to_backend_features(
+            sess,
+            diagnostics,
+            |feature| {
+                to_llvm_features(sess, feature)
+                    .map(|f| SmallVec::<[&str; 2]>::from_iter(f.into_iter()))
+                    .unwrap_or_default()
+            },
+            |feature, enable| {
                 let enable_disable = if enable { '+' } else { '-' };
                 // We run through `to_llvm_features` when
                 // passing requests down to LLVM. This means that all in-language
                 // features also work on the command line instead of having two
                 // different names when the LLVM name and the Rust name differ.
-                let llvm_feature = to_llvm_features(sess, feature)?;
+                let Some(llvm_feature) = to_llvm_features(sess, feature) else { return };
 
-                Some(
+                features.extend(
                     std::iter::once(format!(
                         "{}{}",
                         enable_disable, llvm_feature.llvm_feature_name
@@ -881,27 +751,12 @@ pub(crate) fn global_llvm_features(
                         },
                     )),
                 )
-            })
-            .flatten();
-        features.extend(feats);
-
-        if diagnostics && let Some(f) = check_tied_features(sess, &featsmap) {
-            sess.dcx().emit_err(rustc_codegen_ssa::errors::TargetFeatureDisableOrEnable {
-                features: f,
-                span: None,
-                missing_features: None,
-            });
-        }
+            },
+        );
     }
 
-    // -Zfixed-x18
-    if sess.opts.unstable_opts.fixed_x18 {
-        if sess.target.arch != "aarch64" {
-            sess.dcx().emit_fatal(FixedX18InvalidArch { arch: &sess.target.arch });
-        } else {
-            features.push("+reserve-x18".into());
-        }
-    }
+    // We add this in the "base target" so that these show up in `sess.unstable_target_features`.
+    llvm_features_by_flags(sess, &mut features);
 
     features
 }
diff --git a/compiler/rustc_codegen_llvm/src/mono_item.rs b/compiler/rustc_codegen_llvm/src/mono_item.rs
index 3f38e1e191b..f9edaded60d 100644
--- a/compiler/rustc_codegen_llvm/src/mono_item.rs
+++ b/compiler/rustc_codegen_llvm/src/mono_item.rs
@@ -55,8 +55,8 @@ impl<'tcx> PreDefineCodegenMethods<'tcx> for CodegenCx<'_, 'tcx> {
         let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty());
         let lldecl = self.declare_fn(symbol_name, fn_abi, Some(instance));
         llvm::set_linkage(lldecl, base::linkage_to_llvm(linkage));
-        let attrs = self.tcx.codegen_fn_attrs(instance.def_id());
-        base::set_link_section(lldecl, attrs);
+        let attrs = self.tcx.codegen_instance_attrs(instance.def);
+        base::set_link_section(lldecl, &attrs);
         if (linkage == Linkage::LinkOnceODR || linkage == Linkage::WeakODR)
             && self.tcx.sess.target.supports_comdat()
         {
@@ -131,8 +131,8 @@ impl CodegenCx<'_, '_> {
         }
 
         // Thread-local variables generally don't support copy relocations.
-        let is_thread_local_var = unsafe { llvm::LLVMIsAGlobalVariable(llval) }
-            .is_some_and(|v| unsafe { llvm::LLVMIsThreadLocal(v) } == llvm::True);
+        let is_thread_local_var = llvm::LLVMIsAGlobalVariable(llval)
+            .is_some_and(|v| llvm::LLVMIsThreadLocal(v) == llvm::True);
         if is_thread_local_var {
             return false;
         }
diff --git a/compiler/rustc_codegen_llvm/src/type_.rs b/compiler/rustc_codegen_llvm/src/type_.rs
index 169036f5152..89365503138 100644
--- a/compiler/rustc_codegen_llvm/src/type_.rs
+++ b/compiler/rustc_codegen_llvm/src/type_.rs
@@ -1,7 +1,8 @@
 use std::borrow::Borrow;
+use std::hash::{Hash, Hasher};
 use std::{fmt, ptr};
 
-use libc::{c_char, c_uint};
+use libc::c_uint;
 use rustc_abi::{AddressSpace, Align, Integer, Reg, Size};
 use rustc_codegen_ssa::common::TypeKind;
 use rustc_codegen_ssa::traits::*;
@@ -25,6 +26,14 @@ impl PartialEq for Type {
     }
 }
 
+impl Eq for Type {}
+
+impl Hash for Type {
+    fn hash<H: Hasher>(&self, state: &mut H) {
+        ptr::hash(self, state);
+    }
+}
+
 impl fmt::Debug for Type {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         f.write_str(
@@ -49,13 +58,6 @@ impl<'ll, CX: Borrow<SCx<'ll>>> GenericCx<'ll, CX> {
     pub(crate) fn type_void(&self) -> &'ll Type {
         unsafe { llvm::LLVMVoidTypeInContext(self.llcx()) }
     }
-    pub(crate) fn type_token(&self) -> &'ll Type {
-        unsafe { llvm::LLVMTokenTypeInContext(self.llcx()) }
-    }
-
-    pub(crate) fn type_metadata(&self) -> &'ll Type {
-        unsafe { llvm::LLVMMetadataTypeInContext(self.llcx()) }
-    }
 
     ///x Creates an integer type with the given number of bits, e.g., i24
     pub(crate) fn type_ix(&self, num_bits: u64) -> &'ll Type {
@@ -206,7 +208,7 @@ impl<'ll, CX: Borrow<SCx<'ll>>> BaseTypeCodegenMethods for GenericCx<'ll, CX> {
     }
 
     fn type_ptr(&self) -> &'ll Type {
-        self.type_ptr_ext(AddressSpace::DATA)
+        self.type_ptr_ext(AddressSpace::ZERO)
     }
 
     fn type_ptr_ext(&self, address_space: AddressSpace) -> &'ll Type {
@@ -256,7 +258,7 @@ impl Type {
     }
 
     pub(crate) fn ptr_llcx(llcx: &llvm::Context) -> &Type {
-        unsafe { llvm::LLVMPointerTypeInContext(llcx, AddressSpace::DATA.0) }
+        unsafe { llvm::LLVMPointerTypeInContext(llcx, AddressSpace::ZERO.0) }
     }
 }
 
@@ -296,8 +298,8 @@ impl<'ll, 'tcx> LayoutTypeCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
 }
 
 impl<'ll, 'tcx> TypeMembershipCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
-    fn add_type_metadata(&self, function: &'ll Value, typeid: String) {
-        let typeid_metadata = self.typeid_metadata(typeid).unwrap();
+    fn add_type_metadata(&self, function: &'ll Value, typeid: &[u8]) {
+        let typeid_metadata = self.create_metadata(typeid);
         unsafe {
             let v = [llvm::LLVMValueAsMetadata(self.const_usize(0)), typeid_metadata];
             llvm::LLVMRustGlobalAddMetadata(
@@ -308,8 +310,8 @@ impl<'ll, 'tcx> TypeMembershipCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
         }
     }
 
-    fn set_type_metadata(&self, function: &'ll Value, typeid: String) {
-        let typeid_metadata = self.typeid_metadata(typeid).unwrap();
+    fn set_type_metadata(&self, function: &'ll Value, typeid: &[u8]) {
+        let typeid_metadata = self.create_metadata(typeid);
         unsafe {
             let v = [llvm::LLVMValueAsMetadata(self.const_usize(0)), typeid_metadata];
             llvm::LLVMGlobalSetMetadata(
@@ -320,10 +322,8 @@ impl<'ll, 'tcx> TypeMembershipCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> {
         }
     }
 
-    fn typeid_metadata(&self, typeid: String) -> Option<&'ll Metadata> {
-        Some(unsafe {
-            llvm::LLVMMDStringInContext2(self.llcx, typeid.as_ptr() as *const c_char, typeid.len())
-        })
+    fn typeid_metadata(&self, typeid: &[u8]) -> Option<&'ll Metadata> {
+        Some(self.create_metadata(typeid))
     }
 
     fn add_kcfi_type_metadata(&self, function: &'ll Value, kcfi_typeid: u32) {
diff --git a/compiler/rustc_codegen_llvm/src/va_arg.rs b/compiler/rustc_codegen_llvm/src/va_arg.rs
index 236568590be..ce079f3cb0a 100644
--- a/compiler/rustc_codegen_llvm/src/va_arg.rs
+++ b/compiler/rustc_codegen_llvm/src/va_arg.rs
@@ -45,7 +45,8 @@ fn emit_direct_ptr_va_arg<'ll, 'tcx>(
     let va_list_ty = bx.type_ptr();
     let va_list_addr = list.immediate();
 
-    let ptr = bx.load(va_list_ty, va_list_addr, bx.tcx().data_layout.pointer_align.abi);
+    let ptr_align_abi = bx.tcx().data_layout.pointer_align().abi;
+    let ptr = bx.load(va_list_ty, va_list_addr, ptr_align_abi);
 
     let (addr, addr_align) = if allow_higher_align && align > slot_size {
         (round_pointer_up_to_alignment(bx, ptr, align, bx.type_ptr()), align)
@@ -56,7 +57,7 @@ fn emit_direct_ptr_va_arg<'ll, 'tcx>(
     let aligned_size = size.align_to(slot_size).bytes() as i32;
     let full_direct_size = bx.cx().const_i32(aligned_size);
     let next = bx.inbounds_ptradd(addr, full_direct_size);
-    bx.store(next, va_list_addr, bx.tcx().data_layout.pointer_align.abi);
+    bx.store(next, va_list_addr, ptr_align_abi);
 
     if size.bytes() < slot_size.bytes()
         && bx.tcx().sess.target.endian == Endian::Big
@@ -108,8 +109,8 @@ fn emit_ptr_va_arg<'ll, 'tcx>(
     let (llty, size, align) = if indirect {
         (
             bx.cx.layout_of(Ty::new_imm_ptr(bx.cx.tcx, target_ty)).llvm_type(bx.cx),
-            bx.cx.data_layout().pointer_size,
-            bx.cx.data_layout().pointer_align,
+            bx.cx.data_layout().pointer_size(),
+            bx.cx.data_layout().pointer_align(),
         )
     } else {
         (layout.llvm_type(bx.cx), layout.size, layout.align)
@@ -172,10 +173,10 @@ fn emit_aapcs_va_arg<'ll, 'tcx>(
 
     let gr_type = target_ty.is_any_ptr() || target_ty.is_integral();
     let (reg_off, reg_top, slot_size) = if gr_type {
-        let nreg = (layout.size.bytes() + 7) / 8;
+        let nreg = layout.size.bytes().div_ceil(8);
         (gr_offs, gr_top, nreg * 8)
     } else {
-        let nreg = (layout.size.bytes() + 15) / 16;
+        let nreg = layout.size.bytes().div_ceil(16);
         (vr_offs, vr_top, nreg * 16)
     };
 
@@ -204,7 +205,7 @@ fn emit_aapcs_va_arg<'ll, 'tcx>(
 
     bx.switch_to_block(in_reg);
     let top_type = bx.type_ptr();
-    let top = bx.load(top_type, reg_top, dl.pointer_align.abi);
+    let top = bx.load(top_type, reg_top, dl.pointer_align().abi);
 
     // reg_value = *(@top + reg_off_v);
     let mut reg_addr = bx.ptradd(top, reg_off_v);
@@ -297,6 +298,7 @@ fn emit_powerpc_va_arg<'ll, 'tcx>(
 
     let max_regs = 8u8;
     let use_regs = bx.icmp(IntPredicate::IntULT, num_regs, bx.const_u8(max_regs));
+    let ptr_align_abi = bx.tcx().data_layout.pointer_align().abi;
 
     let in_reg = bx.append_sibling_block("va_arg.in_reg");
     let in_mem = bx.append_sibling_block("va_arg.in_mem");
@@ -308,7 +310,7 @@ fn emit_powerpc_va_arg<'ll, 'tcx>(
         bx.switch_to_block(in_reg);
 
         let reg_safe_area_ptr = bx.inbounds_ptradd(va_list_addr, bx.cx.const_usize(1 + 1 + 2 + 4));
-        let mut reg_addr = bx.load(bx.type_ptr(), reg_safe_area_ptr, dl.pointer_align.abi);
+        let mut reg_addr = bx.load(bx.type_ptr(), reg_safe_area_ptr, ptr_align_abi);
 
         // Floating-point registers start after the general-purpose registers.
         if !is_int && !is_soft_float_abi {
@@ -342,11 +344,11 @@ fn emit_powerpc_va_arg<'ll, 'tcx>(
         let size = if !is_indirect {
             layout.layout.size.align_to(overflow_area_align)
         } else {
-            dl.pointer_size
+            dl.pointer_size()
         };
 
         let overflow_area_ptr = bx.inbounds_ptradd(va_list_addr, bx.cx.const_usize(1 + 1 + 2));
-        let mut overflow_area = bx.load(bx.type_ptr(), overflow_area_ptr, dl.pointer_align.abi);
+        let mut overflow_area = bx.load(bx.type_ptr(), overflow_area_ptr, ptr_align_abi);
 
         // Round up address of argument to alignment
         if layout.layout.align.abi > overflow_area_align {
@@ -362,7 +364,7 @@ fn emit_powerpc_va_arg<'ll, 'tcx>(
 
         // Increase the overflow area.
         overflow_area = bx.inbounds_ptradd(overflow_area, bx.const_usize(size.bytes()));
-        bx.store(overflow_area, overflow_area_ptr, dl.pointer_align.abi);
+        bx.store(overflow_area, overflow_area_ptr, ptr_align_abi);
 
         bx.br(end);
 
@@ -373,11 +375,8 @@ fn emit_powerpc_va_arg<'ll, 'tcx>(
     bx.switch_to_block(end);
     let val_addr = bx.phi(bx.type_ptr(), &[reg_addr, mem_addr], &[in_reg, in_mem]);
     let val_type = layout.llvm_type(bx);
-    let val_addr = if is_indirect {
-        bx.load(bx.cx.type_ptr(), val_addr, dl.pointer_align.abi)
-    } else {
-        val_addr
-    };
+    let val_addr =
+        if is_indirect { bx.load(bx.cx.type_ptr(), val_addr, ptr_align_abi) } else { val_addr };
     bx.load(val_type, val_addr, layout.align.abi)
 }
 
@@ -414,6 +413,7 @@ fn emit_s390x_va_arg<'ll, 'tcx>(
     let in_reg = bx.append_sibling_block("va_arg.in_reg");
     let in_mem = bx.append_sibling_block("va_arg.in_mem");
     let end = bx.append_sibling_block("va_arg.end");
+    let ptr_align_abi = dl.pointer_align().abi;
 
     // FIXME: vector ABI not yet supported.
     let target_ty_size = bx.cx.size_of(target_ty).bytes();
@@ -435,7 +435,7 @@ fn emit_s390x_va_arg<'ll, 'tcx>(
     bx.switch_to_block(in_reg);
 
     // Work out the address of the value in the register save area.
-    let reg_ptr_v = bx.load(bx.type_ptr(), reg_save_area, dl.pointer_align.abi);
+    let reg_ptr_v = bx.load(bx.type_ptr(), reg_save_area, ptr_align_abi);
     let scaled_reg_count = bx.mul(reg_count_v, bx.const_u64(8));
     let reg_off = bx.add(scaled_reg_count, bx.const_u64(reg_save_index * 8 + reg_padding));
     let reg_addr = bx.ptradd(reg_ptr_v, reg_off);
@@ -449,15 +449,14 @@ fn emit_s390x_va_arg<'ll, 'tcx>(
     bx.switch_to_block(in_mem);
 
     // Work out the address of the value in the argument overflow area.
-    let arg_ptr_v =
-        bx.load(bx.type_ptr(), overflow_arg_area, bx.tcx().data_layout.pointer_align.abi);
+    let arg_ptr_v = bx.load(bx.type_ptr(), overflow_arg_area, ptr_align_abi);
     let arg_off = bx.const_u64(padding);
     let mem_addr = bx.ptradd(arg_ptr_v, arg_off);
 
     // Update the argument overflow area pointer.
     let arg_size = bx.cx().const_u64(padded_size);
     let new_arg_ptr_v = bx.inbounds_ptradd(arg_ptr_v, arg_size);
-    bx.store(new_arg_ptr_v, overflow_arg_area, dl.pointer_align.abi);
+    bx.store(new_arg_ptr_v, overflow_arg_area, ptr_align_abi);
     bx.br(end);
 
     // Return the appropriate result.
@@ -465,7 +464,7 @@ fn emit_s390x_va_arg<'ll, 'tcx>(
     let val_addr = bx.phi(bx.type_ptr(), &[reg_addr, mem_addr], &[in_reg, in_mem]);
     let val_type = layout.llvm_type(bx);
     let val_addr =
-        if indirect { bx.load(bx.cx.type_ptr(), val_addr, dl.pointer_align.abi) } else { val_addr };
+        if indirect { bx.load(bx.cx.type_ptr(), val_addr, ptr_align_abi) } else { val_addr };
     bx.load(val_type, val_addr, layout.align.abi)
 }
 
@@ -607,7 +606,7 @@ fn emit_x86_64_sysv64_va_arg<'ll, 'tcx>(
     // loads than necessary. Can we clean this up?
     let reg_save_area_ptr =
         bx.inbounds_ptradd(va_list_addr, bx.cx.const_usize(2 * unsigned_int_offset + ptr_offset));
-    let reg_save_area_v = bx.load(bx.type_ptr(), reg_save_area_ptr, dl.pointer_align.abi);
+    let reg_save_area_v = bx.load(bx.type_ptr(), reg_save_area_ptr, dl.pointer_align().abi);
 
     let reg_addr = match layout.layout.backend_repr() {
         BackendRepr::Scalar(scalar) => match scalar.primitive() {
@@ -749,10 +748,11 @@ fn x86_64_sysv64_va_arg_from_memory<'ll, 'tcx>(
     layout: TyAndLayout<'tcx, Ty<'tcx>>,
 ) -> &'ll Value {
     let dl = bx.cx.data_layout();
+    let ptr_align_abi = dl.data_layout().pointer_align().abi;
 
     let overflow_arg_area_ptr = bx.inbounds_ptradd(va_list_addr, bx.const_usize(8));
 
-    let overflow_arg_area_v = bx.load(bx.type_ptr(), overflow_arg_area_ptr, dl.pointer_align.abi);
+    let overflow_arg_area_v = bx.load(bx.type_ptr(), overflow_arg_area_ptr, ptr_align_abi);
     // AMD64-ABI 3.5.7p5: Step 7. Align l->overflow_arg_area upwards to a 16
     // byte boundary if alignment needed by type exceeds 8 byte boundary.
     // It isn't stated explicitly in the standard, but in practice we use
@@ -771,7 +771,7 @@ fn x86_64_sysv64_va_arg_from_memory<'ll, 'tcx>(
     let size_in_bytes = layout.layout.size().bytes();
     let offset = bx.const_i32(size_in_bytes.next_multiple_of(8) as i32);
     let overflow_arg_area = bx.inbounds_ptradd(overflow_arg_area_v, offset);
-    bx.store(overflow_arg_area, overflow_arg_area_ptr, dl.pointer_align.abi);
+    bx.store(overflow_arg_area, overflow_arg_area_ptr, ptr_align_abi);
 
     mem_addr
 }
@@ -803,6 +803,7 @@ fn emit_xtensa_va_arg<'ll, 'tcx>(
     let from_stack = bx.append_sibling_block("va_arg.from_stack");
     let from_regsave = bx.append_sibling_block("va_arg.from_regsave");
     let end = bx.append_sibling_block("va_arg.end");
+    let ptr_align_abi = bx.tcx().data_layout.pointer_align().abi;
 
     // (*va).va_ndx
     let va_reg_offset = 4;
@@ -825,12 +826,11 @@ fn emit_xtensa_va_arg<'ll, 'tcx>(
 
     bx.switch_to_block(from_regsave);
     // update va_ndx
-    bx.store(offset_next, offset_ptr, bx.tcx().data_layout.pointer_align.abi);
+    bx.store(offset_next, offset_ptr, ptr_align_abi);
 
     // (*va).va_reg
     let regsave_area_ptr = bx.inbounds_ptradd(va_list_addr, bx.cx.const_usize(va_reg_offset));
-    let regsave_area =
-        bx.load(bx.type_ptr(), regsave_area_ptr, bx.tcx().data_layout.pointer_align.abi);
+    let regsave_area = bx.load(bx.type_ptr(), regsave_area_ptr, ptr_align_abi);
     let regsave_value_ptr = bx.inbounds_ptradd(regsave_area, offset);
     bx.br(end);
 
@@ -849,11 +849,11 @@ fn emit_xtensa_va_arg<'ll, 'tcx>(
     // va_ndx = offset_next_corrected;
     let offset_next_corrected = bx.add(offset_next, bx.const_i32(slot_size));
     // update va_ndx
-    bx.store(offset_next_corrected, offset_ptr, bx.tcx().data_layout.pointer_align.abi);
+    bx.store(offset_next_corrected, offset_ptr, ptr_align_abi);
 
     // let stack_value_ptr = unsafe { (*va).va_stk.byte_add(offset_corrected) };
     let stack_area_ptr = bx.inbounds_ptradd(va_list_addr, bx.cx.const_usize(0));
-    let stack_area = bx.load(bx.type_ptr(), stack_area_ptr, bx.tcx().data_layout.pointer_align.abi);
+    let stack_area = bx.load(bx.type_ptr(), stack_area_ptr, ptr_align_abi);
     let stack_value_ptr = bx.inbounds_ptradd(stack_area, offset_corrected);
     bx.br(end);
 
@@ -861,7 +861,7 @@ fn emit_xtensa_va_arg<'ll, 'tcx>(
 
     // On big-endian, for values smaller than the slot size we'd have to align the read to the end
     // of the slot rather than the start. While the ISA and GCC support big-endian, all the Xtensa
-    // targets supported by rustc are litte-endian so don't worry about it.
+    // targets supported by rustc are little-endian so don't worry about it.
 
     // if from_regsave {
     //     unsafe { *regsave_value_ptr }