diff --git a/Cargo.lock b/Cargo.lock
index 325503a0e956..98ca2d85b91d 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -3883,6 +3883,7 @@ dependencies = [
  "rustc_plugin_impl",
  "rustc_privacy",
  "rustc_query_impl",
+ "rustc_query_system",
  "rustc_resolve",
  "rustc_serialize",
  "rustc_session",
diff --git a/compiler/rustc_incremental/src/assert_dep_graph.rs b/compiler/rustc_incremental/src/assert_dep_graph.rs
index 60b48e9bc8a0..5689d8349aa6 100644
--- a/compiler/rustc_incremental/src/assert_dep_graph.rs
+++ b/compiler/rustc_incremental/src/assert_dep_graph.rs
@@ -41,7 +41,7 @@ use rustc_hir as hir;
 use rustc_hir::def_id::DefId;
 use rustc_hir::intravisit::{self, Visitor};
 use rustc_middle::dep_graph::{
-    DepGraphQuery, DepKind, DepNode, DepNodeExt, DepNodeFilter, EdgeFilter,
+    dep_kind, DepGraphQuery, DepNode, DepNodeExt, DepNodeFilter, EdgeFilter,
 };
 use rustc_middle::hir::nested_filter;
 use rustc_middle::ty::TyCtxt;
@@ -128,7 +128,7 @@ impl<'tcx> IfThisChanged<'tcx> {
                 let dep_node_interned = self.argument(attr);
                 let dep_node = match dep_node_interned {
                     None => {
-                        DepNode::from_def_path_hash(self.tcx, def_path_hash, DepKind::hir_owner)
+                        DepNode::from_def_path_hash(self.tcx, def_path_hash, dep_kind::hir_owner)
                     }
                     Some(n) => {
                         match DepNode::from_label_string(self.tcx, n.as_str(), def_path_hash) {
diff --git a/compiler/rustc_interface/Cargo.toml b/compiler/rustc_interface/Cargo.toml
index 29d1cd0e0546..e349ad50d3f5 100644
--- a/compiler/rustc_interface/Cargo.toml
+++ b/compiler/rustc_interface/Cargo.toml
@@ -43,6 +43,7 @@ rustc_lint = { path = "../rustc_lint" }
 rustc_errors = { path = "../rustc_errors" }
 rustc_plugin_impl = { path = "../rustc_plugin_impl" }
 rustc_privacy = { path = "../rustc_privacy" }
+rustc_query_system = { path = "../rustc_query_system" }
 rustc_query_impl = { path = "../rustc_query_impl" }
 rustc_resolve = { path = "../rustc_resolve" }
 rustc_trait_selection = { path = "../rustc_trait_selection" }
diff --git a/compiler/rustc_interface/src/callbacks.rs b/compiler/rustc_interface/src/callbacks.rs
index a18e2d1d6388..b442d999dfce 100644
--- a/compiler/rustc_interface/src/callbacks.rs
+++ b/compiler/rustc_interface/src/callbacks.rs
@@ -9,8 +9,9 @@
 //! The functions in this file should fall back to the default set in their
 //! origin crate when the `TyCtxt` is not present in TLS.
 
-use rustc_errors::{Diagnostic, TRACK_DIAGNOSTICS};
+use rustc_middle::dep_graph::DepNodeExt;
 use rustc_middle::ty::tls;
+use rustc_query_system::dep_graph::{DepKind, DepNode};
 use std::fmt;
 
 fn track_span_parent(def_id: rustc_span::def_id::LocalDefId) {
@@ -23,20 +24,6 @@ fn track_span_parent(def_id: rustc_span::def_id::LocalDefId) {
     })
 }
 
-/// This is a callback from `rustc_ast` as it cannot access the implicit state
-/// in `rustc_middle` otherwise. It is used to when diagnostic messages are
-/// emitted and stores them in the current query, if there is one.
-fn track_diagnostic(diagnostic: &Diagnostic) {
-    tls::with_context_opt(|icx| {
-        if let Some(icx) = icx {
-            if let Some(diagnostics) = icx.diagnostics {
-                let mut diagnostics = diagnostics.lock();
-                diagnostics.extend(Some(diagnostic.clone()));
-            }
-        }
-    })
-}
-
 /// This is a callback from `rustc_hir` as it cannot access the implicit state
 /// in `rustc_middle` otherwise.
 fn def_id_debug(def_id: rustc_hir::def_id::DefId, f: &mut fmt::Formatter<'_>) -> fmt::Result {
@@ -50,10 +37,42 @@ fn def_id_debug(def_id: rustc_hir::def_id::DefId, f: &mut fmt::Formatter<'_>) ->
     write!(f, ")")
 }
 
+fn dep_kind_debug(kind: &DepKind, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+    tls::with_opt(|opt_tcx| {
+        if let Some(tcx) = opt_tcx {
+            write!(f, "{}", tcx.query_name(*kind))
+        } else {
+            write!(f, "DepKind({:?})", kind.index())
+        }
+    })
+}
+
+fn dep_node_debug(node: &DepNode, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+    tls::with_opt(|opt_tcx| {
+        if let Some(tcx) = opt_tcx {
+            write!(f, "{}(", tcx.query_name(node.kind))?;
+
+            if let Some(def_id) = node.extract_def_id(tcx) {
+                write!(f, "{}", tcx.def_path_debug_str(def_id))?;
+            } else if let Some(ref s) = tcx.dep_graph.dep_node_debug_str(*node) {
+                write!(f, "{}", s)?;
+            } else {
+                write!(f, "{}", node.hash)?;
+            }
+
+            write!(f, ")")
+        } else {
+            write!(f, "DepKind({:?})({})", node.kind.index(), node.hash)
+        }
+    })
+}
+
 /// Sets up the callbacks in prior crates which we want to refer to the
 /// TyCtxt in.
 pub fn setup_callbacks() {
     rustc_span::SPAN_TRACK.swap(&(track_span_parent as fn(_)));
     rustc_hir::def_id::DEF_ID_DEBUG.swap(&(def_id_debug as fn(_, &mut fmt::Formatter<'_>) -> _));
-    TRACK_DIAGNOSTICS.swap(&(track_diagnostic as fn(&_)));
+    rustc_query_system::dep_graph::KIND_DEBUG.swap(&(dep_kind_debug as _));
+    rustc_query_system::dep_graph::NODE_DEBUG.swap(&(dep_node_debug as _));
+    rustc_errors::TRACK_DIAGNOSTICS.swap(&(rustc_query_system::tls::track_diagnostic as fn(&_)));
 }
diff --git a/compiler/rustc_interface/src/interface.rs b/compiler/rustc_interface/src/interface.rs
index e7563933c88a..648142a9ea34 100644
--- a/compiler/rustc_interface/src/interface.rs
+++ b/compiler/rustc_interface/src/interface.rs
@@ -344,9 +344,9 @@ pub fn try_print_query_stack(handler: &Handler, num_frames: Option<usize>) {
     // Be careful relying on global state here: this code is called from
     // a panic hook, which means that the global `Handler` may be in a weird
     // state if it was responsible for triggering the panic.
-    let i = ty::tls::with_context_opt(|icx| {
-        if let Some(icx) = icx {
-            QueryCtxt::from_tcx(icx.tcx).try_print_query_stack(icx.query, handler, num_frames)
+    let i = ty::tls::with_opt(|tcx| {
+        if let Some(tcx) = tcx {
+            QueryCtxt::from_tcx(tcx).try_print_query_stack(handler, num_frames)
         } else {
             0
         }
diff --git a/compiler/rustc_interface/src/passes.rs b/compiler/rustc_interface/src/passes.rs
index 2a01b677e33b..5f396deb037a 100644
--- a/compiler/rustc_interface/src/passes.rs
+++ b/compiler/rustc_interface/src/passes.rs
@@ -806,13 +806,12 @@ pub struct QueryContext<'tcx> {
     gcx: &'tcx GlobalCtxt<'tcx>,
 }
 
-impl<'tcx> QueryContext<'tcx> {
+impl<'gcx> QueryContext<'gcx> {
     pub fn enter<F, R>(&mut self, f: F) -> R
     where
-        F: FnOnce(TyCtxt<'tcx>) -> R,
+        F: for<'tcx> FnOnce(TyCtxt<'tcx>) -> R,
     {
-        let icx = ty::tls::ImplicitCtxt::new(self.gcx);
-        ty::tls::enter_context(&icx, |_| f(icx.tcx))
+        ty::tls::enter_context(self.gcx, f)
     }
 }
 
diff --git a/compiler/rustc_interface/src/util.rs b/compiler/rustc_interface/src/util.rs
index 3fa8017dc93c..e2df8ddc6be4 100644
--- a/compiler/rustc_interface/src/util.rs
+++ b/compiler/rustc_interface/src/util.rs
@@ -7,7 +7,7 @@ use rustc_data_structures::jobserver;
 use rustc_data_structures::sync::Lrc;
 use rustc_errors::registry::Registry;
 #[cfg(parallel_compiler)]
-use rustc_middle::ty::tls;
+use rustc_middle::ty::{tls, GlobalCtxt};
 use rustc_parse::validate_attr;
 #[cfg(parallel_compiler)]
 use rustc_query_impl::QueryCtxt;
@@ -166,19 +166,21 @@ pub fn run_in_thread_pool_with_globals<F: FnOnce() -> R + Send, R: Send>(
 unsafe fn handle_deadlock() {
     let registry = rustc_rayon_core::Registry::current();
 
-    let context = tls::get_tlv();
-    assert!(context != 0);
-    rustc_data_structures::sync::assert_sync::<tls::ImplicitCtxt<'_, '_>>();
-    let icx: &tls::ImplicitCtxt<'_, '_> = &*(context as *const tls::ImplicitCtxt<'_, '_>);
+    // We do not need to copy the query ImplicitCtxt since this deadlock handler is not part of a
+    // normal query invocation.
 
     let session_globals = rustc_span::with_session_globals(|sg| sg as *const _);
     let session_globals = &*session_globals;
+
+    // Extend the lifetime of the GlobalCtxt so the new thread can know of it.
+    // The current thread will not free it, it is deadlocked.
+    let tcx: &'static GlobalCtxt<'static> =
+        tls::with(|tcx| &*(*tcx as *const GlobalCtxt<'_>).cast());
+
     thread::spawn(move || {
-        tls::enter_context(icx, |_| {
-            rustc_span::set_session_globals_then(session_globals, || {
-                tls::with(|tcx| QueryCtxt::from_tcx(tcx).deadlock(&registry))
-            })
-        });
+        rustc_span::set_session_globals_then(session_globals, || {
+            tls::enter_context(tcx, |tcx| QueryCtxt::from_tcx(tcx).deadlock(&registry))
+        })
     });
 }
 
diff --git a/compiler/rustc_metadata/src/rmeta/decoder/cstore_impl.rs b/compiler/rustc_metadata/src/rmeta/decoder/cstore_impl.rs
index 63bf929fb863..ca294de473b7 100644
--- a/compiler/rustc_metadata/src/rmeta/decoder/cstore_impl.rs
+++ b/compiler/rustc_metadata/src/rmeta/decoder/cstore_impl.rs
@@ -50,8 +50,8 @@ macro_rules! provide_one {
             // External query providers call `crate_hash` in order to register a dependency
             // on the crate metadata. The exception is `crate_hash` itself, which obviously
             // doesn't need to do this (and can't, as it would cause a query cycle).
-            use rustc_middle::dep_graph::DepKind;
-            if DepKind::$name != DepKind::crate_hash && $tcx.dep_graph.is_fully_enabled() {
+            use rustc_middle::dep_graph::dep_kind;
+            if dep_kind::$name != dep_kind::crate_hash && $tcx.dep_graph.is_fully_enabled() {
                 $tcx.ensure().crate_hash($def_id.krate);
             }
 
diff --git a/compiler/rustc_middle/src/dep_graph/dep_node.rs b/compiler/rustc_middle/src/dep_graph/dep_node.rs
index d20be0a34d2d..fc195fa68862 100644
--- a/compiler/rustc_middle/src/dep_graph/dep_node.rs
+++ b/compiler/rustc_middle/src/dep_graph/dep_node.rs
@@ -63,11 +63,8 @@ use rustc_data_structures::fingerprint::Fingerprint;
 use rustc_hir::def_id::{CrateNum, DefId, LocalDefId, CRATE_DEF_INDEX};
 use rustc_hir::definitions::DefPathHash;
 use rustc_hir::HirId;
-use rustc_query_system::dep_graph::FingerprintStyle;
+use rustc_query_system::dep_graph::{DepKind, DepNode, DepNodeParams, FingerprintStyle};
 use rustc_span::symbol::Symbol;
-use std::hash::Hash;
-
-pub use rustc_query_system::dep_graph::{DepContext, DepNodeParams};
 
 /// This struct stores metadata about each DepKind.
 ///
@@ -89,6 +86,9 @@ pub struct DepKindStruct {
     /// See [DepNodeParams] trait for the behaviour of each key type.
     pub fingerprint_style: FingerprintStyle,
 
+    /// Name of the query for pretty-printing.
+    pub label: &'static str,
+
     /// The red/green evaluation system will try to mark a specific DepNode in the
     /// dependency graph as green by recursively trying to mark the dependencies of
     /// that `DepNode` as green. While doing so, it will sometimes encounter a `DepNode`
@@ -130,16 +130,21 @@ pub struct DepKindStruct {
     pub try_load_from_on_disk_cache: Option<fn(TyCtxt<'_>, DepNode)>,
 }
 
-impl DepKind {
+impl TyCtxt<'_> {
     #[inline(always)]
-    pub fn fingerprint_style(self, tcx: TyCtxt<'_>) -> FingerprintStyle {
+    crate fn query_fingerprint_style(self, dep_kind: DepKind) -> FingerprintStyle {
         // Only fetch the DepKindStruct once.
-        let data = tcx.query_kind(self);
+        let data = self.query_kind(dep_kind);
         if data.is_anon {
             return FingerprintStyle::Opaque;
         }
         data.fingerprint_style
     }
+
+    #[inline(always)]
+    pub fn query_name(self, dep_kind: DepKind) -> &'static str {
+        self.query_kind(dep_kind).label
+    }
 }
 
 macro_rules! define_dep_nodes {
@@ -151,38 +156,40 @@ macro_rules! define_dep_nodes {
     ) => (
         #[macro_export]
         macro_rules! make_dep_kind_array {
-            ($mod:ident) => {[ $($mod::$variant()),* ]};
+            ($mod:ident) => {[ $mod::NULL(), $($mod::$variant()),* ]};
         }
 
-        /// This enum serves as an index into arrays built by `make_dep_kind_array`.
-        #[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, Encodable, Decodable)]
-        #[allow(non_camel_case_types)]
-        pub enum DepKind {
-            $($variant),*
+        /// Definition of the `DepKind`s for the known queries.
+        /// The constants are indices in the arrays produced by `make_dep_kind_array`,
+        /// and should be kept in sync.
+        #[allow(non_upper_case_globals)]
+        pub mod dep_kind {
+            use super::DepKind;
+
+            /// We use this for most things when incr. comp. is turned off.
+            pub const NULL: DepKind = DepKind::NULL;
+            $(pub const $variant: DepKind = DepKind::new(1 + ${index()});)*
         }
 
         fn dep_kind_from_label_string(label: &str) -> Result<DepKind, ()> {
             match label {
-                $(stringify!($variant) => Ok(DepKind::$variant),)*
+                "NULL" => Ok(dep_kind::NULL),
+                $(stringify!($variant) => Ok(dep_kind::$variant),)*
                 _ => Err(()),
             }
         }
 
         /// Contains variant => str representations for constructing
         /// DepNode groups for tests.
-        #[allow(dead_code, non_upper_case_globals)]
+        #[allow(non_upper_case_globals)]
         pub mod label_strs {
-           $(
-                pub const $variant: &str = stringify!($variant);
-            )*
+            pub const NULL: &str = "NULL";
+           $(pub const $variant: &str = stringify!($variant);)*
         }
     );
 }
 
 rustc_dep_node_append!([define_dep_nodes!][ <'tcx>
-    // We use this for most things when incr. comp. is turned off.
-    [] Null,
-
     [anon] TraitSelect,
 
     // WARNING: if `Symbol` is changed, make sure you update `make_compile_codegen_unit` below.
@@ -196,26 +203,15 @@ rustc_dep_node_append!([define_dep_nodes!][ <'tcx>
 // WARNING: `construct` is generic and does not know that `CompileCodegenUnit` takes `Symbol`s as keys.
 // Be very careful changing this type signature!
 crate fn make_compile_codegen_unit(tcx: TyCtxt<'_>, name: Symbol) -> DepNode {
-    DepNode::construct(tcx, DepKind::CompileCodegenUnit, &name)
+    DepNode::construct(tcx, dep_kind::CompileCodegenUnit, &name)
 }
 
 // WARNING: `construct` is generic and does not know that `CompileMonoItem` takes `MonoItem`s as keys.
 // Be very careful changing this type signature!
 crate fn make_compile_mono_item<'tcx>(tcx: TyCtxt<'tcx>, mono_item: &MonoItem<'tcx>) -> DepNode {
-    DepNode::construct(tcx, DepKind::CompileMonoItem, mono_item)
+    DepNode::construct(tcx, dep_kind::CompileMonoItem, mono_item)
 }
 
-pub type DepNode = rustc_query_system::dep_graph::DepNode<DepKind>;
-
-// We keep a lot of `DepNode`s in memory during compilation. It's not
-// required that their size stay the same, but we don't want to change
-// it inadvertently. This assert just ensures we're aware of any change.
-#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
-static_assert_size!(DepNode, 18);
-
-#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
-static_assert_size!(DepNode, 24);
-
 pub trait DepNodeExt: Sized {
     /// Construct a DepNode from the given DepKind and DefPathHash. This
     /// method will assert that the given DepKind actually requires a
@@ -249,8 +245,9 @@ impl DepNodeExt for DepNode {
     /// Construct a DepNode from the given DepKind and DefPathHash. This
     /// method will assert that the given DepKind actually requires a
     /// single DefId/DefPathHash parameter.
+    #[inline]
     fn from_def_path_hash(tcx: TyCtxt<'_>, def_path_hash: DefPathHash, kind: DepKind) -> DepNode {
-        debug_assert!(kind.fingerprint_style(tcx) == FingerprintStyle::DefPathHash);
+        debug_assert!(tcx.query_fingerprint_style(kind) == FingerprintStyle::DefPathHash);
         DepNode { kind, hash: def_path_hash.0.into() }
     }
 
@@ -264,8 +261,9 @@ impl DepNodeExt for DepNode {
     /// DepNode. Condition (2) might not be fulfilled if a DepNode
     /// refers to something from the previous compilation session that
     /// has been removed.
+    #[inline]
     fn extract_def_id<'tcx>(&self, tcx: TyCtxt<'tcx>) -> Option<DefId> {
-        if self.kind.fingerprint_style(tcx) == FingerprintStyle::DefPathHash {
+        if tcx.query_fingerprint_style(self.kind) == FingerprintStyle::DefPathHash {
             Some(tcx.def_path_hash_to_def_id(DefPathHash(self.hash.into()), &mut || {
                 panic!("Failed to extract DefId: {:?} {}", self.kind, self.hash)
             }))
@@ -282,7 +280,7 @@ impl DepNodeExt for DepNode {
     ) -> Result<DepNode, ()> {
         let kind = dep_kind_from_label_string(label)?;
 
-        match kind.fingerprint_style(tcx) {
+        match tcx.query_fingerprint_style(kind) {
             FingerprintStyle::Opaque => Err(()),
             FingerprintStyle::Unit => Ok(DepNode::new_no_params(tcx, kind)),
             FingerprintStyle::DefPathHash => {
@@ -292,6 +290,7 @@ impl DepNodeExt for DepNode {
     }
 
     /// Used in testing
+    #[inline]
     fn has_label_string(label: &str) -> bool {
         dep_kind_from_label_string(label).is_ok()
     }
diff --git a/compiler/rustc_middle/src/dep_graph/mod.rs b/compiler/rustc_middle/src/dep_graph/mod.rs
index 6bfd1b7ffab2..2869c51eb293 100644
--- a/compiler/rustc_middle/src/dep_graph/mod.rs
+++ b/compiler/rustc_middle/src/dep_graph/mod.rs
@@ -1,4 +1,4 @@
-use crate::ty::{self, TyCtxt};
+use crate::ty::TyCtxt;
 use rustc_data_structures::profiling::SelfProfilerRef;
 use rustc_query_system::ich::StableHashingContext;
 use rustc_session::Session;
@@ -7,69 +7,15 @@ use rustc_session::Session;
 mod dep_node;
 
 pub use rustc_query_system::dep_graph::{
-    debug::DepNodeFilter, hash_result, DepContext, DepNodeColor, DepNodeIndex,
-    SerializedDepNodeIndex, WorkProduct, WorkProductId,
+    debug::DepNodeFilter, debug::EdgeFilter, hash_result, DepContext, DepGraph, DepGraphQuery,
+    DepKind, DepNode, DepNodeColor, DepNodeIndex, SerializedDepGraph, SerializedDepNodeIndex,
+    TaskDeps, WorkProduct, WorkProductId,
 };
 
-pub use dep_node::{label_strs, DepKind, DepKindStruct, DepNode, DepNodeExt};
+pub use dep_node::{dep_kind, label_strs, DepKindStruct, DepNodeExt};
 crate use dep_node::{make_compile_codegen_unit, make_compile_mono_item};
 
-pub type DepGraph = rustc_query_system::dep_graph::DepGraph<DepKind>;
-pub type TaskDeps = rustc_query_system::dep_graph::TaskDeps<DepKind>;
-pub type TaskDepsRef<'a> = rustc_query_system::dep_graph::TaskDepsRef<'a, DepKind>;
-pub type DepGraphQuery = rustc_query_system::dep_graph::DepGraphQuery<DepKind>;
-pub type SerializedDepGraph = rustc_query_system::dep_graph::SerializedDepGraph<DepKind>;
-pub type EdgeFilter = rustc_query_system::dep_graph::debug::EdgeFilter<DepKind>;
-
-impl rustc_query_system::dep_graph::DepKind for DepKind {
-    const NULL: Self = DepKind::Null;
-
-    fn debug_node(node: &DepNode, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{:?}(", node.kind)?;
-
-        ty::tls::with_opt(|opt_tcx| {
-            if let Some(tcx) = opt_tcx {
-                if let Some(def_id) = node.extract_def_id(tcx) {
-                    write!(f, "{}", tcx.def_path_debug_str(def_id))?;
-                } else if let Some(ref s) = tcx.dep_graph.dep_node_debug_str(*node) {
-                    write!(f, "{}", s)?;
-                } else {
-                    write!(f, "{}", node.hash)?;
-                }
-            } else {
-                write!(f, "{}", node.hash)?;
-            }
-            Ok(())
-        })?;
-
-        write!(f, ")")
-    }
-
-    fn with_deps<OP, R>(task_deps: TaskDepsRef<'_>, op: OP) -> R
-    where
-        OP: FnOnce() -> R,
-    {
-        ty::tls::with_context(|icx| {
-            let icx = ty::tls::ImplicitCtxt { task_deps, ..icx.clone() };
-
-            ty::tls::enter_context(&icx, |_| op())
-        })
-    }
-
-    fn read_deps<OP>(op: OP)
-    where
-        OP: for<'a> FnOnce(TaskDepsRef<'a>),
-    {
-        ty::tls::with_context_opt(|icx| {
-            let Some(icx) = icx else { return };
-            op(icx.task_deps)
-        })
-    }
-}
-
 impl<'tcx> DepContext for TyCtxt<'tcx> {
-    type DepKind = DepKind;
-
     #[inline]
     fn create_stable_hashing_context(&self) -> StableHashingContext<'_> {
         TyCtxt::create_stable_hashing_context(*self)
@@ -92,7 +38,7 @@ impl<'tcx> DepContext for TyCtxt<'tcx> {
 
     #[inline(always)]
     fn fingerprint_style(&self, kind: DepKind) -> rustc_query_system::dep_graph::FingerprintStyle {
-        kind.fingerprint_style(*self)
+        self.query_fingerprint_style(kind)
     }
 
     #[inline(always)]
@@ -117,8 +63,8 @@ impl<'tcx> DepContext for TyCtxt<'tcx> {
         // hit the cache instead of having to go through `force_from_dep_node`.
         // This assertion makes sure, we actually keep applying the solution above.
         debug_assert!(
-            dep_node.kind != DepKind::codegen_unit,
-            "calling force_from_dep_node() on DepKind::codegen_unit"
+            dep_node.kind != dep_kind::codegen_unit,
+            "calling force_from_dep_node() on dep_kind::codegen_unit"
         );
 
         let cb = self.query_kind(dep_node.kind);
diff --git a/compiler/rustc_middle/src/hir/map/mod.rs b/compiler/rustc_middle/src/hir/map/mod.rs
index 65796fbc698d..8c512b29adf9 100644
--- a/compiler/rustc_middle/src/hir/map/mod.rs
+++ b/compiler/rustc_middle/src/hir/map/mod.rs
@@ -1156,18 +1156,8 @@ fn hir_id_to_string(map: Map<'_>, id: HirId) -> String {
     let id_str = format!(" (hir_id={})", id);
 
     let path_str = || {
-        // This functionality is used for debugging, try to use `TyCtxt` to get
-        // the user-friendly path, otherwise fall back to stringifying `DefPath`.
-        crate::ty::tls::with_opt(|tcx| {
-            if let Some(tcx) = tcx {
-                let def_id = map.local_def_id(id);
-                tcx.def_path_str(def_id.to_def_id())
-            } else if let Some(path) = map.def_path_from_hir_id(id) {
-                path.data.into_iter().map(|elem| elem.to_string()).collect::<Vec<_>>().join("::")
-            } else {
-                String::from("<missing path>")
-            }
-        })
+        let def_id = map.local_def_id(id);
+        map.tcx.def_path_str(def_id.to_def_id())
     };
 
     let span_str = || map.tcx.sess.source_map().span_to_snippet(map.span(id)).unwrap_or_default();
diff --git a/compiler/rustc_middle/src/lib.rs b/compiler/rustc_middle/src/lib.rs
index fd2b5f5335f6..e424528cc1fa 100644
--- a/compiler/rustc_middle/src/lib.rs
+++ b/compiler/rustc_middle/src/lib.rs
@@ -43,6 +43,7 @@
 #![feature(once_cell)]
 #![feature(let_chains)]
 #![feature(let_else)]
+#![feature(macro_metavar_expr)]
 #![feature(min_specialization)]
 #![feature(trusted_len)]
 #![feature(type_alias_impl_trait)]
diff --git a/compiler/rustc_middle/src/ty/context.rs b/compiler/rustc_middle/src/ty/context.rs
index 4e6be84ad7f8..5926bc19e868 100644
--- a/compiler/rustc_middle/src/ty/context.rs
+++ b/compiler/rustc_middle/src/ty/context.rs
@@ -1176,7 +1176,7 @@ impl<'tcx> TyCtxt<'tcx> {
     }
 
     crate fn query_kind(self, k: DepKind) -> &'tcx DepKindStruct {
-        &self.query_kinds[k as usize]
+        &self.query_kinds[k.index()]
     }
 
     /// Constructs a `TyKind::Error` type and registers a `delay_span_bug` to ensure it gets used.
@@ -1681,174 +1681,62 @@ nop_list_lift! {substs; GenericArg<'a> => GenericArg<'tcx>}
 CloneLiftImpls! { for<'tcx> { Constness, traits::WellFormedLoc, } }
 
 pub mod tls {
-    use super::{ptr_eq, GlobalCtxt, TyCtxt};
-
-    use crate::dep_graph::TaskDepsRef;
-    use crate::ty::query;
-    use rustc_data_structures::sync::{self, Lock};
-    use rustc_data_structures::thin_vec::ThinVec;
-    use rustc_errors::Diagnostic;
-    use std::mem;
-
-    #[cfg(not(parallel_compiler))]
+    use super::{GlobalCtxt, TyCtxt};
+    use rustc_data_structures::sync;
     use std::cell::Cell;
 
-    #[cfg(parallel_compiler)]
-    use rustc_rayon_core as rayon_core;
-
-    /// This is the implicit state of rustc. It contains the current
-    /// `TyCtxt` and query. It is updated when creating a local interner or
-    /// executing a new query. Whenever there's a `TyCtxt` value available
-    /// you should also have access to an `ImplicitCtxt` through the functions
-    /// in this module.
-    #[derive(Clone)]
-    pub struct ImplicitCtxt<'a, 'tcx> {
-        /// The current `TyCtxt`.
-        pub tcx: TyCtxt<'tcx>,
-
-        /// The current query job, if any. This is updated by `JobOwner::start` in
-        /// `ty::query::plumbing` when executing a query.
-        pub query: Option<query::QueryJobId>,
-
-        /// Where to store diagnostics for the current query job, if any.
-        /// This is updated by `JobOwner::start` in `ty::query::plumbing` when executing a query.
-        pub diagnostics: Option<&'a Lock<ThinVec<Diagnostic>>>,
-
-        /// Used to prevent layout from recursing too deeply.
-        pub layout_depth: usize,
-
-        /// The current dep graph task. This is used to add dependencies to queries
-        /// when executing them.
-        pub task_deps: TaskDepsRef<'a>,
-    }
-
-    impl<'a, 'tcx> ImplicitCtxt<'a, 'tcx> {
-        pub fn new(gcx: &'tcx GlobalCtxt<'tcx>) -> Self {
-            let tcx = TyCtxt { gcx };
-            ImplicitCtxt {
-                tcx,
-                query: None,
-                diagnostics: None,
-                layout_depth: 0,
-                task_deps: TaskDepsRef::Ignore,
-            }
-        }
-    }
-
-    /// Sets Rayon's thread-local variable, which is preserved for Rayon jobs
-    /// to `value` during the call to `f`. It is restored to its previous value after.
-    /// This is used to set the pointer to the new `ImplicitCtxt`.
-    #[cfg(parallel_compiler)]
-    #[inline]
-    fn set_tlv<F: FnOnce() -> R, R>(value: usize, f: F) -> R {
-        rayon_core::tlv::with(value, f)
-    }
-
-    /// Gets Rayon's thread-local variable, which is preserved for Rayon jobs.
-    /// This is used to get the pointer to the current `ImplicitCtxt`.
-    #[cfg(parallel_compiler)]
-    #[inline]
-    pub fn get_tlv() -> usize {
-        rayon_core::tlv::get()
-    }
-
-    #[cfg(not(parallel_compiler))]
     thread_local! {
-        /// A thread local variable that stores a pointer to the current `ImplicitCtxt`.
-        static TLV: Cell<usize> = const { Cell::new(0) };
+        /// This is the implicit state of rustc. It contains the `TyCtxt`. Whenever there's a
+        /// `TyCtxt` value available you should also have access to it through the functions in this
+        /// module.
+        static TLV: Cell<*const GlobalCtxt<'static>> = const { Cell::new(std::ptr::null()) };
     }
 
-    /// Sets TLV to `value` during the call to `f`.
-    /// It is restored to its previous value after.
-    /// This is used to set the pointer to the new `ImplicitCtxt`.
-    #[cfg(not(parallel_compiler))]
+    /// Sets `context` as the new current `TyCtxt` for the duration of the function `f`.
     #[inline]
-    fn set_tlv<F: FnOnce() -> R, R>(value: usize, f: F) -> R {
-        let old = get_tlv();
-        let _reset = rustc_data_structures::OnDrop(move || TLV.with(|tlv| tlv.set(old)));
-        TLV.with(|tlv| tlv.set(value));
-        f()
-    }
-
-    /// Gets the pointer to the current `ImplicitCtxt`.
-    #[cfg(not(parallel_compiler))]
-    #[inline]
-    fn get_tlv() -> usize {
-        TLV.with(|tlv| tlv.get())
-    }
-
-    /// Sets `context` as the new current `ImplicitCtxt` for the duration of the function `f`.
-    #[inline]
-    pub fn enter_context<'a, 'tcx, F, R>(context: &ImplicitCtxt<'a, 'tcx>, f: F) -> R
+    pub fn enter_context<'gcx, F, R>(gcx: &'gcx GlobalCtxt<'gcx>, f: F) -> R
     where
-        F: FnOnce(&ImplicitCtxt<'a, 'tcx>) -> R,
-    {
-        set_tlv(context as *const _ as usize, || f(&context))
-    }
-
-    /// Allows access to the current `ImplicitCtxt` in a closure if one is available.
-    #[inline]
-    pub fn with_context_opt<F, R>(f: F) -> R
-    where
-        F: for<'a, 'tcx> FnOnce(Option<&ImplicitCtxt<'a, 'tcx>>) -> R,
-    {
-        let context = get_tlv();
-        if context == 0 {
-            f(None)
-        } else {
-            // We could get an `ImplicitCtxt` pointer from another thread.
-            // Ensure that `ImplicitCtxt` is `Sync`.
-            sync::assert_sync::<ImplicitCtxt<'_, '_>>();
-
-            unsafe { f(Some(&*(context as *const ImplicitCtxt<'_, '_>))) }
-        }
-    }
-
-    /// Allows access to the current `ImplicitCtxt`.
-    /// Panics if there is no `ImplicitCtxt` available.
-    #[inline]
-    pub fn with_context<F, R>(f: F) -> R
-    where
-        F: for<'a, 'tcx> FnOnce(&ImplicitCtxt<'a, 'tcx>) -> R,
+        F: for<'tcx> FnOnce(TyCtxt<'tcx>) -> R,
     {
-        with_context_opt(|opt_context| f(opt_context.expect("no ImplicitCtxt stored in tls")))
-    }
+        // We are storing `GlobalCtxt` as a global.
+        // Ensure that it is `Sync`.
+        sync::assert_sync::<GlobalCtxt<'_>>();
 
-    /// Allows access to the current `ImplicitCtxt` whose tcx field is the same as the tcx argument
-    /// passed in. This means the closure is given an `ImplicitCtxt` with the same `'tcx` lifetime
-    /// as the `TyCtxt` passed in.
-    /// This will panic if you pass it a `TyCtxt` which is different from the current
-    /// `ImplicitCtxt`'s `tcx` field.
-    #[inline]
-    pub fn with_related_context<'tcx, F, R>(tcx: TyCtxt<'tcx>, f: F) -> R
-    where
-        F: FnOnce(&ImplicitCtxt<'_, 'tcx>) -> R,
-    {
-        with_context(|context| unsafe {
-            assert!(ptr_eq(context.tcx.gcx, tcx.gcx));
-            let context: &ImplicitCtxt<'_, '_> = mem::transmute(context);
-            f(context)
-        })
+        let address = (gcx as *const GlobalCtxt<'_>).cast();
+        let old = TLV.with(|tlv| tlv.replace(address));
+        let _reset = rustc_data_structures::OnDrop(move || TLV.with(|tlv| tlv.set(old)));
+        debug_assert!(
+            old == address || old == std::ptr::null(),
+            "There can only be one GlobalCtxt."
+        );
+        f(TyCtxt { gcx })
     }
 
-    /// Allows access to the `TyCtxt` in the current `ImplicitCtxt`.
-    /// Panics if there is no `ImplicitCtxt` available.
+    /// Allows access to the current `TyCtxt`.
+    /// Panics if there is no `TyCtxt` available.
     #[inline]
     pub fn with<F, R>(f: F) -> R
     where
         F: for<'tcx> FnOnce(TyCtxt<'tcx>) -> R,
     {
-        with_context(|context| f(context.tcx))
+        with_opt(|opt_tcx| f(opt_tcx.expect("no TyCtxt stored in tls")))
     }
 
-    /// Allows access to the `TyCtxt` in the current `ImplicitCtxt`.
-    /// The closure is passed None if there is no `ImplicitCtxt` available.
+    /// Allows access to the current `TyCtxt`.
+    /// The closure is passed None if there is no `TyCtxt` available.
     #[inline]
     pub fn with_opt<F, R>(f: F) -> R
     where
         F: for<'tcx> FnOnce(Option<TyCtxt<'tcx>>) -> R,
     {
-        with_context_opt(|opt_context| f(opt_context.map(|context| context.tcx)))
+        let context = TLV.with(|tlv| tlv.get());
+        let tcx = if context == std::ptr::null() {
+            None
+        } else {
+            let gcx = unsafe { &*context.cast::<GlobalCtxt<'_>>() };
+            Some(TyCtxt { gcx })
+        };
+        f(tcx)
     }
 }
 
@@ -2872,12 +2760,6 @@ impl<T, R, E> InternIteratorElement<T, R> for Result<T, E> {
     }
 }
 
-// We are comparing types with different invariant lifetimes, so `ptr::eq`
-// won't work for us.
-fn ptr_eq<T, U>(t: *const T, u: *const U) -> bool {
-    t as *const () == u as *const ()
-}
-
 pub fn provide(providers: &mut ty::query::Providers) {
     providers.resolutions = |tcx, ()| &tcx.untracked_resolutions;
     providers.module_reexports =
diff --git a/compiler/rustc_middle/src/ty/layout.rs b/compiler/rustc_middle/src/ty/layout.rs
index f6d139fe59d3..67fe2b930de4 100644
--- a/compiler/rustc_middle/src/ty/layout.rs
+++ b/compiler/rustc_middle/src/ty/layout.rs
@@ -9,6 +9,7 @@ use rustc_hir as hir;
 use rustc_hir::lang_items::LangItem;
 use rustc_index::bit_set::BitSet;
 use rustc_index::vec::{Idx, IndexVec};
+use rustc_query_system::tls;
 use rustc_session::{config::OptLevel, DataTypeKind, FieldInfo, SizeKind, VariantInfo};
 use rustc_span::symbol::Symbol;
 use rustc_span::{Span, DUMMY_SP};
@@ -225,51 +226,46 @@ fn layout_of<'tcx>(
     tcx: TyCtxt<'tcx>,
     query: ty::ParamEnvAnd<'tcx, Ty<'tcx>>,
 ) -> Result<TyAndLayout<'tcx>, LayoutError<'tcx>> {
-    ty::tls::with_related_context(tcx, move |icx| {
+    tls::with_increased_layout_depth(move |layout_depth| {
         let (param_env, ty) = query.into_parts();
         debug!(?ty);
 
-        if !tcx.recursion_limit().value_within_limit(icx.layout_depth) {
+        if !tcx.recursion_limit().value_within_limit(layout_depth) {
             tcx.sess.fatal(&format!("overflow representing the type `{}`", ty));
         }
 
-        // Update the ImplicitCtxt to increase the layout_depth
-        let icx = ty::tls::ImplicitCtxt { layout_depth: icx.layout_depth + 1, ..icx.clone() };
-
-        ty::tls::enter_context(&icx, |_| {
-            let param_env = param_env.with_reveal_all_normalized(tcx);
-            let unnormalized_ty = ty;
-
-            // FIXME: We might want to have two different versions of `layout_of`:
-            // One that can be called after typecheck has completed and can use
-            // `normalize_erasing_regions` here and another one that can be called
-            // before typecheck has completed and uses `try_normalize_erasing_regions`.
-            let ty = match tcx.try_normalize_erasing_regions(param_env, ty) {
-                Ok(t) => t,
-                Err(normalization_error) => {
-                    return Err(LayoutError::NormalizationFailure(ty, normalization_error));
-                }
-            };
-
-            if ty != unnormalized_ty {
-                // Ensure this layout is also cached for the normalized type.
-                return tcx.layout_of(param_env.and(ty));
+        let param_env = param_env.with_reveal_all_normalized(tcx);
+        let unnormalized_ty = ty;
+
+        // FIXME: We might want to have two different versions of `layout_of`:
+        // One that can be called after typecheck has completed and can use
+        // `normalize_erasing_regions` here and another one that can be called
+        // before typecheck has completed and uses `try_normalize_erasing_regions`.
+        let ty = match tcx.try_normalize_erasing_regions(param_env, ty) {
+            Ok(t) => t,
+            Err(normalization_error) => {
+                return Err(LayoutError::NormalizationFailure(ty, normalization_error));
             }
+        };
+
+        if ty != unnormalized_ty {
+            // Ensure this layout is also cached for the normalized type.
+            return tcx.layout_of(param_env.and(ty));
+        }
 
-            let cx = LayoutCx { tcx, param_env };
+        let cx = LayoutCx { tcx, param_env };
 
-            let layout = cx.layout_of_uncached(ty)?;
-            let layout = TyAndLayout { ty, layout };
+        let layout = cx.layout_of_uncached(ty)?;
+        let layout = TyAndLayout { ty, layout };
 
-            cx.record_layout_for_printing(layout);
+        cx.record_layout_for_printing(layout);
 
-            // Type-level uninhabitedness should always imply ABI uninhabitedness.
-            if tcx.conservative_is_privately_uninhabited(param_env.and(ty)) {
-                assert!(layout.abi.is_uninhabited());
-            }
+        // Type-level uninhabitedness should always imply ABI uninhabitedness.
+        if tcx.conservative_is_privately_uninhabited(param_env.and(ty)) {
+            assert!(layout.abi.is_uninhabited());
+        }
 
-            Ok(layout)
-        })
+        Ok(layout)
     })
 }
 
diff --git a/compiler/rustc_middle/src/ty/query.rs b/compiler/rustc_middle/src/ty/query.rs
index 9e48c569c253..db14fc04172f 100644
--- a/compiler/rustc_middle/src/ty/query.rs
+++ b/compiler/rustc_middle/src/ty/query.rs
@@ -60,7 +60,6 @@ use std::ops::Deref;
 use std::path::PathBuf;
 use std::sync::Arc;
 
-pub(crate) use rustc_query_system::query::QueryJobId;
 use rustc_query_system::query::*;
 
 #[derive(Copy, Clone)]
diff --git a/compiler/rustc_query_impl/src/lib.rs b/compiler/rustc_query_impl/src/lib.rs
index 6ebff5388f45..437be15514d9 100644
--- a/compiler/rustc_query_impl/src/lib.rs
+++ b/compiler/rustc_query_impl/src/lib.rs
@@ -17,7 +17,7 @@ extern crate rustc_middle;
 use rustc_data_structures::stable_hasher::{HashStable, StableHasher};
 use rustc_data_structures::sync::AtomicU64;
 use rustc_middle::arena::Arena;
-use rustc_middle::dep_graph::{self, DepKindStruct, SerializedDepNodeIndex};
+use rustc_middle::dep_graph::{self, label_strs, DepKindStruct, SerializedDepNodeIndex};
 use rustc_middle::ty::query::{query_keys, query_storage, query_stored, query_values};
 use rustc_middle::ty::query::{ExternProviders, Providers, QueryEngine};
 use rustc_middle::ty::{self, TyCtxt};
diff --git a/compiler/rustc_query_impl/src/plumbing.rs b/compiler/rustc_query_impl/src/plumbing.rs
index bc82b0053b9e..1645245804c3 100644
--- a/compiler/rustc_query_impl/src/plumbing.rs
+++ b/compiler/rustc_query_impl/src/plumbing.rs
@@ -3,15 +3,11 @@
 //! manage the caches, and so forth.
 
 use crate::{on_disk_cache, Queries};
+use rustc_errors::Handler;
 use rustc_middle::dep_graph::{DepNodeIndex, SerializedDepNodeIndex};
-use rustc_middle::ty::tls::{self, ImplicitCtxt};
 use rustc_middle::ty::TyCtxt;
 use rustc_query_system::dep_graph::HasDepContext;
 use rustc_query_system::query::{QueryContext, QueryJobId, QueryMap, QuerySideEffects};
-
-use rustc_data_structures::sync::Lock;
-use rustc_data_structures::thin_vec::ThinVec;
-use rustc_errors::{Diagnostic, Handler};
 use rustc_serialize::opaque;
 
 use std::any::Any;
@@ -33,7 +29,6 @@ impl<'tcx> std::ops::Deref for QueryCtxt<'tcx> {
 }
 
 impl<'tcx> HasDepContext for QueryCtxt<'tcx> {
-    type DepKind = rustc_middle::dep_graph::DepKind;
     type DepContext = TyCtxt<'tcx>;
 
     #[inline]
@@ -52,10 +47,6 @@ impl QueryContext for QueryCtxt<'_> {
         )
     }
 
-    fn current_query_job(&self) -> Option<QueryJobId> {
-        tls::with_related_context(**self, |icx| icx.query)
-    }
-
     fn try_collect_active_jobs(&self) -> Option<QueryMap> {
         self.queries.try_collect_active_jobs(**self)
     }
@@ -84,36 +75,6 @@ impl QueryContext for QueryCtxt<'_> {
             c.store_side_effects_for_anon_node(dep_node_index, side_effects)
         }
     }
-
-    /// Executes a job by changing the `ImplicitCtxt` to point to the
-    /// new query job while it executes. It returns the diagnostics
-    /// captured during execution and the actual result.
-    #[inline(always)]
-    fn start_query<R>(
-        &self,
-        token: QueryJobId,
-        diagnostics: Option<&Lock<ThinVec<Diagnostic>>>,
-        compute: impl FnOnce() -> R,
-    ) -> R {
-        // The `TyCtxt` stored in TLS has the same global interner lifetime
-        // as `self`, so we use `with_related_context` to relate the 'tcx lifetimes
-        // when accessing the `ImplicitCtxt`.
-        tls::with_related_context(**self, move |current_icx| {
-            // Update the `ImplicitCtxt` to point to our new query job.
-            let new_icx = ImplicitCtxt {
-                tcx: **self,
-                query: Some(token),
-                diagnostics,
-                layout_depth: current_icx.layout_depth,
-                task_deps: current_icx.task_deps,
-            };
-
-            // Use the `ImplicitCtxt` while we execute the query.
-            tls::enter_context(&new_icx, |_| {
-                rustc_data_structures::stack::ensure_sufficient_stack(compute)
-            })
-        })
-    }
 }
 
 impl<'tcx> QueryCtxt<'tcx> {
@@ -160,13 +121,8 @@ impl<'tcx> QueryCtxt<'tcx> {
         Ok(())
     }
 
-    pub fn try_print_query_stack(
-        self,
-        query: Option<QueryJobId>,
-        handler: &Handler,
-        num_frames: Option<usize>,
-    ) -> usize {
-        rustc_query_system::query::print_query_stack(self, query, handler, num_frames)
+    pub fn try_print_query_stack(self, handler: &Handler, num_frames: Option<usize>) -> usize {
+        rustc_query_system::query::print_query_stack(self, handler, num_frames)
     }
 }
 
@@ -268,7 +224,7 @@ macro_rules! define_queries {
             $(#[allow(nonstandard_style)] $(#[$attr])*
             pub fn $name<$tcx>(tcx: QueryCtxt<$tcx>, key: query_keys::$name<$tcx>) -> QueryStackFrame {
                 opt_remap_env_constness!([$($modifiers)*][key]);
-                let kind = dep_graph::DepKind::$name;
+                let kind = dep_graph::dep_kind::$name;
                 let name = stringify!($name);
                 // Disable visible paths printing for performance reasons.
                 // Showing visible path instead of any path is not that important in production.
@@ -283,7 +239,7 @@ macro_rules! define_queries {
                 } else {
                     description
                 };
-                let span = if kind == dep_graph::DepKind::def_span {
+                let span = if kind == dep_graph::dep_kind::def_span {
                     // The `def_span` query is used to calculate `default_span`,
                     // so exit to avoid infinite recursion.
                     None
@@ -298,7 +254,7 @@ macro_rules! define_queries {
                 let hash = || {
                     let mut hcx = tcx.create_stable_hashing_context();
                     let mut hasher = StableHasher::new();
-                    std::mem::discriminant(&kind).hash_stable(&mut hcx, &mut hasher);
+                    kind.index().hash_stable(&mut hcx, &mut hasher);
                     key.hash_stable(&mut hcx, &mut hasher);
                     hasher.finish::<u64>()
                 };
@@ -351,7 +307,7 @@ macro_rules! define_queries {
                 QueryVtable {
                     anon: is_anon!([$($modifiers)*]),
                     eval_always: is_eval_always!([$($modifiers)*]),
-                    dep_kind: dep_graph::DepKind::$name,
+                    dep_kind: dep_graph::dep_kind::$name,
                     hash_result: hash_result!([$($modifiers)*]),
                     handle_cycle_error: |tcx, mut error| handle_cycle_error!([$($modifiers)*][tcx, error]),
                     compute,
@@ -371,8 +327,9 @@ macro_rules! define_queries {
             use rustc_query_system::dep_graph::FingerprintStyle;
 
             // We use this for most things when incr. comp. is turned off.
-            pub fn Null() -> DepKindStruct {
+            pub fn NULL() -> DepKindStruct {
                 DepKindStruct {
+                    label: label_strs::NULL,
                     is_anon: false,
                     is_eval_always: false,
                     fingerprint_style: FingerprintStyle::Unit,
@@ -383,6 +340,7 @@ macro_rules! define_queries {
 
             pub fn TraitSelect() -> DepKindStruct {
                 DepKindStruct {
+                    label: label_strs::TraitSelect,
                     is_anon: true,
                     is_eval_always: false,
                     fingerprint_style: FingerprintStyle::Unit,
@@ -393,6 +351,7 @@ macro_rules! define_queries {
 
             pub fn CompileCodegenUnit() -> DepKindStruct {
                 DepKindStruct {
+                    label: label_strs::CompileCodegenUnit,
                     is_anon: false,
                     is_eval_always: false,
                     fingerprint_style: FingerprintStyle::Opaque,
@@ -403,6 +362,7 @@ macro_rules! define_queries {
 
             pub fn CompileMonoItem() -> DepKindStruct {
                 DepKindStruct {
+                    label: label_strs::CompileMonoItem,
                     is_anon: false,
                     is_eval_always: false,
                     fingerprint_style: FingerprintStyle::Opaque,
@@ -420,6 +380,7 @@ macro_rules! define_queries {
 
                 if is_anon || !fingerprint_style.reconstructible() {
                     return DepKindStruct {
+                        label: label_strs::$name,
                         is_anon,
                         is_eval_always,
                         fingerprint_style,
@@ -453,6 +414,7 @@ macro_rules! define_queries {
                 }
 
                 DepKindStruct {
+                    label: stringify!($name),
                     is_anon,
                     is_eval_always,
                     fingerprint_style,
diff --git a/compiler/rustc_query_system/src/dep_graph/debug.rs b/compiler/rustc_query_system/src/dep_graph/debug.rs
index a544ac2c343a..968ca20655db 100644
--- a/compiler/rustc_query_system/src/dep_graph/debug.rs
+++ b/compiler/rustc_query_system/src/dep_graph/debug.rs
@@ -1,6 +1,6 @@
 //! Code for debugging the dep-graph.
 
-use super::{DepKind, DepNode, DepNodeIndex};
+use super::{DepNode, DepNodeIndex};
 use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::sync::Lock;
 use std::error::Error;
@@ -28,7 +28,7 @@ impl DepNodeFilter {
     }
 
     /// Tests whether `node` meets the filter, returning true if so.
-    pub fn test<K: DepKind>(&self, node: &DepNode<K>) -> bool {
+    pub fn test(&self, node: &DepNode) -> bool {
         let debug_str = format!("{:?}", node);
         self.text.split('&').map(|s| s.trim()).all(|f| debug_str.contains(f))
     }
@@ -36,14 +36,14 @@ impl DepNodeFilter {
 
 /// A filter like `F -> G` where `F` and `G` are valid dep-node
 /// filters. This can be used to test the source/target independently.
-pub struct EdgeFilter<K: DepKind> {
+pub struct EdgeFilter {
     pub source: DepNodeFilter,
     pub target: DepNodeFilter,
-    pub index_to_node: Lock<FxHashMap<DepNodeIndex, DepNode<K>>>,
+    pub index_to_node: Lock<FxHashMap<DepNodeIndex, DepNode>>,
 }
 
-impl<K: DepKind> EdgeFilter<K> {
-    pub fn new(test: &str) -> Result<EdgeFilter<K>, Box<dyn Error>> {
+impl EdgeFilter {
+    pub fn new(test: &str) -> Result<EdgeFilter, Box<dyn Error>> {
         let parts: Vec<_> = test.split("->").collect();
         if parts.len() != 2 {
             Err(format!("expected a filter like `a&b -> c&d`, not `{}`", test).into())
@@ -57,7 +57,7 @@ impl<K: DepKind> EdgeFilter<K> {
     }
 
     #[cfg(debug_assertions)]
-    pub fn test(&self, source: &DepNode<K>, target: &DepNode<K>) -> bool {
+    pub fn test(&self, source: &DepNode, target: &DepNode) -> bool {
         self.source.test(source) && self.target.test(target)
     }
 }
diff --git a/compiler/rustc_query_system/src/dep_graph/dep_node.rs b/compiler/rustc_query_system/src/dep_graph/dep_node.rs
index c274c2cc26c1..2f89f5328b94 100644
--- a/compiler/rustc_query_system/src/dep_graph/dep_node.rs
+++ b/compiler/rustc_query_system/src/dep_graph/dep_node.rs
@@ -42,35 +42,55 @@
 //!   `DefId` it was computed from. In other cases, too much information gets
 //!   lost during fingerprint computation.
 
-use super::{DepContext, DepKind, FingerprintStyle};
+use super::{DepContext, FingerprintStyle};
 use crate::ich::StableHashingContext;
 
 use rustc_data_structures::fingerprint::{Fingerprint, PackedFingerprint};
 use rustc_data_structures::stable_hasher::{HashStable, StableHasher};
+use rustc_data_structures::AtomicRef;
 use std::fmt;
 use std::hash::Hash;
 
-#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Encodable, Decodable)]
-pub struct DepNode<K> {
-    pub kind: K,
+#[derive(Clone, Copy, PartialEq, Eq, Hash, Encodable, Decodable)]
+pub struct DepKind(u16);
+
+impl DepKind {
+    /// We use this for most things when incr. comp. is turned off.
+    pub const NULL: Self = DepKind(0);
+
+    #[inline]
+    pub const fn new(index: u16) -> Self {
+        DepKind(index)
+    }
+
+    #[inline]
+    pub const fn index(self) -> usize {
+        self.0 as usize
+    }
+}
+
+#[derive(Clone, Copy, PartialEq, Eq, Hash, Encodable, Decodable)]
+pub struct DepNode {
+    pub kind: DepKind,
     pub hash: PackedFingerprint,
 }
 
-impl<K: DepKind> DepNode<K> {
+impl DepNode {
     /// Creates a new, parameterless DepNode. This method will assert
     /// that the DepNode corresponding to the given DepKind actually
     /// does not require any parameters.
-    pub fn new_no_params<Ctxt>(tcx: Ctxt, kind: K) -> DepNode<K>
+    #[inline]
+    pub fn new_no_params<Ctxt>(tcx: Ctxt, kind: DepKind) -> DepNode
     where
-        Ctxt: super::DepContext<DepKind = K>,
+        Ctxt: super::DepContext,
     {
         debug_assert_eq!(tcx.fingerprint_style(kind), FingerprintStyle::Unit);
         DepNode { kind, hash: Fingerprint::ZERO.into() }
     }
 
-    pub fn construct<Ctxt, Key>(tcx: Ctxt, kind: K, arg: &Key) -> DepNode<K>
+    pub fn construct<Ctxt, Key>(tcx: Ctxt, kind: DepKind, arg: &Key) -> DepNode
     where
-        Ctxt: super::DepContext<DepKind = K>,
+        Ctxt: super::DepContext,
         Key: DepNodeParams<Ctxt>,
     {
         let hash = arg.to_fingerprint(tcx);
@@ -90,12 +110,44 @@ impl<K: DepKind> DepNode<K> {
     }
 }
 
-impl<K: DepKind> fmt::Debug for DepNode<K> {
+impl fmt::Debug for DepKind {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        K::debug_node(self, f)
+        (*KIND_DEBUG)(self, f)
     }
 }
 
+pub static KIND_DEBUG: AtomicRef<fn(&DepKind, &mut fmt::Formatter<'_>) -> fmt::Result> =
+    AtomicRef::new(&(default_kind_debug as _));
+
+fn default_kind_debug(kind: &DepKind, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+    write!(f, "{:?}", kind.0)
+}
+
+impl fmt::Debug for DepNode {
+    #[inline]
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        (*NODE_DEBUG)(self, f)
+    }
+}
+
+pub static NODE_DEBUG: AtomicRef<fn(&DepNode, &mut fmt::Formatter<'_>) -> fmt::Result> =
+    AtomicRef::new(&(default_node_debug as _));
+
+fn default_node_debug(node: &DepNode, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+    write!(f, "{:?}({})", node.kind.0, node.hash)
+}
+
+// We keep a lot of `DepNode`s in memory during compilation. It's not
+// required that their size stay the same, but we don't want to change
+// it inadvertently. This assert just ensures we're aware of any change.
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+static_assert_size!(DepNode, 18);
+
+#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
+static_assert_size!(DepNode, 24);
+
+static_assert_size!(DepKind, 2);
+
 pub trait DepNodeParams<Ctxt: DepContext>: fmt::Debug + Sized {
     fn fingerprint_style() -> FingerprintStyle;
 
@@ -117,7 +169,7 @@ pub trait DepNodeParams<Ctxt: DepContext>: fmt::Debug + Sized {
     /// `fingerprint_style()` is not `FingerprintStyle::Opaque`.
     /// It is always valid to return `None` here, in which case incremental
     /// compilation will treat the query as having changed instead of forcing it.
-    fn recover(tcx: Ctxt, dep_node: &DepNode<Ctxt::DepKind>) -> Option<Self>;
+    fn recover(tcx: Ctxt, dep_node: &DepNode) -> Option<Self>;
 }
 
 impl<Ctxt: DepContext, T> DepNodeParams<Ctxt> for T
@@ -145,7 +197,7 @@ where
     }
 
     #[inline(always)]
-    default fn recover(_: Ctxt, _: &DepNode<Ctxt::DepKind>) -> Option<Self> {
+    default fn recover(_: Ctxt, _: &DepNode) -> Option<Self> {
         None
     }
 }
@@ -162,6 +214,7 @@ pub struct WorkProductId {
 }
 
 impl WorkProductId {
+    #[inline]
     pub fn from_cgu_name(cgu_name: &str) -> WorkProductId {
         let mut hasher = StableHasher::new();
         cgu_name.len().hash(&mut hasher);
diff --git a/compiler/rustc_query_system/src/dep_graph/graph.rs b/compiler/rustc_query_system/src/dep_graph/graph.rs
index f7655e55d348..4027d607cac2 100644
--- a/compiler/rustc_query_system/src/dep_graph/graph.rs
+++ b/compiler/rustc_query_system/src/dep_graph/graph.rs
@@ -26,8 +26,8 @@ use crate::query::{QueryContext, QuerySideEffects};
 use {super::debug::EdgeFilter, std::env};
 
 #[derive(Clone)]
-pub struct DepGraph<K: DepKind> {
-    data: Option<Lrc<DepGraphData<K>>>,
+pub struct DepGraph {
+    data: Option<Lrc<DepGraphData>>,
 
     /// This field is used for assigning DepNodeIndices when running in
     /// non-incremental mode. Even in non-incremental mode we make sure that
@@ -67,16 +67,16 @@ impl DepNodeColor {
     }
 }
 
-struct DepGraphData<K: DepKind> {
+struct DepGraphData {
     /// The new encoding of the dependency graph, optimized for red/green
     /// tracking. The `current` field is the dependency graph of only the
     /// current compilation session: We don't merge the previous dep-graph into
     /// current one anymore, but we do reference shared data to save space.
-    current: CurrentDepGraph<K>,
+    current: CurrentDepGraph,
 
     /// The dep-graph from the previous compilation session. It contains all
     /// nodes and edges as well as all fingerprints of nodes that have them.
-    previous: SerializedDepGraph<K>,
+    previous: SerializedDepGraph,
 
     colors: DepNodeColorMap,
 
@@ -88,12 +88,12 @@ struct DepGraphData<K: DepKind> {
     /// this map. We can later look for and extract that data.
     previous_work_products: FxHashMap<WorkProductId, WorkProduct>,
 
-    dep_node_debug: Lock<FxHashMap<DepNode<K>, String>>,
+    dep_node_debug: Lock<FxHashMap<DepNode, String>>,
 
     /// Used by incremental compilation tests to assert that
     /// a particular query result was decoded from disk
     /// (not just marked green)
-    debug_loaded_from_disk: Lock<FxHashSet<DepNode<K>>>,
+    debug_loaded_from_disk: Lock<FxHashSet<DepNode>>,
 }
 
 pub fn hash_result<R>(hcx: &mut StableHashingContext<'_>, result: &R) -> Fingerprint
@@ -105,15 +105,15 @@ where
     stable_hasher.finish()
 }
 
-impl<K: DepKind> DepGraph<K> {
+impl DepGraph {
     pub fn new(
         profiler: &SelfProfilerRef,
-        prev_graph: SerializedDepGraph<K>,
+        prev_graph: SerializedDepGraph,
         prev_work_products: FxHashMap<WorkProductId, WorkProduct>,
         encoder: FileEncoder,
         record_graph: bool,
         record_stats: bool,
-    ) -> DepGraph<K> {
+    ) -> DepGraph {
         let prev_graph_node_count = prev_graph.node_count();
 
         let current = CurrentDepGraph::new(
@@ -147,7 +147,8 @@ impl<K: DepKind> DepGraph<K> {
         }
     }
 
-    pub fn new_disabled() -> DepGraph<K> {
+    #[inline]
+    pub fn new_disabled() -> DepGraph {
         DepGraph { data: None, virtual_dep_node_index: Lrc::new(AtomicU32::new(0)) }
     }
 
@@ -157,15 +158,16 @@ impl<K: DepKind> DepGraph<K> {
         self.data.is_some()
     }
 
-    pub fn with_query(&self, f: impl Fn(&DepGraphQuery<K>)) {
+    pub fn with_query(&self, f: impl Fn(&DepGraphQuery)) {
         if let Some(data) = &self.data {
             data.current.encoder.borrow().with_query(f)
         }
     }
 
+    #[inline]
     pub fn assert_ignored(&self) {
         if let Some(..) = self.data {
-            K::read_deps(|task_deps| {
+            crate::tls::read_deps(|task_deps| {
                 assert_matches!(
                     task_deps,
                     TaskDepsRef::Ignore,
@@ -179,7 +181,7 @@ impl<K: DepKind> DepGraph<K> {
     where
         OP: FnOnce() -> R,
     {
-        K::with_deps(TaskDepsRef::Ignore, op)
+        crate::tls::with_deps(TaskDepsRef::Ignore, op)
     }
 
     /// Used to wrap the deserialization of a query result from disk,
@@ -232,7 +234,7 @@ impl<K: DepKind> DepGraph<K> {
     where
         OP: FnOnce() -> R,
     {
-        K::with_deps(TaskDepsRef::Forbid, op)
+        crate::tls::with_deps(TaskDepsRef::Forbid, op)
     }
 
     /// Starts a new dep-graph task. Dep-graph tasks are specified
@@ -262,9 +264,9 @@ impl<K: DepKind> DepGraph<K> {
     ///   `arg` parameter.
     ///
     /// [rustc dev guide]: https://rustc-dev-guide.rust-lang.org/incremental-compilation.html
-    pub fn with_task<Ctxt: HasDepContext<DepKind = K>, A: Debug, R>(
+    pub fn with_task<Ctxt: HasDepContext, A: Debug, R>(
         &self,
-        key: DepNode<K>,
+        key: DepNode,
         cx: Ctxt,
         arg: A,
         task: fn(Ctxt, A) -> R,
@@ -281,9 +283,9 @@ impl<K: DepKind> DepGraph<K> {
         }
     }
 
-    fn with_task_impl<Ctxt: HasDepContext<DepKind = K>, A: Debug, R>(
+    fn with_task_impl<Ctxt: HasDepContext, A: Debug, R>(
         &self,
-        key: DepNode<K>,
+        key: DepNode,
         cx: Ctxt,
         arg: A,
         task: fn(Ctxt, A) -> R,
@@ -323,7 +325,7 @@ impl<K: DepKind> DepGraph<K> {
             None => TaskDepsRef::Ignore,
         };
 
-        let result = K::with_deps(task_deps_ref, || task(cx, arg));
+        let result = crate::tls::with_deps(task_deps_ref, || task(cx, arg));
         let edges = task_deps.map_or_else(|| smallvec![], |lock| lock.into_inner().reads);
 
         let dcx = cx.dep_context();
@@ -363,10 +365,10 @@ impl<K: DepKind> DepGraph<K> {
 
     /// Executes something within an "anonymous" task, that is, a task the
     /// `DepNode` of which is determined by the list of inputs it read from.
-    pub fn with_anon_task<Ctxt: DepContext<DepKind = K>, OP, R>(
+    pub fn with_anon_task<Ctxt: DepContext, OP, R>(
         &self,
         cx: Ctxt,
-        dep_kind: K,
+        dep_kind: DepKind,
         op: OP,
     ) -> (R, DepNodeIndex)
     where
@@ -376,7 +378,7 @@ impl<K: DepKind> DepGraph<K> {
 
         if let Some(ref data) = self.data {
             let task_deps = Lock::new(TaskDeps::default());
-            let result = K::with_deps(TaskDepsRef::Allow(&task_deps), op);
+            let result = crate::tls::with_deps(TaskDepsRef::Allow(&task_deps), op);
             let task_deps = task_deps.into_inner();
             let task_deps = task_deps.reads;
 
@@ -428,7 +430,7 @@ impl<K: DepKind> DepGraph<K> {
     #[inline]
     pub fn read_index(&self, dep_node_index: DepNodeIndex) {
         if let Some(ref data) = self.data {
-            K::read_deps(|task_deps| {
+            crate::tls::read_deps(|task_deps| {
                 let mut task_deps = match task_deps {
                     TaskDepsRef::Allow(deps) => deps.lock(),
                     TaskDepsRef::Ignore => return,
@@ -476,12 +478,12 @@ impl<K: DepKind> DepGraph<K> {
     }
 
     #[inline]
-    pub fn dep_node_index_of(&self, dep_node: &DepNode<K>) -> DepNodeIndex {
+    pub fn dep_node_index_of(&self, dep_node: &DepNode) -> DepNodeIndex {
         self.dep_node_index_of_opt(dep_node).unwrap()
     }
 
     #[inline]
-    pub fn dep_node_index_of_opt(&self, dep_node: &DepNode<K>) -> Option<DepNodeIndex> {
+    pub fn dep_node_index_of_opt(&self, dep_node: &DepNode) -> Option<DepNodeIndex> {
         let data = self.data.as_ref().unwrap();
         let current = &data.current;
 
@@ -493,36 +495,39 @@ impl<K: DepKind> DepGraph<K> {
     }
 
     #[inline]
-    pub fn dep_node_exists(&self, dep_node: &DepNode<K>) -> bool {
+    pub fn dep_node_exists(&self, dep_node: &DepNode) -> bool {
         self.data.is_some() && self.dep_node_index_of_opt(dep_node).is_some()
     }
 
-    pub fn prev_fingerprint_of(&self, dep_node: &DepNode<K>) -> Option<Fingerprint> {
+    #[inline]
+    pub fn prev_fingerprint_of(&self, dep_node: &DepNode) -> Option<Fingerprint> {
         self.data.as_ref().unwrap().previous.fingerprint_of(dep_node)
     }
 
     /// Checks whether a previous work product exists for `v` and, if
     /// so, return the path that leads to it. Used to skip doing work.
+    #[inline]
     pub fn previous_work_product(&self, v: &WorkProductId) -> Option<WorkProduct> {
         self.data.as_ref().and_then(|data| data.previous_work_products.get(v).cloned())
     }
 
     /// Access the map of work-products created during the cached run. Only
     /// used during saving of the dep-graph.
+    #[inline]
     pub fn previous_work_products(&self) -> &FxHashMap<WorkProductId, WorkProduct> {
         &self.data.as_ref().unwrap().previous_work_products
     }
 
-    pub fn mark_debug_loaded_from_disk(&self, dep_node: DepNode<K>) {
+    pub fn mark_debug_loaded_from_disk(&self, dep_node: DepNode) {
         self.data.as_ref().unwrap().debug_loaded_from_disk.lock().insert(dep_node);
     }
 
-    pub fn debug_was_loaded_from_disk(&self, dep_node: DepNode<K>) -> bool {
+    pub fn debug_was_loaded_from_disk(&self, dep_node: DepNode) -> bool {
         self.data.as_ref().unwrap().debug_loaded_from_disk.lock().contains(&dep_node)
     }
 
     #[inline(always)]
-    pub fn register_dep_node_debug_str<F>(&self, dep_node: DepNode<K>, debug_str_gen: F)
+    pub fn register_dep_node_debug_str<F>(&self, dep_node: DepNode, debug_str_gen: F)
     where
         F: FnOnce() -> String,
     {
@@ -535,11 +540,13 @@ impl<K: DepKind> DepGraph<K> {
         dep_node_debug.borrow_mut().insert(dep_node, debug_str);
     }
 
-    pub fn dep_node_debug_str(&self, dep_node: DepNode<K>) -> Option<String> {
+    #[inline]
+    pub fn dep_node_debug_str(&self, dep_node: DepNode) -> Option<String> {
         self.data.as_ref()?.dep_node_debug.borrow().get(&dep_node).cloned()
     }
 
-    fn node_color(&self, dep_node: &DepNode<K>) -> Option<DepNodeColor> {
+    #[inline]
+    fn node_color(&self, dep_node: &DepNode) -> Option<DepNodeColor> {
         if let Some(ref data) = self.data {
             if let Some(prev_index) = data.previous.node_to_index_opt(dep_node) {
                 return data.colors.get(prev_index);
@@ -557,10 +564,10 @@ impl<K: DepKind> DepGraph<K> {
     /// A node will have an index, when it's already been marked green, or when we can mark it
     /// green. This function will mark the current task as a reader of the specified node, when
     /// a node index can be found for that node.
-    pub fn try_mark_green<Ctxt: QueryContext<DepKind = K>>(
+    pub fn try_mark_green<Ctxt: QueryContext>(
         &self,
         tcx: Ctxt,
-        dep_node: &DepNode<K>,
+        dep_node: &DepNode,
     ) -> Option<(SerializedDepNodeIndex, DepNodeIndex)> {
         debug_assert!(!tcx.dep_context().is_eval_always(dep_node.kind));
 
@@ -584,12 +591,12 @@ impl<K: DepKind> DepGraph<K> {
         }
     }
 
-    fn try_mark_parent_green<Ctxt: QueryContext<DepKind = K>>(
+    fn try_mark_parent_green<Ctxt: QueryContext>(
         &self,
         tcx: Ctxt,
-        data: &DepGraphData<K>,
+        data: &DepGraphData,
         parent_dep_node_index: SerializedDepNodeIndex,
-        dep_node: &DepNode<K>,
+        dep_node: &DepNode,
     ) -> Option<()> {
         let dep_dep_node_color = data.colors.get(parent_dep_node_index);
         let dep_dep_node = &data.previous.index_to_node(parent_dep_node_index);
@@ -696,12 +703,12 @@ impl<K: DepKind> DepGraph<K> {
     }
 
     /// Try to mark a dep-node which existed in the previous compilation session as green.
-    fn try_mark_previous_green<Ctxt: QueryContext<DepKind = K>>(
+    fn try_mark_previous_green<Ctxt: QueryContext>(
         &self,
         tcx: Ctxt,
-        data: &DepGraphData<K>,
+        data: &DepGraphData,
         prev_dep_node_index: SerializedDepNodeIndex,
-        dep_node: &DepNode<K>,
+        dep_node: &DepNode,
     ) -> Option<DepNodeIndex> {
         debug!("try_mark_previous_green({:?}) - BEGIN", dep_node);
 
@@ -766,10 +773,10 @@ impl<K: DepKind> DepGraph<K> {
     /// This may be called concurrently on multiple threads for the same dep node.
     #[cold]
     #[inline(never)]
-    fn emit_side_effects<Ctxt: QueryContext<DepKind = K>>(
+    fn emit_side_effects<Ctxt: QueryContext>(
         &self,
         tcx: Ctxt,
-        data: &DepGraphData<K>,
+        data: &DepGraphData,
         dep_node_index: DepNodeIndex,
         side_effects: QuerySideEffects,
     ) {
@@ -792,13 +799,15 @@ impl<K: DepKind> DepGraph<K> {
 
     // Returns true if the given node has been marked as red during the
     // current compilation session. Used in various assertions
-    pub fn is_red(&self, dep_node: &DepNode<K>) -> bool {
+    #[inline]
+    pub fn is_red(&self, dep_node: &DepNode) -> bool {
         self.node_color(dep_node) == Some(DepNodeColor::Red)
     }
 
     // Returns true if the given node has been marked as green during the
     // current compilation session. Used in various assertions
-    pub fn is_green(&self, dep_node: &DepNode<K>) -> bool {
+    #[inline]
+    pub fn is_green(&self, dep_node: &DepNode) -> bool {
         self.node_color(dep_node).map_or(false, |c| c.is_green())
     }
 
@@ -810,7 +819,7 @@ impl<K: DepKind> DepGraph<K> {
     //
     // This method will only load queries that will end up in the disk cache.
     // Other queries will not be executed.
-    pub fn exec_cache_promotions<Ctxt: DepContext<DepKind = K>>(&self, tcx: Ctxt) {
+    pub fn exec_cache_promotions<Ctxt: DepContext>(&self, tcx: Ctxt) {
         let _prof_timer = tcx.profiler().generic_activity("incr_comp_query_cache_promotion");
 
         let data = self.data.as_ref().unwrap();
@@ -838,6 +847,7 @@ impl<K: DepKind> DepGraph<K> {
         }
     }
 
+    #[inline]
     pub fn encode(&self, profiler: &SelfProfilerRef) -> FileEncodeResult {
         if let Some(data) = &self.data {
             data.current.encoder.steal().finish(profiler)
@@ -846,6 +856,7 @@ impl<K: DepKind> DepGraph<K> {
         }
     }
 
+    #[inline]
     pub(crate) fn next_virtual_depnode_index(&self) -> DepNodeIndex {
         let index = self.virtual_dep_node_index.fetch_add(1, Relaxed);
         DepNodeIndex::from_u32(index)
@@ -918,15 +929,15 @@ rustc_index::newtype_index! {
 /// `new_node_to_index` and `data`, or `prev_index_to_index` and `data`. When
 /// manipulating both, we acquire `new_node_to_index` or `prev_index_to_index`
 /// first, and `data` second.
-pub(super) struct CurrentDepGraph<K: DepKind> {
-    encoder: Steal<GraphEncoder<K>>,
-    new_node_to_index: Sharded<FxHashMap<DepNode<K>, DepNodeIndex>>,
+pub(super) struct CurrentDepGraph {
+    encoder: Steal<GraphEncoder>,
+    new_node_to_index: Sharded<FxHashMap<DepNode, DepNodeIndex>>,
     prev_index_to_index: Lock<IndexVec<SerializedDepNodeIndex, Option<DepNodeIndex>>>,
 
     /// Used to trap when a specific edge is added to the graph.
     /// This is used for debug purposes and is only active with `debug_assertions`.
     #[cfg(debug_assertions)]
-    forbidden_edge: Option<EdgeFilter<K>>,
+    forbidden_edge: Option<EdgeFilter>,
 
     /// Anonymous `DepNode`s are nodes whose IDs we compute from the list of
     /// their edges. This has the beneficial side-effect that multiple anonymous
@@ -953,14 +964,14 @@ pub(super) struct CurrentDepGraph<K: DepKind> {
     node_intern_event_id: Option<EventId>,
 }
 
-impl<K: DepKind> CurrentDepGraph<K> {
+impl CurrentDepGraph {
     fn new(
         profiler: &SelfProfilerRef,
         prev_graph_node_count: usize,
         encoder: FileEncoder,
         record_graph: bool,
         record_stats: bool,
-    ) -> CurrentDepGraph<K> {
+    ) -> CurrentDepGraph {
         use std::time::{SystemTime, UNIX_EPOCH};
 
         let duration = SystemTime::now().duration_since(UNIX_EPOCH).unwrap();
@@ -1012,7 +1023,8 @@ impl<K: DepKind> CurrentDepGraph<K> {
     }
 
     #[cfg(debug_assertions)]
-    fn record_edge(&self, dep_node_index: DepNodeIndex, key: DepNode<K>) {
+    #[inline]
+    fn record_edge(&self, dep_node_index: DepNodeIndex, key: DepNode) {
         if let Some(forbidden_edge) = &self.forbidden_edge {
             forbidden_edge.index_to_node.lock().insert(dep_node_index, key);
         }
@@ -1020,10 +1032,11 @@ impl<K: DepKind> CurrentDepGraph<K> {
 
     /// Writes the node to the current dep-graph and allocates a `DepNodeIndex` for it.
     /// Assumes that this is a node that has no equivalent in the previous dep-graph.
+    #[inline]
     fn intern_new_node(
         &self,
         profiler: &SelfProfilerRef,
-        key: DepNode<K>,
+        key: DepNode,
         edges: EdgesVec,
         current_fingerprint: Fingerprint,
     ) -> DepNodeIndex {
@@ -1040,11 +1053,12 @@ impl<K: DepKind> CurrentDepGraph<K> {
         }
     }
 
+    #[inline]
     fn intern_node(
         &self,
         profiler: &SelfProfilerRef,
-        prev_graph: &SerializedDepGraph<K>,
-        key: DepNode<K>,
+        prev_graph: &SerializedDepGraph,
+        key: DepNode,
         edges: EdgesVec,
         fingerprint: Option<Fingerprint>,
         print_status: bool,
@@ -1142,10 +1156,11 @@ impl<K: DepKind> CurrentDepGraph<K> {
         }
     }
 
+    #[inline]
     fn promote_node_and_deps_to_current(
         &self,
         profiler: &SelfProfilerRef,
-        prev_graph: &SerializedDepGraph<K>,
+        prev_graph: &SerializedDepGraph,
         prev_index: SerializedDepNodeIndex,
     ) -> DepNodeIndex {
         self.debug_assert_not_in_new_nodes(prev_graph, prev_index);
@@ -1177,7 +1192,7 @@ impl<K: DepKind> CurrentDepGraph<K> {
     #[inline]
     fn debug_assert_not_in_new_nodes(
         &self,
-        prev_graph: &SerializedDepGraph<K>,
+        prev_graph: &SerializedDepGraph,
         prev_index: SerializedDepNodeIndex,
     ) {
         let node = &prev_graph.index_to_node(prev_index);
@@ -1193,11 +1208,11 @@ const TASK_DEPS_READS_CAP: usize = 8;
 type EdgesVec = SmallVec<[DepNodeIndex; TASK_DEPS_READS_CAP]>;
 
 #[derive(Debug, Clone, Copy)]
-pub enum TaskDepsRef<'a, K: DepKind> {
+pub enum TaskDepsRef<'a> {
     /// New dependencies can be added to the
     /// `TaskDeps`. This is used when executing a 'normal' query
     /// (no `eval_always` modifier)
-    Allow(&'a Lock<TaskDeps<K>>),
+    Allow(&'a Lock<TaskDeps>),
     /// New dependencies are ignored. This is used when
     /// executing an `eval_always` query, since there's no
     /// need to track dependencies for a query that's always
@@ -1210,25 +1225,20 @@ pub enum TaskDepsRef<'a, K: DepKind> {
     Forbid,
 }
 
-#[derive(Debug)]
-pub struct TaskDeps<K: DepKind> {
+impl Default for TaskDepsRef<'_> {
+    #[inline]
+    fn default() -> Self {
+        TaskDepsRef::Ignore
+    }
+}
+
+#[derive(Default, Debug)]
+pub struct TaskDeps {
     #[cfg(debug_assertions)]
-    node: Option<DepNode<K>>,
+    node: Option<DepNode>,
     reads: EdgesVec,
     read_set: FxHashSet<DepNodeIndex>,
-    phantom_data: PhantomData<DepNode<K>>,
-}
-
-impl<K: DepKind> Default for TaskDeps<K> {
-    fn default() -> Self {
-        Self {
-            #[cfg(debug_assertions)]
-            node: None,
-            reads: EdgesVec::new(),
-            read_set: FxHashSet::default(),
-            phantom_data: PhantomData,
-        }
-    }
+    phantom_data: PhantomData<DepNode>,
 }
 
 // A data structure that stores Option<DepNodeColor> values as a contiguous
@@ -1242,6 +1252,7 @@ const COMPRESSED_RED: u32 = 1;
 const COMPRESSED_FIRST_GREEN: u32 = 2;
 
 impl DepNodeColorMap {
+    #[inline]
     fn new(size: usize) -> DepNodeColorMap {
         DepNodeColorMap { values: (0..size).map(|_| AtomicU32::new(COMPRESSED_NONE)).collect() }
     }
@@ -1257,6 +1268,7 @@ impl DepNodeColorMap {
         }
     }
 
+    #[inline]
     fn insert(&self, index: SerializedDepNodeIndex, color: DepNodeColor) {
         self.values[index].store(
             match color {
diff --git a/compiler/rustc_query_system/src/dep_graph/mod.rs b/compiler/rustc_query_system/src/dep_graph/mod.rs
index 5907ae309ca3..b951c236ccc0 100644
--- a/compiler/rustc_query_system/src/dep_graph/mod.rs
+++ b/compiler/rustc_query_system/src/dep_graph/mod.rs
@@ -4,7 +4,7 @@ mod graph;
 mod query;
 mod serialized;
 
-pub use dep_node::{DepNode, DepNodeParams, WorkProductId};
+pub use dep_node::{DepKind, DepNode, DepNodeParams, WorkProductId, KIND_DEBUG, NODE_DEBUG};
 pub use graph::{
     hash_result, DepGraph, DepNodeColor, DepNodeIndex, TaskDeps, TaskDepsRef, WorkProduct,
 };
@@ -13,20 +13,14 @@ pub use serialized::{SerializedDepGraph, SerializedDepNodeIndex};
 
 use crate::ich::StableHashingContext;
 use rustc_data_structures::profiling::SelfProfilerRef;
-use rustc_serialize::{opaque::FileEncoder, Encodable};
 use rustc_session::Session;
 
-use std::fmt;
-use std::hash::Hash;
-
 pub trait DepContext: Copy {
-    type DepKind: self::DepKind;
-
     /// Create a hashing context for hashing new results.
     fn create_stable_hashing_context(&self) -> StableHashingContext<'_>;
 
     /// Access the DepGraph.
-    fn dep_graph(&self) -> &DepGraph<Self::DepKind>;
+    fn dep_graph(&self) -> &DepGraph;
 
     /// Access the profiler.
     fn profiler(&self) -> &SelfProfilerRef;
@@ -35,26 +29,24 @@ pub trait DepContext: Copy {
     fn sess(&self) -> &Session;
 
     /// Return whether this kind always require evaluation.
-    fn is_eval_always(&self, kind: Self::DepKind) -> bool;
+    fn is_eval_always(&self, kind: DepKind) -> bool;
 
-    fn fingerprint_style(&self, kind: Self::DepKind) -> FingerprintStyle;
+    fn fingerprint_style(&self, kind: DepKind) -> FingerprintStyle;
 
     /// Try to force a dep node to execute and see if it's green.
-    fn try_force_from_dep_node(&self, dep_node: DepNode<Self::DepKind>) -> bool;
+    fn try_force_from_dep_node(&self, dep_node: DepNode) -> bool;
 
     /// Load data from the on-disk cache.
-    fn try_load_from_on_disk_cache(&self, dep_node: DepNode<Self::DepKind>);
+    fn try_load_from_on_disk_cache(&self, dep_node: DepNode);
 }
 
 pub trait HasDepContext: Copy {
-    type DepKind: self::DepKind;
-    type DepContext: self::DepContext<DepKind = Self::DepKind>;
+    type DepContext: self::DepContext;
 
     fn dep_context(&self) -> &Self::DepContext;
 }
 
 impl<T: DepContext> HasDepContext for T {
-    type DepKind = T::DepKind;
     type DepContext = Self;
 
     fn dep_context(&self) -> &Self::DepContext {
@@ -82,21 +74,3 @@ impl FingerprintStyle {
         }
     }
 }
-
-/// Describe the different families of dependency nodes.
-pub trait DepKind: Copy + fmt::Debug + Eq + Hash + Send + Encodable<FileEncoder> + 'static {
-    const NULL: Self;
-
-    /// Implementation of `std::fmt::Debug` for `DepNode`.
-    fn debug_node(node: &DepNode<Self>, f: &mut fmt::Formatter<'_>) -> fmt::Result;
-
-    /// Execute the operation with provided dependencies.
-    fn with_deps<OP, R>(deps: TaskDepsRef<'_, Self>, op: OP) -> R
-    where
-        OP: FnOnce() -> R;
-
-    /// Access dependencies from current implicit context.
-    fn read_deps<OP>(op: OP)
-    where
-        OP: for<'a> FnOnce(TaskDepsRef<'a, Self>);
-}
diff --git a/compiler/rustc_query_system/src/dep_graph/query.rs b/compiler/rustc_query_system/src/dep_graph/query.rs
index 27b3b5e13667..380b9c8d9151 100644
--- a/compiler/rustc_query_system/src/dep_graph/query.rs
+++ b/compiler/rustc_query_system/src/dep_graph/query.rs
@@ -2,16 +2,16 @@ use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::graph::implementation::{Direction, Graph, NodeIndex, INCOMING};
 use rustc_index::vec::IndexVec;
 
-use super::{DepKind, DepNode, DepNodeIndex};
+use super::{DepNode, DepNodeIndex};
 
-pub struct DepGraphQuery<K> {
-    pub graph: Graph<DepNode<K>, ()>,
-    pub indices: FxHashMap<DepNode<K>, NodeIndex>,
+pub struct DepGraphQuery {
+    pub graph: Graph<DepNode, ()>,
+    pub indices: FxHashMap<DepNode, NodeIndex>,
     pub dep_index_to_index: IndexVec<DepNodeIndex, Option<NodeIndex>>,
 }
 
-impl<K: DepKind> DepGraphQuery<K> {
-    pub fn new(prev_node_count: usize) -> DepGraphQuery<K> {
+impl DepGraphQuery {
+    pub fn new(prev_node_count: usize) -> DepGraphQuery {
         let node_count = prev_node_count + prev_node_count / 4;
         let edge_count = 6 * node_count;
 
@@ -22,7 +22,7 @@ impl<K: DepKind> DepGraphQuery<K> {
         DepGraphQuery { graph, indices, dep_index_to_index }
     }
 
-    pub fn push(&mut self, index: DepNodeIndex, node: DepNode<K>, edges: &[DepNodeIndex]) {
+    pub fn push(&mut self, index: DepNodeIndex, node: DepNode, edges: &[DepNodeIndex]) {
         let source = self.graph.add_node(node);
         if index.index() >= self.dep_index_to_index.len() {
             self.dep_index_to_index.resize(index.index() + 1, None);
@@ -40,11 +40,11 @@ impl<K: DepKind> DepGraphQuery<K> {
         }
     }
 
-    pub fn nodes(&self) -> Vec<&DepNode<K>> {
+    pub fn nodes(&self) -> Vec<&DepNode> {
         self.graph.all_nodes().iter().map(|n| &n.data).collect()
     }
 
-    pub fn edges(&self) -> Vec<(&DepNode<K>, &DepNode<K>)> {
+    pub fn edges(&self) -> Vec<(&DepNode, &DepNode)> {
         self.graph
             .all_edges()
             .iter()
@@ -53,7 +53,7 @@ impl<K: DepKind> DepGraphQuery<K> {
             .collect()
     }
 
-    fn reachable_nodes(&self, node: &DepNode<K>, direction: Direction) -> Vec<&DepNode<K>> {
+    fn reachable_nodes(&self, node: &DepNode, direction: Direction) -> Vec<&DepNode> {
         if let Some(&index) = self.indices.get(node) {
             self.graph.depth_traverse(index, direction).map(|s| self.graph.node_data(s)).collect()
         } else {
@@ -62,7 +62,7 @@ impl<K: DepKind> DepGraphQuery<K> {
     }
 
     /// All nodes that can reach `node`.
-    pub fn transitive_predecessors(&self, node: &DepNode<K>) -> Vec<&DepNode<K>> {
+    pub fn transitive_predecessors(&self, node: &DepNode) -> Vec<&DepNode> {
         self.reachable_nodes(node, INCOMING)
     }
 }
diff --git a/compiler/rustc_query_system/src/dep_graph/serialized.rs b/compiler/rustc_query_system/src/dep_graph/serialized.rs
index 1f8d87a7e913..bd72478086cf 100644
--- a/compiler/rustc_query_system/src/dep_graph/serialized.rs
+++ b/compiler/rustc_query_system/src/dep_graph/serialized.rs
@@ -35,9 +35,9 @@ rustc_index::newtype_index! {
 
 /// Data for use when recompiling the **current crate**.
 #[derive(Debug)]
-pub struct SerializedDepGraph<K: DepKind> {
+pub struct SerializedDepGraph {
     /// The set of all DepNodes in the graph
-    nodes: IndexVec<SerializedDepNodeIndex, DepNode<K>>,
+    nodes: IndexVec<SerializedDepNodeIndex, DepNode>,
     /// The set of all Fingerprints in the graph. Each Fingerprint corresponds to
     /// the DepNode at the same index in the nodes vector.
     fingerprints: IndexVec<SerializedDepNodeIndex, Fingerprint>,
@@ -49,10 +49,11 @@ pub struct SerializedDepGraph<K: DepKind> {
     /// implicit in edge_list_indices.
     edge_list_data: Vec<SerializedDepNodeIndex>,
     /// Reciprocal map to `nodes`.
-    index: FxHashMap<DepNode<K>, SerializedDepNodeIndex>,
+    index: FxHashMap<DepNode, SerializedDepNodeIndex>,
 }
 
-impl<K: DepKind> Default for SerializedDepGraph<K> {
+impl Default for SerializedDepGraph {
+    #[inline]
     fn default() -> Self {
         SerializedDepGraph {
             nodes: Default::default(),
@@ -64,7 +65,7 @@ impl<K: DepKind> Default for SerializedDepGraph<K> {
     }
 }
 
-impl<K: DepKind> SerializedDepGraph<K> {
+impl SerializedDepGraph {
     #[inline]
     pub fn edge_targets_from(&self, source: SerializedDepNodeIndex) -> &[SerializedDepNodeIndex] {
         let targets = self.edge_list_indices[source];
@@ -72,17 +73,17 @@ impl<K: DepKind> SerializedDepGraph<K> {
     }
 
     #[inline]
-    pub fn index_to_node(&self, dep_node_index: SerializedDepNodeIndex) -> DepNode<K> {
+    pub fn index_to_node(&self, dep_node_index: SerializedDepNodeIndex) -> DepNode {
         self.nodes[dep_node_index]
     }
 
     #[inline]
-    pub fn node_to_index_opt(&self, dep_node: &DepNode<K>) -> Option<SerializedDepNodeIndex> {
+    pub fn node_to_index_opt(&self, dep_node: &DepNode) -> Option<SerializedDepNodeIndex> {
         self.index.get(dep_node).cloned()
     }
 
     #[inline]
-    pub fn fingerprint_of(&self, dep_node: &DepNode<K>) -> Option<Fingerprint> {
+    pub fn fingerprint_of(&self, dep_node: &DepNode) -> Option<Fingerprint> {
         self.index.get(dep_node).map(|&node_index| self.fingerprints[node_index])
     }
 
@@ -91,16 +92,15 @@ impl<K: DepKind> SerializedDepGraph<K> {
         self.fingerprints[dep_node_index]
     }
 
+    #[inline]
     pub fn node_count(&self) -> usize {
         self.index.len()
     }
 }
 
-impl<'a, K: DepKind + Decodable<opaque::Decoder<'a>>> Decodable<opaque::Decoder<'a>>
-    for SerializedDepGraph<K>
-{
+impl<'a> Decodable<opaque::Decoder<'a>> for SerializedDepGraph {
     #[instrument(level = "debug", skip(d))]
-    fn decode(d: &mut opaque::Decoder<'a>) -> SerializedDepGraph<K> {
+    fn decode(d: &mut opaque::Decoder<'a>) -> SerializedDepGraph {
         let start_position = d.position();
 
         // The last 16 bytes are the node count and edge count.
@@ -122,7 +122,7 @@ impl<'a, K: DepKind + Decodable<opaque::Decoder<'a>>> Decodable<opaque::Decoder<
         let mut edge_list_data = Vec::with_capacity(edge_count);
 
         for _index in 0..node_count {
-            let dep_node: DepNode<K> = Decodable::decode(d);
+            let dep_node: DepNode = Decodable::decode(d);
             let _i: SerializedDepNodeIndex = nodes.push(dep_node);
             debug_assert_eq!(_i.index(), _index);
 
@@ -149,28 +149,29 @@ impl<'a, K: DepKind + Decodable<opaque::Decoder<'a>>> Decodable<opaque::Decoder<
     }
 }
 
-#[derive(Debug, Encodable, Decodable)]
-pub struct NodeInfo<K: DepKind> {
-    node: DepNode<K>,
+#[derive(Debug, Encodable)]
+pub struct NodeInfo {
+    node: DepNode,
     fingerprint: Fingerprint,
     edges: SmallVec<[DepNodeIndex; 8]>,
 }
 
-struct Stat<K: DepKind> {
-    kind: K,
+struct Stat {
+    kind: DepKind,
     node_counter: u64,
     edge_counter: u64,
 }
 
-struct EncoderState<K: DepKind> {
+struct EncoderState {
     encoder: FileEncoder,
     total_node_count: usize,
     total_edge_count: usize,
     result: FileEncodeResult,
-    stats: Option<FxHashMap<K, Stat<K>>>,
+    stats: Option<FxHashMap<DepKind, Stat>>,
 }
 
-impl<K: DepKind> EncoderState<K> {
+impl EncoderState {
+    #[inline]
     fn new(encoder: FileEncoder, record_stats: bool) -> Self {
         Self {
             encoder,
@@ -183,8 +184,8 @@ impl<K: DepKind> EncoderState<K> {
 
     fn encode_node(
         &mut self,
-        node: &NodeInfo<K>,
-        record_graph: &Option<Lock<DepGraphQuery<K>>>,
+        node: &NodeInfo,
+        record_graph: &Option<Lock<DepGraphQuery>>,
     ) -> DepNodeIndex {
         let index = DepNodeIndex::new(self.total_node_count);
         self.total_node_count += 1;
@@ -235,12 +236,13 @@ impl<K: DepKind> EncoderState<K> {
     }
 }
 
-pub struct GraphEncoder<K: DepKind> {
-    status: Lock<EncoderState<K>>,
-    record_graph: Option<Lock<DepGraphQuery<K>>>,
+pub struct GraphEncoder {
+    status: Lock<EncoderState>,
+    record_graph: Option<Lock<DepGraphQuery>>,
 }
 
-impl<K: DepKind + Encodable<FileEncoder>> GraphEncoder<K> {
+impl GraphEncoder {
+    #[inline]
     pub fn new(
         encoder: FileEncoder,
         prev_node_count: usize,
@@ -253,7 +255,7 @@ impl<K: DepKind + Encodable<FileEncoder>> GraphEncoder<K> {
         GraphEncoder { status, record_graph }
     }
 
-    pub(crate) fn with_query(&self, f: impl Fn(&DepGraphQuery<K>)) {
+    pub(crate) fn with_query(&self, f: impl Fn(&DepGraphQuery)) {
         if let Some(record_graph) = &self.record_graph {
             f(&record_graph.lock())
         }
@@ -314,10 +316,11 @@ impl<K: DepKind + Encodable<FileEncoder>> GraphEncoder<K> {
         }
     }
 
+    #[inline]
     pub(crate) fn send(
         &self,
         profiler: &SelfProfilerRef,
-        node: DepNode<K>,
+        node: DepNode,
         fingerprint: Fingerprint,
         edges: SmallVec<[DepNodeIndex; 8]>,
     ) -> DepNodeIndex {
@@ -326,6 +329,7 @@ impl<K: DepKind + Encodable<FileEncoder>> GraphEncoder<K> {
         self.status.lock().encode_node(&node, &self.record_graph)
     }
 
+    #[inline]
     pub fn finish(self, profiler: &SelfProfilerRef) -> FileEncodeResult {
         let _prof_timer = profiler.generic_activity("incr_comp_encode_dep_graph");
         self.status.into_inner().finish(profiler)
diff --git a/compiler/rustc_query_system/src/lib.rs b/compiler/rustc_query_system/src/lib.rs
index 6b70e9342fa6..dcfb35e22186 100644
--- a/compiler/rustc_query_system/src/lib.rs
+++ b/compiler/rustc_query_system/src/lib.rs
@@ -18,3 +18,5 @@ pub mod cache;
 pub mod dep_graph;
 pub mod ich;
 pub mod query;
+
+pub mod tls;
diff --git a/compiler/rustc_query_system/src/query/config.rs b/compiler/rustc_query_system/src/query/config.rs
index 7ca668f8a1f0..e7816616ce0d 100644
--- a/compiler/rustc_query_system/src/query/config.rs
+++ b/compiler/rustc_query_system/src/query/config.rs
@@ -1,5 +1,6 @@
 //! Query configuration and description traits.
 
+use crate::dep_graph::DepKind;
 use crate::dep_graph::DepNode;
 use crate::dep_graph::SerializedDepNodeIndex;
 use crate::ich::StableHashingContext;
@@ -21,7 +22,7 @@ pub trait QueryConfig {
 
 pub struct QueryVtable<CTX: QueryContext, K, V> {
     pub anon: bool,
-    pub dep_kind: CTX::DepKind,
+    pub dep_kind: DepKind,
     pub eval_always: bool,
     pub cache_on_disk: bool,
 
@@ -32,7 +33,7 @@ pub struct QueryVtable<CTX: QueryContext, K, V> {
 }
 
 impl<CTX: QueryContext, K, V> QueryVtable<CTX, K, V> {
-    pub(crate) fn to_dep_node(&self, tcx: CTX::DepContext, key: &K) -> DepNode<CTX::DepKind>
+    pub(crate) fn to_dep_node(&self, tcx: CTX::DepContext, key: &K) -> DepNode
     where
         K: crate::dep_graph::DepNodeParams<CTX::DepContext>,
     {
diff --git a/compiler/rustc_query_system/src/query/job.rs b/compiler/rustc_query_system/src/query/job.rs
index 2a07d9b7f809..0f20dac3b880 100644
--- a/compiler/rustc_query_system/src/query/job.rs
+++ b/compiler/rustc_query_system/src/query/job.rs
@@ -41,21 +41,25 @@ pub type QueryMap = FxHashMap<QueryJobId, QueryJobInfo>;
 pub struct QueryJobId(pub NonZeroU64);
 
 impl QueryJobId {
+    #[inline]
     fn query(self, map: &QueryMap) -> QueryStackFrame {
         map.get(&self).unwrap().query.clone()
     }
 
     #[cfg(parallel_compiler)]
+    #[inline]
     fn span(self, map: &QueryMap) -> Span {
         map.get(&self).unwrap().job.span
     }
 
     #[cfg(parallel_compiler)]
+    #[inline]
     fn parent(self, map: &QueryMap) -> Option<QueryJobId> {
         map.get(&self).unwrap().job.parent
     }
 
     #[cfg(parallel_compiler)]
+    #[inline]
     fn latch<'a>(self, map: &'a QueryMap) -> Option<&'a QueryLatch> {
         map.get(&self).unwrap().job.latch.as_ref()
     }
@@ -84,6 +88,7 @@ pub struct QueryJob {
 
 impl QueryJob {
     /// Creates a new query job.
+    #[inline]
     pub fn new(id: QueryJobId, span: Span, parent: Option<QueryJobId>) -> Self {
         QueryJob {
             id,
@@ -95,6 +100,7 @@ impl QueryJob {
     }
 
     #[cfg(parallel_compiler)]
+    #[inline]
     pub(super) fn latch(&mut self) -> QueryLatch {
         if self.latch.is_none() {
             self.latch = Some(QueryLatch::new());
@@ -106,6 +112,7 @@ impl QueryJob {
     ///
     /// This does nothing for single threaded rustc,
     /// as there are no concurrent jobs which could be waiting on us
+    #[inline]
     pub fn signal_complete(self) {
         #[cfg(parallel_compiler)]
         {
@@ -188,6 +195,7 @@ pub(super) struct QueryLatch {
 
 #[cfg(parallel_compiler)]
 impl QueryLatch {
+    #[inline]
     fn new() -> Self {
         QueryLatch {
             info: Lrc::new(Mutex::new(QueryLatchInfo { complete: false, waiters: Vec::new() })),
@@ -195,6 +203,7 @@ impl QueryLatch {
     }
 
     /// Awaits for the query job to complete.
+    #[inline]
     pub(super) fn wait_on(&self, query: Option<QueryJobId>, span: Span) -> Result<(), CycleError> {
         let waiter =
             Lrc::new(QueryWaiter { query, span, cycle: Lock::new(None), condvar: Condvar::new() });
@@ -210,6 +219,7 @@ impl QueryLatch {
     }
 
     /// Awaits the caller on this latch by blocking the current thread.
+    #[inline]
     fn wait_on_inner(&self, waiter: &Lrc<QueryWaiter>) {
         let mut info = self.info.lock();
         if !info.complete {
@@ -232,6 +242,7 @@ impl QueryLatch {
     }
 
     /// Sets the latch and resumes all waiters on it
+    #[inline]
     fn set(&self) {
         let mut info = self.info.lock();
         debug_assert!(!info.complete);
@@ -244,6 +255,7 @@ impl QueryLatch {
 
     /// Removes a single waiter from the list of waiters.
     /// This is used to break query cycles.
+    #[inline]
     fn extract_waiter(&self, waiter: usize) -> Lrc<QueryWaiter> {
         let mut info = self.info.lock();
         debug_assert!(!info.complete);
@@ -584,10 +596,11 @@ pub(crate) fn report_cycle<'a>(
 
 pub fn print_query_stack<CTX: QueryContext>(
     tcx: CTX,
-    mut current_query: Option<QueryJobId>,
     handler: &Handler,
     num_frames: Option<usize>,
 ) -> usize {
+    let mut current_query = crate::tls::current_query_job();
+
     // Be careful relying on global state here: this code is called from
     // a panic hook, which means that the global `Handler` may be in a weird
     // state if it was responsible for triggering the panic.
diff --git a/compiler/rustc_query_system/src/query/mod.rs b/compiler/rustc_query_system/src/query/mod.rs
index de64ebb62030..9341e9c2ca16 100644
--- a/compiler/rustc_query_system/src/query/mod.rs
+++ b/compiler/rustc_query_system/src/query/mod.rs
@@ -16,7 +16,6 @@ pub use self::config::{QueryConfig, QueryDescription, QueryVtable};
 
 use crate::dep_graph::{DepNodeIndex, HasDepContext, SerializedDepNodeIndex};
 
-use rustc_data_structures::sync::Lock;
 use rustc_data_structures::thin_vec::ThinVec;
 use rustc_errors::Diagnostic;
 use rustc_hir::def::DefKind;
@@ -94,9 +93,6 @@ impl QuerySideEffects {
 pub trait QueryContext: HasDepContext {
     fn next_job_id(&self) -> QueryJobId;
 
-    /// Get the query information from the TLS context.
-    fn current_query_job(&self) -> Option<QueryJobId>;
-
     fn try_collect_active_jobs(&self) -> Option<QueryMap>;
 
     /// Load side effects associated to the node in the previous session.
@@ -111,14 +107,4 @@ pub trait QueryContext: HasDepContext {
         dep_node_index: DepNodeIndex,
         side_effects: QuerySideEffects,
     );
-
-    /// Executes a job by changing the `ImplicitCtxt` to point to the
-    /// new query job while it executes. It returns the diagnostics
-    /// captured during execution and the actual result.
-    fn start_query<R>(
-        &self,
-        token: QueryJobId,
-        diagnostics: Option<&Lock<ThinVec<Diagnostic>>>,
-        compute: impl FnOnce() -> R,
-    ) -> R;
 }
diff --git a/compiler/rustc_query_system/src/query/plumbing.rs b/compiler/rustc_query_system/src/query/plumbing.rs
index 700290d67a41..d94430e26a82 100644
--- a/compiler/rustc_query_system/src/query/plumbing.rs
+++ b/compiler/rustc_query_system/src/query/plumbing.rs
@@ -162,7 +162,7 @@ where
         match lock.entry(key) {
             Entry::Vacant(entry) => {
                 let id = tcx.next_job_id();
-                let job = tcx.current_query_job();
+                let job = crate::tls::current_query_job();
                 let job = QueryJob::new(id, span, job);
 
                 let key = entry.key().clone();
@@ -182,7 +182,7 @@ where
                         // so we just return the error.
                         return TryGetJob::Cycle(id.find_cycle_in_stack(
                             tcx.try_collect_active_jobs().unwrap(),
-                            &tcx.current_query_job(),
+                            &crate::tls::current_query_job(),
                             span,
                         ));
                     }
@@ -200,7 +200,7 @@ where
 
                         // With parallel queries we might just have to wait on some other
                         // thread.
-                        let result = latch.wait_on(tcx.current_query_job(), span);
+                        let result = latch.wait_on(crate::tls::current_query_job(), span);
 
                         match result {
                             Ok(()) => TryGetJob::JobCompleted(query_blocked_prof_timer),
@@ -330,7 +330,7 @@ fn try_execute_query<CTX, C>(
     cache: &C,
     span: Span,
     key: C::Key,
-    dep_node: Option<DepNode<CTX::DepKind>>,
+    dep_node: Option<DepNode>,
     query: &QueryVtable<CTX, C::Key, C::Value>,
 ) -> (C::Stored, Option<DepNodeIndex>)
 where
@@ -367,7 +367,7 @@ where
 fn execute_job<CTX, K, V>(
     tcx: CTX,
     key: K,
-    mut dep_node_opt: Option<DepNode<CTX::DepKind>>,
+    mut dep_node_opt: Option<DepNode>,
     query: &QueryVtable<CTX, K, V>,
     job_id: QueryJobId,
 ) -> (V, DepNodeIndex)
@@ -381,7 +381,8 @@ where
     // Fast path for when incr. comp. is off.
     if !dep_graph.is_fully_enabled() {
         let prof_timer = tcx.dep_context().profiler().query_provider();
-        let result = tcx.start_query(job_id, None, || query.compute(*tcx.dep_context(), key));
+        let result =
+            crate::tls::start_query(job_id, None, || query.compute(*tcx.dep_context(), key));
         let dep_node_index = dep_graph.next_virtual_depnode_index();
         prof_timer.finish_with_query_invocation_id(dep_node_index.into());
         return (result, dep_node_index);
@@ -394,7 +395,7 @@ where
 
         // The diagnostics for this query will be promoted to the current session during
         // `try_mark_green()`, so we can ignore them here.
-        if let Some(ret) = tcx.start_query(job_id, None, || {
+        if let Some(ret) = crate::tls::start_query(job_id, None, || {
             try_load_from_disk_and_cache_in_memory(tcx, &key, &dep_node, query)
         }) {
             return ret;
@@ -404,7 +405,7 @@ where
     let prof_timer = tcx.dep_context().profiler().query_provider();
     let diagnostics = Lock::new(ThinVec::new());
 
-    let (result, dep_node_index) = tcx.start_query(job_id, Some(&diagnostics), || {
+    let (result, dep_node_index) = crate::tls::start_query(job_id, Some(&diagnostics), || {
         if query.anon {
             return dep_graph.with_anon_task(*tcx.dep_context(), query.dep_kind, || {
                 query.compute(*tcx.dep_context(), key)
@@ -436,11 +437,11 @@ where
 fn try_load_from_disk_and_cache_in_memory<CTX, K, V>(
     tcx: CTX,
     key: &K,
-    dep_node: &DepNode<CTX::DepKind>,
+    dep_node: &DepNode,
     query: &QueryVtable<CTX, K, V>,
 ) -> Option<(V, DepNodeIndex)>
 where
-    K: Clone,
+    K: Clone + DepNodeParams<CTX::DepContext>,
     CTX: QueryContext,
     V: Debug,
 {
@@ -527,7 +528,7 @@ where
 fn incremental_verify_ich<CTX, K, V: Debug>(
     tcx: CTX::DepContext,
     result: &V,
-    dep_node: &DepNode<CTX::DepKind>,
+    dep_node: &DepNode,
     query: &QueryVtable<CTX, K, V>,
 ) where
     CTX: QueryContext,
@@ -642,7 +643,7 @@ fn ensure_must_run<CTX, K, V>(
     tcx: CTX,
     key: &K,
     query: &QueryVtable<CTX, K, V>,
-) -> (bool, Option<DepNode<CTX::DepKind>>)
+) -> (bool, Option<DepNode>)
 where
     K: crate::dep_graph::DepNodeParams<CTX::DepContext>,
     CTX: QueryContext,
@@ -713,7 +714,7 @@ where
     Some(result)
 }
 
-pub fn force_query<Q, CTX>(tcx: CTX, key: Q::Key, dep_node: DepNode<CTX::DepKind>)
+pub fn force_query<Q, CTX>(tcx: CTX, key: Q::Key, dep_node: DepNode)
 where
     Q: QueryDescription<CTX>,
     Q::Key: DepNodeParams<CTX::DepContext>,
diff --git a/compiler/rustc_query_system/src/tls.rs b/compiler/rustc_query_system/src/tls.rs
new file mode 100644
index 000000000000..a4cb605dee0c
--- /dev/null
+++ b/compiler/rustc_query_system/src/tls.rs
@@ -0,0 +1,164 @@
+use crate::dep_graph::TaskDepsRef;
+use crate::query::QueryJobId;
+use rustc_data_structures::sync::{self, Lock};
+use rustc_data_structures::thin_vec::ThinVec;
+use rustc_errors::Diagnostic;
+
+#[cfg(not(parallel_compiler))]
+use std::cell::Cell;
+
+#[cfg(parallel_compiler)]
+use rustc_rayon_core as rayon_core;
+
+/// This is the implicit state of rustc. It contains the current
+/// query. It is updated when
+/// executing a new query.
+#[derive(Clone, Default)]
+struct ImplicitCtxt<'a> {
+    /// The current query job, if any. This is updated by `JobOwner::start` in
+    /// `ty::query::plumbing` when executing a query.
+    query: Option<QueryJobId>,
+
+    /// Where to store diagnostics for the current query job, if any.
+    /// This is updated by `JobOwner::start` in `ty::query::plumbing` when executing a query.
+    diagnostics: Option<&'a Lock<ThinVec<Diagnostic>>>,
+
+    /// Used to prevent layout from recursing too deeply.
+    layout_depth: usize,
+
+    /// The current dep graph task. This is used to add dependencies to queries
+    /// when executing them.
+    task_deps: TaskDepsRef<'a>,
+}
+
+/// Sets Rayon's thread-local variable, which is preserved for Rayon jobs
+/// to `value` during the call to `f`. It is restored to its previous value after.
+/// This is used to set the pointer to the new `ImplicitCtxt`.
+#[cfg(parallel_compiler)]
+#[inline]
+fn set_tlv<F: FnOnce() -> R, R>(value: usize, f: F) -> R {
+    rayon_core::tlv::with(value, f)
+}
+
+/// Gets Rayon's thread-local variable, which is preserved for Rayon jobs.
+/// This is used to get the pointer to the current `ImplicitCtxt`.
+#[cfg(parallel_compiler)]
+#[inline]
+pub fn get_tlv() -> usize {
+    rayon_core::tlv::get()
+}
+
+#[cfg(not(parallel_compiler))]
+thread_local! {
+    /// A thread local variable that stores a pointer to the current `ImplicitCtxt`.
+    static TLV: Cell<usize> = const { Cell::new(0) };
+}
+
+/// Sets TLV to `value` during the call to `f`.
+/// It is restored to its previous value after.
+/// This is used to set the pointer to the new `ImplicitCtxt`.
+#[cfg(not(parallel_compiler))]
+#[inline]
+fn set_tlv<F: FnOnce() -> R, R>(value: usize, f: F) -> R {
+    let old = get_tlv();
+    let _reset = rustc_data_structures::OnDrop(move || TLV.with(|tlv| tlv.set(old)));
+    TLV.with(|tlv| tlv.set(value));
+    f()
+}
+
+/// Gets the pointer to the current `ImplicitCtxt`.
+#[cfg(not(parallel_compiler))]
+#[inline]
+fn get_tlv() -> usize {
+    TLV.with(|tlv| tlv.get())
+}
+
+/// Sets `context` as the new current `ImplicitCtxt` for the duration of the function `f`.
+#[inline]
+fn enter_context<'a, F, R>(context: &ImplicitCtxt<'a>, f: F) -> R
+where
+    F: FnOnce() -> R,
+{
+    set_tlv(context as *const _ as usize, || f())
+}
+
+/// Allows access to the current `ImplicitCtxt` in a closure if one is available.
+#[inline]
+fn with_context_opt<F, R>(f: F) -> R
+where
+    F: for<'a> FnOnce(Option<&ImplicitCtxt<'a>>) -> R,
+{
+    // We could get a `ImplicitCtxt` pointer from another thread.
+    // Ensure that `ImplicitCtxt` is `Sync`.
+    sync::assert_sync::<ImplicitCtxt<'_>>();
+
+    let context = get_tlv();
+    let context =
+        if context == 0 { None } else { unsafe { Some(&*(context as *const ImplicitCtxt<'_>)) } };
+    f(context)
+}
+
+/// This is a callback from `rustc_ast` as it cannot access the implicit state
+/// in `rustc_middle` otherwise. It is used to when diagnostic messages are
+/// emitted and stores them in the current query, if there is one.
+pub fn track_diagnostic(diagnostic: &Diagnostic) {
+    with_context_opt(|icx| {
+        if let Some(icx) = icx {
+            if let Some(ref diagnostics) = icx.diagnostics {
+                let mut diagnostics = diagnostics.lock();
+                diagnostics.extend(Some(diagnostic.clone()));
+            }
+        }
+    })
+}
+
+pub fn with_deps<R>(task_deps: TaskDepsRef<'_>, op: impl FnOnce() -> R) -> R {
+    crate::tls::with_context_opt(|icx| {
+        let icx = crate::tls::ImplicitCtxt { task_deps, ..icx.cloned().unwrap_or_default() };
+
+        crate::tls::enter_context(&icx, op)
+    })
+}
+
+pub fn read_deps(op: impl FnOnce(TaskDepsRef<'_>)) {
+    crate::tls::with_context_opt(|icx| if let Some(icx) = icx { op(icx.task_deps) } else { return })
+}
+
+/// Get the query information from the TLS context.
+#[inline(always)]
+pub fn current_query_job() -> Option<QueryJobId> {
+    with_context_opt(|icx| icx?.query)
+}
+
+/// Executes a job by changing the `ImplicitCtxt` to point to the
+/// new query job while it executes. It returns the diagnostics
+/// captured during execution and the actual result.
+#[inline(always)]
+pub fn start_query<R>(
+    token: QueryJobId,
+    diagnostics: Option<&Lock<ThinVec<Diagnostic>>>,
+    compute: impl FnOnce() -> R,
+) -> R {
+    with_context_opt(move |current_icx| {
+        let task_deps = current_icx.map_or(TaskDepsRef::Ignore, |icx| icx.task_deps);
+        let layout_depth = current_icx.map_or(0, |icx| icx.layout_depth);
+
+        // Update the `ImplicitCtxt` to point to our new query job.
+        let new_icx = ImplicitCtxt { query: Some(token), diagnostics, task_deps, layout_depth };
+
+        // Use the `ImplicitCtxt` while we execute the query.
+        enter_context(&new_icx, || rustc_data_structures::stack::ensure_sufficient_stack(compute))
+    })
+}
+
+pub fn with_increased_layout_depth<R>(f: impl FnOnce(usize) -> R) -> R {
+    with_context_opt(|icx| {
+        let layout_depth = icx.map_or(0, |icx| icx.layout_depth);
+
+        // Update the ImplicitCtxt to increase the layout_depth
+        let icx =
+            ImplicitCtxt { layout_depth: layout_depth + 1, ..icx.cloned().unwrap_or_default() };
+
+        enter_context(&icx, || f(layout_depth))
+    })
+}
diff --git a/compiler/rustc_session/src/cstore.rs b/compiler/rustc_session/src/cstore.rs
index c1fd3c7c61b9..658e969fcfe9 100644
--- a/compiler/rustc_session/src/cstore.rs
+++ b/compiler/rustc_session/src/cstore.rs
@@ -215,4 +215,4 @@ pub trait CrateStore: std::fmt::Debug {
     fn import_source_files(&self, sess: &Session, cnum: CrateNum);
 }
 
-pub type CrateStoreDyn = dyn CrateStore + sync::Sync;
+pub type CrateStoreDyn = dyn CrateStore + sync::Sync + sync::Send;
diff --git a/compiler/rustc_trait_selection/src/traits/select/mod.rs b/compiler/rustc_trait_selection/src/traits/select/mod.rs
index e69bf6eca90e..75ea252a73a7 100644
--- a/compiler/rustc_trait_selection/src/traits/select/mod.rs
+++ b/compiler/rustc_trait_selection/src/traits/select/mod.rs
@@ -30,7 +30,7 @@ use rustc_errors::{Diagnostic, ErrorGuaranteed};
 use rustc_hir as hir;
 use rustc_hir::def_id::DefId;
 use rustc_infer::infer::LateBoundRegionConversionTime;
-use rustc_middle::dep_graph::{DepKind, DepNodeIndex};
+use rustc_middle::dep_graph::{dep_kind, DepNodeIndex};
 use rustc_middle::mir::interpret::ErrorHandled;
 use rustc_middle::thir::abstract_const::NotConstEvaluatable;
 use rustc_middle::ty::fast_reject::{self, TreatParams};
@@ -1128,7 +1128,7 @@ impl<'cx, 'tcx> SelectionContext<'cx, 'tcx> {
         OP: FnOnce(&mut Self) -> R,
     {
         let (result, dep_node) =
-            self.tcx().dep_graph.with_anon_task(self.tcx(), DepKind::TraitSelect, || op(self));
+            self.tcx().dep_graph.with_anon_task(self.tcx(), dep_kind::TraitSelect, || op(self));
         self.tcx().dep_graph.read_index(dep_node);
         (result, dep_node)
     }