Auto merge of #67397 - michaelwoerister:query-keys-in-self-profiling,…

… r=wesleywiser

self-profiling: Support recording query keys

This PR makes self-profiling able to record query keys. The implementation is not as efficient as it could be yet (all query keys except for `DefId`s cause string data to be duplicated) and the rendered strings could be nicer too. But the implementation is functional and introduces the basic framework for emitting per-query-invocation event data.

I tried to add proper documentation on how everything works. Let me know if more documentation is needed.

r? @wesleywiser

@Mark-Simulacrum, heads up: This updates `measureme` to 0.7.0 which means that `summarize` on perf.rlo needs to be update accordingly once this is merged.

Cargo.lock

@@ -1995,9 +1995,9 @@ dependencies = [
 
 [[package]]
 name = "measureme"
-version = "0.5.0"
+version = "0.7.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c420bbc064623934620b5ab2dc0cf96451b34163329e82f95e7fa1b7b99a6ac8"
+checksum = "fef709d3257013bba7cff14fc504e07e80631d3fe0f6d38ce63b8f6510ccb932"
 dependencies = [
  "byteorder",
  "memmap",
@@ -3079,6 +3079,7 @@ dependencies = [
  "graphviz",
  "jobserver",
  "log",
+ "measureme",
  "parking_lot",
  "polonius-engine",
  "rustc-rayon",

src/librustc/Cargo.toml

@@ -36,5 +36,6 @@ parking_lot = "0.9"
 byteorder = { version = "1.3" }
 chalk-engine = { version = "0.9.0", default-features=false }
 smallvec = { version = "1.0", features = ["union", "may_dangle"] }
+measureme = "0.7.1"
 rustc_error_codes = { path = "../librustc_error_codes" }
 rustc_session = { path = "../librustc_session" }

src/librustc/dep_graph/graph.rs

@@ -2,6 +2,7 @@ use crate::ty::{self, TyCtxt};
 use errors::Diagnostic;
 use parking_lot::{Condvar, Mutex};
 use rustc_data_structures::fx::{FxHashMap, FxHashSet};
+use rustc_data_structures::profiling::QueryInvocationId;
 use rustc_data_structures::sharded::{self, Sharded};
 use rustc_data_structures::stable_hasher::{HashStable, StableHasher};
 use rustc_data_structures::sync::{AtomicU32, AtomicU64, Lock, Lrc, Ordering};
@@ -11,7 +12,7 @@ use std::collections::hash_map::Entry;
 use std::env;
 use std::hash::Hash;
 use std::mem;
-use std::sync::atomic::Ordering::SeqCst;
+use std::sync::atomic::Ordering::Relaxed;
 
 use crate::ich::{Fingerprint, StableHashingContext, StableHashingContextProvider};
 
@@ -25,6 +26,12 @@ use super::serialized::{SerializedDepGraph, SerializedDepNodeIndex};
 #[derive(Clone)]
 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
+    /// each task has a `DepNodeIndex` that uniquely identifies it. This unique
+    /// ID is used for self-profiling.
+    virtual_dep_node_index: Lrc<AtomicU32>,
 }
 
 rustc_index::newtype_index! {
@@ -35,6 +42,13 @@ impl DepNodeIndex {
     pub const INVALID: DepNodeIndex = DepNodeIndex::MAX;
 }
 
+impl std::convert::From<DepNodeIndex> for QueryInvocationId {
+    #[inline]
+    fn from(dep_node_index: DepNodeIndex) -> Self {
+        QueryInvocationId(dep_node_index.as_u32())
+    }
+}
+
 #[derive(PartialEq)]
 pub enum DepNodeColor {
     Red,
@@ -105,11 +119,12 @@ impl DepGraph {
                 previous: prev_graph,
                 colors: DepNodeColorMap::new(prev_graph_node_count),
             })),
+            virtual_dep_node_index: Lrc::new(AtomicU32::new(0)),
         }
     }
 
     pub fn new_disabled() -> DepGraph {
-        DepGraph { data: None }
+        DepGraph { data: None, virtual_dep_node_index: Lrc::new(AtomicU32::new(0)) }
     }
 
     /// Returns `true` if we are actually building the full dep-graph, and `false` otherwise.
@@ -322,7 +337,7 @@ impl DepGraph {
 
             (result, dep_node_index)
         } else {
-            (task(cx, arg), DepNodeIndex::INVALID)
+            (task(cx, arg), self.next_virtual_depnode_index())
         }
     }
 
@@ -352,7 +367,7 @@ impl DepGraph {
             let dep_node_index = data.current.complete_anon_task(dep_kind, task_deps);
             (result, dep_node_index)
         } else {
-            (op(), DepNodeIndex::INVALID)
+            (op(), self.next_virtual_depnode_index())
         }
     }
 
@@ -478,8 +493,8 @@ impl DepGraph {
             let current_dep_graph = &self.data.as_ref().unwrap().current;
 
             Some((
-                current_dep_graph.total_read_count.load(SeqCst),
-                current_dep_graph.total_duplicate_read_count.load(SeqCst),
+                current_dep_graph.total_read_count.load(Relaxed),
+                current_dep_graph.total_duplicate_read_count.load(Relaxed),
             ))
         } else {
             None
@@ -877,6 +892,11 @@ impl DepGraph {
             }
         }
     }
+
+    fn next_virtual_depnode_index(&self) -> DepNodeIndex {
+        let index = self.virtual_dep_node_index.fetch_add(1, Relaxed);
+        DepNodeIndex::from_u32(index)
+    }
 }
 
 /// A "work product" is an intermediate result that we save into the
@@ -1087,7 +1107,7 @@ impl DepGraphData {
             if let Some(task_deps) = icx.task_deps {
                 let mut task_deps = task_deps.lock();
                 if cfg!(debug_assertions) {
-                    self.current.total_read_count.fetch_add(1, SeqCst);
+                    self.current.total_read_count.fetch_add(1, Relaxed);
                 }
                 if task_deps.read_set.insert(source) {
                     task_deps.reads.push(source);
@@ -1105,7 +1125,7 @@ impl DepGraphData {
                         }
                     }
                 } else if cfg!(debug_assertions) {
-                    self.current.total_duplicate_read_count.fetch_add(1, SeqCst);
+                    self.current.total_duplicate_read_count.fetch_add(1, Relaxed);
                 }
             }
         })

src/librustc/ty/query/config.rs

@@ -2,8 +2,7 @@ use crate::dep_graph::SerializedDepNodeIndex;
 use crate::dep_graph::{DepKind, DepNode};
 use crate::ty::query::plumbing::CycleError;
 use crate::ty::query::queries;
-use crate::ty::query::QueryCache;
-use crate::ty::query::{Query, QueryName};
+use crate::ty::query::{Query, QueryCache};
 use crate::ty::TyCtxt;
 use rustc_data_structures::profiling::ProfileCategory;
 use rustc_hir::def_id::{CrateNum, DefId};
@@ -20,7 +19,7 @@ use std::hash::Hash;
 // FIXME(eddyb) false positive, the lifetime parameter is used for `Key`/`Value`.
 #[allow(unused_lifetimes)]
 pub trait QueryConfig<'tcx> {
-    const NAME: QueryName;
+    const NAME: &'static str;
     const CATEGORY: ProfileCategory;
 
     type Key: Eq + Hash + Clone + Debug;

src/librustc/ty/query/mod.rs

@@ -81,6 +81,9 @@ pub(crate) use self::config::QueryDescription;
 mod on_disk_cache;
 pub use self::on_disk_cache::OnDiskCache;
 
+mod profiling_support;
+pub use self::profiling_support::{IntoSelfProfilingString, QueryKeyStringBuilder};
+
 // Each of these queries corresponds to a function pointer field in the
 // `Providers` struct for requesting a value of that type, and a method
 // on `tcx: TyCtxt` (and `tcx.at(span)`) for doing that request in a way