rustc_span: Symbol perf experiments

Author: petrochenkov

Cargo.lock

@@ -3910,6 +3910,7 @@ dependencies = [
 name = "rustc_macros"
 version = "0.1.0"
 dependencies = [
+ "phf_codegen",
  "proc-macro2",
  "quote",
  "syn",
@@ -4197,6 +4198,7 @@ version = "0.0.0"
 dependencies = [
  "cfg-if 0.1.10",
  "md-5",
+ "phf",
  "rustc_arena",
  "rustc_data_structures",
  "rustc_index",

compiler/rustc_macros/Cargo.toml

@@ -10,5 +10,6 @@ proc-macro = true
 [dependencies]
 synstructure = "0.12.1"
 syn = { version = "1", features = ["full"] }
+phf_codegen = "0.8"
 proc-macro2 = "1"
 quote = "1"

compiler/rustc_macros/src/symbols.rs

@@ -127,6 +127,7 @@ fn symbols_with_errors(input: TokenStream) -> (TokenStream, Vec<syn::Error>) {
     let mut keyword_stream = quote! {};
     let mut symbols_stream = quote! {};
     let mut prefill_stream = quote! {};
+    let mut symbol_strings = Vec::new();
     let mut counter = 0u32;
     let mut keys =
         HashMap::<String, Span>::with_capacity(input.keywords.len() + input.symbols.len() + 10);
@@ -163,6 +164,7 @@ fn symbols_with_errors(input: TokenStream) -> (TokenStream, Vec<syn::Error>) {
         keyword_stream.extend(quote! {
             pub const #name: Symbol = Symbol::new(#counter);
         });
+        symbol_strings.push(value_string);
         counter += 1;
     }
 
@@ -182,6 +184,7 @@ fn symbols_with_errors(input: TokenStream) -> (TokenStream, Vec<syn::Error>) {
         symbols_stream.extend(quote! {
             pub const #name: Symbol = Symbol::new(#counter);
         });
+        symbol_strings.push(value);
         counter += 1;
     }
 
@@ -194,9 +197,19 @@ fn symbols_with_errors(input: TokenStream) -> (TokenStream, Vec<syn::Error>) {
         prefill_stream.extend(quote! {
             #n,
         });
+        symbol_strings.push(n);
     }
     let _ = counter; // for future use
 
+    // Build the PHF map. This translates from strings to Symbol values.
+    let mut phf_map = phf_codegen::Map::<&str>::new();
+    for (symbol_index, symbol) in symbol_strings.iter().enumerate() {
+        phf_map.entry(symbol, format!("Symbol::new({})", symbol_index as u32).as_str());
+    }
+    let phf_map_built = phf_map.build();
+    let phf_map_text = phf_map_built.to_string();
+    let phf_map_expr = syn::parse_str::<syn::Expr>(&phf_map_text).unwrap();
+
     let output = quote! {
         const SYMBOL_DIGITS_BASE: u32 = #digits_base;
 
@@ -222,6 +235,8 @@ fn symbols_with_errors(input: TokenStream) -> (TokenStream, Vec<syn::Error>) {
                 ])
             }
         }
+
+        static STATIC_SYMBOLS_PHF: ::phf::Map<&'static str, Symbol> = #phf_map_expr;
     };
 
     (output, errors.list)

compiler/rustc_span/Cargo.toml

@@ -20,3 +20,4 @@ tracing = "0.1"
 sha-1 = "0.9"
 sha2 = "0.9"
 md-5 = "0.9"
+phf = "0.8"

compiler/rustc_span/src/symbol.rs

@@ -1453,7 +1453,11 @@ impl Symbol {
 
     /// Maps a string to its interned representation.
     pub fn intern(string: &str) -> Self {
-        with_interner(|interner| interner.intern(string))
+        if let Some(symbol) = STATIC_SYMBOLS_PHF.get(string) {
+            *symbol
+        } else {
+            with_interner(|interner| interner.intern(string))
+        }
     }
 
     /// Convert to a `SymbolStr`. This is a slowish operation because it

src/tools/tidy/src/deps.rs

@@ -132,6 +132,10 @@ const PERMITTED_DEPENDENCIES: &[&str] = &[
     "parking_lot",
     "parking_lot_core",
     "pathdiff",
+    "phf",
+    "phf_codegen",
+    "phf_generator",
+    "phf_shared",
     "pkg-config",
     "polonius-engine",
     "ppv-lite86",
@@ -164,6 +168,7 @@ const PERMITTED_DEPENDENCIES: &[&str] = &[
     "sha-1",
     "sha2",
     "smallvec",
+    "siphasher",
     "snap",
     "stable_deref_trait",
     "stacker",