Auto merge of rust-lang#127040 - blyxyas:xxh3-hasher, r=<try>

[Perf Experiment] Change FxHash to GxHash

I've done some experiments comparing hashing libraries that can be found in crates.io, and it while FxHash is faster than the default hasher, Gxhash is still about 10 times faster than fxhash.

This is just with the default implementation, the library permits customization using CPU-specific optimizations. But we can ask the maintainers for that if we were to verify that the optimization is big enough.

Author: bors

Cargo.lock

@@ -1382,7 +1382,7 @@ dependencies = [
  "fluent-syntax",
  "intl-memoizer",
  "intl_pluralrules",
- "rustc-hash",
+ "rustc-hash 1.1.0",
  "self_cell 0.10.3",
  "smallvec",
  "unic-langid",
@@ -1646,6 +1646,15 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "gxhash"
+version = "3.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a197c9b654827513cf53842c5c6d3da2b4b35a785f8e0eff78bdf8e445aba1bb"
+dependencies = [
+ "rustversion",
+]
+
 [[package]]
 name = "handlebars"
 version = "5.1.2"
@@ -2074,7 +2083,7 @@ dependencies = [
  "anyhow",
  "clap",
  "fs-err",
- "rustc-hash",
+ "rustc-hash 1.1.0",
  "rustdoc-json-types",
  "serde",
  "serde_json",
@@ -2406,7 +2415,7 @@ dependencies = [
  "memmap2",
  "parking_lot",
  "perf-event-open-sys",
- "rustc-hash",
+ "rustc-hash 1.1.0",
  "smallvec",
 ]
 
@@ -3016,7 +3025,7 @@ checksum = "c4e8e505342045d397d0b6674dcb82d6faf5cf40484d30eeb88fc82ef14e903f"
 dependencies = [
  "datafrog",
  "log",
- "rustc-hash",
+ "rustc-hash 1.1.0",
 ]
 
 [[package]]
@@ -3448,6 +3457,12 @@ version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "08d43f7aa6b08d49f382cde6a7982047c3426db949b1424bc4b7ec9ae12c6ce2"
 
+[[package]]
+name = "rustc-hash"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "583034fd73374156e66797ed8e5b0d5690409c9226b22d87cb7f19821c05d152"
+
 [[package]]
 name = "rustc-main"
 version = "0.0.0"
@@ -3817,14 +3832,15 @@ dependencies = [
  "either",
  "elsa",
  "ena",
+ "gxhash",
  "indexmap",
  "jobserver",
  "libc",
  "measureme",
  "memmap2",
  "parking_lot",
  "portable-atomic",
- "rustc-hash",
+ "rustc-hash 2.0.0",
  "rustc-rayon",
  "rustc_arena",
  "rustc_graphviz",
@@ -4512,7 +4528,7 @@ dependencies = [
 name = "rustc_pattern_analysis"
 version = "0.0.0"
 dependencies = [
- "rustc-hash",
+ "rustc-hash 1.1.0",
  "rustc_apfloat",
  "rustc_arena",
  "rustc_data_structures",
@@ -4904,7 +4920,7 @@ name = "rustdoc-json-types"
 version = "0.1.0"
 dependencies = [
  "bincode",
- "rustc-hash",
+ "rustc-hash 1.1.0",
  "serde",
  "serde_json",
 ]
@@ -5632,7 +5648,7 @@ dependencies = [
  "ignore",
  "miropt-test-tools",
  "regex",
- "rustc-hash",
+ "rustc-hash 1.1.0",
  "semver",
  "similar",
  "termcolor",
@@ -5886,7 +5902,7 @@ version = "0.5.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "deb68604048ff8fa93347f02441e4487594adc20bb8a084f9e564d2b827a0a9f"
 dependencies = [
- "rustc-hash",
+ "rustc-hash 1.1.0",
 ]
 
 [[package]]

blah.txt

@@ -0,0 +1 @@
+hola1

compiler/rustc_ast/src/format.rs

@@ -1,6 +1,6 @@
 use crate::ptr::P;
 use crate::Expr;
-use rustc_data_structures::fx::FxHashMap;
+use rustc_data_structures::gx::GxHashMap;
 use rustc_macros::{Decodable, Encodable};
 use rustc_span::symbol::{Ident, Symbol};
 use rustc_span::Span;
@@ -65,14 +65,14 @@ pub struct FormatArguments {
     arguments: Vec<FormatArgument>,
     num_unnamed_args: usize,
     num_explicit_args: usize,
-    names: FxHashMap<Symbol, usize>,
+    names: GxHashMap<Symbol, usize>,
 }
 
 impl FormatArguments {
     pub fn new() -> Self {
         Self {
             arguments: Vec::new(),
-            names: FxHashMap::default(),
+            names: GxHashMap::default(),
             num_unnamed_args: 0,
             num_explicit_args: 0,
         }

compiler/rustc_ast_lowering/src/asm.rs

@@ -11,7 +11,7 @@ use super::LoweringContext;
 
 use rustc_ast::ptr::P;
 use rustc_ast::*;
-use rustc_data_structures::fx::{FxHashMap, FxHashSet, FxIndexMap};
+use rustc_data_structures::gx::{GxHashMap, GxHashSet, GxIndexMap};
 use rustc_hir as hir;
 use rustc_hir::def::{DefKind, Res};
 use rustc_session::parse::feature_err;
@@ -68,7 +68,7 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
                 .emit();
         }
 
-        let mut clobber_abis = FxIndexMap::default();
+        let mut clobber_abis = GxIndexMap::default();
         if let Some(asm_arch) = asm_arch {
             for (abi_name, abi_span) in &asm.clobber_abis {
                 match asm::InlineAsmClobberAbi::parse(asm_arch, &self.tcx.sess.target, *abi_name) {
@@ -318,8 +318,8 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
             }
         }
 
-        let mut used_input_regs = FxHashMap::default();
-        let mut used_output_regs = FxHashMap::default();
+        let mut used_input_regs = GxHashMap::default();
+        let mut used_output_regs = GxHashMap::default();
 
         for (idx, &(ref op, op_sp)) in operands.iter().enumerate() {
             if let Some(reg) = op.reg() {
@@ -362,7 +362,7 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
                     // Flag to output the error only once per operand
                     let mut skip = false;
 
-                    let mut check = |used_regs: &mut FxHashMap<asm::InlineAsmReg, usize>,
+                    let mut check = |used_regs: &mut GxHashMap<asm::InlineAsmReg, usize>,
                                      input,
                                      r: asm::InlineAsmReg| {
                         match used_regs.entry(r) {
@@ -436,7 +436,7 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
 
         // If a clobber_abi is specified, add the necessary clobbers to the
         // operands list.
-        let mut clobbered = FxHashSet::default();
+        let mut clobbered = GxHashSet::default();
         for (abi, (_, abi_span)) in clobber_abis {
             for &clobber in abi.clobbered_regs() {
                 // Don't emit a clobber for a register already clobbered

compiler/rustc_ast_lowering/src/format.rs

@@ -3,7 +3,7 @@ use core::ops::ControlFlow;
 use rustc_ast as ast;
 use rustc_ast::visit::Visitor;
 use rustc_ast::*;
-use rustc_data_structures::fx::FxIndexMap;
+use rustc_data_structures::gx::GxIndexMap;
 use rustc_hir as hir;
 use rustc_span::{
     sym,
@@ -282,7 +282,7 @@ fn make_count<'hir>(
     ctx: &mut LoweringContext<'_, 'hir>,
     sp: Span,
     count: &Option<FormatCount>,
-    argmap: &mut FxIndexMap<(usize, ArgumentType), Option<Span>>,
+    argmap: &mut GxIndexMap<(usize, ArgumentType), Option<Span>>,
 ) -> hir::Expr<'hir> {
     match count {
         Some(FormatCount::Literal(n)) => {
@@ -335,7 +335,7 @@ fn make_format_spec<'hir>(
     ctx: &mut LoweringContext<'_, 'hir>,
     sp: Span,
     placeholder: &FormatPlaceholder,
-    argmap: &mut FxIndexMap<(usize, ArgumentType), Option<Span>>,
+    argmap: &mut GxIndexMap<(usize, ArgumentType), Option<Span>>,
 ) -> hir::Expr<'hir> {
     let position = match placeholder.argument.index {
         Ok(arg_index) => {
@@ -432,7 +432,7 @@ fn expand_format_args<'hir>(
 
     // Create a list of all _unique_ (argument, format trait) combinations.
     // E.g. "{0} {0:x} {0} {1}" -> [(0, Display), (0, LowerHex), (1, Display)]
-    let mut argmap = FxIndexMap::default();
+    let mut argmap = GxIndexMap::default();
     for piece in &fmt.template {
         let FormatArgsPiece::Placeholder(placeholder) = piece else { continue };
         if placeholder.format_options != Default::default() {

compiler/rustc_ast_lowering/src/lib.rs

@@ -46,7 +46,7 @@ use rustc_ast::{self as ast, *};
 use rustc_ast_pretty::pprust;
 use rustc_data_structures::captures::Captures;
 use rustc_data_structures::fingerprint::Fingerprint;
-use rustc_data_structures::fx::FxIndexSet;
+use rustc_data_structures::gx::GxIndexSet;
 use rustc_data_structures::sorted_map::SortedMap;
 use rustc_data_structures::stable_hasher::{HashStable, StableHasher};
 use rustc_data_structures::sync::Lrc;
@@ -1630,7 +1630,7 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
         opaque_ty_node_id: NodeId,
         origin: hir::OpaqueTyOrigin,
         in_trait: bool,
-        captured_lifetimes_to_duplicate: FxIndexSet<Lifetime>,
+        captured_lifetimes_to_duplicate: GxIndexSet<Lifetime>,
         span: Span,
         opaque_ty_span: Span,
         lower_item_bounds: impl FnOnce(&mut Self) -> &'hir [hir::GenericBound<'hir>],

compiler/rustc_ast_lowering/src/lifetime_collector.rs

@@ -1,7 +1,7 @@
 use super::ResolverAstLoweringExt;
 use rustc_ast::visit::{self, BoundKind, LifetimeCtxt, Visitor};
 use rustc_ast::{GenericBounds, Lifetime, NodeId, PathSegment, PolyTraitRef, Ty, TyKind};
-use rustc_data_structures::fx::FxIndexSet;
+use rustc_data_structures::gx::GxIndexSet;
 use rustc_hir::def::{DefKind, LifetimeRes, Res};
 use rustc_middle::span_bug;
 use rustc_middle::ty::ResolverAstLowering;
@@ -11,12 +11,12 @@ use rustc_span::Span;
 struct LifetimeCollectVisitor<'ast> {
     resolver: &'ast ResolverAstLowering,
     current_binders: Vec<NodeId>,
-    collected_lifetimes: FxIndexSet<Lifetime>,
+    collected_lifetimes: GxIndexSet<Lifetime>,
 }
 
 impl<'ast> LifetimeCollectVisitor<'ast> {
     fn new(resolver: &'ast ResolverAstLowering) -> Self {
-        Self { resolver, current_binders: Vec::new(), collected_lifetimes: FxIndexSet::default() }
+        Self { resolver, current_binders: Vec::new(), collected_lifetimes: GxIndexSet::default() }
     }
 
     fn record_lifetime_use(&mut self, lifetime: Lifetime) {
@@ -108,7 +108,7 @@ impl<'ast> Visitor<'ast> for LifetimeCollectVisitor<'ast> {
 pub(crate) fn lifetimes_in_bounds(
     resolver: &ResolverAstLowering,
     bounds: &GenericBounds,
-) -> FxIndexSet<Lifetime> {
+) -> GxIndexSet<Lifetime> {
     let mut visitor = LifetimeCollectVisitor::new(resolver);
     for bound in bounds {
         visitor.visit_param_bound(bound, BoundKind::Bound);

compiler/rustc_ast_passes/src/ast_validation.rs

@@ -11,7 +11,7 @@ use rustc_ast::ptr::P;
 use rustc_ast::visit::{walk_list, AssocCtxt, BoundKind, FnCtxt, FnKind, Visitor};
 use rustc_ast::*;
 use rustc_ast_pretty::pprust::{self, State};
-use rustc_data_structures::fx::FxIndexMap;
+use rustc_data_structures::gx::GxIndexMap;
 use rustc_errors::DiagCtxtHandle;
 use rustc_feature::Features;
 use rustc_parse::validate_attr;
@@ -828,7 +828,7 @@ impl<'a> AstValidator<'a> {
 /// which is lifetimes, then types and then consts. (`<'a, T, const N: usize>`)
 fn validate_generic_param_order(dcx: DiagCtxtHandle<'_>, generics: &[GenericParam], span: Span) {
     let mut max_param: Option<ParamKindOrd> = None;
-    let mut out_of_order = FxIndexMap::default();
+    let mut out_of_order = GxIndexMap::default();
     let mut param_idents = Vec::with_capacity(generics.len());
 
     for (idx, param) in generics.iter().enumerate() {

compiler/rustc_borrowck/src/borrow_set.rs

@@ -1,7 +1,7 @@
 use crate::path_utils::allow_two_phase_borrow;
 use crate::place_ext::PlaceExt;
 use crate::BorrowIndex;
-use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
+use rustc_data_structures::gx::{GxIndexMap, GxIndexSet};
 use rustc_index::bit_set::BitSet;
 use rustc_middle::mir::traversal;
 use rustc_middle::mir::visit::{MutatingUseContext, NonUseContext, PlaceContext, Visitor};
@@ -18,16 +18,16 @@ pub struct BorrowSet<'tcx> {
     /// by the `Location` of the assignment statement in which it
     /// appears on the right hand side. Thus the location is the map
     /// key, and its position in the map corresponds to `BorrowIndex`.
-    pub location_map: FxIndexMap<Location, BorrowData<'tcx>>,
+    pub location_map: GxIndexMap<Location, BorrowData<'tcx>>,
 
     /// Locations which activate borrows.
     /// NOTE: a given location may activate more than one borrow in the future
     /// when more general two-phase borrow support is introduced, but for now we
     /// only need to store one borrow index.
-    pub activation_map: FxIndexMap<Location, Vec<BorrowIndex>>,
+    pub activation_map: GxIndexMap<Location, Vec<BorrowIndex>>,
 
     /// Map from local to all the borrows on that local.
-    pub local_map: FxIndexMap<mir::Local, FxIndexSet<BorrowIndex>>,
+    pub local_map: GxIndexMap<mir::Local, GxIndexSet<BorrowIndex>>,
 
     pub locals_state_at_exit: LocalsStateAtExit,
 }
@@ -174,9 +174,9 @@ impl<'tcx> BorrowSet<'tcx> {
 struct GatherBorrows<'a, 'tcx> {
     tcx: TyCtxt<'tcx>,
     body: &'a Body<'tcx>,
-    location_map: FxIndexMap<Location, BorrowData<'tcx>>,
-    activation_map: FxIndexMap<Location, Vec<BorrowIndex>>,
-    local_map: FxIndexMap<mir::Local, FxIndexSet<BorrowIndex>>,
+    location_map: GxIndexMap<Location, BorrowData<'tcx>>,
+    activation_map: GxIndexMap<Location, Vec<BorrowIndex>>,
+    local_map: GxIndexMap<mir::Local, GxIndexSet<BorrowIndex>>,
 
     /// When we encounter a 2-phase borrow statement, it will always
     /// be assigning into a temporary TEMP:
@@ -186,7 +186,7 @@ struct GatherBorrows<'a, 'tcx> {
     /// We add TEMP into this map with `b`, where `b` is the index of
     /// the borrow. When we find a later use of this activation, we
     /// remove from the map (and add to the "tombstone" set below).
-    pending_activations: FxIndexMap<mir::Local, BorrowIndex>,
+    pending_activations: GxIndexMap<mir::Local, BorrowIndex>,
 
     locals_state_at_exit: LocalsStateAtExit,
 }