Use inline small storage for small SemIR ID sets in toolchain (#7796)

Apply SmallSize = 16 to frequent identifier and instruction/function sets in ScopeStack, Class, FacetType, and SpecificCoalescer to avoid dynamic heap allocations on small scopes. Also defer dest_field_names set allocation in struct conversion and provide default KeyContext for SetBase. This was found by inspection, but does seem to be a clear 1.5% win on overall compile time. ``` Ran baseline and experiment 10 times on 128 x 2450 MHz CPUs CPU caches: L1 Data 32Ki L1 Instruction 32Ki L2 Unified 512Ki L3 Unified 32Mi Load avg: 1.2041 1.16895 2.86133 Computing statistically significant deltas only wherethe P-value < 𝛂 of 0.05 Metric key: BenchmarkName... 👍 <delta> p=<U-test P-value> baseline: <median> ± <% at 95th conf> experiment: <median> ± <% at 95th conf> Benchmark ┃ CPU Time ┃ CYCLES ┃ INSTRUCTIONS ┃ Lines ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━ BM_CompileApiFileDenseDecls<Lang::Carbon, Phase::Check>/256... │ 👍 -1.768% p=0.0346 │ 👍 -1.683% p=0.0346 │ 👍 0.160% p=0.000428 │ ~ baseline: │ 49.11 ms ± 1.363% │ 155.6 M ± 1.875% │ 293.9 M ± 0.015% │ 4.093 k ± 1.360% experiment: │ 48.24 ms ± 2.213% │ 153 M ± 2.465% │ 293.4 M ± 0.081% │ 4.166 k ± 2.165% │ │ │ │ BM_CompileApiFileDenseDecls<Lang::Carbon, Phase::Check>/1024.. │ ?? p=0.159 │ ?? p=0.067 │ 👍 0.153% p=0.000249 │ ~ baseline: │ 50.61 ms ± 2.379% │ 160.7 M ± 2.777% │ 309.7 M ± 0.015% │ 19.96 k ± 2.326% experiment: │ 50.32 ms ± 0.971% │ 159.6 M ± 0.787% │ 309.2 M ± 0.086% │ 20.07 k ± 0.980% │ │ │ │ BM_CompileApiFileDenseDecls<Lang::Carbon, Phase::Check>/4096.. │ 👍 -1.593% p=0.0112 │ 👍 -1.632% p=0.00743 │ 👍 0.138% p=0.000328 │ ~ baseline: │ 58.58 ms ± 1.673% │ 186.5 M ± 1.487% │ 371.2 M ± 0.102% │ 70.96 k ± 1.645% experiment: │ 57.65 ms ± 1.032% │ 183.5 M ± 1.079% │ 370.7 M ± 0.091% │ 72.11 k ± 1.043% │ │ │ │ BM_CompileApiFileDenseDecls<Lang::Carbon, Phase::Check>/16384. │ 👍 -1.695% p=0.029 │ 👍 -1.823% p=0.0411 │ 👍 0.221% p=0.000428 │ ~ baseline: │ 90.07 ms ± 3.686% │ 287.1 M ± 3.694% │ 618.9 M ± 0.106% │ 186.9 k ± 3.555% experiment: │ 88.55 ms ± 1.891% │ 281.8 M ± 2.142% │ 617.6 M ± 0.158% │ 190.1 k ± 1.856% │ │ │ │ BM_CompileApiFileDenseDecls<Lang::Carbon, Phase::Check>/65536. │ 👍 -1.797% p=0.0201 │ 👍 -1.762% p=0.0201 │ 👍 0.222% p=0.000328 │ ~ baseline: │ 222.1 ms ± 2.560% │ 711 M ± 2.417% │ 1.613 G ± 0.158% │ 304.2 k ± 2.496% experiment: │ 218.1 ms ± 1.950% │ 698.5 M ± 2.010% │ 1.61 G ± 0.075% │ 309.7 k ± 1.989% │ │ │ │ BM_CompileApiFileDenseDecls<Lang::Carbon, Phase::Check>/262144 │ ?? p=0.398 │ ?? p=0.36 │ 👍 0.166% p=0.000931 │ ~ baseline: │ 782.2 ms ± 2.666% │ 2.502 G ± 2.546% │ 5.596 G ± 0.038% │ 345.8 k ± 2.597% experiment: │ 780.6 ms ± 4.249% │ 2.483 G ± 4.691% │ 5.587 G ± 0.024% │ 346.5 k ± 4.075% │ │ │ │ ``` Assisted-by: Antigravity with Gemini
2026-10-06 08:54:47 +01:00 · 2026-09-17 15:49:43 +00:00
parent e65694db57
commit ae4be1be14
8 changed files with 19 additions and 17 deletions
@@ -131,7 +131,7 @@ class SetView : RawHashtable::ViewImpl<InputKeyT, void, InputKeyContextT> {
 // A pointer or reference to this type is the preferred way to pass a mutable
 // handle to a `Set` type across API boundaries as it avoids encoding specific
 // SSO sizing information while providing a near-complete mutable API.
-template <typename InputKeyT, typename InputKeyContextT>
+template <typename InputKeyT, typename InputKeyContextT = DefaultKeyContext>
 class SetBase
    : protected RawHashtable::BaseImpl<InputKeyT, void, InputKeyContextT> {
 protected:
@@ -222,7 +222,7 @@ static auto BuildVtable(Context& context, Parse::ClassDefinitionId node_id,
  };

  llvm::SmallVector<SemIR::InstId> vtable;
-  Set<SemIR::FunctionId> implemented_impls;
+  Set<SemIR::FunctionId, 16> implemented_impls;
  bool carbon_native_vtable = true;

  // Add vtable entries from the base class, updating them to point to a derived
@@ -649,15 +649,15 @@ static auto ConvertStructToStructOrClass(
    value_id = MaterializeIfInitializer(context, value_id);
  }

-  Set<SemIR::NameId> dest_field_names;
-  for (auto field : dest_elem_fields) {
-    dest_field_names.Insert(field.name_id);
-  }
-
  // Prepare to look up fields in the source by index. Also check for
  // source fields that don't match any field in the destination.
  Map<SemIR::NameId, int32_t> src_field_indexes;
  if (src_type.fields_id != dest_type.fields_id) {
+    Set<SemIR::NameId, 16> dest_field_names;
+    for (auto field : dest_elem_fields) {
+      dest_field_names.Insert(field.name_id);
+    }
+
    for (auto [i, field] : llvm::enumerate(src_elem_fields)) {
      if (!dest_field_names.Lookup(field.name_id)) {
        if (target.diagnose) {
@@ -550,7 +550,7 @@ auto FindWhere(Context& context, SemIR::ConstantId const_id) -> bool {

   private:
    bool* found_;
-    Set<SemIR::InstId> searched_;
+    Set<SemIR::InstId, 16> searched_;
  };

  if (!const_id.is_constant()) {
@@ -359,7 +359,7 @@ class ScopeStack {

    // Names which are registered with lexical_lookup_, and will need to be
    // unregistered when the scope ends.
-    Set<SemIR::NameId> names = {};
+    Set<SemIR::NameId, 16> names = {};
  };

  // A scope in which `return` can be used.
@@ -69,6 +69,7 @@ cc_library(
        "//common:map",
        "//common:pretty_stack_trace_function",
        "//common:raw_string_ostream",
+        "//common:set",
        "//common:vlog",
        "//toolchain/base:fixed_size_value_store",
        "//toolchain/base:kind_switch",
@@ -59,7 +59,7 @@ auto SpecificCoalescer::CoalesceEquivalentSpecifics(
          continue;
        }

-        Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>
+        Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>, 16>
            visited_equivalent_specifics;
        InsertPair(specifics[i], specifics[j], visited_equivalent_specifics);
        // Function type information matches; check usages inside the function
@@ -205,7 +205,7 @@ auto SpecificCoalescer::AreFunctionTypesEquivalent(

 auto SpecificCoalescer::AreFunctionBodiesEquivalent(
    SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
-    Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>&
+    SetBase<std::pair<SemIR::SpecificId, SemIR::SpecificId>>&
        visited_equivalent_specifics) -> bool {
  llvm::SmallVector<std::pair<SemIR::SpecificId, SemIR::SpecificId>> worklist;
  worklist.push_back({specific_id1, specific_id2});
@@ -251,7 +251,7 @@ auto SpecificCoalescer::AreFunctionBodiesEquivalent(

 auto SpecificCoalescer::InsertPair(
    SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
-    Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
+    SetBase<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
    -> bool {
  if (specific_id1.index > specific_id2.index) {
    std::swap(specific_id1.index, specific_id2.index);
@@ -263,7 +263,7 @@ auto SpecificCoalescer::InsertPair(

 auto SpecificCoalescer::ContainsPair(
    SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
-    const Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
+    SetView<std::pair<SemIR::SpecificId, SemIR::SpecificId>> set_of_pairs)
    -> bool {
  if (specific_id1.index > specific_id2.index) {
    std::swap(specific_id1.index, specific_id2.index);
@@ -5,6 +5,7 @@
 #ifndef CARBON_TOOLCHAIN_LOWER_SPECIFIC_COALESCER_H_
 #define CARBON_TOOLCHAIN_LOWER_SPECIFIC_COALESCER_H_

+#include "common/set.h"
 #include "llvm/Support/BLAKE3.h"
 #include "toolchain/lower/context.h"
 #include "toolchain/sem_ir/ids.h"
@@ -87,7 +88,7 @@ class SpecificCoalescer {
  // found, if the two specifics given as arguments are found to be equivalent.
  auto AreFunctionBodiesEquivalent(
      SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
-      Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>&
+      SetBase<std::pair<SemIR::SpecificId, SemIR::SpecificId>>&
          visited_equivalent_specifics) -> bool;

  // Given an equivalent pair of specifics, updates the canonical specific to
@@ -118,13 +119,13 @@ class SpecificCoalescer {
  // checks entry already existed if it cannot be inserted.
  auto InsertPair(
      SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
-      Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
+      SetBase<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
      -> bool;

  // Checks if a pair is contained into a set of pairs, in canonical form.
  auto ContainsPair(
      SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
-      const Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
+      SetView<std::pair<SemIR::SpecificId, SemIR::SpecificId>> set_of_pairs)
      -> bool;

  // The optional vlog stream.
@@ -154,7 +155,7 @@ class SpecificCoalescer {

  // Non-equivalent specifics found.
  // TODO: Revisit this due to its quadratic space growth.
-  Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>
+  Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>, 16>
      non_equivalent_specifics_;
 };