Slight perf benefit from using swiss tables

absl::flat_hash_{set,map} are called "swiss tables", and they provide a
small perf win for us.

Author: Quincunx271

extras/analyze/CMakeLists.txt

@@ -3,6 +3,7 @@ cmake_minimum_required(VERSION 3.20.3)
 project(EventAnalyze)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 include(FetchContent)
 
@@ -20,6 +21,14 @@ FetchContent_Declare(
 )
 FetchContent_MakeAvailable(pybind11)
 
+FetchContent_Declare(
+  abseil
+  GIT_REPOSITORY https://github.com/abseil/abseil-cpp.git
+  GIT_TAG        f39e6ad4753e06d4a0d6a9bf6310478757479984
+)
+set(BUILD_TESTING OFF)
+FetchContent_MakeAvailable(abseil)
+
 FetchContent_Declare(
   FindTBB
   GIT_REPOSITORY https://github.com/justusc/FindTBB.git
@@ -45,4 +54,4 @@ file(GLOB_RECURSE sources CONFIGURE_DEPENDS "src/*.cpp")
 pybind11_add_module(eventanalyze ${sources})
 target_include_directories(eventanalyze PUBLIC include)
 target_compile_features(eventanalyze PUBLIC cxx_std_20)
-target_link_libraries(eventanalyze PRIVATE mio::mio)
+target_link_libraries(eventanalyze PRIVATE mio::mio absl::base absl::flat_hash_map)

extras/analyze/include/types.hpp

@@ -14,6 +14,8 @@
 
 #include "py.hpp"
 
+#include <absl/container/flat_hash_map.h>
+#include <absl/container/flat_hash_set.h>
 #include <mio/mmap.hpp>
 
 namespace ev {
@@ -91,7 +93,7 @@ struct EventIdEq {
 };
 
 using BlockEventMap =
-    std::unordered_set<std::vector<Event>, EventIdHash, EventIdEq>;
+    absl::flat_hash_set<std::vector<Event>, EventIdHash, EventIdEq>;
 
 struct Logs;
 struct Benchmark;

extras/analyze/src/parse.cpp

@@ -10,8 +10,6 @@
 #include <ranges>
 #include <string>
 #include <string_view>
-#include <unordered_map>
-#include <unordered_set>
 #include <utility>
 #include <vector>
 
@@ -65,9 +63,9 @@ static EventSchema parseEventSchema(
 // Schemas are globally loaded.
 // This static/thread_local dance is to make it appropriately thread safe but
 // still fast.
-static std::unordered_set<EventSchema, EventIdHash, EventIdEq> MasterSchemas;
+static absl::flat_hash_set<EventSchema, EventIdHash, EventIdEq> MasterSchemas;
 static std::mutex MasterSchemaMutex;
-thread_local std::unordered_set<EventSchema, EventIdHash, EventIdEq> Schemas;
+thread_local absl::flat_hash_set<EventSchema, EventIdHash, EventIdEq> Schemas;
 
 static void updateSchemaStructures(EventId Id, EventSchema schema) {
   std::scoped_lock Lock(MasterSchemaMutex);
@@ -167,7 +165,7 @@ static const std::boyer_moore_horspool_searcher
     EventTagSearcher(EventTag.begin(), EventTag.end());
 
 static BlockEventMap parseEvents(const std::string_view BlockLog) {
-  std::unordered_map<EventId, std::vector<Event>, EventIdHash, EventIdEq>
+  absl::flat_hash_map<EventId, std::vector<Event>, EventIdHash, EventIdEq>
       Result;
 
   const auto E = BlockLog.end();
@@ -205,7 +203,9 @@ static Block parseBlock(ev::Benchmark *Bench, const std::string_view BlockLog) {
       .RawLog = BlockLog,
       .UniqueId = std::move(UniqueId),
       .Bench = Bench,
-      .File = "", // TODO: Get this information too
+      // Extracting file info costs quite a bit of time, and we never use it
+      // anyway.
+      .File = "",
   };
 }