Apply function multi versioning higher in call stack to save branches

These are a relic of when we used forEachInRange for checkMaxBetween

Bench.cpp

@@ -258,4 +258,107 @@ void benchConflictSet() {
   }
 }
 
-int main(void) { benchConflictSet(); }
+constexpr int kKeyLenForWorstCase = 50;
+
+ConflictSet worstCaseConflictSetForRadixRangeRead(int cardinality) {
+  ConflictSet cs{0};
+
+  for (int i = 0; i < kKeyLenForWorstCase; ++i) {
+    for (int j = 0; j < cardinality; ++j) {
+      auto b = std::vector<uint8_t>(i, 0);
+      b.push_back(j);
+      auto e = std::vector<uint8_t>(i, 255);
+      e.push_back(255 - j);
+      weaselab::ConflictSet::WriteRange w[] = {{
+                                                   {b.data(), int(b.size())},
+                                                   {nullptr, 0},
+                                               },
+                                               {
+                                                   {e.data(), int(e.size())},
+                                                   {nullptr, 0},
+                                               }};
+      std::sort(std::begin(w), std::end(w),
+                [](const auto &lhs, const auto &rhs) {
+                  int cl = std::min(lhs.begin.len, rhs.begin.len);
+                  if (cl > 0) {
+                    int c = memcmp(lhs.begin.p, rhs.begin.p, cl);
+                    if (c != 0) {
+                      return c < 0;
+                    }
+                  }
+                  return lhs.begin.len < rhs.begin.len;
+                });
+      cs.addWrites(w, sizeof(w) / sizeof(w[0]), 0);
+    }
+  }
+
+  // Defeat short-circuiting on the left
+  {
+    auto k = std::vector<uint8_t>(kKeyLenForWorstCase, 0);
+    weaselab::ConflictSet::WriteRange w[] = {
+        {
+            {k.data(), int(k.size())},
+            {nullptr, 0},
+        },
+    };
+    cs.addWrites(w, sizeof(w) / sizeof(w[0]), 1);
+  }
+
+  // Defeat short-circuiting on the right
+  {
+    auto k = std::vector<uint8_t>(kKeyLenForWorstCase, 255);
+    weaselab::ConflictSet::WriteRange w[] = {
+        {
+            {k.data(), int(k.size())},
+            {nullptr, 0},
+        },
+    };
+    cs.addWrites(w, sizeof(w) / sizeof(w[0]), 1);
+  }
+
+  return cs;
+}
+
+void benchWorstCaseForRadixRangeRead() {
+  ankerl::nanobench::Bench bench;
+
+  std::unique_ptr<ConflictSet> cs[256];
+  for (int i = 0; i < 256; ++i) {
+    cs[i] =
+        std::make_unique<ConflictSet>(worstCaseConflictSetForRadixRangeRead(i));
+  }
+
+  auto begin = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 0);
+  begin.push_back(1);
+  auto end = std::vector<uint8_t>(kKeyLenForWorstCase - 1, 255);
+  end.push_back(254);
+
+  weaselab::ConflictSet::Result result;
+  weaselab::ConflictSet::ReadRange r{
+      {begin.data(), int(begin.size())}, {end.data(), int(end.size())}, 0};
+
+  bench.run("worst case for radix tree", [&]() {
+    for (int i = 0; i < 256; ++i) {
+      result = weaselab::ConflictSet::TooOld;
+      cs[i]->check(&r, &result, 1);
+      if (result != weaselab::ConflictSet::Commit) {
+        abort();
+      }
+    }
+  });
+
+  // for (int i = 0; i < 256; ++i) {
+  //   bench.run("worst case for radix tree, span " + std::to_string(i), [&]() {
+  //     result = weaselab::ConflictSet::TooOld;
+  //     cs[i]->check(&r, &result, 1);
+  //     if (result != weaselab::ConflictSet::Commit) {
+  //       abort();
+  //     }
+  //   });
+  // }
+}
+
+int main(void) {
+  benchConflictSet();
+  benchWorstCaseForRadixRangeRead();
+}

CMakeLists.txt

@@ -1,14 +1,16 @@
 cmake_minimum_required(VERSION 3.18)
 project(
   conflict-set
-  VERSION 0.0.5
+  VERSION 0.0.7
   DESCRIPTION
     "A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys."
   HOMEPAGE_URL "https://git.weaselab.dev/weaselab/conflict-set"
   LANGUAGES C CXX)
 set(CMAKE_CXX_STANDARD 20)
 
-file(WRITE ${CMAKE_BINARY_DIR}/version.txt ${PROJECT_VERSION})
+file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/version.txt ${PROJECT_VERSION})
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/version.txt.in
+               ${CMAKE_CURRENT_SOURCE_DIR}/paper/version.txt)
 
 include(CMakePushCheckState)
 include(CheckCXXCompilerFlag)
@@ -47,7 +49,8 @@ if(HAS_FULL_RELRO)
 endif()
 cmake_pop_check_state()
 
-set(version_script_flags LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+set(version_script_flags
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 cmake_push_check_state()
 list(APPEND CMAKE_REQUIRED_LINK_OPTIONS ${version_script_flags})
 check_cxx_source_compiles("int main(){}" HAS_VERSION_SCRIPT FAIL_REGEX
@@ -56,10 +59,14 @@ cmake_pop_check_state()
 
 option(USE_SIMD_FALLBACK
        "Use fallback implementations of functions that use SIMD" OFF)
+option(
+  USE_32_BIT_VERSIONS
+  "Store 32 bit versions internally, and rely on versions never being different by more than 2e9"
+  OFF)
 
 # This is encouraged according to
 # https://valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.clientreq
-include_directories(SYSTEM ${CMAKE_SOURCE_DIR}/third_party/valgrind)
+include_directories(SYSTEM ${CMAKE_CURRENT_SOURCE_DIR}/third_party/valgrind)
 
 add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-Wno-invalid-offsetof>)
 
@@ -100,25 +107,30 @@ if(NOT USE_SIMD_FALLBACK)
   endif()
 endif()
 
+if(USE_32_BIT_VERSIONS)
+  add_compile_definitions(INTERNAL_VERSION_32_BIT=1)
+endif()
+
 set(CMAKE_CXX_IMPLICIT_LINK_LIBRARIES "")
 
 add_library(${PROJECT_NAME}-object OBJECT ConflictSet.cpp)
 target_compile_options(${PROJECT_NAME}-object PRIVATE -fno-exceptions
                                                       -fvisibility=hidden)
 target_include_directories(${PROJECT_NAME}-object
-                           PRIVATE ${CMAKE_SOURCE_DIR}/include)
+                           PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
 
 add_library(${PROJECT_NAME} SHARED $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 set_target_properties(
   ${PROJECT_NAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                             "${CMAKE_BINARY_DIR}/radix_tree")
+                             "${CMAKE_CURRENT_BINARY_DIR}/radix_tree")
 if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE C)
 endif()
 
 if(HAS_VERSION_SCRIPT)
-  target_link_options(${PROJECT_NAME} PRIVATE
-                      LINKER:--version-script=${CMAKE_SOURCE_DIR}/linker.map)
+  target_link_options(
+    ${PROJECT_NAME} PRIVATE
+    LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 endif()
 
 add_library(${PROJECT_NAME}-static STATIC
@@ -131,7 +143,7 @@ if(APPLE)
   add_custom_command(
     TARGET ${PROJECT_NAME}-static
     PRE_LINK
-    COMMAND ${CMAKE_SOURCE_DIR}/privatize_symbols_macos.sh
+    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/privatize_symbols_macos.sh
             $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 else()
   add_custom_command(
@@ -139,21 +151,22 @@ else()
     POST_BUILD
     COMMAND
       ${CMAKE_OBJCOPY}
-      --keep-global-symbols=${CMAKE_SOURCE_DIR}/symbol-exports.txt
+      --keep-global-symbols=${CMAKE_CURRENT_SOURCE_DIR}/symbol-exports.txt
       $<TARGET_FILE:${PROJECT_NAME}-static> || echo
       "Proceeding with all symbols global in static library")
 endif()
 
-set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
-
 include(CTest)
 
-if(BUILD_TESTING)
+# disable tests if this is being used through e.g. FetchContent
+if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
+
+  set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)
 
   # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
   # coverage with a small number of tests.
 
-  file(GLOB CORPUS_TESTS ${CMAKE_SOURCE_DIR}/corpus/*)
+  file(GLOB CORPUS_TESTS ${CMAKE_CURRENT_SOURCE_DIR}/corpus/*)
 
   # extra testing that relies on shared libraries, which aren't available with
   # wasm
@@ -162,9 +175,11 @@ if(BUILD_TESTING)
     add_library(skip_list SHARED SkipList.cpp)
     target_compile_options(skip_list PRIVATE -fno-exceptions
                                              -fvisibility=hidden)
-    target_include_directories(skip_list PUBLIC ${CMAKE_SOURCE_DIR}/include)
-    set_target_properties(skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                                               "${CMAKE_BINARY_DIR}/skip_list")
+    target_include_directories(skip_list
+                               PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+    set_target_properties(
+      skip_list PROPERTIES LIBRARY_OUTPUT_DIRECTORY
+                           "${CMAKE_CURRENT_BINARY_DIR}/skip_list")
     set_target_properties(skip_list PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
     set_target_properties(
       skip_list PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
@@ -175,10 +190,11 @@ if(BUILD_TESTING)
     add_library(hash_table SHARED HashTable.cpp)
     target_compile_options(hash_table PRIVATE -fno-exceptions
                                               -fvisibility=hidden)
-    target_include_directories(hash_table PUBLIC ${CMAKE_SOURCE_DIR}/include)
+    target_include_directories(hash_table
+                               PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
     set_target_properties(
       hash_table PROPERTIES LIBRARY_OUTPUT_DIRECTORY
-                            "${CMAKE_BINARY_DIR}/hash_table")
+                            "${CMAKE_CURRENT_BINARY_DIR}/hash_table")
     set_target_properties(hash_table PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
     set_target_properties(
       hash_table PROPERTIES VERSION ${PROJECT_VERSION} SOVERSION
@@ -266,15 +282,19 @@ if(BUILD_TESTING)
     set_property(
       DIRECTORY
       APPEND
-      PROPERTY CMAKE_CONFIGURE_DEPENDS ${CMAKE_SOURCE_DIR}/test_conflict_set.py)
+      PROPERTY CMAKE_CONFIGURE_DEPENDS
+               ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py)
     execute_process(
-      COMMAND ${Python3_EXECUTABLE} ${CMAKE_SOURCE_DIR}/test_conflict_set.py
-              list OUTPUT_VARIABLE SCRIPT_TESTS)
+      COMMAND ${Python3_EXECUTABLE}
+              ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py list
+      OUTPUT_VARIABLE SCRIPT_TESTS)
     foreach(TEST ${SCRIPT_TESTS})
       add_test(
         NAME script_test_${TEST}
-        COMMAND ${Python3_EXECUTABLE} ${CMAKE_SOURCE_DIR}/test_conflict_set.py
-                test ${TEST} --build-dir ${CMAKE_BINARY_DIR})
+        COMMAND
+          ${Python3_EXECUTABLE}
+          ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
+          --build-dir ${CMAKE_CURRENT_BINARY_DIR})
     endforeach()
   endif()
 
@@ -308,25 +328,26 @@ if(BUILD_TESTING)
   # symbol visibility tests
   if(NOT WASM AND NOT CMAKE_BUILD_TYPE STREQUAL Debug)
     if(APPLE)
-      set(symbol_exports ${CMAKE_SOURCE_DIR}/apple-symbol-exports.txt)
-      set(symbol_imports ${CMAKE_SOURCE_DIR}/apple-symbol-imports.txt)
+      set(symbol_exports ${CMAKE_CURRENT_SOURCE_DIR}/apple-symbol-exports.txt)
+      set(symbol_imports ${CMAKE_CURRENT_SOURCE_DIR}/apple-symbol-imports.txt)
     else()
-      set(symbol_exports ${CMAKE_SOURCE_DIR}/symbol-exports.txt)
+      set(symbol_exports ${CMAKE_CURRENT_SOURCE_DIR}/symbol-exports.txt)
       if(CMAKE_SYSTEM_PROCESSOR STREQUAL aarch64)
-        set(symbol_imports ${CMAKE_SOURCE_DIR}/aarch64-symbol-imports.txt)
+        set(symbol_imports
+            ${CMAKE_CURRENT_SOURCE_DIR}/aarch64-symbol-imports.txt)
       else()
-        set(symbol_imports ${CMAKE_SOURCE_DIR}/symbol-imports.txt)
+        set(symbol_imports ${CMAKE_CURRENT_SOURCE_DIR}/symbol-imports.txt)
       endif()
     endif()
     add_test(
       NAME conflict_set_shared_symbols
       COMMAND
-        ${CMAKE_SOURCE_DIR}/test_symbols.sh $<TARGET_FILE:${PROJECT_NAME}>
-        ${symbol_exports} ${symbol_imports})
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
+        $<TARGET_FILE:${PROJECT_NAME}> ${symbol_exports} ${symbol_imports})
     add_test(
       NAME conflict_set_static_symbols
       COMMAND
-        ${CMAKE_SOURCE_DIR}/test_symbols.sh
+        ${CMAKE_CURRENT_SOURCE_DIR}/test_symbols.sh
         $<TARGET_FILE:${PROJECT_NAME}-static> ${symbol_exports}
         ${symbol_imports})
   endif()
@@ -369,13 +390,13 @@ set(CMAKE_OSX_DEPLOYMENT_TARGET 11.0)
 if(APPLE)
   find_program(PANDOC_EXE pandoc)
   if(PANDOC_EXE)
-    execute_process(COMMAND ${PANDOC_EXE} ${CMAKE_SOURCE_DIR}/README.md -o
-                            ${CMAKE_BINARY_DIR}/README.txt)
-    set(CPACK_RESOURCE_FILE_README ${CMAKE_BINARY_DIR}/README.txt)
+    execute_process(COMMAND ${PANDOC_EXE} ${CMAKE_CURRENT_SOURCE_DIR}/README.md
+                            -o ${CMAKE_CURRENT_BINARY_DIR}/README.txt)
+    set(CPACK_RESOURCE_FILE_README ${CMAKE_CURRENT_BINARY_DIR}/README.txt)
   endif()
-  configure_file(${CMAKE_SOURCE_DIR}/LICENSE ${CMAKE_BINARY_DIR}/LICENSE.txt
-                 COPYONLY)
-  set(CPACK_RESOURCE_FILE_LICENSE ${CMAKE_BINARY_DIR}/LICENSE.txt)
+  configure_file(${CMAKE_CURRENT_SOURCE_DIR}/LICENSE
+                 ${CMAKE_CURRENT_BINARY_DIR}/LICENSE.txt COPYONLY)
+  set(CPACK_RESOURCE_FILE_LICENSE ${CMAKE_CURRENT_BINARY_DIR}/LICENSE.txt)
 endif()
 
 include(CPack)

Internal.h

@@ -578,8 +578,8 @@ template <class ConflictSetImpl> struct TestDriver {
   explicit TestDriver(const uint8_t *data, size_t size)
       : arbitrary({data, size}) {}
 
-  int64_t writeVersion = 100;
-  int64_t oldestVersion = 0;
+  int64_t oldestVersion = arbitrary.bounded(2) ? 0 : 0xfffffff0;
+  int64_t writeVersion = oldestVersion + 100;
   ConflictSetImpl cs{oldestVersion};
   ReferenceImpl refImpl{oldestVersion};
 

Jenkinsfile

@@ -59,6 +59,17 @@ pipeline {
                 CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
             }
         }
+        stage('32-bit versions') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DUSE_32_BIT_VERSIONS=ON")
+            }
+        }
         stage('Release [gcc]') {
             agent {
                 dockerfile {

README.md

@@ -58,27 +58,29 @@ Performance counters:
 
 ## Skip list
 
-|  ns/op |         op/s | err% | total | benchmark                           |
-| -----: | -----------: | ---: | ----: | :---------------------------------- |
-| 246.99 | 4,048,700.59 | 0.2% |  0.01 | `point reads`                       |
-| 260.16 | 3,843,784.65 | 0.1% |  0.01 | `prefix reads`                      |
-| 493.35 | 2,026,953.19 | 0.1% |  0.01 | `range reads`                       |
-| 462.05 | 2,164,289.23 | 0.6% |  0.01 | `point writes`                      |
-| 448.19 | 2,231,205.25 | 0.9% |  0.01 | `prefix writes`                     |
-| 255.83 | 3,908,845.72 | 1.5% |  0.02 | `range writes`                      |
-| 582.63 | 1,716,349.02 | 1.3% |  0.01 | `monotonic increasing point writes` |
+|               ns/op |                op/s |    err% |     total | benchmark
+|--------------------:|--------------------:|--------:|----------:|:----------
+|              256.89 |        3,892,784.92 |    0.3% |      0.01 | `point reads`
+|              272.90 |        3,664,395.04 |    0.2% |      0.01 | `prefix reads`
+|              507.22 |        1,971,549.50 |    0.7% |      0.01 | `range reads`
+|              452.66 |        2,209,181.91 |    0.5% |      0.01 | `point writes`
+|              438.09 |        2,282,619.96 |    0.4% |      0.01 | `prefix writes`
+|              253.33 |        3,947,420.36 |    2.5% |      0.02 | `range writes`
+|              574.07 |        1,741,936.71 |    0.3% |      0.01 | `monotonic increasing point writes`
+|          151,562.50 |            6,597.94 |    1.5% |      0.01 | `worst case for radix tree`
 
 ## Radix tree (this implementation)
 
-|  ns/op |          op/s | err% | total | benchmark                           |
-| -----: | ------------: | ---: | ----: | :---------------------------------- |
-|  19.42 | 51,483,206.67 | 0.3% |  0.01 | `point reads`                       |
-|  58.43 | 17,115,612.57 | 0.1% |  0.01 | `prefix reads`                      |
-| 216.09 |  4,627,766.60 | 0.2% |  0.01 | `range reads`                       |
-|  28.35 | 35,267,567.72 | 0.2% |  0.01 | `point writes`                      |
-|  43.43 | 23,026,226.17 | 0.2% |  0.01 | `prefix writes`                     |
-|  50.00 | 20,000,000.00 | 0.0% |  0.01 | `range writes`                      |
-|  92.38 | 10,824,863.69 | 4.1% |  0.01 | `monotonic increasing point writes` |
+|               ns/op |                op/s |    err% |     total | benchmark
+|--------------------:|--------------------:|--------:|----------:|:----------
+|               19.83 |       50,420,955.28 |    0.1% |      0.01 | `point reads`
+|               55.95 |       17,872,542.40 |    0.5% |      0.01 | `prefix reads`
+|               88.28 |       11,327,709.50 |    0.4% |      0.01 | `range reads`
+|               29.15 |       34,309,531.64 |    0.5% |      0.01 | `point writes`
+|               42.36 |       23,607,424.27 |    1.1% |      0.01 | `prefix writes`
+|               50.00 |       20,000,000.00 |    0.0% |      0.01 | `range writes`
+|               93.52 |       10,692,413.79 |    3.3% |      0.01 | `monotonic increasing point writes`
+|        2,388,417.00 |              418.69 |    0.4% |      0.03 | `worst case for radix tree`
 
 # "Real data" test
 

apple-symbol-imports.txt

@@ -1,3 +1,5 @@
+___stack_chk_fail
+___stack_chk_guard
 __tlv_bootstrap
 _abort
 _bzero

conflict_set.py

@@ -115,7 +115,9 @@ class ConflictSet:
 
     def check(self, *reads: ReadRange) -> list[Result]:
         r = (ctypes.c_int * len(reads))()
-        self._lib.ConflictSet_check(self.p, *reads, r, 1)
+        self._lib.ConflictSet_check(
+            self.p, (ReadRange * len(reads))(*reads), r, len(reads)
+        )
         return [Result(x) for x in r]
 
     def setOldestVersion(self, version: int) -> None: