Reordering seems to improve codegen

WIP
Use preserve_none and put continuation array in CheckAll
2024-09-23 15:28:51 -07:00 · 2024-09-23 15:11:48 -07:00 · 2024-09-23 14:53:16 -07:00 · 2024-09-23 13:32:56 -07:00 · 2024-09-23 12:52:30 -07:00 · 2024-09-21 14:28:15 -07:00
191 changed files with 949 additions and 759 deletions
@@ -1,306 +0,0 @@
-name: CI
-
-on: [push, pull_request]
-
-env:
-  CC: clang
-  CXX: clang++
-
-jobs:
-  pre-commit:
-    runs-on: ubuntu-latest-amd64
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-amd64-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install dependencies
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -y git nodejs pre-commit
-
-      - uses: actions/cache@v4
-        with:
-          path: ~/.cache/pre-commit
-          key: pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
-
-      - name: Run pre-commit
-        run: |
-          git config --global --add safe.directory "$PWD"
-          pre-commit run --all-files --show-diff-on-failure
-
-  test:
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - name: 64-bit-versions
-            cmake_args: -DCMAKE_CXX_FLAGS=-DUSE_64_BIT=1
-          - name: debug
-            cmake_args: -DCMAKE_BUILD_TYPE=Debug
-          - name: simd-fallback
-            cmake_args: -DUSE_SIMD_FALLBACK=ON
-          - name: gcc
-            cmake_args: -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++
-    runs-on: ubuntu-latest-amd64
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-amd64-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install common dependencies
-        run: |
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            build-essential ccache cmake gcc g++ libc6-dbg \
-            ninja-build python3 valgrind zstd
-          sudo curl -Ls "https://minio.weaselab.dev/public/$(uname -m)/mc.RELEASE.2025-08-13T08-35-41Z" \
-            -o /usr/local/bin/mc && sudo chmod +x /usr/local/bin/mc
-
-      - name: Install LLVM toolchain
-        if: matrix.name != 'gcc'
-        run: |
-          . /etc/os-release
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc
-          echo "deb http://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            clang-21 llvm-21 lld-21 mold
-          for tool in clang clang++ llvm-ar llvm-nm llvm-ranlib llvm-objcopy llvm-cov llvm-symbolizer lld ld.lld; do
-            sudo update-alternatives --install /usr/bin/${tool} ${tool} /usr/bin/${tool}-21 100
-          done
-
-      - name: Build
-        run: |
-          export CCACHE_DIR="$GITHUB_WORKSPACE/.ccache"
-          rm -rf build
-          cmake -S . -B build -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ${{ matrix.cmake_args }}
-          ninja -C build
-          ccache -s
-
-      - name: Test
-        run: |
-          cd build
-          ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 -T Test -j "$(nproc)" --timeout 90 > /dev/null
-
-      - name: Upload test results to MinIO
-        if: always()
-        env:
-          MINIO_ACCESS_KEY: ${{ secrets.MINIO_ACCESS_KEY }}
-          MC_HOST_minio: https://${{ secrets.MINIO_ACCESS_KEY }}:${{ secrets.MINIO_SECRET_KEY }}@minio.weaselab.dev
-        run: |
-          if [ -z "$MINIO_ACCESS_KEY" ]; then
-            echo "MinIO credentials not configured; skipping upload"
-            exit 0
-          fi
-          zstd build/Testing/*/Test.xml
-          mc cp build/Testing/*/Test.xml.zst "minio/jenkins/conflict-set/${{ gitea.run_number }}/${{ matrix.name }}/"
-
-      - name: Test summary
-        if: always()
-        run: |
-          python3 ctest_summary.py build/Testing/*/Test.xml \
-            --link "https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/${{ matrix.name }}/Test.xml.zst" \
-            | tee -a "$GITHUB_STEP_SUMMARY"
-
-  release:
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - runner: ubuntu-latest-amd64
-            arch: amd64
-          - runner: ubuntu-latest-arm64
-            arch: arm64
-    runs-on: ${{ matrix.runner }}
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-${{ matrix.arch }}-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install dependencies
-        run: |
-          . /etc/os-release
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc
-          echo "deb http://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            build-essential ccache clang-21 cmake devscripts \
-            libc6-dbg llvm-21 lld-21 mold ninja-build rpm \
-            valgrind zstd
-          sudo curl -Ls "https://minio.weaselab.dev/public/$(uname -m)/mc.RELEASE.2025-08-13T08-35-41Z" \
-            -o /usr/local/bin/mc && sudo chmod +x /usr/local/bin/mc
-          for tool in clang clang++ llvm-ar llvm-nm llvm-ranlib llvm-objcopy llvm-cov llvm-symbolizer lld ld.lld; do
-            sudo update-alternatives --install /usr/bin/${tool} ${tool} /usr/bin/${tool}-21 100
-          done
-
-      - name: Install paper build dependencies
-        if: matrix.arch == 'amd64'
-        run: |
-          sudo apt-get install -y \
-            biber latexmk \
-            texlive-bibtex-extra texlive-fonts-recommended \
-            texlive-latex-extra texlive-pictures
-
-      - uses: actions/cache@v4
-        with:
-          path: .ccache
-          key: ccache-release-${{ matrix.arch }}-${{ gitea.sha }}
-          restore-keys: |
-            ccache-release-${{ matrix.arch }}-
-
-      - name: Build
-        run: |
-          export CCACHE_DIR="$GITHUB_WORKSPACE/.ccache"
-          rm -rf build
-          cmake -S . -B build -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_FLAGS=-DNVALGRIND
-          ninja -C build
-          ccache -s
-
-      - name: Test
-        run: |
-          cd build
-          # On arm64, valgrind needs the MAKE_MEM_DEFINED client requests for
-          # https://git.weaselab.dev/weaselab/conflict-set/issues/39, but this
-          # build has -DNVALGRIND, which compiles them out. Skip valgrind
-          # tests here; they run annotated in the test job.
-          ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 ${{ matrix.arch == 'arm64' && '-E valgrind' || '' }} -T Test -j "$(nproc)" --timeout 90 > /dev/null
-
-      - name: Package
-        run: |
-          cd build
-          cpack -G DEB
-          cpack -G RPM
-
-      - name: Build paper
-        if: matrix.arch == 'amd64'
-        run: |
-          cd paper
-          make
-
-      - name: Upload artifacts to MinIO
-        if: always()
-        shell: bash
-        env:
-          MINIO_ACCESS_KEY: ${{ secrets.MINIO_ACCESS_KEY }}
-          MC_HOST_minio: https://${{ secrets.MINIO_ACCESS_KEY }}:${{ secrets.MINIO_SECRET_KEY }}@minio.weaselab.dev
-        run: |
-          if [ -z "$MINIO_ACCESS_KEY" ]; then
-            echo "MinIO credentials not configured; skipping upload"
-            exit 0
-          fi
-          dest="minio/jenkins/conflict-set/${{ gitea.run_number }}/release-${{ matrix.arch }}/"
-          zstd build/Testing/*/Test.xml
-          mc cp build/Testing/*/Test.xml.zst "$dest"
-          if compgen -G "build/*.deb" > /dev/null; then
-            mc cp build/*.deb "$dest"
-          fi
-          if compgen -G "build/*.rpm" > /dev/null; then
-            mc cp build/*.rpm "$dest"
-          fi
-          if compgen -G "paper/*.pdf" > /dev/null; then
-            mc cp paper/*.pdf "$dest"
-          fi
-
-      - name: Test summary
-        if: always()
-        run: |
-          python3 ctest_summary.py build/Testing/*/Test.xml \
-            --link "https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/release-${{ matrix.arch }}/Test.xml.zst" \
-            | tee -a "$GITHUB_STEP_SUMMARY"
-
-  coverage:
-    runs-on: ubuntu-latest-amd64
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/cache@v4
-        with:
-          path: /var/cache/apt/archives
-          key: apt-amd64-${{ hashFiles('.gitea/workflows/ci.yml') }}
-
-      - name: Install dependencies
-        run: |
-          . /etc/os-release
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc
-          echo "deb http://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo apt-get update -qq
-          sudo apt-get install -y \
-            build-essential ccache clang-21 cmake gcovr \
-            libc6-dbg llvm-21 lld-21 mold ninja-build python3 valgrind zstd
-          sudo curl -Ls "https://minio.weaselab.dev/public/$(uname -m)/mc.RELEASE.2025-08-13T08-35-41Z" \
-            -o /usr/local/bin/mc && sudo chmod +x /usr/local/bin/mc
-          for tool in clang clang++ llvm-ar llvm-nm llvm-ranlib llvm-objcopy llvm-cov llvm-symbolizer lld ld.lld; do
-            sudo update-alternatives --install /usr/bin/${tool} ${tool} /usr/bin/${tool}-21 100
-          done
-
-      - uses: actions/cache@v4
-        with:
-          path: .ccache
-          key: ccache-coverage-${{ gitea.sha }}
-          restore-keys: |
-            ccache-coverage-
-
-      - name: Build
-        run: |
-          export CCACHE_DIR="$GITHUB_WORKSPACE/.ccache"
-          rm -rf build
-          cmake -S . -B build -G Ninja -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-            -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage \
-            -DCMAKE_BUILD_TYPE=Debug -DDISABLE_TSAN=ON
-          ninja -C build
-          ccache -s
-
-      - name: Test
-        run: |
-          cd build
-          ctest --no-compress-output --test-output-size-passed 100000 --test-output-size-failed 100000 -T Test -j "$(nproc)" --timeout 90 > /dev/null
-
-      - name: Coverage report
-        shell: bash
-        run: |
-          gcov_args=(-f ConflictSet.cpp -f LongestCommonPrefix.h -f Metrics.h
-                     --gcov-executable "llvm-cov gcov" --exclude-noncode-lines)
-          gcovr "${gcov_args[@]}" --cobertura > build/coverage.xml
-          gcovr "${gcov_args[@]}"
-          mkdir -p build/coverage_html
-          gcovr "${gcov_args[@]}" --html-details build/coverage_html/index.html
-          gcovr "${gcov_args[@]}" --fail-under-line 100 > /dev/null
-
-      - name: Upload results to MinIO
-        if: always()
-        env:
-          MINIO_ACCESS_KEY: ${{ secrets.MINIO_ACCESS_KEY }}
-          MC_HOST_minio: https://${{ secrets.MINIO_ACCESS_KEY }}:${{ secrets.MINIO_SECRET_KEY }}@minio.weaselab.dev
-        run: |
-          if [ -z "$MINIO_ACCESS_KEY" ]; then
-            echo "MinIO credentials not configured; skipping upload"
-            exit 0
-          fi
-          dest="minio/jenkins/conflict-set/${{ gitea.run_number }}/coverage/"
-          zstd build/Testing/*/Test.xml
-          mc cp build/Testing/*/Test.xml.zst "$dest"
-          if [ -e build/coverage.xml ]; then
-            mc cp build/coverage.xml "$dest"
-          fi
-          if [ -d build/coverage_html ]; then
-            mc cp -r build/coverage_html "$dest"
-          fi
-
-      - name: Test summary
-        if: always()
-        run: |
-          python3 ctest_summary.py build/Testing/*/Test.xml \
-            --link "https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/coverage/Test.xml.zst" \
-            | tee -a "$GITHUB_STEP_SUMMARY"
-          echo "" | tee -a "$GITHUB_STEP_SUMMARY"
-          echo "📊 [Coverage report](https://minio.weaselab.dev/jenkins/conflict-set/${{ gitea.run_number }}/coverage/coverage_html/index.html)" | tee -a "$GITHUB_STEP_SUMMARY"
@@ -24,6 +24,14 @@ repos:
        entry: "^#define SHOW_MEMORY 1$"
        language: pygrep
        types: [c++]
+  - repo: local
+    hooks:
+      - id: sim cache misses check
+        name: disallow checking in SIM_CACHE_MISSES=1
+        description: disallow checking in SIM_CACHE_MISSES=1
+        entry: "^#define SIM_CACHE_MISSES 1$"
+        language: pygrep
+        types: [c++]
  - repo: https://github.com/shellcheck-py/shellcheck-py
    rev: a23f6b85d0fdd5bb9d564e2579e678033debbdff # frozen: v0.10.0.1
    hooks:
@@ -7,7 +7,6 @@
 void showMemory(const ConflictSet &cs);
 #endif

-#define ANKERL_NANOBENCH_IMPLEMENT
 #include "third_party/nanobench.h"

 constexpr int kNumKeys = 1000000;
@@ -72,12 +72,6 @@ else()
  add_link_options(-Wl,--gc-sections)
 endif()

-if(EMSCRIPTEN)
-  # https://github.com/emscripten-core/emscripten/issues/15377#issuecomment-1285167486
-  add_link_options(-lnodefs.js -lnoderawfs.js)
-  add_link_options(-s ALLOW_MEMORY_GROWTH)
-endif()
-
 if(NOT USE_SIMD_FALLBACK)
  cmake_push_check_state()
  list(APPEND CMAKE_REQUIRED_FLAGS -mavx)
@@ -144,6 +138,8 @@ include(CTest)
 # disable tests if this is being used through e.g. FetchContent
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)

+  add_library(nanobench ${CMAKE_CURRENT_SOURCE_DIR}/nanobench.cpp)
+
  set(TEST_FLAGS -Wall -Wextra -Wunreachable-code -Wpedantic -UNDEBUG)

  # corpus tests, which are tests curated by libfuzzer. The goal is to get broad
@@ -191,6 +187,7 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
  target_include_directories(conflict_set_main
                             PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
  target_compile_definitions(conflict_set_main PRIVATE ENABLE_MAIN)
+  target_link_libraries(conflict_set_main PRIVATE nanobench)

  if(NOT APPLE)
    # libfuzzer target, to generate/manage corpus
@@ -336,7 +333,7 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)

  # bench
  add_executable(conflict_set_bench Bench.cpp)
-  target_link_libraries(conflict_set_bench PRIVATE ${PROJECT_NAME})
+  target_link_libraries(conflict_set_bench PRIVATE ${PROJECT_NAME} nanobench)
  set_target_properties(conflict_set_bench PROPERTIES SKIP_BUILD_RPATH ON)
  add_executable(real_data_bench RealDataBench.cpp)
  target_link_libraries(real_data_bench PRIVATE ${PROJECT_NAME})
@@ -351,6 +348,14 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
  add_executable(server_bench ServerBench.cpp)
  target_link_libraries(server_bench PRIVATE ${PROJECT_NAME})
  set_target_properties(server_bench PROPERTIES SKIP_BUILD_RPATH ON)
+
+  add_executable(interleaving_test InterleavingTest.cpp)
+  # work around lack of musttail for gcc
+  if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_BUILD_TYPE STREQUAL "Debug")
+    target_compile_options(interleaving_test PRIVATE -Og
+                                                     -foptimize-sibling-calls)
+  endif()
+  target_link_libraries(interleaving_test PRIVATE nanobench)
 endif()

 # packaging
@@ -17,9 +17,9 @@ limitations under the License.
 #include "ConflictSet.h"
 #include "Internal.h"
 #include "LongestCommonPrefix.h"
+#include "Metrics.h"

 #include <algorithm>
-#include <atomic>
 #include <bit>
 #include <cassert>
 #include <cstddef>
@@ -48,6 +48,17 @@ limitations under the License.
 #endif
 #endif

+#define SIM_CACHE_MISSES 0
+#if SIM_CACHE_MISSES
+constexpr void simCacheMiss(void *x) {
+  if (x) {
+    _mm_clflush(x);
+  }
+}
+#else
+constexpr void simCacheMiss(void *) {}
+#endif
+
 #include <memcheck.h>

 using namespace weaselab;
@@ -578,39 +589,6 @@ static_assert(kBytesPerKey - sizeof(Node0) >= kMinNodeSurplus);

 constexpr int64_t kFreeListMaxMemory = 1 << 20;

-struct Metric {
-  Metric *prev;
-  const char *name;
-  const char *help;
-  ConflictSet::MetricsV1::Type type;
-  std::atomic<int64_t> value;
-
-protected:
-  Metric(ConflictSet::Impl *impl, const char *name, const char *help,
-         ConflictSet::MetricsV1::Type type);
-};
-
-struct Gauge : private Metric {
-  Gauge(ConflictSet::Impl *impl, const char *name, const char *help)
-      : Metric(impl, name, help, ConflictSet::MetricsV1::Gauge) {}
-
-  void set(int64_t value) {
-    this->value.store(value, std::memory_order_relaxed);
-  }
-};
-
-struct Counter : private Metric {
-  Counter(ConflictSet::Impl *impl, const char *name, const char *help)
-      : Metric(impl, name, help, ConflictSet::MetricsV1::Counter) {}
-  // Expensive. Accumulate locally and then call add instead of repeatedly
-  // calling add.
-  void add(int64_t value) {
-    assert(value >= 0);
-    static_assert(std::atomic<int64_t>::is_always_lock_free);
-    this->value.fetch_add(value, std::memory_order_relaxed);
-  }
-};
-
 template <class T> struct BoundedFreeListAllocator {

  static_assert(sizeof(T) >= sizeof(void *));
@@ -869,21 +847,29 @@ int getNodeIndex(Node16 *self, uint8_t index) {

 // Precondition - an entry for index must exist in the node
 Node *&getChildExists(Node3 *self, uint8_t index) {
-  return self->children[getNodeIndex(self, index)];
+  auto &result = self->children[getNodeIndex(self, index)];
+  simCacheMiss(result);
+  return result;
 }
 // Precondition - an entry for index must exist in the node
 Node *&getChildExists(Node16 *self, uint8_t index) {
-  return self->children[getNodeIndex(self, index)];
+  auto &result = self->children[getNodeIndex(self, index)];
+  simCacheMiss(result);
+  return result;
 }
 // Precondition - an entry for index must exist in the node
 Node *&getChildExists(Node48 *self, uint8_t index) {
  assert(self->bitSet.test(index));
-  return self->children[self->index[index]];
+  auto &result = self->children[self->index[index]];
+  simCacheMiss(result);
+  return result;
 }
 // Precondition - an entry for index must exist in the node
 Node *&getChildExists(Node256 *self, uint8_t index) {
  assert(self->bitSet.test(index));
-  return self->children[index];
+  auto &result = self->children[index];
+  simCacheMiss(result);
+  return result;
 }

 // Precondition - an entry for index must exist in the node
@@ -1063,6 +1049,7 @@ ChildAndMaxVersion getChildAndMaxVersion(Node3 *self, uint8_t index) {
  if (i < 0) {
    return {};
  }
+  simCacheMiss(self->children[i]);
  return {self->children[i], self->childMaxVersion[i]};
 }
 ChildAndMaxVersion getChildAndMaxVersion(Node16 *self, uint8_t index) {
@@ -1070,6 +1057,7 @@ ChildAndMaxVersion getChildAndMaxVersion(Node16 *self, uint8_t index) {
  if (i < 0) {
    return {};
  }
+  simCacheMiss(self->children[i]);
  return {self->children[i], self->childMaxVersion[i]};
 }
 ChildAndMaxVersion getChildAndMaxVersion(Node48 *self, uint8_t index) {
@@ -1077,9 +1065,11 @@ ChildAndMaxVersion getChildAndMaxVersion(Node48 *self, uint8_t index) {
  if (i < 0) {
    return {};
  }
+  simCacheMiss(self->children[i]);
  return {self->children[i], self->childMaxVersion[i]};
 }
 ChildAndMaxVersion getChildAndMaxVersion(Node256 *self, uint8_t index) {
+  simCacheMiss(self->children[index]);
  return {self->children[index], self->childMaxVersion[index]};
 }

@@ -1105,6 +1095,7 @@ Node *getChildGeq(Node0 *, int) { return nullptr; }
 Node *getChildGeq(Node3 *n, int child) {
  for (int i = 0; i < n->numChildren; ++i) {
    if (n->index[i] >= child) {
+      simCacheMiss(n->children[i]);
      return n->children[i];
    }
  }
@@ -1123,7 +1114,10 @@ Node *getChildGeq(Node16 *self, int child) {
  __m128i results = _mm_cmpeq_epi8(key_vec, _mm_min_epu8(key_vec, indices));
  int mask = (1 << self->numChildren) - 1;
  uint32_t bitfield = _mm_movemask_epi8(results) & mask;
-  return bitfield == 0 ? nullptr : self->children[std::countr_zero(bitfield)];
+  auto *result =
+      bitfield == 0 ? nullptr : self->children[std::countr_zero(bitfield)];
+  simCacheMiss(result);
+  return result;
 #elif defined(HAS_ARM_NEON)
  uint8x16_t indices;
  memcpy(&indices, self->index, sizeof(self->index));
@@ -1159,13 +1153,16 @@ Node *getChildGeq(Node48 *self, int child) {
  if (c < 0) {
    return nullptr;
  }
-  return self->children[self->index[c]];
+  auto *result = self->children[self->index[c]];
+  simCacheMiss(result);
+  return result;
 }
 Node *getChildGeq(Node256 *self, int child) {
  int c = self->bitSet.firstSetGeq(child);
  if (c < 0) {
    return nullptr;
  }
+  simCacheMiss(self->children[c]);
  return self->children[c];
 }

@@ -1189,20 +1186,26 @@ Node *getChildGeq(Node *self, int child) {
 // Precondition: self has a child
 Node *getFirstChildExists(Node3 *self) {
  assert(self->numChildren > 0);
+  simCacheMiss(self->children[0]);
  return self->children[0];
 }
 // Precondition: self has a child
 Node *getFirstChildExists(Node16 *self) {
  assert(self->numChildren > 0);
+  simCacheMiss(self->children[0]);
  return self->children[0];
 }
 // Precondition: self has a child
 Node *getFirstChildExists(Node48 *self) {
-  return self->children[self->index[self->bitSet.firstSetGeq(0)]];
+  auto *result = self->children[self->index[self->bitSet.firstSetGeq(0)]];
+  simCacheMiss(result);
+  return result;
 }
 // Precondition: self has a child
 Node *getFirstChildExists(Node256 *self) {
-  return self->children[self->bitSet.firstSetGeq(0)];
+  auto *result = self->children[self->bitSet.firstSetGeq(0)];
+  simCacheMiss(result);
+  return result;
 }

 // Precondition: self has a child
@@ -2543,38 +2546,19 @@ downLeftSpine:
 namespace {
 // Return true if the max version among all keys that start with key[:prefixLen]
 // that are >= key is <= readVersion
-struct CheckRangeLeftSide {
-  CheckRangeLeftSide(Node *n, std::span<const uint8_t> key, int prefixLen,
-                     InternalVersionT readVersion, ReadContext *tls)
-      : n(n), remaining(key), prefixLen(prefixLen), readVersion(readVersion),
-        impl(tls->impl), tls(tls) {
-#if DEBUG_VERBOSE && !defined(NDEBUG)
-    fprintf(stderr, "Check range left side from %s for keys starting with %s\n",
-            printable(key).c_str(),
-            printable(key.subspan(0, prefixLen)).c_str());
-#endif
-  }
-
-  Node *n;
-  std::span<const uint8_t> remaining;
-  int prefixLen;
-  InternalVersionT readVersion;
-  ConflictSet::Impl *impl;
-  ReadContext *tls;
+bool checkRangeLeftSide(Node *n, std::span<const uint8_t> key, int prefixLen,
+                        InternalVersionT readVersion, ReadContext *tls) {
+  auto remaining = key;
  int searchPathLen = 0;
-  bool ok;
-
-  bool step() {
+  for (;; ++tls->range_read_iterations_accum) {
    if (remaining.size() == 0) {
      assert(searchPathLen >= prefixLen);
-      ok = maxVersion(n) <= readVersion;
-      return true;
+      return maxVersion(n) <= readVersion;
    }

    if (searchPathLen >= prefixLen) {
      if (!checkMaxBetweenExclusive(n, remaining[0], 256, readVersion, tls)) {
-        ok = false;
-        return true;
+        return false;
      }
    }

@@ -2584,18 +2568,16 @@ struct CheckRangeLeftSide {
      if (c != nullptr) {
        if (searchPathLen < prefixLen) {
          n = c;
-          return downLeftSpine();
+          goto downLeftSpine;
        }
        n = c;
-        ok = maxVersion(n) <= readVersion;
-        return true;
+        return maxVersion(n) <= readVersion;
      } else {
        n = nextSibling(n);
        if (n == nullptr) {
-          ok = true;
          return true;
        }
-        return downLeftSpine();
+        goto downLeftSpine;
      }
    }

@@ -2611,21 +2593,18 @@ struct CheckRangeLeftSide {
        auto c = n->partialKey()[i] <=> remaining[i];
        if (c > 0) {
          if (searchPathLen < prefixLen) {
-            return downLeftSpine();
+            goto downLeftSpine;
          }
          if (n->entryPresent && n->entry.rangeVersion > readVersion) {
-            ok = false;
-            return true;
+            return false;
          }
-          ok = maxVersion(n) <= readVersion;
-          return true;
+          return maxVersion(n) <= readVersion;
        } else {
          n = nextSibling(n);
          if (n == nullptr) {
-            ok = true;
            return true;
          }
-          return downLeftSpine();
+          goto downLeftSpine;
        }
      }
      if (commonLen == n->partialKeyLen) {
@@ -2634,83 +2613,47 @@ struct CheckRangeLeftSide {
      } else if (n->partialKeyLen > int(remaining.size())) {
        assert(searchPathLen >= prefixLen);
        if (n->entryPresent && n->entry.rangeVersion > readVersion) {
-          ok = false;
-          return true;
+          return false;
        }
-        ok = maxVersion(n) <= readVersion;
-        return true;
+        return maxVersion(n) <= readVersion;
      }
    }
    if (maxV <= readVersion) {
-      ok = true;
      return true;
    }
-    return false;
  }
-
-  bool downLeftSpine() {
+downLeftSpine:
  for (; !n->entryPresent; n = getFirstChildExists(n)) {
  }
-    ok = n->entry.rangeVersion <= readVersion;
-    return true;
+  return n->entry.rangeVersion <= readVersion;
 }
-};

 // Return true if the max version among all keys that start with key[:prefixLen]
 // that are < key is <= readVersion
-struct CheckRangeRightSide {
-  CheckRangeRightSide(Node *n, std::span<const uint8_t> key, int prefixLen,
-                      InternalVersionT readVersion, ReadContext *tls)
-      : n(n), key(key), remaining(key), prefixLen(prefixLen),
-        readVersion(readVersion), impl(tls->impl), tls(tls) {
-#if DEBUG_VERBOSE && !defined(NDEBUG)
-    fprintf(stderr, "Check range right side to %s for keys starting with %s\n",
-            printable(key).c_str(),
-            printable(key.subspan(0, prefixLen)).c_str());
-#endif
-  }
-
-  Node *n;
-  std::span<const uint8_t> key;
-  std::span<const uint8_t> remaining;
-  int prefixLen;
-  InternalVersionT readVersion;
-  ConflictSet::Impl *impl;
-  ReadContext *tls;
+bool checkRangeRightSide(Node *n, std::span<const uint8_t> key, int prefixLen,
+                         InternalVersionT readVersion, ReadContext *tls) {
+  auto remaining = key;
  int searchPathLen = 0;
-  bool ok;
-
-  bool step() {
-#if DEBUG_VERBOSE && !defined(NDEBUG)
-    fprintf(stderr,
-            "Search path: %s, searchPathLen: %d, prefixLen: %d, remaining: "
-            "%s\n",
-            getSearchPathPrintable(n).c_str(), searchPathLen, prefixLen,
-            printable(remaining).c_str());
-#endif

+  for (;; ++tls->range_read_iterations_accum) {
    assert(searchPathLen <= int(key.size()));
-
    if (remaining.size() == 0) {
-      return downLeftSpine();
+      goto downLeftSpine;
    }

    if (searchPathLen >= prefixLen) {
      if (n->entryPresent && n->entry.pointVersion > readVersion) {
-        ok = false;
-        return true;
+        return false;
      }

      if (!checkMaxBetweenExclusive(n, -1, remaining[0], readVersion, tls)) {
-        ok = false;
-        return true;
+        return false;
      }
    }

    if (searchPathLen > prefixLen && n->entryPresent &&
        n->entry.rangeVersion > readVersion) {
-      ok = false;
-      return true;
+      return false;
    }

    auto *child = getChild(n, remaining[0]);
@@ -2718,9 +2661,9 @@ struct CheckRangeRightSide {
      auto c = getChildGeq(n, remaining[0]);
      if (c != nullptr) {
        n = c;
-        return downLeftSpine();
+        goto downLeftSpine;
      } else {
-        return backtrack();
+        goto backtrack;
      }
    }

@@ -2736,35 +2679,30 @@ struct CheckRangeRightSide {
        ++searchPathLen;
        auto c = n->partialKey()[i] <=> remaining[i];
        if (c > 0) {
-          return downLeftSpine();
+          goto downLeftSpine;
        } else {
          if (searchPathLen > prefixLen && n->entryPresent &&
              n->entry.rangeVersion > readVersion) {
-            ok = false;
-            return true;
+            return false;
          }
-          return backtrack();
+          goto backtrack;
        }
      }
      if (commonLen == n->partialKeyLen) {
        // partial key matches
        remaining = remaining.subspan(commonLen, remaining.size() - commonLen);
      } else if (n->partialKeyLen > int(remaining.size())) {
-        return downLeftSpine();
+        goto downLeftSpine;
      }
    }
-    return false;
  }
-
-  bool backtrack() {
+backtrack:
  for (;;) {
    // searchPathLen > prefixLen implies n is not the root
    if (searchPathLen > prefixLen && maxVersion(n) > readVersion) {
-        ok = false;
-        return true;
+      return false;
    }
    if (n->parent == nullptr) {
-        ok = true;
      return true;
    }
    auto next = getChildGeq(n->parent, n->parentsIndex + 1);
@@ -2775,18 +2713,14 @@ struct CheckRangeRightSide {
      searchPathLen -= n->partialKeyLen;
      n = next;
      searchPathLen += n->partialKeyLen;
-        return downLeftSpine();
+      goto downLeftSpine;
    }
  }
-  }
-
-  bool downLeftSpine() {
+downLeftSpine:
  for (; !n->entryPresent; n = getFirstChildExists(n)) {
  }
-    ok = n->entry.rangeVersion <= readVersion;
-    return true;
+  return n->entry.rangeVersion <= readVersion;
 }
-};
 } // namespace

 bool checkRangeRead(Node *n, std::span<const uint8_t> begin,
@@ -2807,8 +2741,8 @@ bool checkRangeRead(Node *n, std::span<const uint8_t> begin,

  auto remaining = begin.subspan(0, lcp);
  Arena arena;
-  // If the common prefix isn't a prefix of any physical entry in the tree, we
-  // can go to "downLeftSpine"
+
+  // Advance down common prefix, but stay on a physical path in the tree
  for (;; ++tls->range_read_iterations_accum) {
    assert(getSearchPath(arena, n) <=>
               begin.subspan(0, lcp - remaining.size()) ==
@@ -2849,47 +2783,17 @@ bool checkRangeRead(Node *n, std::span<const uint8_t> begin,
  lcp -= consumed;

  if (lcp == int(begin.size())) {
-    CheckRangeRightSide checkRangeRightSide{n, end, lcp, readVersion, tls};
-    while (!checkRangeRightSide.step())
-      ;
-    return checkRangeRightSide.ok;
+    return checkRangeRightSide(n, end, lcp, readVersion, tls);
  }

-  if (!checkRangeStartsWith(n, begin.subspan(0, lcp), begin[lcp], end[lcp],
-                            readVersion, tls)) {
-    return false;
-  }
-
-  // This makes it safe to check maxVersion within CheckRangeLeftSide. If this
+  // This makes it safe to check maxVersion within checkRangeLeftSide. If this
  // were false, then we would have returned above since lcp == begin.size().
  assert(!(n->parent == nullptr && begin.size() == 0));
-  CheckRangeLeftSide checkRangeLeftSide{n, begin, lcp + 1, readVersion, tls};
-  CheckRangeRightSide checkRangeRightSide{n, end, lcp + 1, readVersion, tls};

-  for (;;) {
-    bool leftDone = checkRangeLeftSide.step();
-    bool rightDone = checkRangeRightSide.step();
-    if (!leftDone && !rightDone) {
-      tls->range_read_iterations_accum += 2;
-      continue;
-    }
-    if (leftDone && rightDone) {
-      break;
-    } else if (leftDone) {
-      while (!checkRangeRightSide.step()) {
-        ++tls->range_read_iterations_accum;
-      }
-      break;
-    } else {
-      assert(rightDone);
-      while (!checkRangeLeftSide.step()) {
-        ++tls->range_read_iterations_accum;
-      }
-    }
-    break;
-  }
-
-  return checkRangeLeftSide.ok && checkRangeRightSide.ok;
+  return checkRangeStartsWith(n, begin.subspan(0, lcp), begin[lcp], end[lcp],
+                              readVersion, tls) &&
+         checkRangeLeftSide(n, begin, lcp + 1, readVersion, tls) &&
+         checkRangeRightSide(n, end, lcp + 1, readVersion, tls);
 }

 #ifdef __x86_64__
@@ -3141,34 +3045,288 @@ Node *firstGeqPhysical(Node *n, const std::span<const uint8_t> key) {
  }
 }

+#ifndef __has_attribute
+#define __has_attribute(x) 0
+#endif
+
+#if __has_attribute(musttail)
+#define MUSTTAIL __attribute__((musttail))
+#else
+#define MUSTTAIL
+#endif
+
+#if __has_attribute(preserve_none)
+#define CONTINUATION_CALLING_CONVENTION __attribute__((preserve_none))
+#else
+#define CONTINUATION_CALLING_CONVENTION
+#endif
+
+typedef CONTINUATION_CALLING_CONVENTION void (*continuation)(struct CheckAll *,
+                                                             int64_t prevJob,
+                                                             int64_t job,
+                                                             int64_t started,
+                                                             int64_t count);
+
+// State relevant to a particular query
+struct CheckJob {
+  void setResult(bool ok) {
+    *result = ok ? ConflictSet::Commit : ConflictSet::Conflict;
+  }
+
+  [[nodiscard]] continuation init(const ConflictSet::ReadRange *read,
+                                  ConflictSet::Result *result, Node *root,
+                                  int64_t oldestVersionFullPrecision,
+                                  ReadContext *tls);
+
+  Node *n;
+  ChildAndMaxVersion childAndVersion;
+  std::span<const uint8_t> begin;
+  InternalVersionT readVersion;
+  ConflictSet::Result *result;
+};
+
+// State relevant to all queries
+struct CheckAll {
+  constexpr static int kConcurrent = 32;
+  CheckJob inProgress[kConcurrent];
+  continuation next[kConcurrent];
+  int nextJob[kConcurrent];
+  Node *root;
+  int64_t oldestVersionFullPrecision;
+  ReadContext *tls;
+  const ConflictSet::ReadRange *queries;
+  ConflictSet::Result *results;
+};
+
+CONTINUATION_CALLING_CONVENTION void keepGoing(CheckAll *context,
+                                               int64_t prevJob, int64_t job,
+                                               int64_t started, int64_t count) {
+  prevJob = job;
+  job = context->nextJob[job];
+  MUSTTAIL return context->next[job](context, prevJob, job, started, count);
+}
+
+CONTINUATION_CALLING_CONVENTION void complete(CheckAll *context,
+                                              int64_t prevJob, int64_t job,
+                                              int64_t started, int64_t count) {
+  if (started == count) {
+    if (prevJob == job) {
+      return;
+    }
+    context->nextJob[prevJob] = context->nextJob[job];
+    job = prevJob;
+  } else {
+    int temp = started++;
+    context->next[job] = context->inProgress[job].init(
+        context->queries + temp, context->results + temp, context->root,
+        context->oldestVersionFullPrecision, context->tls);
+  }
+  MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+}
+
+namespace check_point_read_state_machine {
+
+CONTINUATION_CALLING_CONVENTION void
+down_left_spine(struct CheckAll *, int64_t prevJob, int64_t job,
+                int64_t started, int64_t count);
+CONTINUATION_CALLING_CONVENTION void iter(struct CheckAll *, int64_t prevJob,
+                                          int64_t job, int64_t started,
+                                          int64_t count);
+CONTINUATION_CALLING_CONVENTION void begin(struct CheckAll *, int64_t prevJob,
+                                           int64_t job, int64_t started,
+                                           int64_t count);
+
+void begin(struct CheckAll *context, int64_t prevJob, int64_t job,
+           int64_t started, int64_t count) {
+  ++context->tls->point_read_accum;
+#if DEBUG_VERBOSE && !defined(NDEBUG)
+  fprintf(stderr, "Check point read: %s\n", printable(key).c_str());
+#endif
+  auto *j = context->inProgress + job;
+
+  if (j->begin.size() == 0) {
+    if (j->n->entryPresent) {
+      j->setResult(j->n->entry.pointVersion <= j->readVersion);
+      MUSTTAIL return complete(context, prevJob, job, started, count);
+    }
+    j->n = getFirstChildExists(j->n);
+    context->next[job] = down_left_spine;
+    __builtin_prefetch(j->n);
+    MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+  }
+
+  j->childAndVersion = getChildAndMaxVersion(j->n, j->begin[0]);
+  context->next[job] = iter;
+  __builtin_prefetch(j->childAndVersion.child);
+  MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+}
+
+void iter(struct CheckAll *context, int64_t prevJob, int64_t job,
+          int64_t started, int64_t count) {
+  auto *j = context->inProgress + job;
+  if (j->childAndVersion.child == nullptr) {
+    auto c = getChildGeq(j->n, j->begin[0]);
+    if (c != nullptr) {
+      j->n = c;
+      context->next[job] = down_left_spine;
+      __builtin_prefetch(j->n);
+      MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+    } else {
+      j->n = nextSibling(j->n);
+      if (j->n == nullptr) {
+        j->setResult(true);
+        MUSTTAIL return complete(context, prevJob, job, started, count);
+      }
+      context->next[job] = down_left_spine;
+      __builtin_prefetch(j->n);
+      MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+    }
+  }
+
+  j->n = j->childAndVersion.child;
+  j->begin = j->begin.subspan(1, j->begin.size() - 1);
+
+  if (j->n->partialKeyLen > 0) {
+    int commonLen = std::min<int>(j->n->partialKeyLen, j->begin.size());
+    int i = longestCommonPrefix(j->n->partialKey(), j->begin.data(), commonLen);
+    if (i < commonLen) {
+      auto c = j->n->partialKey()[i] <=> j->begin[i];
+      if (c > 0) {
+        context->next[job] = down_left_spine;
+        MUSTTAIL return down_left_spine(context, prevJob, job, started, count);
+      } else {
+        j->n = nextSibling(j->n);
+        if (j->n == nullptr) {
+          j->setResult(true);
+          MUSTTAIL return complete(context, prevJob, job, started, count);
+        }
+        context->next[job] = down_left_spine;
+        __builtin_prefetch(j->n);
+        MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+      }
+    }
+    if (commonLen == j->n->partialKeyLen) {
+      // partial key matches
+      j->begin = j->begin.subspan(commonLen, j->begin.size() - commonLen);
+    } else if (j->n->partialKeyLen > int(j->begin.size())) {
+      // n is the first physical node greater than remaining, and there's no
+      // eq node
+      context->next[job] = down_left_spine;
+      MUSTTAIL return down_left_spine(context, prevJob, job, started, count);
+    }
+  }
+
+  if (j->childAndVersion.maxVersion <= j->readVersion) {
+    ++context->tls->point_read_short_circuit_accum;
+    j->setResult(true);
+    MUSTTAIL return complete(context, prevJob, job, started, count);
+  }
+
+  ++context->tls->point_read_iterations_accum;
+
+  if (j->begin.size() == 0) {
+    if (j->n->entryPresent) {
+      j->setResult(j->n->entry.pointVersion <= j->readVersion);
+      MUSTTAIL return complete(context, prevJob, job, started, count);
+    }
+    j->n = getFirstChildExists(j->n);
+    context->next[job] = down_left_spine;
+    __builtin_prefetch(j->n);
+    MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+  }
+
+  j->childAndVersion = getChildAndMaxVersion(j->n, j->begin[0]);
+  __builtin_prefetch(j->childAndVersion.child);
+  // j->next is already iter
+  MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+}
+
+void down_left_spine(struct CheckAll *context, int64_t prevJob, int64_t job,
+                     int64_t started, int64_t count) {
+  auto *j = context->inProgress + job;
+  if (j->n->entryPresent) {
+    j->setResult(j->n->entry.rangeVersion <= j->readVersion);
+    MUSTTAIL return complete(context, prevJob, job, started, count);
+  }
+  j->n = getFirstChildExists(j->n);
+  __builtin_prefetch(j->n);
+  // j->next is already down_left_spine
+  MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+}
+
+} // namespace check_point_read_state_machine
+
+continuation CheckJob::init(const ConflictSet::ReadRange *read,
+                            ConflictSet::Result *result, Node *root,
+                            int64_t oldestVersionFullPrecision,
+                            ReadContext *tls) {
+  auto begin = std::span<const uint8_t>(read->begin.p, read->begin.len);
+  auto end = std::span<const uint8_t>(read->end.p, read->end.len);
+  if (read->readVersion < oldestVersionFullPrecision) {
+    *result = ConflictSet::TooOld;
+    return complete;
+  } else if (end.size() == 0) {
+    this->begin = begin;
+    this->n = root;
+    this->readVersion = InternalVersionT(read->readVersion);
+    this->result = result;
+    return check_point_read_state_machine::begin;
+    // *result =
+    //     checkPointRead(root, begin, InternalVersionT(read->readVersion), tls)
+    //         ? ConflictSet::Commit
+    //         : ConflictSet::Conflict;
+    // return complete;
+  } else {
+    *result = checkRangeRead(root, begin, end,
+                             InternalVersionT(read->readVersion), tls)
+                  ? ConflictSet::Commit
+                  : ConflictSet::Conflict;
+    return complete;
+  }
+}
+
 struct __attribute__((visibility("hidden"))) ConflictSet::Impl {

  void check(const ReadRange *reads, Result *result, int count) {
+    assert(oldestVersionFullPrecision >=
+           newestVersionFullPrecision - kNominalVersionWindow);
+
+    if (count == 0) {
+      return;
+    }
+
    ReadContext tls;
    tls.impl = this;
    int64_t check_byte_accum = 0;
+
+    CheckAll context;
+    context.oldestVersionFullPrecision = oldestVersionFullPrecision;
+    context.queries = reads;
+    context.results = result;
+    context.root = root;
+    context.tls = &tls;
+
+    int64_t started = std::min(context.kConcurrent, count);
+    for (int i = 0; i < started; i++) {
+      context.next[i] = context.inProgress[i].init(
+          reads + i, result + i, root, oldestVersionFullPrecision, &tls);
+      context.nextJob[i] = i + 1;
+    }
+    context.nextJob[started - 1] = 0;
+    int prevJob = started - 1;
+    int job = 0;
+    context.next[job](&context, prevJob, job, started, count);
+
    for (int i = 0; i < count; ++i) {
      assert(reads[i].readVersion >= 0);
      assert(reads[i].readVersion <= newestVersionFullPrecision);
      const auto &r = reads[i];
      check_byte_accum += r.begin.len + r.end.len;
-      auto begin = std::span<const uint8_t>(r.begin.p, r.begin.len);
-      auto end = std::span<const uint8_t>(r.end.p, r.end.len);
-      assert(oldestVersionFullPrecision >=
-             newestVersionFullPrecision - kNominalVersionWindow);
-      result[i] =
-          reads[i].readVersion < oldestVersionFullPrecision ? TooOld
-          : (end.size() > 0
-                 ? checkRangeRead(root, begin, end,
-                                  InternalVersionT(reads[i].readVersion), &tls)
-                 : checkPointRead(root, begin,
-                                  InternalVersionT(reads[i].readVersion), &tls))
-              ? Commit
-              : Conflict;
      tls.commits_accum += result[i] == Commit;
      tls.conflicts_accum += result[i] == Conflict;
      tls.too_olds_accum += result[i] == TooOld;
    }
+
    point_read_total.add(tls.point_read_accum);
    prefix_read_total.add(tls.prefix_read_accum);
    range_read_total.add(tls.range_read_accum);
@@ -3242,7 +3400,6 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
                          0) *
        2;

-    memory_bytes.set(totalBytes);
    point_writes_total.add(tls.accum.point_writes);
    range_writes_total.add(tls.accum.range_writes);
    nodes_allocated_total.add(tls.accum.nodes_allocated);
@@ -3331,7 +3488,6 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
 #endif
    keyUpdates = gcScanStep(keyUpdates);

-    memory_bytes.set(totalBytes);
    nodes_allocated_total.add(std::exchange(tls.accum.nodes_allocated, 0));
    nodes_released_total.add(std::exchange(tls.accum.nodes_released, 0));
    entries_inserted_total.add(std::exchange(tls.accum.entries_inserted, 0));
@@ -3379,7 +3535,7 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
  explicit Impl(int64_t oldestVersion) {
    assert(oldestVersion >= 0);
    init(oldestVersion);
-    initMetrics();
+    metrics = initMetrics(metricsList, metricsCount);
  }
  ~Impl() {
    eraseTree(root, &tls);
@@ -3402,23 +3558,12 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {

  MetricsV1 *metrics;
  int metricsCount = 0;
-  void initMetrics() {
-    metrics = (MetricsV1 *)safe_malloc(metricsCount * sizeof(metrics[0]));
-    for (auto [i, m] = std::make_tuple(metricsCount - 1, metricList); i >= 0;
-         --i, m = m->prev) {
-      metrics[i].name = m->name;
-      metrics[i].help = m->help;
-      metrics[i].p = m;
-      metrics[i].type = m->type;
-    }
-  }
-
-  Metric *metricList = nullptr;
+  Metric *metricsList = nullptr;

 #define GAUGE(name, help)                                                      \
-  Gauge name { this, #name, help }
+  Gauge name { metricsList, metricsCount, #name, help }
 #define COUNTER(name, help)                                                    \
-  Counter name { this, #name, help }
+  Counter name { metricsList, metricsCount, #name, help }
  // ==================== METRICS DEFINITIONS ====================
  COUNTER(point_read_total, "Total number of point reads checked");
  COUNTER(point_read_short_circuit_total,
@@ -3484,13 +3629,6 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
  }
 };

-Metric::Metric(ConflictSet::Impl *impl, const char *name, const char *help,
-               ConflictSet::MetricsV1::Type type)
-    : prev(std::exchange(impl->metricList, this)), name(name), help(help),
-      type(type), value(0) {
-  ++impl->metricsCount;
-}
-
 Node *&getInTree(Node *n, ConflictSet::Impl *impl) {
  return n->parent == nullptr ? impl->root
                              : getChildExists(n->parent, n->parentsIndex);
@@ -3509,6 +3647,7 @@ void internal_addWrites(ConflictSet::Impl *impl,
  mallocBytesDelta = 0;
  impl->addWrites(writes, count, writeVersion);
  impl->totalBytes += mallocBytesDelta;
+  impl->memory_bytes.set(impl->totalBytes);
 #if SHOW_MEMORY
  if (impl->totalBytes != mallocBytes) {
    abort();
@@ -3520,6 +3659,7 @@ void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
  mallocBytesDelta = 0;
  impl->setOldestVersion(oldestVersion);
  impl->totalBytes += mallocBytesDelta;
+  impl->memory_bytes.set(impl->totalBytes);
 #if SHOW_MEMORY
  if (impl->totalBytes != mallocBytes) {
    abort();
@@ -3648,12 +3788,7 @@ ConflictSet::ConflictSet(ConflictSet &&other) noexcept
    : impl(std::exchange(other.impl, nullptr)) {}

 ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
-  if (this != &other) {
-    if (impl) {
-      internal_destroy(impl);
-    }
  impl = std::exchange(other.impl, nullptr);
-  }
  return *this;
 }

@@ -4078,6 +4213,72 @@ struct __attribute__((visibility("default"))) PeakPrinter {

 #ifdef ENABLE_MAIN

+#include "third_party/nanobench.h"
+
+template <int kN> void benchRezero() {
+  static_assert(kN % 16 == 0);
+  ankerl::nanobench::Bench bench;
+  InternalVersionT vs[kN];
+  InternalVersionT zero;
+  bench.run("rezero" + std::to_string(kN), [&]() {
+    bench.doNotOptimizeAway(vs);
+    bench.doNotOptimizeAway(zero);
+    for (int i = 0; i < kN; i += 16) {
+      rezero16(vs + i, zero);
+    }
+  });
+}
+
+template <int kN> void benchScan1() {
+  static_assert(kN % 16 == 0);
+  ankerl::nanobench::Bench bench;
+  InternalVersionT vs[kN];
+  uint8_t is[kN];
+  uint8_t begin;
+  uint8_t end;
+  InternalVersionT v;
+  bench.run("scan" + std::to_string(kN), [&]() {
+    bench.doNotOptimizeAway(vs);
+    bench.doNotOptimizeAway(is);
+    bench.doNotOptimizeAway(begin);
+    bench.doNotOptimizeAway(end);
+    bench.doNotOptimizeAway(v);
+    for (int i = 0; i < kN; i += 16) {
+      scan16</*kAVX512=*/true>(vs + i, is + i, begin, end, v);
+    }
+  });
+}
+
+template <int kN> void benchScan2() {
+  static_assert(kN % 16 == 0);
+  ankerl::nanobench::Bench bench;
+  InternalVersionT vs[kN];
+  uint8_t is[kN];
+  uint8_t begin;
+  uint8_t end;
+  InternalVersionT v;
+  bench.run("scan" + std::to_string(kN), [&]() {
+    bench.doNotOptimizeAway(vs);
+    bench.doNotOptimizeAway(begin);
+    bench.doNotOptimizeAway(end);
+    bench.doNotOptimizeAway(v);
+    for (int i = 0; i < kN; i += 16) {
+      scan16</*kAVX512=*/true>(vs + i, begin, end, v);
+    }
+  });
+}
+
+void benchLCP(int len) {
+  ankerl::nanobench::Bench bench;
+  std::vector<uint8_t> lhs(len);
+  std::vector<uint8_t> rhs(len);
+  bench.run("lcp " + std::to_string(len), [&]() {
+    bench.doNotOptimizeAway(lhs);
+    bench.doNotOptimizeAway(rhs);
+    bench.doNotOptimizeAway(longestCommonPrefix(lhs.data(), rhs.data(), len));
+  });
+}
+
 void printTree() {
  int64_t writeVersion = 0;
  ConflictSet::Impl cs{writeVersion};
@@ -4099,7 +4300,11 @@ void printTree() {
  debugPrintDot(stdout, cs.root, &cs);
 }

-int main(void) { printTree(); }
+int main(void) {
+  for (int i = 0; i < 256; ++i) {
+    benchLCP(i);
+  }
+}
 #endif

 #ifdef ENABLE_FUZZ
@@ -119,13 +119,7 @@ ConflictSet::ConflictSet(ConflictSet &&other) noexcept
    : impl(std::exchange(other.impl, nullptr)) {}

 ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
-  if (this != &other) {
-    if (impl) {
-      impl->~Impl();
-      safe_free(impl, sizeof(Impl));
-    }
  impl = std::exchange(other.impl, nullptr);
-  }
  return *this;
 }

@@ -0,0 +1,256 @@
+#include <alloca.h>
+#include <cassert>
+#ifdef __x86_64__
+#include <immintrin.h>
+#endif
+
+#include "third_party/nanobench.h"
+
+struct Job {
+  int *input;
+  // Returned void* is a function pointer to the next continuation. We have to
+  // use void* because otherwise the type would be recursive.
+  typedef void *(*continuation)(Job *);
+  continuation next;
+};
+
+void *stepJob(Job *j) {
+  auto done = --(*j->input) == 0;
+#ifdef __x86_64__
+  _mm_clflush(j->input);
+#endif
+  return done ? nullptr : (void *)stepJob;
+}
+
+void sequential(Job **jobs, int count) {
+  for (int i = 0; i < count; ++i) {
+    do {
+      jobs[i]->next = (Job::continuation)jobs[i]->next(jobs[i]);
+    } while (jobs[i]->next);
+  }
+}
+
+void sequentialNoFuncPtr(Job **jobs, int count) {
+  for (int i = 0; i < count; ++i) {
+    while (stepJob(jobs[i]))
+      ;
+  }
+}
+
+void interleaveSwapping(Job **jobs, int remaining) {
+  int current = 0;
+  while (remaining > 0) {
+    auto next = (Job::continuation)jobs[current]->next(jobs[current]);
+    jobs[current]->next = next;
+    if (next == nullptr) {
+      jobs[current] = jobs[remaining - 1];
+      --remaining;
+    } else {
+      ++current;
+    }
+    if (current == remaining) {
+      current = 0;
+    }
+  }
+}
+
+void interleaveBoundedCyclicList(Job **jobs, int count) {
+  if (count == 0) {
+    return;
+  }
+
+  constexpr int kConcurrent = 32;
+  Job *inProgress[kConcurrent];
+  int nextJob[kConcurrent];
+
+  int started = std::min(kConcurrent, count);
+  for (int i = 0; i < started; i++) {
+    inProgress[i] = jobs[i];
+    nextJob[i] = i + 1;
+  }
+  nextJob[started - 1] = 0;
+
+  int prevJob = started - 1;
+  int job = 0;
+  for (;;) {
+    auto next = (Job::continuation)inProgress[job]->next(inProgress[job]);
+    inProgress[job]->next = next;
+    if (next == nullptr) {
+      if (started == count) {
+        if (prevJob == job)
+          break;
+        nextJob[prevJob] = nextJob[job];
+        job = prevJob;
+      } else {
+        int temp = started++;
+        inProgress[job] = jobs[temp];
+      }
+    }
+    prevJob = job;
+    job = nextJob[job];
+  }
+}
+
+#ifndef __has_attribute
+#define __has_attribute(x) 0
+#endif
+
+#if __has_attribute(musttail)
+#define MUSTTAIL __attribute__((musttail))
+#else
+#define MUSTTAIL
+#endif
+
+struct Context {
+  constexpr static int kConcurrent = 32;
+  Job **jobs;
+  Job *inProgress[kConcurrent];
+  void (*continuation[kConcurrent])(Context *, int64_t prevJob, int64_t job,
+                                    int64_t started, int64_t count);
+  int nextJob[kConcurrent];
+};
+
+void keepGoing(Context *context, int64_t prevJob, int64_t job, int64_t started,
+               int64_t count) {
+  prevJob = job;
+  job = context->nextJob[job];
+  MUSTTAIL return context->continuation[job](context, prevJob, job, started,
+                                             count);
+}
+
+void stepJobTailCall(Context *context, int64_t prevJob, int64_t job,
+                     int64_t started, int64_t count);
+
+void complete(Context *context, int64_t prevJob, int64_t job, int64_t started,
+              int64_t count) {
+  if (started == count) {
+    if (prevJob == job) {
+      return;
+    }
+    context->nextJob[prevJob] = context->nextJob[job];
+    job = prevJob;
+  } else {
+    context->inProgress[job] = context->jobs[started++];
+    context->continuation[job] = stepJobTailCall;
+  }
+  prevJob = job;
+  job = context->nextJob[job];
+  MUSTTAIL return context->continuation[job](context, prevJob, job, started,
+                                             count);
+}
+
+void stepJobTailCall(Context *context, int64_t prevJob, int64_t job,
+                     int64_t started, int64_t count) {
+  auto *j = context->inProgress[job];
+  auto done = --(*j->input) == 0;
+#ifdef __x86_64__
+  _mm_clflush(j->input);
+#endif
+  if (done) {
+    MUSTTAIL return complete(context, prevJob, job, started, count);
+  } else {
+    context->continuation[job] = stepJobTailCall;
+    MUSTTAIL return keepGoing(context, prevJob, job, started, count);
+  }
+}
+
+void useTailCalls(Job **jobs, int count) {
+  if (count == 0) {
+    return;
+  }
+  Context context;
+  context.jobs = jobs;
+  int64_t started = std::min(Context::kConcurrent, count);
+  for (int i = 0; i < started; i++) {
+    context.inProgress[i] = jobs[i];
+    context.nextJob[i] = i + 1;
+    context.continuation[i] = stepJobTailCall;
+  }
+  context.nextJob[started - 1] = 0;
+  int prevJob = started - 1;
+  int job = 0;
+  return context.continuation[job](&context, prevJob, job, started, count);
+}
+
+void interleaveCyclicList(Job **jobs, int count) {
+  auto *nextJob = (int *)alloca(sizeof(int) * count);
+
+  for (int i = 0; i < count - 1; ++i) {
+    nextJob[i] = i + 1;
+  }
+  nextJob[count - 1] = 0;
+
+  int prevJob = count - 1;
+  int job = 0;
+  for (;;) {
+    auto next = (Job::continuation)jobs[job]->next(jobs[job]);
+    jobs[job]->next = next;
+    if (next == nullptr) {
+      if (prevJob == job)
+        break;
+      nextJob[prevJob] = nextJob[job];
+      job = prevJob;
+    }
+    prevJob = job;
+    job = nextJob[job];
+  }
+}
+
+int main() {
+  ankerl::nanobench::Bench bench;
+
+  constexpr int kNumJobs = 10000;
+  bench.relative(true);
+
+  Job jobs[kNumJobs];
+  Job jobsCopy[kNumJobs];
+  int iters = 0;
+  int originalInput[kNumJobs];
+  for (int i = 0; i < kNumJobs; ++i) {
+    originalInput[i] = rand() % 5 + 3;
+    jobs[i].input = new int{originalInput[i]};
+    jobs[i].next = stepJob;
+    iters += *jobs[i].input;
+  }
+  bench.batch(iters);
+
+  for (auto [scheduler, name] :
+       {std::make_pair(sequentialNoFuncPtr, "sequentialNoFuncPtr"),
+        std::make_pair(sequential, "sequential"),
+        std::make_pair(useTailCalls, "useTailCalls"),
+        std::make_pair(interleaveSwapping, "interleavingSwapping"),
+        std::make_pair(interleaveBoundedCyclicList,
+                       "interleaveBoundedCyclicList"),
+        std::make_pair(interleaveCyclicList, "interleaveCyclicList")}) {
+    for (int i = 0; i < kNumJobs; ++i) {
+      *jobs[i].input = originalInput[i];
+    }
+    memcpy(jobsCopy, jobs, sizeof(jobs));
+    Job *ps[kNumJobs];
+    for (int i = 0; i < kNumJobs; ++i) {
+      ps[i] = jobsCopy + i;
+    }
+    scheduler(ps, kNumJobs);
+    for (int i = 0; i < kNumJobs; ++i) {
+      if (*jobsCopy[i].input != 0) {
+        fprintf(stderr, "%s failed\n", name);
+        abort();
+      }
+    }
+
+    bench.run(name, [&]() {
+      for (int i = 0; i < kNumJobs; ++i) {
+        *jobs[i].input = originalInput[i];
+      }
+      memcpy(jobsCopy, jobs, sizeof(jobs));
+      Job *ps[kNumJobs];
+      for (int i = 0; i < kNumJobs; ++i) {
+        ps[i] = jobsCopy + i;
+      }
+      scheduler(ps, kNumJobs);
+    });
+  }
+  for (int i = 0; i < kNumJobs; ++i) {
+    delete jobs[i].input;
+  }
+}
@@ -129,7 +129,7 @@ pipeline {
            }
            steps {
                script {
-                    filter_args = "-f ConflictSet.cpp -f LongestCommonPrefix.h"
+                    filter_args = "-f ConflictSet.cpp -f LongestCommonPrefix.h -f Metrics.h"
                }
                CleanBuildAndTest("-DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_FLAGS=--coverage -DCMAKE_CXX_FLAGS=--coverage -DCMAKE_BUILD_TYPE=Debug -DDISABLE_TSAN=ON")
                sh """
@@ -129,7 +129,7 @@ longestCommonPrefix(const uint8_t *ap, const uint8_t *bp, int cl) {
  }

  int i = 0;
-  int end; // GCOVR_EXCL_LINE
+  int end;

  // kStride * kUnrollCount at a time
  end = cl & ~(kStride * kUnrollFactor - 1);
@@ -0,0 +1,64 @@
+#pragma once
+
+#include "ConflictSet.h"
+#include "Internal.h"
+
+#include <assert.h>
+#include <atomic>
+#include <tuple>
+
+struct Metric {
+  Metric *prev;
+  const char *name;
+  const char *help;
+  weaselab::ConflictSet::MetricsV1::Type type;
+  std::atomic<int64_t> value;
+
+protected:
+  Metric(Metric *&metricList, int &metricsCount, const char *name,
+         const char *help, weaselab::ConflictSet::MetricsV1::Type type)
+      : prev(std::exchange(metricList, this)), name(name), help(help),
+        type(type), value(0) {
+    ++metricsCount;
+  }
+};
+
+struct Gauge : private Metric {
+  Gauge(Metric *&metricList, int &metricsCount, const char *name,
+        const char *help)
+      : Metric(metricList, metricsCount, name, help,
+               weaselab::ConflictSet::MetricsV1::Gauge) {}
+
+  void set(int64_t value) {
+    this->value.store(value, std::memory_order_relaxed);
+  }
+};
+
+struct Counter : private Metric {
+  Counter(Metric *&metricList, int &metricsCount, const char *name,
+          const char *help)
+      : Metric(metricList, metricsCount, name, help,
+               weaselab::ConflictSet::MetricsV1::Counter) {}
+  // Expensive. Accumulate locally and then call add instead of repeatedly
+  // calling add.
+  void add(int64_t value) {
+    assert(value >= 0);
+    static_assert(std::atomic<int64_t>::is_always_lock_free);
+    this->value.fetch_add(value, std::memory_order_relaxed);
+  }
+};
+
+inline weaselab::ConflictSet::MetricsV1 *initMetrics(Metric *metricsList,
+                                                     int metricsCount) {
+  weaselab::ConflictSet::MetricsV1 *metrics =
+      (weaselab::ConflictSet::MetricsV1 *)safe_malloc(metricsCount *
+                                                      sizeof(metrics[0]));
+  for (auto [i, m] = std::make_tuple(metricsCount - 1, metricsList); i >= 0;
+       --i, m = m->prev) {
+    metrics[i].name = m->name;
+    metrics[i].help = m->help;
+    metrics[i].p = m;
+    metrics[i].type = m->type;
+  }
+  return metrics;
+}
@@ -2,16 +2,7 @@ A data structure for optimistic concurrency control on ranges of bitwise-lexicog

 Intended as an alternative to FoundationDB's skip list.

-Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM.
-
-```
-$ clang++ --version
-
-Ubuntu clang version 21.1.8 (6ubuntu1)
-Target: x86_64-pc-linux-gnu
-Thread model: posix
-InstalledDir: /usr/lib/llvm-21/bin
-```
+Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM

 # Microbenchmark

@@ -19,45 +10,44 @@ InstalledDir: /usr/lib/llvm-21/bin

 |               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
-|              171.37 |        5,835,467.28 |    0.5% |        2,915.46 |          629.31 |  4.633 |         502.91 |    0.0% |      0.01 | `point reads`
-|              168.16 |        5,946,774.62 |    0.5% |        2,859.73 |          619.03 |  4.620 |         488.59 |    0.0% |      0.01 | `prefix reads`
-|              245.86 |        4,067,294.63 |    0.3% |        3,508.97 |          904.82 |  3.878 |         627.73 |    0.0% |      0.01 | `range reads`
-|              381.14 |        2,623,691.76 |    0.2% |        5,127.16 |        1,402.60 |  3.655 |         830.59 |    1.7% |      0.01 | `point writes`
-|              376.24 |        2,657,887.07 |    0.3% |        5,062.85 |        1,381.07 |  3.666 |         811.68 |    1.6% |      0.01 | `prefix writes`
-|              257.09 |        3,889,688.44 |    3.8% |        3,010.09 |          949.42 |  3.170 |         521.34 |    2.9% |      0.01 | `range writes`
-|              629.71 |        1,588,026.28 |    4.0% |        6,809.60 |        2,320.32 |  2.935 |       1,247.08 |    1.3% |      0.01 | `monotonic increasing point writes`
-|          126,164.50 |            7,926.16 |    1.7% |      772,460.25 |      463,264.67 |  1.667 |     141,512.00 |    0.2% |      0.01 | `worst case for radix tree`
-|               15.33 |       65,217,669.23 |    0.4% |          299.00 |           56.45 |  5.297 |          64.00 |    0.0% |      0.01 | `create and destroy`
+|              172.03 |        5,812,791.77 |    0.4% |        3,130.62 |          879.00 |  3.562 |         509.23 |    0.0% |      0.01 | `point reads`
+|              167.44 |        5,972,130.71 |    0.2% |        3,065.14 |          862.27 |  3.555 |         494.30 |    0.0% |      0.01 | `prefix reads`
+|              238.77 |        4,188,130.84 |    0.9% |        3,589.93 |        1,259.30 |  2.851 |         637.12 |    0.0% |      0.01 | `range reads`
+|              424.01 |        2,358,426.70 |    0.2% |        5,620.05 |        2,242.35 |  2.506 |         854.80 |    1.7% |      0.01 | `point writes`
+|              418.45 |        2,389,780.56 |    0.4% |        5,525.07 |        2,211.05 |  2.499 |         831.71 |    1.7% |      0.01 | `prefix writes`
+|              254.87 |        3,923,568.88 |    2.6% |        3,187.01 |        1,366.50 |  2.332 |         529.11 |    2.7% |      0.02 | `range writes`
+|              675.96 |        1,479,374.50 |    3.3% |        7,735.41 |        3,468.60 |  2.230 |       1,386.02 |    1.8% |      0.01 | `monotonic increasing point writes`
+|          137,986.20 |            7,247.10 |    0.6% |      789,752.33 |      699,462.00 |  1.129 |     144,824.14 |    0.0% |      0.01 | `worst case for radix tree`
+|               21.63 |       46,231,564.03 |    1.0% |          448.00 |          107.14 |  4.181 |          84.00 |    0.0% |      0.01 | `create and destroy`

 ## Radix tree (this implementation)

-
 |               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
 |--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
-|               12.81 |       78,088,759.03 |    0.7% |          187.38 |           47.15 |  3.974 |          33.70 |    0.5% |      0.01 | `point reads`
-|               16.77 |       59,626,501.14 |    0.4% |          277.61 |           61.74 |  4.497 |          44.76 |    0.4% |      0.01 | `prefix reads`
-|               42.45 |       23,556,818.87 |    0.4% |          856.96 |          156.09 |  5.490 |         126.22 |    0.2% |      0.01 | `range reads`
-|               15.76 |       63,453,196.94 |    0.3% |          249.89 |           58.04 |  4.306 |          33.21 |    0.5% |      0.01 | `point writes`
-|               29.71 |       33,662,499.39 |    0.1% |          509.14 |          109.33 |  4.657 |          70.26 |    0.5% |      0.01 | `prefix writes`
-|               35.77 |       27,956,388.03 |    0.8% |          614.18 |          131.72 |  4.663 |          85.57 |    0.1% |      0.01 | `range writes`
-|               80.13 |       12,480,456.47 |    1.5% |        1,279.04 |          294.90 |  4.337 |         234.68 |    0.1% |      0.01 | `monotonic increasing point writes`
-|          360,748.00 |            2,772.02 |    2.6% |    4,401,835.00 |    1,318,754.00 |  3.338 |     728,537.00 |    0.0% |      0.01 | `worst case for radix tree`
-|               90.69 |       11,027,103.45 |    1.0% |        1,686.00 |          333.80 |  5.051 |         284.00 |    0.0% |      0.01 | `create and destroy`
+|               12.88 |       77,653,350.77 |    0.5% |          185.37 |           64.45 |  2.876 |          41.51 |    0.4% |      0.01 | `point reads`
+|               14.67 |       68,179,354.49 |    0.1% |          271.44 |           73.40 |  3.698 |          53.70 |    0.3% |      0.01 | `prefix reads`
+|               34.84 |       28,701,444.36 |    0.3% |          715.74 |          175.27 |  4.084 |         127.30 |    0.2% |      0.01 | `range reads`
+|               17.12 |       58,422,988.28 |    0.2% |          314.30 |           86.11 |  3.650 |          39.82 |    0.4% |      0.01 | `point writes`
+|               31.42 |       31,830,804.65 |    0.1% |          591.06 |          158.07 |  3.739 |          82.67 |    0.2% |      0.01 | `prefix writes`
+|               37.37 |       26,759,432.70 |    2.2% |          681.98 |          188.95 |  3.609 |          96.10 |    0.1% |      0.01 | `range writes`
+|               76.72 |       13,035,140.63 |    2.3% |        1,421.28 |          387.17 |  3.671 |         257.76 |    0.1% |      0.01 | `monotonic increasing point writes`
+|          297,452.00 |            3,361.89 |    0.9% |    3,508,083.00 |    1,500,834.67 |  2.337 |     727,525.33 |    0.1% |      0.01 | `worst case for radix tree`
+|               87.70 |       11,402,490.60 |    1.0% |        1,795.00 |          442.09 |  4.060 |         297.00 |    0.0% |      0.01 | `create and destroy`

 # "Real data" test

-Point queries only. Gc ratio is the ratio of time spent doing garbage collection to time spent adding writes or doing garbage collection. Lower is better.
+Point queries only, best of three runs. Gc ratio is the ratio of time spent doing garbage collection to time spent adding writes or doing garbage collection. Lower is better.

 ## skip list

 ```
-Check: 4.42779 seconds, 368.253 MB/s, Add: 4.64942 seconds, 120.586 MB/s, Gc ratio: 28.8784%, Peak idle memory: 5.51735e+06
+Check: 4.47891 seconds, 364.05 MB/s, Add: 4.55599 seconds, 123.058 MB/s, Gc ratio: 37.1145%
 ```

 ## radix tree

 ```
-Check: 0.891163 seconds, 1829.69 MB/s, Add: 1.23064 seconds, 455.579 MB/s, Gc ratio: 42.9417%, Peak idle memory: 2.28333e+06
+Check: 0.953012 seconds, 1710.94 MB/s, Add: 1.30025 seconds, 431.188 MB/s, Gc ratio: 43.9816%, Peak idle memory: 2.28375e+06
 ```

 ## hash table
@@ -65,6 +55,5 @@ Check: 0.891163 seconds, 1829.69 MB/s, Add: 1.23064 seconds, 455.579 MB/s, Gc ra
 (The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be)

 ```
-Check: 0.858484 seconds, 1899.33 MB/s, Add: 0.632886 seconds, 885.868 MB/s, Gc ratio: 41.252%, Peak idle memory: 0
+Check: 0.804094 seconds, 2027.81 MB/s, Add: 0.652952 seconds, 858.645 MB/s, Gc ratio: 35.3885%
 ```
-
@@ -1,4 +1,5 @@
 #include <atomic>
+#include <cstdint>
 #include <errno.h>
 #include <netdb.h>
 #include <stdio.h>
@@ -21,78 +22,55 @@

 std::atomic<int64_t> transactions;

-constexpr int kBaseSearchDepth = 32;
 constexpr int kWindowSize = 10000000;

-std::string numToKey(int64_t num) {
+constexpr int kNumPrefixes = 250000;
+
+std::string makeKey(int64_t num, int suffixLen) {
  std::string result;
-  result.resize(kBaseSearchDepth + sizeof(int64_t));
-  memset(result.data(), 0, kBaseSearchDepth);
+  result.resize(sizeof(int64_t) + suffixLen);
  int64_t be = __builtin_bswap64(num);
-  memcpy(result.data() + kBaseSearchDepth, &be, sizeof(int64_t));
+  memcpy(result.data(), &be, sizeof(int64_t));
+  memset(result.data() + sizeof(int64_t), 0, suffixLen);
  return result;
 }

 void workload(weaselab::ConflictSet *cs) {
  int64_t version = kWindowSize;
-  cs->addWrites(nullptr, 0, version);
-  for (;; transactions.fetch_add(1, std::memory_order_relaxed)) {
-    // Reads
-    {
-      auto beginK = numToKey(version - kWindowSize);
-      auto endK = numToKey(version - 1);
-      auto pointRv = version - kWindowSize + rand() % kWindowSize + 1;
-      auto pointK = numToKey(pointRv);
-      weaselab::ConflictSet::ReadRange reads[] = {
-          {
-              {(const uint8_t *)pointK.data(), int(pointK.size())},
-              {nullptr, 0},
-              pointRv,
-          },
-          {
-              {(const uint8_t *)beginK.data(), int(beginK.size())},
-              {(const uint8_t *)endK.data(), int(endK.size())},
-              version - 2,
-          },
-      };
-      weaselab::ConflictSet::Result result[sizeof(reads) / sizeof(reads[0])];
-      cs->check(reads, result, sizeof(reads) / sizeof(reads[0]));
-      // for (int i = 0; i < sizeof(reads) / sizeof(reads[0]); ++i) {
-      //   if (result[i] != weaselab::ConflictSet::Commit) {
-      //     fprintf(stderr, "Unexpected conflict: [%s, %s) @ %" PRId64 "\n",
-      //             printable(reads[i].begin).c_str(),
-      //             printable(reads[i].end).c_str(), reads[i].readVersion);
-      //     abort();
-      //   }
-      // }
-    }
-    // Writes
-    {
-      weaselab::ConflictSet::WriteRange w;
-      auto k = numToKey(version);
-      w.begin.p = (const uint8_t *)k.data();
-      w.end.len = 0;
-      if (version % (kWindowSize / 2) == 0) {
-        for (int l = 0; l <= k.size(); ++l) {
-          w.begin.len = l;
-          cs->addWrites(&w, 1, version);
-        }
-      } else {
-        w.begin.len = k.size();
-        cs->addWrites(&w, 1, version);
-        int64_t beginN = version - kWindowSize + rand() % kWindowSize;
-        auto b = numToKey(beginN);
-        auto e = numToKey(beginN + 1000);
-        w.begin.p = (const uint8_t *)b.data();
-        w.begin.len = b.size();
-        w.end.p = (const uint8_t *)e.data();
-        w.end.len = e.size();
-        cs->addWrites(&w, 1, version);
+  for (int i = 0; i < kNumPrefixes; ++i) {
+    for (int j = 0; j < 50; ++j) {
+      weaselab::ConflictSet::WriteRange wr;
+      auto k = makeKey(i, j);
+      wr.begin.p = (const uint8_t *)k.data();
+      wr.begin.len = k.size();
+      wr.end.len = 0;
+      cs->addWrites(&wr, 1, version);
    }
  }
-    // GC
-    cs->setOldestVersion(version - kWindowSize);
  ++version;
+  for (int i = 0; i < kNumPrefixes; ++i) {
+    weaselab::ConflictSet::WriteRange wr;
+    auto k = makeKey(i, 50);
+    wr.begin.p = (const uint8_t *)k.data();
+    wr.begin.len = k.size();
+    wr.end.len = 0;
+    cs->addWrites(&wr, 1, version);
+  }
+
+  std::vector<weaselab::ConflictSet::Result> results(10);
+  for (;; transactions.fetch_add(1, std::memory_order_relaxed)) {
+    std::vector<std::string> keys(10);
+    for (auto &k : keys) {
+      k = makeKey(rand() % kNumPrefixes, 49);
+    }
+    std::vector<weaselab::ConflictSet::ReadRange> reads(10);
+    for (int i = 0; i < reads.size(); ++i) {
+      reads[i].begin.p = (const uint8_t *)(keys[i].data());
+      reads[i].begin.len = keys[i].size();
+      reads[i].end.len = 0;
+      reads[i].readVersion = version - 1;
+    }
+    cs->check(reads.data(), results.data(), 10);
  }
 }

@@ -22,6 +22,7 @@

 #include "ConflictSet.h"
 #include "Internal.h"
+#include "Metrics.h"

 #include <algorithm>
 #include <span>
@@ -434,13 +435,14 @@ public:
    return result;
  }

-  void detectConflicts(ReadConflictRange *ranges, int count,
+  // Return number of iterations of main loop
+  int detectConflicts(ReadConflictRange *ranges, int count,
                      ConflictSet::Result *transactionConflictStatus) const {
    const int M = 16;
    int nextJob[M];
    CheckMax inProgress[M];
    if (!count)
-      return;
+      return 0;

    int started = std::min(M, count);
    for (int i = 0; i < started; i++) {
@@ -451,8 +453,9 @@ public:

    int prevJob = started - 1;
    int job = 0;
+    int iters = 0;
    // vtune: 340 parts
-    while (true) {
+    for (;; ++iters) {
      if (inProgress[job].advance()) {
        if (started == count) {
          if (prevJob == job)
@@ -468,6 +471,7 @@ public:
      prevJob = job;
      job = nextJob[job];
    }
+    return iters;
  }

  void find(const StringRef *values, Finger *results, int *temp, int count) {
@@ -702,15 +706,27 @@ private:
  };
 };

+struct ReadContext {
+  int64_t commits_accum = 0;
+  int64_t conflicts_accum = 0;
+  int64_t too_olds_accum = 0;
+  int64_t check_bytes_accum = 0;
+};
+
 struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
  Impl(int64_t oldestVersion)
      : oldestVersion(oldestVersion), newestVersion(oldestVersion),
-        skipList(oldestVersion) {}
+        skipList(oldestVersion) {
+    metrics = initMetrics(metricsList, metricsCount);
+  }
+  ~Impl() { safe_free(metrics, metricsCount * sizeof(metrics[0])); }
  void check(const ConflictSet::ReadRange *reads, ConflictSet::Result *results,
-             int count) const {
+             int count) {
+    ReadContext tls;
    Arena arena;
    auto *ranges = new (arena) ReadConflictRange[count];
    for (int i = 0; i < count; ++i) {
+      tls.check_bytes_accum += reads[i].begin.len + reads[i].end.len;
      ranges[i].begin = {reads[i].begin.p, size_t(reads[i].begin.len)};
      ranges[i].end = reads[i].end.len > 0
                          ? StringRef{reads[i].end.p, size_t(reads[i].end.len)}
@@ -718,13 +734,22 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
      ranges[i].version = reads[i].readVersion;
      results[i] = ConflictSet::Commit;
    }
-    skipList.detectConflicts(ranges, count, results);
+    int iters = skipList.detectConflicts(ranges, count, results);
    for (int i = 0; i < count; ++i) {
      if (reads[i].readVersion < oldestVersion ||
          reads[i].readVersion < newestVersion - 2e9) {
        results[i] = TooOld;
      }
+      tls.commits_accum += results[i] == Commit;
+      tls.conflicts_accum += results[i] == Conflict;
+      tls.too_olds_accum += results[i] == TooOld;
    }
+    range_read_iterations_total.add(iters);
+    range_read_total.add(count);
+    commits_total.add(tls.commits_accum);
+    conflicts_total.add(tls.conflicts_accum);
+    too_olds_total.add(tls.too_olds_accum);
+    check_bytes_total.add(tls.check_bytes_accum);
  }

  void addWrites(const ConflictSet::WriteRange *writes, int count,
@@ -788,6 +813,9 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {
  }

  void setOldestVersion(int64_t oldestVersion) {
+    // This isn't 100% accurate. It overcounts if you hit the end
+    gc_iterations_total.add(keyUpdates);
+
    assert(oldestVersion >= this->oldestVersion);
    this->oldestVersion = oldestVersion;
    SkipList::Finger finger;
@@ -802,6 +830,54 @@ struct __attribute__((visibility("hidden"))) ConflictSet::Impl {

  int64_t totalBytes = 0;

+  MetricsV1 *metrics;
+  int metricsCount = 0;
+  Metric *metricsList = nullptr;
+
+#define GAUGE(name, help)                                                      \
+  Gauge name { metricsList, metricsCount, #name, help }
+#define COUNTER(name, help)                                                    \
+  Counter name { metricsList, metricsCount, #name, help }
+  // ==================== METRICS DEFINITIONS ====================
+  COUNTER(range_read_total, "Total number of range reads checked");
+  COUNTER(range_read_iterations_total,
+          "Total number of iterations of the main loops for range read checks");
+  COUNTER(commits_total,
+          "Total number of checks where the result is \"commit\"");
+  COUNTER(conflicts_total,
+          "Total number of checks where the result is \"conflict\"");
+  COUNTER(too_olds_total,
+          "Total number of checks where the result is \"too old\"");
+  COUNTER(check_bytes_total, "Total number of key bytes checked");
+  GAUGE(memory_bytes, "Total number of bytes in use");
+  COUNTER(nodes_allocated_total,
+          "The total number of physical tree nodes allocated");
+  COUNTER(nodes_released_total,
+          "The total number of physical tree nodes released");
+  COUNTER(insert_iterations_total,
+          "The total number of iterations of the main loop for insertion. "
+          "Includes searches where the entry already existed, and so insertion "
+          "did not take place");
+  COUNTER(entries_inserted_total,
+          "The total number of entries inserted in the tree");
+  COUNTER(entries_erased_total,
+          "The total number of entries erased from the tree");
+  COUNTER(
+      gc_iterations_total,
+      "The total number of iterations of the main loop for garbage collection");
+  COUNTER(write_bytes_total, "Total number of key bytes in calls to addWrites");
+  GAUGE(oldest_version,
+        "The lowest version that doesn't result in \"TooOld\" for checks");
+  GAUGE(newest_version, "The version of the most recent call to addWrites");
+  // ==================== END METRICS DEFINITIONS ====================
+#undef GAUGE
+#undef COUNTER
+
+  void getMetricsV1(MetricsV1 **metrics, int *count) {
+    *metrics = this->metrics;
+    *count = metricsCount;
+  }
+
 private:
  int64_t keyUpdates = 0;
  Arena removalArena;
@@ -824,6 +900,7 @@ void internal_addWrites(ConflictSet::Impl *impl,
  mallocBytesDelta = 0;
  impl->addWrites(writes, count, writeVersion);
  impl->totalBytes += mallocBytesDelta;
+  impl->memory_bytes.set(impl->totalBytes);
 #if SHOW_MEMORY
  if (impl->totalBytes != mallocBytes) {
    abort();
@@ -835,6 +912,7 @@ void internal_setOldestVersion(ConflictSet::Impl *impl, int64_t oldestVersion) {
  mallocBytesDelta = 0;
  impl->setOldestVersion(oldestVersion);
  impl->totalBytes += mallocBytesDelta;
+  impl->memory_bytes.set(impl->totalBytes);
 #if SHOW_MEMORY
  if (impl->totalBytes != mallocBytes) {
    abort();
@@ -858,12 +936,11 @@ int64_t internal_getBytes(ConflictSet::Impl *impl) { return impl->totalBytes; }

 void internal_getMetricsV1(ConflictSet::Impl *impl,
                           ConflictSet::MetricsV1 **metrics, int *count) {
-  *metrics = nullptr;
-  *count = 0;
+  return impl->getMetricsV1(metrics, count);
 }

 double internal_getMetricValue(const ConflictSet::MetricsV1 *metric) {
-  return 0;
+  return ((Metric *)metric->p)->value.load(std::memory_order_relaxed);
 }

 void ConflictSet::check(const ReadRange *reads, Result *results,
@@ -903,12 +980,7 @@ ConflictSet::ConflictSet(ConflictSet &&other) noexcept
    : impl(std::exchange(other.impl, nullptr)) {}

 ConflictSet &ConflictSet::operator=(ConflictSet &&other) noexcept {
-  if (this != &other) {
-    if (impl) {
-      internal_destroy(impl);
-    }
  impl = std::exchange(other.impl, nullptr);
-  }
  return *this;
 }

--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
andrew	ed67486077	Reordering seems to improve codegen Tests / Clang total: 3244, passed: 3244 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 3244, passed: 3244 Details Tests / Debug total: 3242, passed: 3242 Details Tests / SIMD fallback total: 3244, passed: 3244 Details Tests / Release [gcc] total: 3244, passed: 3244 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2419, passed: 2419 Details Tests / Coverage total: 2437, passed: 2437 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 98.98% (1938/1958) * Branch Coverage: 68.67% (1497/2180) * Complexity Density: 0.00 * Lines of Code: 1958 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head There was a failure building this commit Details	2024-09-23 15:28:51 -07:00
andrew	b376f6fdd5	WIP Tests / Clang total: 3244, passed: 3244 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 3244, passed: 3244 Details Tests / Debug total: 3242, passed: 3242 Details Tests / SIMD fallback total: 3244, passed: 3244 Details Tests / Release [gcc] total: 3244, passed: 3244 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2419, passed: 2419 Details Tests / Coverage total: 2437, passed: 2437 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 98.98% (1938/1958) * Branch Coverage: 68.67% (1497/2180) * Complexity Density: 0.00 * Lines of Code: 1958 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head There was a failure building this commit Details	2024-09-23 15:11:48 -07:00
andrew	6de63dd3fe	Use preserve_none and put continuation array in CheckAll	2024-09-23 14:53:16 -07:00
andrew	3e5f13bf54	WIP - tests pass	2024-09-23 13:32:56 -07:00
andrew	e7e1d1f7f5	Add tail-call based interleaving approach	2024-09-23 12:52:30 -07:00
andrew	442658e983	Target ~1GB memory usage in server bench	2024-09-21 14:28:15 -07:00
andrew	26f602215e	Accentuate cache misses for point reads in server_bench	2024-09-14 22:42:40 -07:00
andrew	98236f81cb	Add missing __builtin_prefetch	2024-09-14 22:41:58 -07:00
andrew	3593b72880	Disallow checking SIM_CACHE_MISSES=1	2024-09-10 22:23:37 -07:00
andrew	814aac4ea7	Experiment with causing cache misses	2024-09-10 22:06:00 -07:00
andrew	0550fa0016	Add "iter" state	2024-09-10 17:22:10 -07:00
andrew	fe5cfb0336	Remove redundant cast	2024-09-10 17:06:45 -07:00
andrew	82203515a0	check_point_read_state_machine::down_left_spine	2024-09-10 17:05:09 -07:00
andrew	465372c734	Scaffolding to prepare for interleaving checks	2024-09-10 16:10:57 -07:00
andrew	867136ff1b	Return pointer to next function	2024-09-09 21:59:49 -07:00
andrew	4b8f7320d3	Call function pointer in job	2024-09-09 21:00:31 -07:00
andrew	6628092384	Tinker with interleaveBoundedCyclicList	2024-09-09 20:25:40 -07:00
andrew	a0a4f1afea	Only compile nanobench once	2024-09-09 20:10:55 -07:00
andrew	ca479c03ce	Induce a cache miss in interleaving test	2024-09-09 17:55:35 -07:00
andrew	0a2e133ab9	Add InterleavingTest to explore #5	2024-09-09 17:27:58 -07:00
andrew	b0b31419b0	Remove vestigial emscripten from cmake Tests / Clang total: 3244, passed: 3244 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 3244, passed: 3244 Details Tests / Debug total: 3242, passed: 3242 Details Tests / SIMD fallback total: 3244, passed: 3244 Details Tests / Release [gcc] total: 3244, passed: 3244 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2419, passed: 2419 Details Tests / Coverage total: 2437, passed: 2437 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 99.23% (1803/1817) * Branch Coverage: 68.36% (1426/2086) * Complexity Density: 0.00 * Lines of Code: 1817 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head This commit looks good Details	2024-09-06 20:29:59 -07:00
andrew	5c0cc1edf5	Add longestCommonPrefix bench Tests / Clang total: 3244, passed: 3244 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 3244, passed: 3244 Details Tests / Debug total: 3242, passed: 3242 Details Tests / SIMD fallback total: 3244, passed: 3244 Details Tests / Release [gcc] total: 3244, passed: 3244 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2419, passed: 2419 Details Tests / Coverage total: 2437, passed: 2437 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 99.23% (1803/1817) * Branch Coverage: 68.36% (1426/2086) * Complexity Density: 0.00 * Lines of Code: 1817 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head This commit looks good Details	2024-09-05 16:30:58 -07:00
andrew	de47aa53b0	Add to corpus	2024-09-05 16:24:41 -07:00
andrew	56893f9702	Update stale comments Tests / Clang total: 2879, passed: 2879 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 2879, passed: 2879 Details Tests / Debug total: 2877, passed: 2877 Details Tests / SIMD fallback total: 2879, passed: 2879 Details Tests / Release [gcc] total: 2879, passed: 2879 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2146, passed: 2146 Details Tests / Coverage total: 2163, passed: 2163 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 99.23% (1803/1817) * Branch Coverage: 68.36% (1426/2086) * Complexity Density: 0.00 * Lines of Code: 1817 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head This commit looks good Details	2024-09-05 10:06:02 -07:00
andrew	e2234be10f	Update README.md	2024-09-04 13:08:20 -07:00
andrew	ce853680f2	Finish checkRangeRightSide conversion	2024-09-04 12:36:43 -07:00
andrew	5c39c1d64f	CheckRangeRightSide conversion WIP	2024-09-04 12:25:16 -07:00
andrew	55b73c8ddb	Convert CheckRangeRightSide WIP	2024-09-04 12:22:36 -07:00
andrew	b9503f8258	Converting CheckRangeRightSide WIP	2024-09-04 12:19:53 -07:00
andrew	c4c4531bd3	Simplify	2024-09-04 12:12:54 -07:00
andrew	2037d37c66	checkRangeLeftSide	2024-09-04 12:11:16 -07:00
andrew	6fe6a244af	Get CheckRangeLeftSide closer, WIP	2024-09-04 12:08:16 -07:00
andrew	8a4b370e2a	Make CheckRange{Left,Right}Side into normal functions, WIP	2024-09-04 12:04:58 -07:00
andrew	394f09f9fb	WIP adding metrics to skiplist	2024-08-30 16:06:43 -07:00
andrew	5e06a30357	Add optional microbenchmarks for rezero/scan avx512 Tests / Clang total: 2879, passed: 2879 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 2879, passed: 2879 Details Tests / Debug total: 2877, passed: 2877 Details Tests / SIMD fallback total: 2879, passed: 2879 Details Tests / Release [gcc] total: 2879, passed: 2879 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2146, passed: 2146 Details Tests / Coverage total: 2163, passed: 2163 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 98.92% (1834/1854) * Branch Coverage: 66.85% (1496/2238) * Complexity Density: 0.00 * Lines of Code: 1854 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head This commit looks good Details	2024-08-29 12:12:24 -07:00
andrew	cb6e4292f2	Add to corpus Tests / Clang total: 2879, passed: 2879 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 2879, passed: 2879 Details Tests / Debug total: 2877, passed: 2877 Details Tests / SIMD fallback total: 2879, passed: 2879 Details Tests / Release [gcc] total: 2879, passed: 2879 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2146, passed: 2146 Details Tests / Coverage total: 2163, passed: 2163 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 98.92% (1834/1854) * Branch Coverage: 66.85% (1496/2238) * Complexity Density: 0.00 * Lines of Code: 1854 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head This commit looks good Details	2024-08-26 14:33:16 -07:00
andrew	154a48ded0	Bump version Tests / Clang total: 2843, passed: 2843 Details Clang \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / 64 bit versions total: 2843, passed: 2843 Details Tests / Debug total: 2841, passed: 2841 Details Tests / SIMD fallback total: 2843, passed: 2843 Details Tests / Release [gcc] total: 2843, passed: 2843 Details GNU C Compiler (gcc) \|Total\|New\|Outstanding\|Fixed\|Trend \|:-:\|:-:\|:-:\|:-:\|:-: \|0\|0\|0\|0\|:clap: Details Tests / Release [gcc,aarch64] total: 2119, passed: 2119 Details Tests / Coverage total: 2136, passed: 2136 Details Code Coverage #### Project Overview No changes detected, that affect the code coverage. * Line Coverage: 98.92% (1834/1854) * Branch Coverage: 66.85% (1496/2238) * Complexity Density: 0.00 * Lines of Code: 1854 #### Quality Gates Summary Output truncated. Details weaselab/conflict-set/pipeline/head This commit looks good Details	2024-08-26 14:25:13 -07:00