Opt-in to rpm/deb default package filenames

Closes #13

Bench.cpp

@@ -34,7 +34,7 @@ ConflictSet::ReadRange singleton(Arena &arena, std::span<const uint8_t> key) {
       std::span<uint8_t>(new (arena) uint8_t[key.size() + 1], key.size() + 1);
   memcpy(r.data(), key.data(), key.size());
   r[key.size()] = 0;
-  return {key.data(), int(key.size()), r.data(), int(r.size())};
+  return {{key.data(), int(key.size())}, {r.data(), int(r.size())}, 0};
 }
 
 ConflictSet::ReadRange prefixRange(Arena &arena, std::span<const uint8_t> key) {
@@ -52,7 +52,7 @@ ConflictSet::ReadRange prefixRange(Arena &arena, std::span<const uint8_t> key) {
   auto r = std::span<uint8_t>(new (arena) uint8_t[index + 1], index + 1);
   memcpy(r.data(), key.data(), index + 1);
   r[r.size() - 1]++;
-  return {key.data(), int(key.size()), r.data(), int(r.size())};
+  return {{key.data(), int(key.size())}, {r.data(), int(r.size())}, 0};
 }
 
 void benchConflictSet() {
@@ -258,4 +258,4 @@ void benchConflictSet() {
   }
 }
 
-int main(void) { benchConflictSet(); }
+int main(void) { benchConflictSet(); }

CMakeLists.txt

@@ -25,6 +25,9 @@ endif()
 add_compile_options(-fdata-sections -ffunction-sections -Wswitch-enum
                     -Werror=switch-enum)
 
+option(USE_SIMD_FALLBACK
+       "Use fallback implementations of functions that use SIMD" 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)
@@ -43,18 +46,20 @@ endif()
 include(CheckIncludeFileCXX)
 include(CMakePushCheckState)
 
-cmake_push_check_state()
+if(NOT USE_SIMD_FALLBACK)
-list(APPEND CMAKE_REQUIRED_FLAGS -mavx)
+  cmake_push_check_state()
-check_include_file_cxx("immintrin.h" HAS_AVX)
+  list(APPEND CMAKE_REQUIRED_FLAGS -mavx)
-if(HAS_AVX)
+  check_include_file_cxx("immintrin.h" HAS_AVX)
-  add_compile_options(-mavx)
+  if(HAS_AVX)
-  add_compile_definitions(HAS_AVX)
+    add_compile_options(-mavx)
-endif()
+    add_compile_definitions(HAS_AVX)
-cmake_pop_check_state()
+  endif()
+  cmake_pop_check_state()
 
-check_include_file_cxx("arm_neon.h" HAS_ARM_NEON)
+  check_include_file_cxx("arm_neon.h" HAS_ARM_NEON)
-if(HAS_ARM_NEON)
+  if(HAS_ARM_NEON)
-  add_compile_definitions(HAS_ARM_NEON)
+    add_compile_definitions(HAS_ARM_NEON)
+  endif()
 endif()
 
 set(CMAKE_CXX_IMPLICIT_LINK_LIBRARIES "")
@@ -267,6 +272,10 @@ set(CPACK_RESOURCE_FILE_README "${CMAKE_CURRENT_SOURCE_DIR}/README.md")
 set(CPACK_RPM_PACKAGE_ARCHITECTURE ${CMAKE_SYSTEM_PROCESSOR})
 set(CPACK_RPM_SPEC_INSTALL_POST "/bin/true") # avoid stripping
 set(CPACK_RPM_PACKAGE_LICENSE "Apache 2.0")
+set(CPACK_RPM_FILE_NAME RPM-DEFAULT)
+
+# deb
+set(CPACK_DEBIAN_FILE_NAME DEB-DEFAULT)
 
 include(CPack)
 

ConflictSet.cpp

@@ -29,6 +29,7 @@ limitations under the License.
 #include <span>
 #include <string>
 #include <string_view>
+#include <type_traits>
 #include <utility>
 
 #ifdef HAS_AVX
@@ -954,6 +955,42 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     assert(self->getType() == Type_Node16);
 
     ++self->numChildren;
+#ifdef HAS_AVX
+    __m128i key_vec = _mm_set1_epi8(index);
+    __m128i indices;
+    memcpy(&indices, self16->index, sizeof(self16->index));
+    __m128i results = _mm_cmpeq_epi8(key_vec, _mm_min_epu8(key_vec, indices));
+    int mask = (1 << (self->numChildren - 1)) - 1;
+    uint32_t bitfield = _mm_movemask_epi8(results) & mask;
+    bitfield |= uint32_t(1) << (self->numChildren - 1);
+    int i = std::countr_zero(bitfield);
+    if (i < self->numChildren - 1) {
+      memmove(self16->index + i + 1, self16->index + i,
+              self->numChildren - (i + 1));
+      memmove(self16->children + i + 1, self16->children + i,
+              (self->numChildren - (i + 1)) * sizeof(Child));
+    }
+#elif defined(HAS_ARM_NEON)
+    uint8x16_t indices;
+    memcpy(&indices, self16->index, sizeof(self16->index));
+    // 0xff for each leq
+    auto results = vcleq_u8(vdupq_n_u8(index), indices);
+    uint64_t mask = (uint64_t(1) << ((self->numChildren - 1) * 4)) - 1;
+    // 0xf for each 0xff (within mask)
+    uint64_t bitfield =
+        vget_lane_u64(
+            vreinterpret_u64_u8(vshrn_n_u16(vreinterpretq_u16_u8(results), 4)),
+            0) &
+        mask;
+    bitfield |= uint64_t(0xf) << ((self->numChildren - 1) * 4);
+    int i = std::countr_zero(bitfield) / 4;
+    if (i < self->numChildren - 1) {
+      memmove(self16->index + i + 1, self16->index + i,
+              self->numChildren - (i + 1));
+      memmove(self16->children + i + 1, self16->children + i,
+              (self->numChildren - (i + 1)) * sizeof(Child));
+    }
+#else
     int i = 0;
     for (; i < int(self->numChildren) - 1; ++i) {
       if (int(self16->index[i]) > int(index)) {
@@ -964,6 +1001,7 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
         break;
       }
     }
+#endif
     self16->index[i] = index;
     auto &result = self16->children[i].child;
     result = nullptr;
@@ -979,8 +1017,8 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
       self = newSelf;
       goto insert256;
     }
-  insert48:
 
+  insert48:
     auto *self48 = static_cast<Node48 *>(self);
     self48->bitSet.set(index);
     ++self->numChildren;
@@ -992,6 +1030,7 @@ Node *&getOrCreateChild(Node *&self, uint8_t index,
     return result;
   }
   case Type_Node256: {
+
   insert256:
     auto *self256 = static_cast<Node256 *>(self);
     ++self->numChildren;
@@ -2858,8 +2897,8 @@ Iterator firstGeq(Node *n, std::string_view key) {
   }
 }
 
-bool checkCorrectness(Node *node, int64_t oldestVersion,
+[[maybe_unused]] bool checkCorrectness(Node *node, int64_t oldestVersion,
-                      ConflictSet::Impl *impl) {
+                                       ConflictSet::Impl *impl) {
   bool success = true;
 
   checkParentPointers(node, success);

Jenkinsfile

@@ -48,6 +48,17 @@ pipeline {
                 recordIssues(tools: [clang()])
             }
         }
+        stage('SIMD fallback') {
+            agent {
+                dockerfile {
+                    args '-v /home/jenkins/ccache:/ccache'
+                    reuseNode true
+                }
+            }
+            steps {
+                CleanBuildAndTest("-DUSE_SIMD_FALLBACK=ON")
+            }
+        }
         stage('Release [gcc]') {
             agent {
                 dockerfile {

README.md

@@ -9,23 +9,23 @@ Hardware for all benchmarks is a mac m1 2020.
 ## Skip list
 
 ```
-New conflict set: 1.962 sec
+New conflict set: 1.957 sec
-                  0.637 Mtransactions/sec
+                  0.639 Mtransactions/sec
-                  2.548 Mkeys/sec
+                  2.555 Mkeys/sec
-Detect only:      1.842 sec
+Detect only:      1.845 sec
-                  0.679 Mtransactions/sec
+                  0.678 Mtransactions/sec
-                  2.714 Mkeys/sec
+                  2.710 Mkeys/sec
-Skiplist only:    1.261 sec
+Skiplist only:    1.263 sec
-                  0.991 Mtransactions/sec
+                  0.990 Mtransactions/sec
-                  3.964 Mkeys/sec
+                  3.960 Mkeys/sec
 Performance counters:
-               Build: 0.0597
+               Build: 0.0546
-                 Add: 0.0587
+                 Add: 0.0563
               Detect: 1.84
-              D.Sort: 0.411
+              D.Sort: 0.412
-           D.Combine: 0.0136
+           D.Combine: 0.0141
-         D.CheckRead: 0.67
+         D.CheckRead: 0.671
-   D.CheckIntraBatch: 0.00671
+   D.CheckIntraBatch: 0.0068
         D.MergeWrite: 0.592
       D.RemoveBefore: 0.146
 ```
@@ -33,25 +33,25 @@ Performance counters:
 ## Radix tree (this implementation)
 
 ```
-New conflict set: 1.660 sec
+New conflict set: 1.366 sec
-                  0.753 Mtransactions/sec
+                  0.915 Mtransactions/sec
-                  3.012 Mkeys/sec
+                  3.660 Mkeys/sec
-Detect only:      1.524 sec
+Detect only:      1.248 sec
-                  0.820 Mtransactions/sec
+                  1.002 Mtransactions/sec
-                  3.280 Mkeys/sec
+                  4.007 Mkeys/sec
-Skiplist only:    0.844 sec
+Skiplist only:    0.573 sec
-                  1.481 Mtransactions/sec
+                  2.182 Mtransactions/sec
-                  5.926 Mkeys/sec
+                  8.730 Mkeys/sec
 Performance counters:
-               Build: 0.0699
+               Build: 0.0594
-                 Add: 0.0628
+                 Add: 0.0572
-              Detect: 1.52
+              Detect: 1.25
-              D.Sort: 0.435
+              D.Sort: 0.418
-           D.Combine: 0.0183
+           D.Combine: 0.0149
-         D.CheckRead: 0.37
+         D.CheckRead: 0.232
-   D.CheckIntraBatch: 0.00801
+   D.CheckIntraBatch: 0.0067
-        D.MergeWrite: 0.473
+        D.MergeWrite: 0.341
-      D.RemoveBefore: 0.215
+      D.RemoveBefore: 0.232
 ```
 
 # Our benchmark

RealDataBench.cpp

@@ -30,7 +30,7 @@ constexpr inline size_t rightAlign(size_t offset, size_t alignment) {
 
 int main(int argc, const char **argv) {
   // Use with this dataset https://snap.stanford.edu/data/memetracker9.html
-  // Preprocess the files with `sed -i '' '/^Q/d'`
+  // Preprocess the files with `sed -i'' '/^Q/d'`
 
   double checkTime = 0;
   double addTime = 0;