Reordering seems to improve codegen

.pre-commit-config.yaml

@@ -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:

CMakeLists.txt

@@ -95,23 +95,12 @@ target_compile_options(${PROJECT_NAME}-object PRIVATE -fno-exceptions
                                                       -fvisibility=hidden)
 target_include_directories(${PROJECT_NAME}-object
                            PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
-if(NOT LD_EXE)
-  set(LD_EXE ld)
-endif()
-add_custom_command(
-  OUTPUT ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o
-  COMMAND ${LD_EXE} -r $<TARGET_OBJECTS:${PROJECT_NAME}-object> -o
-          ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o
-  DEPENDS $<TARGET_OBJECTS:${PROJECT_NAME}-object>
-  COMMAND_EXPAND_LISTS)
 
-add_library(${PROJECT_NAME} SHARED ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o)
+add_library(${PROJECT_NAME} SHARED $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
 set_target_properties(
   ${PROJECT_NAME} PROPERTIES LIBRARY_OUTPUT_DIRECTORY
                              "${CMAKE_CURRENT_BINARY_DIR}/radix_tree")
-if(CMAKE_BUILD_TYPE STREQUAL Debug)
-  set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE CXX)
-else()
+if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME} PROPERTIES LINKER_LANGUAGE C)
 endif()
 
@@ -121,13 +110,19 @@ if(HAS_VERSION_SCRIPT)
     LINKER:--version-script=${CMAKE_CURRENT_SOURCE_DIR}/linker.map)
 endif()
 
-add_library(${PROJECT_NAME}-static STATIC ${CMAKE_BINARY_DIR}/${PROJECT_NAME}.o)
-if(CMAKE_BUILD_TYPE STREQUAL Debug)
-  set_target_properties(${PROJECT_NAME}-static PROPERTIES LINKER_LANGUAGE CXX)
-else()
+add_library(${PROJECT_NAME}-static STATIC
+            $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
+if(NOT CMAKE_BUILD_TYPE STREQUAL Debug)
   set_target_properties(${PROJECT_NAME}-static PROPERTIES LINKER_LANGUAGE C)
 endif()
-if(NOT APPLE)
+
+if(APPLE)
+  add_custom_command(
+    TARGET ${PROJECT_NAME}-static
+    PRE_LINK
+    COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/privatize_symbols_macos.sh
+            $<TARGET_OBJECTS:${PROJECT_NAME}-object>)
+else()
   add_custom_command(
     TARGET ${PROJECT_NAME}-static
     POST_BUILD
@@ -253,19 +248,6 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
     add_test(NAME conflict_set_blackbox_${hash} COMMAND driver ${TEST})
   endforeach()
 
-  find_program(VALGRIND_EXE valgrind)
-  if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
-    list(LENGTH CORPUS_TESTS len)
-    math(EXPR last "${len} - 1")
-    set(partition_size 100)
-    foreach(i RANGE 0 ${last} ${partition_size})
-      list(SUBLIST CORPUS_TESTS ${i} ${partition_size} partition)
-      add_test(NAME conflict_set_blackbox_valgrind_${i}
-               COMMAND ${VALGRIND_EXE} --error-exitcode=99 --
-                       $<TARGET_FILE:driver> ${partition})
-    endforeach()
-  endif()
-
   # scripted tests. Written manually to fill in anything libfuzzer couldn't
   # find.
   if(NOT CMAKE_CROSSCOMPILING)
@@ -286,14 +268,19 @@ if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR AND BUILD_TESTING)
           ${Python3_EXECUTABLE}
           ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
           --build-dir ${CMAKE_CURRENT_BINARY_DIR})
-      if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
-        add_test(
-          NAME script_test_${TEST}_valgrind
-          COMMAND
-            ${VALGRIND_EXE} ${Python3_EXECUTABLE}
-            ${CMAKE_CURRENT_SOURCE_DIR}/test_conflict_set.py test ${TEST}
-            --build-dir ${CMAKE_CURRENT_BINARY_DIR})
-      endif()
+    endforeach()
+  endif()
+
+  find_program(VALGRIND_EXE valgrind)
+  if(VALGRIND_EXE AND NOT CMAKE_CROSSCOMPILING)
+    list(LENGTH CORPUS_TESTS len)
+    math(EXPR last "${len} - 1")
+    set(partition_size 100)
+    foreach(i RANGE 0 ${last} ${partition_size})
+      list(SUBLIST CORPUS_TESTS ${i} ${partition_size} partition)
+      add_test(NAME conflict_set_blackbox_valgrind_${i}
+               COMMAND ${VALGRIND_EXE} --error-exitcode=99 --
+                       $<TARGET_FILE:driver> ${partition})
     endforeach()
   endif()
 

Dockerfile

@@ -8,7 +8,6 @@ RUN chmod -R 777 /tmp
 RUN apt-get update
 RUN apt-get upgrade -y
 RUN TZ=America/Los_Angeles DEBIAN_FRONTEND=noninteractive apt-get install -y \
-    binutils-aarch64-linux-gnu \
     build-essential \
     ccache \
     clang \

ServerBench.cpp

@@ -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 = 115;
 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 (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);
+    }
+  }
+  ++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)) {
-    // 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();
-      //   }
-      // }
+    std::vector<std::string> keys(10);
+    for (auto &k : keys) {
+      k = makeKey(rand() % kNumPrefixes, 49);
     }
-    // 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);
-      }
+    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;
     }
-    // GC
-    cs->setOldestVersion(version - kWindowSize);
-    ++version;
+    cs->check(reads.data(), results.data(), 10);
   }
 }
 

aarch64-toolchain.cmake

@@ -5,4 +5,3 @@ set(CMAKE_CXX_COMPILER "/usr/bin/aarch64-linux-gnu-g++")
 set(CMAKE_FIND_ROOT_PATH /usr/aarch64-linux-gnu)
 set(CMAKE_CROSSCOMPILING_EMULATOR "qemu-aarch64;-L;/usr/aarch64-linux-gnu/")
 set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE arm64)
-set(LD_EXE "/usr/bin/aarch64-linux-gnu-ld")