Merge pull request #28 from DominikHorn/feature/update-dependencies-a…

…nd-fix-build

Update dependencies and improve local development experience

CMakeLists.txt

@@ -5,11 +5,14 @@ if(UNIX AND NOT APPLE)
     set(LINUX TRUE)
 endif()
 
-if (${LINUX})
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math -Wall -Wfatal-errors -march=native -fopenmp")
-else ()
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math -Wall -Wfatal-errors -march=native")
-endif ()
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math -Wall -Wfatal-errors -march=native")
+
+# Enable OpenMP if available
+include(CheckCXXCompilerFlag)
+check_cxx_compiler_flag(-fopenmp HAS_OPENMP)
+if (HAS_OPENMP)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
+endif()
 
 set(CMAKE_CXX_STANDARD 17)
 

README.md

@@ -33,11 +33,11 @@ We provide a number of scripts to automate things. Each is located in the `scrip
 
 ## Running the benchmark
 
-* `./scripts/download.sh` downloads and stores required data from the Internet
-* `./scripts/build_rmis.sh` compiles and builds the RMIs for each dataset
-  * `./scripts/download_rmis.sh` will download pre-built RMIs instead, which may be faster. You'll need to run `build_rmis.sh` if you want to measure build times on your platform.
-* `./scripts/prepare.sh` constructs query workloads and compiles the benchmark
-* `./scripts/execute.sh` executes the benchmark on each workload, storing the results in `results`. You can use the `-c` flag to output a .csv file of results rather than a .txt.
+- `./scripts/download.sh` downloads and stores required data from the Internet
+- `./scripts/build_rmis.sh` compiles and builds the RMIs for each dataset. If you run into the error message `error: no override and no default toolchain set`, try running `rustup install stable`.
+  - `./scripts/download_rmis.sh` will download pre-built RMIs instead, which may be faster. You'll need to run `build_rmis.sh` if you want to measure build times on your platform.
+- `./scripts/prepare.sh` constructs query workloads and compiles the benchmark
+- `./scripts/execute.sh` executes the benchmark on each workload, storing the results in `results`. You can use the `-c` flag to output a .csv file of results rather than a .txt.
 
 Build times can be long, as we make aggressive use of templates to ensure we do not accidentally measure vtable lookup time. For development, this can be annoying: you can set `USE_FAST_MODE` in `config.h` to disable some features and get a faster build time.
 

competitors/stanford_hash.h

@@ -80,6 +80,7 @@ class CuckooHashMap {
     uint32_t i1 = alt_mod(hash, num_buckets_);
     Bucket* b1 = &buckets_[i1];
 
+#ifdef __AVX__
     __m256i vkey = _mm256_set1_epi32(key);
     __m256i vbucket = _mm256_load_si256((const __m256i*)&b1->keys);
     __m256i cmp = _mm256_cmpeq_epi32(vkey, vbucket);
@@ -88,20 +89,31 @@ class CuckooHashMap {
       int index = __builtin_ctz(mask) / 4;
       return {true, b1->values[index]};
     }
+#else
+    for (size_t i = 0; i < BUCKET_SIZE; i++)
+      if (b1->keys[i] == key)
+        return {true, b1->values[i]};
+#endif
 
     uint32_t i2 = alt_mod(hash32(key ^ hash), num_buckets_);
     if (i2 == i1) {
       i2 = (i1 == num_buckets_ - 1) ? 0 : i1 + 1;
     }
     Bucket* b2 = &buckets_[i2];
 
+#ifdef __AVX__
     vbucket = _mm256_load_si256((const __m256i*)&b2->keys);
     cmp = _mm256_cmpeq_epi32(vkey, vbucket);
     mask = _mm256_movemask_epi8(cmp);
     if (mask != 0) {
       int index = __builtin_ctz(mask) / 4;
       return {true, b2->values[index]};
     }
+#else
+    for (size_t i = 0; i < BUCKET_SIZE; i++)
+      if (b2->keys[i] == key)
+        return {true, b2->values[i]};
+#endif
 
     return {false, uninitialized_value_};
   }
@@ -127,6 +139,7 @@ class CuckooHashMap {
     Bucket* b2 = &buckets_[i2];
 
     // Update old value if the key is already in the table
+#ifdef __AVX__
     __m256i vkey = _mm256_set1_epi32(key);
     __m256i vbucket = _mm256_load_si256((const __m256i*)&b1->keys);
     __m256i cmp = _mm256_cmpeq_epi32(vkey, vbucket);
@@ -136,7 +149,16 @@ class CuckooHashMap {
       b1->values[index] = value;
       return;
     }
+#else
+    for (size_t i = 0; i < BUCKET_SIZE; i++) {
+      if (b1->keys[i] == key) {
+        b1->values[i] = value;
+        return;
+      }
+    }
+#endif
 
+#ifdef __AVX__
     vbucket = _mm256_load_si256((const __m256i*)&b2->keys);
     cmp = _mm256_cmpeq_epi32(vkey, vbucket);
     mask = _mm256_movemask_epi8(cmp);
@@ -145,6 +167,14 @@ class CuckooHashMap {
       b2->values[index] = value;
       return;
     }
+#else
+    for (size_t i = 0; i < BUCKET_SIZE; i++) {
+      if (b2->keys[i] == key) {
+        b2->values[i] = value;
+        return;
+      }
+    }
+#endif
 
     if (!is_reinsert) {
       size_++;

competitors/ts.h

@@ -18,9 +18,7 @@ class TS : public Competitor {
         min = data.front().key;
         max = data.back().key;
       }
-      ts::Builder<KeyType> tsb(min, max, config_.spline_max_error,
-                               config_.num_bins, config_.tree_max_error,
-                               /*single_pass=*/false, /*use_cache=*/false);
+      ts::Builder<KeyType> tsb(min, max, spline_max_error_);
       for (const auto& key_and_value : data) tsb.AddKey(key_and_value.key);
       ts_ = tsb.Finalize();
     });
@@ -49,44 +47,29 @@ class TS : public Competitor {
   int variant() const { return size_scale; }
 
  private:
-  struct TSConfig {
-    size_t spline_max_error;
-    size_t num_bins;
-    size_t tree_max_error;
-  };
-
   bool SetParameters(const std::string& dataset) {
     assert(size_scale >= 1 && size_scale <= 10);
-    std::vector<TSConfig> configs;
+    std::vector<size_t> configs;
 
     if (dataset == "books_200M_uint64") {
-      configs = {{512, 128, 2},  {256, 256, 2},  {128, 256, 16}, {64, 1024, 4},
-                 {32, 1024, 16}, {16, 1024, 16}, {16, 1024, 8},  {4, 256, 8},
-                 {2, 512, 8},    {2, 1024, 8}};
+      configs = {500, 200, 150, 60, 50, 25, 25, 4, 2, 1};
     } else if (dataset == "fb_200M_uint64") {
-      configs = {{1024, 1024, 16}, {1024, 1024, 16}, {1024, 512, 8},
-                 {256, 512, 8},    {128, 512, 8},    {16, 128, 16},
-                 {16, 1024, 16},   {8, 1024, 16},    {4, 256, 16},
-                 {2, 256, 16}};
+      configs = {225, 225, 225, 225, 100, 32, 16, 8, 8, 2};
     } else if (dataset == "osm_cellids_200M_uint64") {
-      configs = {{1024, 32, 16}, {1024, 32, 16}, {512, 32, 16}, {128, 128, 16},
-                 {64, 128, 16},  {16, 64, 16},   {8, 32, 16},   {8, 256, 16},
-                 {2, 256, 16},   {2, 512, 16}};
+      configs = {150, 150, 150, 150, 80, 25, 8, 8, 4, 1};
     } else if (dataset == "wiki_ts_200M_uint64") {
-      configs = {{1024, 128, 4}, {128, 128, 8}, {64, 256, 8}, {32, 1024, 8},
-                 {16, 1024, 8},  {16, 1024, 4}, {4, 128, 16}, {8, 128, 2},
-                 {2, 512, 8},    {2, 128, 2}};
+      configs = {175, 175, 90, 32, 25, 16, 16, 4, 2, 1};
     } else {
       // No config.
       return false;
     }
 
-    config_ = configs[size_scale - 1];
+    spline_max_error_ = configs[size_scale - 1];
     parameters_set_ = true;
     return true;
   }
 
   ts::TrieSpline<KeyType> ts_;
-  TSConfig config_;
+  size_t spline_max_error_;
   bool parameters_set_ = false;
 };