latest version that's being performance tested

GraphStreamingProject · Mar 11, 2024 · 7a3b793 · 7a3b793
1 parent 6b82f64
commit 7a3b793
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 13 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -8,6 +8,7 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS ON)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
 set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
+set(BUILD_BENCH TRUE)
 
 # Make the default build type Release. If user or another
 # project sets a different value than use that

diff --git a/include/bucket.h b/include/bucket.h
@@ -59,7 +59,7 @@ inline col_hash_t Bucket_Boruvka::get_index_depth(const vec_t update_idx, const
                                                   const vec_hash_t max_depth) {
   col_hash_t depth_hash = col_hash(&update_idx, sizeof(vec_t), seed_and_col);
   depth_hash |= (1ull << max_depth); // assert not > max_depth by ORing
-  return __builtin_ctzll(depth_hash) >> 1;
+  return __builtin_ctzll(depth_hash);
 }
 
 inline vec_hash_t Bucket_Boruvka::get_index_hash(const vec_t update_idx, const long sketch_seed) {

diff --git a/src/sketch.cpp b/src/sketch.cpp
@@ -111,20 +111,19 @@ SketchSample Sketch::fast_sample() {
 
     Bucket* current_column = buckets + ((idx * cols_per_sample + col)  * bkt_per_col);
 
-    // TODO - currently this is only set up for a single column. Should be pretty minimal changes
-    for (size_t idx=0; idx < 5; idx++) {
+    for (size_t idx=0; idx < 3; idx++) {
       if (Bucket_Boruvka::is_good(current_column[idx], checksum_seed()))
         return {current_column[idx].alpha, GOOD};
     }    
     size_t lo=1, hi=bkt_per_col;
 
-    while (lo + window_size < bkt_per_col && !(
-      !Bucket_Boruvka::is_zero(current_column[lo]) && Bucket_Boruvka::is_zero(current_column[lo+window_size])
-      )) {
-        lo *= 2;
-    }
-    hi = std::min(lo+2*window_size, hi);
-    lo /= 2;
+    // while (lo + window_size < bkt_per_col && !(
+    //   !Bucket_Boruvka::is_zero(current_column[lo]) && Bucket_Boruvka::is_zero(current_column[lo+window_size])
+    //   )) {
+    //     lo *= 2;
+    // }
+    // hi = std::min(lo+2*window_size, hi);
+    // lo /= 2;
     size_t midpt = (lo+hi)/2;
     do {
       if (!Bucket_Boruvka::is_zero(current_column[midpt]) && Bucket_Boruvka::is_zero(current_column[midpt+window_size])) {
@@ -144,10 +143,11 @@ SketchSample Sketch::fast_sample() {
         lo = midpt;
       }
       midpt = (lo+hi)/2;
-    } while (hi-lo >= 3*window_size);
+    } while (hi-lo >= 2*window_size);
     lo = std::max((size_t) 0, lo);
     hi = std::min(hi, bkt_per_col);
-    for (size_t i=lo; i < hi; i++) {
+    // for (size_t i=lo; i < hi; i++) {
+    for (size_t i=hi-1; i <= lo; i--) {
       if (Bucket_Boruvka::is_good(current_column[i], checksum_seed()))
         return {current_column[i].alpha, GOOD};
     }

diff --git a/tools/benchmark/graphcc_bench.cpp b/tools/benchmark/graphcc_bench.cpp
@@ -220,7 +220,7 @@ BENCHMARK(BM_Sketch_Update)->RangeMultiplier(4)->Ranges({{KB << 4, MB << 4}});
 
 // Benchmark the speed of querying sketches
 static void BM_Sketch_Query(benchmark::State& state) {
-  constexpr size_t vec_size = KB << 5;
+  constexpr size_t vec_size = KB << 22;
   constexpr size_t num_sketches = 100;
   double density = ((double)state.range(0)) / 100;