gistfile1.txt

commit 9bf51d73c66e36c5f5d26eb161683507d5717069 (HEAD)
Author: Todd Lipcon <todd@apache.org>
Date:   Mon Jul 15 23:52:10 2019 -0700

    wip: wire protocol opt
    
    Change-Id: I118b01ae2acfd8dd290e7f67445796635d305351

diff --git a/src/kudu/common/wire_protocol-test.cc b/src/kudu/common/wire_protocol-test.cc
index 0317cab47..68e22bdc8 100644
--- a/src/kudu/common/wire_protocol-test.cc
+++ b/src/kudu/common/wire_protocol-test.cc
@@ -151,22 +151,28 @@ class WireProtocolTest : public KuduTest,
   void RunBenchmark(int column_count, double select_rate) {
     ResetBenchmarkSchema(column_count);
     Arena arena(1024);
-    const int kNumTrials = AllowSlowTests() ? 100 : 10;
-    RowBlock block(&benchmark_schema_, 10000, &arena);
+    RowBlock block(&benchmark_schema_, 1000, &arena);
+    // Regardless of the config, use a constant number of cells for the test by
+    // looping the conversion an appropriate number of times.
+    const int64_t kNumCellsToConvert = AllowSlowTests() ? 100000000 : 1000000;
+    const int kNumTrials = kNumCellsToConvert / select_rate / column_count / block.nrows();
     FillRowBlockForBenchmark(&block);
     SelectRandomRowsWithRate(&block, select_rate);
 
     RowwiseRowBlockPB pb;
     faststring direct, indirect;
-    LOG_TIMING(INFO, Substitute("Converting to PB with column count $0 and row select rate $1 ",
-                                column_count, select_rate)) {
-      for (int i = 0; i < kNumTrials; ++i) {
-        pb.Clear();
-        direct.clear();
-        indirect.clear();
-        SerializeRowBlock(block, &pb, nullptr, &direct, &indirect);
-      }
+    int64_t cycle_start = CycleClock::Now();
+    for (int i = 0; i < kNumTrials; ++i) {
+      pb.Clear();
+      direct.clear();
+      indirect.clear();
+      SerializeRowBlock(block, &pb, nullptr, &direct, &indirect);
     }
+    int64_t cycle_end = CycleClock::Now();
+    LOG(INFO) << Substitute(
+        "Converting to PB with column count $0 and row select rate $1: $2 cycles/cell",
+        column_count, select_rate,
+        static_cast<double>(cycle_end - cycle_start) / kNumCellsToConvert);
   }
  protected:
   Schema schema_;
diff --git a/src/kudu/common/wire_protocol.cc b/src/kudu/common/wire_protocol.cc
index 30de63f93..3a6be2b00 100644
--- a/src/kudu/common/wire_protocol.cc
+++ b/src/kudu/common/wire_protocol.cc
@@ -892,7 +892,7 @@ void AppendRowToString<RowBlockRow>(const RowBlockRow& row, string* buf) {
 // be copied to column 'dst_col_idx' in the output protobuf; otherwise,
 // dst_col_idx must be equal to col_idx.
 template<bool IS_NULLABLE, bool IS_VARLEN>
-static void CopyColumn(const ColumnBlock& column_block, int dst_col_idx, uint8_t* dst_base,
+static void CopyColumn(const ColumnBlock& column_block, int dst_col_idx, uint8_t* __restrict__ dst_base,
                        faststring* indirect_data, const Schema* dst_schema, size_t row_stride,
                        size_t schema_byte_size, size_t column_offset,
                        const vector<int>& row_idx_select) {
@@ -930,25 +930,28 @@ static void CopyColumn(const ColumnBlock& column_block, int dst_col_idx, uint8_t
 
 void convert_bitmap_to_vector(const SelectionVector* sel_vec, size_t nrows,
                               vector<int>* select_row_idx) {
-  int row_idx = 0;
-  int run_size = 0;
-  bool selected = false;
-  if (!sel_vec->AnySelected()) {
+
+  int n_selected = sel_vec->CountSelected();
+  if (n_selected == 0) {
     return;
   }
-  if (sel_vec->CountSelected() == nrows) {
-    select_row_idx->resize(nrows);
+  select_row_idx->resize(n_selected);
+  int row_idx = 0;
+  if (n_selected == nrows) {
     std::iota(select_row_idx->begin(), select_row_idx->end(), row_idx);
     return;
   }
 
-  BitmapIterator selected_row_iter(sel_vec->bitmap(), nrows);
-  while ((run_size = selected_row_iter.Next(&selected))) {
-    if (selected) {
-      select_row_idx->resize(select_row_idx->size() + run_size);
-      std::iota(select_row_idx->end() - run_size, select_row_idx->end(), row_idx);
+  int n_bytes = KUDU_ALIGN_UP(nrows, 8) / 8;
+  const uint8_t* bitmap = sel_vec->bitmap();
+  int* dst = select_row_idx->data();
+  for (int i = 0; i < n_bytes; i++) {
+    uint8_t bm = *bitmap++;
+    while (bm != 0) {
+      int bit = Bits::FindLSBSetNonZero(bm);
+      *dst++ = (i * 8) + bit;
+      bm ^= (1 << bit);
     }
-    row_idx += run_size;
   }
 }