sync set cover

Author: lperron

Diff for: ortools/algorithms/BUILD.bazel

@@ -120,11 +120,14 @@ cc_test(
         "//ortools/base:dump_vars",
         "//ortools/base:gmock_main",
         "//ortools/base:mathutil",
+        "//ortools/base:timer",
         "@com_google_absl//absl/algorithm:container",
         "@com_google_absl//absl/log",
+        "@com_google_absl//absl/log:check",
         "@com_google_absl//absl/random",
         "@com_google_absl//absl/random:bit_gen_ref",
         "@com_google_absl//absl/random:distributions",
+        "@com_google_absl//absl/time",
         "@com_google_absl//absl/types:span",
         "@com_google_benchmark//:benchmark",
     ],
@@ -286,6 +289,7 @@ cc_library(
     srcs = ["set_cover_model.cc"],
     hdrs = ["set_cover_model.h"],
     deps = [
+        ":radix_sort",
         ":set_cover_cc_proto",
         "//ortools/base:intops",
         "//ortools/base:strong_vector",
@@ -296,6 +300,7 @@ cc_library(
         "@com_google_absl//absl/random:distributions",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/strings:str_format",
+        "@com_google_absl//absl/types:span",
     ],
 )
 
@@ -495,17 +500,19 @@ cc_library(
         ":dynamic_partition",
         ":dynamic_permutation",
         ":sparse_permutation",
-        "//ortools/base:dump_vars",
         "//ortools/base:murmur",
         "//ortools/graph",
         "//ortools/graph:iterators",
         "//ortools/graph:util",
         "//ortools/util:stats",
         "//ortools/util:time_limit",
         "@com_google_absl//absl/algorithm:container",
+        "@com_google_absl//absl/base:log_severity",
+        "@com_google_absl//absl/container:flat_hash_map",
         "@com_google_absl//absl/container:flat_hash_set",
         "@com_google_absl//absl/flags:flag",
-        "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/log",
+        "@com_google_absl//absl/log:check",
         "@com_google_absl//absl/numeric:int128",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/strings",

Diff for: ortools/algorithms/find_graph_symmetries.cc

@@ -14,18 +14,21 @@
 #include "ortools/algorithms/find_graph_symmetries.h"
 
 #include <algorithm>
+#include <cmath>
 #include <cstdint>
-#include <limits>
 #include <memory>
 #include <numeric>
 #include <string>
 #include <utility>
 #include <vector>
 
 #include "absl/algorithm/container.h"
+#include "absl/base/log_severity.h"
+#include "absl/container/flat_hash_map.h"
 #include "absl/container/flat_hash_set.h"
 #include "absl/flags/flag.h"
-#include "absl/memory/memory.h"
+#include "absl/log/check.h"
+#include "absl/numeric/int128.h"
 #include "absl/status/status.h"
 #include "absl/strings/str_format.h"
 #include "absl/strings/str_join.h"
@@ -36,6 +39,8 @@
 #include "ortools/algorithms/dynamic_partition.h"
 #include "ortools/algorithms/dynamic_permutation.h"
 #include "ortools/algorithms/sparse_permutation.h"
+#include "ortools/base/logging.h"
+#include "ortools/graph/graph.h"
 #include "ortools/graph/iterators.h"
 #include "ortools/graph/util.h"
 
@@ -699,9 +704,17 @@ GraphSymmetryFinder::FindOneSuitablePermutation(
   // DCHECKs() and statistics.
   ScopedTimeDistributionUpdater search_time_updater(&stats_.search_time);
   DCHECK_EQ("", tmp_dynamic_permutation_.DebugString());
-  DCHECK_EQ(
-      base_partition->DebugString(/*sort_parts_lexicographically=*/false),
-      image_partition->DebugString(/*sort_parts_lexicographically=*/false));
+  DCHECK_EQ(base_partition->NumParts(), image_partition->NumParts());
+  if (DEBUG_MODE) {
+    for (int i = 0; i < base_partition->NumParts(); ++i) {
+      DCHECK_EQ(base_partition->FprintOfPart(i),
+                image_partition->FprintOfPart(i))
+          << base_partition->DebugString(/*sort_parts_lexicographically=*/false)
+          << " "
+          << image_partition->DebugString(
+                 /*sort_parts_lexicographically=*/false);
+    }
+  }
   DCHECK(search_states_.empty());
 
   // These will be used during the search. See their usage.

Diff for: ortools/algorithms/set_cover.proto

@@ -44,16 +44,21 @@ message SetCoverSolutionResponse {
   enum Status {
     // Undefined.
     UNDEFINED = 0;
+
     // The solver found the proven optimal solution.
     OPTIMAL = 1;
+
     // The solver had enough time to find some solution that satisfied all
     // constraints, but it did not reach the optimal.
     FEASIBLE = 2;
+
     // The model does not have any solution.
     INFEASIBLE = 3;
+
     // The model is invalid.
     INVALID = 4;
   }
+
   // For future use. TODO(user): Implement.
   optional Status status = 1;
 

Diff for: ortools/algorithms/set_cover_heuristics.cc

@@ -289,10 +289,9 @@ void IncreasingCountingSort(uint32_t radix, int shift, std::vector<Key>& keys,
   const auto num_keys = keys.size();
   // In this order for stability.
   for (int64_t i = num_keys - 1; i >= 0; --i) {
-    Counter& c = counts[Bucket(keys[i], shift, radix)];
-    scratch_keys[c - 1] = keys[i];
-    scratch_payloads[c - 1] = payloads[i];
-    --c;
+    Counter c = --counts[Bucket(keys[i], shift, radix)];
+    scratch_keys[c] = keys[i];
+    scratch_payloads[c] = payloads[i];
   }
   std::swap(keys, scratch_keys);
   std::swap(payloads, scratch_payloads);
@@ -301,14 +300,14 @@ void IncreasingCountingSort(uint32_t radix, int shift, std::vector<Key>& keys,
 
 template <typename Key, typename Payload>
 void RadixSort(int radix_log, std::vector<Key>& keys,
-               std::vector<Payload>& payloads, Key min_key, Key max_key) {
+               std::vector<Payload>& payloads, Key /*min_key*/, Key max_key) {
   // range_log is the number of bits necessary to represent the max_key
   // We could as well use max_key - min_key, but it is more expensive to
   // compute.
   const int range_log = internal::NumBitsToRepresent(max_key);
   DCHECK_EQ(internal::NumBitsToRepresent(0), 0);
   DCHECK_LE(internal::NumBitsToRepresent(std::numeric_limits<Key>::max()),
-            sizeof(Key) * CHAR_BIT);
+            std::numeric_limits<Key>::digits);
   const int radix = 1 << radix_log;  // By definition.
   std::vector<uint32_t> counters(radix, 0);
   std::vector<Key> scratch_keys(keys.size());
@@ -330,7 +329,7 @@ std::vector<ElementIndex> GetUncoveredElementsSortedByDegree(
   keys.reserve(num_elements);
   const SparseRowView& rows = inv->model()->rows();
   BaseInt max_degree = 0;
-  for (ElementIndex element : inv->model()->ElementRange()) {
+  for (const ElementIndex element : inv->model()->ElementRange()) {
     // Already covered elements should not be considered.
     if (inv->coverage()[element] != 0) continue;
     degree_sorted_elements.push_back(element);
@@ -729,8 +728,8 @@ bool GuidedLocalSearch::NextSolution(int num_iterations) {
 }
 
 Cost GuidedLocalSearch::ComputeDelta(SubsetIndex subset) const {
-  float delta = (penalization_factor_ * penalties_[subset] +
-                 inv_->model()->subset_costs()[subset]);
+  const float delta = (penalization_factor_ * penalties_[subset] +
+                       inv_->model()->subset_costs()[subset]);
   if (inv_->is_selected()[subset] && inv_->ComputeIsRedundant(subset)) {
     return delta;
   } else if (!inv_->is_selected()[subset]) {

Diff for: ortools/algorithms/set_cover_model.cc

@@ -29,6 +29,8 @@
 #include "absl/random/distributions.h"
 #include "absl/random/random.h"
 #include "absl/strings/str_format.h"
+#include "absl/types/span.h"
+#include "ortools/algorithms/radix_sort.h"
 #include "ortools/algorithms/set_cover.pb.h"
 #include "ortools/base/logging.h"
 
@@ -322,7 +324,9 @@ void SetCoverModel::ReserveNumElementsInSubset(ElementIndex num_elements,
 
 void SetCoverModel::SortElementsInSubsets() {
   for (const SubsetIndex subset : SubsetRange()) {
-    std::sort(columns_[subset].begin(), columns_[subset].end());
+    // std::sort(columns_[subset].begin(), columns_[subset].end());
+    BaseInt* data = reinterpret_cast<BaseInt*>(columns_[subset].data());
+    RadixSort(absl::MakeSpan(data, columns_[subset].size()));
   }
   elements_in_subsets_are_sorted_ = true;
 }
@@ -336,7 +340,10 @@ void SetCoverModel::CreateSparseRowView() {
   for (const SubsetIndex subset : SubsetRange()) {
     // Sort the columns. It's not super-critical to improve performance here
     // as this needs to be done only once.
-    std::sort(columns_[subset].begin(), columns_[subset].end());
+    // std::sort(columns_[subset].begin(), columns_[subset].end());
+    BaseInt* data = reinterpret_cast<BaseInt*>(columns_[subset].data());
+    RadixSort(absl::MakeSpan(data, columns_[subset].size()));
+
     for (const ElementIndex element : columns_[subset]) {
       ++row_sizes[element];
     }
@@ -346,7 +353,7 @@ void SetCoverModel::CreateSparseRowView() {
   }
   for (const SubsetIndex subset : SubsetRange()) {
     for (const ElementIndex element : columns_[subset]) {
-      rows_[element].push_back(subset);
+      rows_[element].emplace_back(subset);
     }
   }
   row_view_is_valid_ = true;
@@ -392,8 +399,10 @@ SetCoverProto SetCoverModel::ExportModelAsProto() const {
                            100.0 * subset.value() / num_subsets());
     SetCoverProto::Subset* subset_proto = message.add_subset();
     subset_proto->set_cost(subset_costs_[subset]);
-    SparseColumn column = columns_[subset];
-    std::sort(column.begin(), column.end());
+    SparseColumn column = columns_[subset];  // Copy is intentional.
+    // std::sort(column.begin(), column.end());
+    BaseInt* data = reinterpret_cast<BaseInt*>(column.data());
+    RadixSort(absl::MakeSpan(data, column.size()));
     for (const ElementIndex element : column) {
       subset_proto->add_element(element.value());
     }