Some improvements on correct behaviour.

Still some edge cases found.

.vscode/settings.json

@@ -71,6 +71,7 @@
         "thread": "cpp",
         "cinttypes": "cpp",
         "typeinfo": "cpp",
-        "variant": "cpp"
+        "variant": "cpp",
+        "bitset": "cpp"
     }
 }

eval.py

@@ -9,7 +9,7 @@ def run_experiment(output_file, build_dir):
     # The number of threads is not currently used, it's just here in case you want to parallelize your code.
     for threads in [1]:
         # for size in [1e2, 1e3, 1e4 + 1, 1e5, 1e6 - 1, 1e7]:
-        for size in [1e2]:
+        for size in [1e2, 1e3, 1e4 + 1, 1e5, 1e7]:
             print("Measuring p=" + str(threads) + " n=" + str(size))
             executable = Path(build_dir) / "Sorter"
             returncode = subprocess.call([executable, str(size), str(threads)], stdout=output_file)

result.txt

@@ -0,0 +1,5 @@
+RESULT name=sort n=100 t=1 iterations=545 durationNanoseconds=1835 totalDurationNanoseconds=1000511 constructorNanoseconds=232 totalConstructorNanoseconds=126950
+RESULT name=sort n=1000 t=1 iterations=29 durationNanoseconds=35382 totalDurationNanoseconds=1026091 constructorNanoseconds=773 totalConstructorNanoseconds=22440
+RESULT name=sort n=10001 t=1 iterations=2 durationNanoseconds=592405 totalDurationNanoseconds=1184811 constructorNanoseconds=31980 totalConstructorNanoseconds=63960
+RESULT name=sort n=100000 t=1 iterations=1 durationNanoseconds=24243178 totalDurationNanoseconds=24243178 constructorNanoseconds=562170 totalConstructorNanoseconds=562170
+RESULT name=sort n=10000000 t=1 iterations=1 durationNanoseconds=219559500741 totalDurationNanoseconds=219559500741 constructorNanoseconds=53379598 totalConstructorNanoseconds=53379598

src/sorter.cpp

@@ -7,15 +7,29 @@
 #include <bitset>
 #include <climits>
 
+#define DEBUG false
+
 namespace ae {
 
 void sorter::sort(container& data) {
-  // TODO Implement your sorting algorithm
   for (auto i = 1uz; i < data.placeholder_.size(); ++i) {
     std::ranges::copy(data.placeholder_[i], std::back_inserter(data.placeholder_[0]));
     data.placeholder_[i].clear();
   }
+  #if DEBUG
+    std::vector<container::element_type> copy;
+    std::ranges::copy(data.placeholder_[0], std::back_inserter(copy));
+    std::sort(copy.begin(), copy.end());
+  #endif
   sorter::msd_inplace_radix_sort(data.placeholder_[0], 0, [&](auto span) {sorter::robin_hood_sort(span);});
+
+  #if DEBUG
+    for (int i = 0; i < copy.size(); i++) {
+      if (copy[i] != data.placeholder_[0][i])
+      std::cerr << i << " " << "sorted: " << copy[i] << " actual:" << data.placeholder_[0][i] << std::endl;
+    }
+  #endif
+
 }
 
 void sorter::msd_inplace_radix_sort(
@@ -23,10 +37,22 @@ void sorter::msd_inplace_radix_sort(
   size_t passes, 
   const std::function<void(std::span<container::element_type> bucket)>& bucket_sort
 ) {
-  std::cerr << "passes: " << passes << "size: " << range.size() << std::endl;
+  if (std::begin(range) >= std::end(range)) {
+    return;
+  }
   if (sorter::RADIX_ITERATIONS == passes) {
-    std::cerr << "going to robin " << *std::begin(range) << std::endl;
+    switch (range.size()) {
-    if (std::begin(range) + 1 < std::end(range)) {
+      case 1: return;
+      case 2: 
+        if (range[0] >= range[1]) {
+          std::swap(range[0], range[1]);
+        }
+        return;
+      default:
+        bucket_sort(range);
+        return;
+    }
+    if (range.size() > 1) {
       bucket_sort(range);
     }
     return;
@@ -34,60 +60,81 @@ void sorter::msd_inplace_radix_sort(
 
   auto lower = std::begin(range);
   auto upper = std::end(range);
-
+  while (lower < upper) {
-  for (auto element = lower; element < upper;) {
+    
-    std::cerr << *element << " " << &*element << " " << &*lower << " " << &*upper << std::endl;
+    if (*lower & (1L << (sizeof(container::element_type) * CHAR_BIT - passes - 1))) {
-    // Mask out the <sorter::RADIX_ITERATIONS>-last bit and check if it is set
+      // The <passes>-left bit is set, so move to the beginning of the end section and decrement the upper iterator
-    // std::cerr << std::bitset<64>((1L << (sizeof(container::element_type) * CHAR_BIT - passes - 1))) << std::endl;
-    // fprintf(stderr, "%lx\n", (1L << (sizeof(container::element_type) * CHAR_BIT - passes - 1)));
-    if ((*element & (1L << (sizeof(container::element_type) * CHAR_BIT - passes - 1)))) {
-      // The <passes> bit is set, so move to the beginning of the end section and decrement the upper iterator
-      std::swap(*upper, *element);
       --upper;
+      std::swap(*upper, *lower);
     } else {
-      // The <passes> bit is unset, so move to the end of the beginning section and increment the upper iterator
-      std::swap(*lower, *element);
       ++lower;
-      ++element;
     }
   }
-  // assert(lower == upper);
+  #if DEBUG
+    std::cerr << "pass: " << passes << " begin: " << &*std::begin(range) << " end: " << &*std::end(range) << " lower: " << &*lower << std::endl;
+  #endif
   sorter::msd_inplace_radix_sort(std::span<container::element_type> (std::begin(range), lower), passes + 1, bucket_sort);
-  sorter::msd_inplace_radix_sort(std::span<container::element_type> (upper, std::end(range)), passes + 1, bucket_sort);
+  sorter::msd_inplace_radix_sort(std::span<container::element_type> (lower, std::end(range)), passes + 1, bucket_sort);
 }
 void sorter::robin_hood_sort(std::span<container::element_type> bucket) {
-  std::cerr << "robin hood lol" << std::endl;
   const auto size = bucket.size() + sorter::OVERHEAD_SIZE;
   const auto mask = ((1L) << (sizeof(container::element_type) * CHAR_BIT - sorter::RADIX_ITERATIONS)) - 1;
-  container::element_type space[size];
+  std::vector<container::element_type> space(size, -1L);
-  std::cerr << "robin hood before loop" << std::endl;
   for (auto element : bucket) {
     auto masked_element = (element & mask);
-    std::cerr << "robin hood elements " << element << " " << masked_element << " " << size << " " << mask << std::endl;
+    auto index = ((masked_element) * bucket.size()) / mask;
-    auto index = (masked_element * size) / mask;
-    std::cerr << "robin hood index " << index << " " << size << std::endl;
     if (space[index] == -1) {
-      space[index] = masked_element;
+      space[index] = element;
     } else {
+      #if DEBUG
+        std::cerr << "Linear probing of " << element << " at index " << index << ". Current element " << space[index] << std::endl;
+      #endif
       auto i = index;
-      while (i < size && space[index] != -1) {++i;};
+      // linear probing
-      space[i] = masked_element;
+      while (i < size - 1 && space[i] != -1) {++i;};
+      #if DEBUG
+        std::cerr << "Inserting " << element << " at index " << i << " instead of " << index << std::endl;
+      #endif
+      space[i] = element;
     }
   }
-  std::cerr << "robin hood after loop" << std::endl;
+
+  #if DEBUG
+    std::cerr << "Unsorted\n";
+    for (auto element : space) {
+      std::cerr << element << " ";
+    }
+    std::cerr << std::endl;
+  #endif
 
   // One final pass to correct linear probing errors
   for (auto i = 1; i < size; ++i) {
     auto j = i;
-    while (space[j-1] > space[j] && j > 0) {
+    while ((uint64_t) space[j-1] > space[j] && j > 0) {
       std::swap((space[j]),space[j-1]);
+      j--;
     }
   }
 
+  #if DEBUG
+    std::cerr << "Original\n";
+    for (auto element : bucket) {
+      std::cerr << element << " ";
+    }
+    std::cerr << std::endl;
+
+    std::cerr << "Checking if sorted\n";
+    for (auto element : space) {
+      std::cerr << element << " ";
+    }
+    std::cerr << std::endl;
+  #endif
+
   // copy data back into original range
   auto i = 0;
   for (auto element = std::begin(bucket); element < std::end(bucket); ++element) {
     *element = space[i];
+    ++i;
   }
 }
 

src/sorter.hpp

@@ -17,7 +17,7 @@ class sorter {
   );
 
   const uint32_t OVERHEAD_SIZE = 10L;
-  const uint32_t RADIX_ITERATIONS = 6;
+  const uint32_t RADIX_ITERATIONS = 8;
 
   void robin_hood_sort(std::span<container::element_type> range);
 };