little optimization to gpts and mine space partition bucket sort

gptsort.h

@@ -2,6 +2,7 @@
 #include <vector>
 #include <algorithm>
 
+// ChatGPT and me did this space partitioning bucket sort
 void gpt_bucket_sort(uint32_t* array, int n) {
   // Calculate the number of buckets to use
   int num_buckets = std::sqrt(n);
@@ -36,3 +37,40 @@ void gpt_bucket_sort(uint32_t* array, int n) {
     }
   }
 }
+
+// Further optimizations (no chatGPT)
+void my_bucket_sort(uint32_t* array, int n) {
+  // Calculate the number of buckets to use
+  int num_buckets = std::sqrt(n);
+
+  // Create a vector of buckets
+  std::vector<std::vector<uint32_t>> buckets(num_buckets);
+
+  // Calculate the range of values that each bucket can hold
+  auto mm = std::minmax_element(array, array + n);
+  uint32_t min_value = *mm.first;
+  uint32_t max_value = *mm.second;
+  uint32_t range = max_value - min_value + 1;
+  uint32_t bucket_size = range / num_buckets + 1;
+
+  // Distribute the elements of the array into the buckets
+  for (int i = 0; i < n; i++) {
+    // Calculate the bucket index for this element
+    // using the range of values and the bucket size as the divisor
+    int bucket_index = (array[i] - min_value) / bucket_size;
+    buckets[bucket_index].push_back(array[i]);
+  }
+
+  // Sort the elements in each bucket using std::sort
+  for (int i = 0; i < num_buckets; i++) {
+    std::sort(buckets[i].begin(), buckets[i].end());
+  }
+
+  // Concatenate the buckets to get the sorted array
+  int k = 0;
+  for (int i = 0; i < num_buckets; i++) {
+    for (int j = 0; j < buckets[i].size(); j++) {
+      array[k++] = buckets[i][j];
+    }
+  }
+}

makefile

@@ -31,5 +31,8 @@ clang_release: test.cpp magyarsort.h
 clang_release3: test.cpp magyarsort.h
 	clang++ test.cpp -DNDEBUG -std=c++17 -O3 -o test.out
 
+clang_release_ypsu: ypsu.cpp magyarsort.h
+	clang++ ypsu.cpp -DNDEBUG -std=c++17 -O2 -o ypsu.out
+
 clean: test.out
 	rm test.out

ypsu.cpp

@@ -399,7 +399,8 @@
     fflush(stdout);
     std::vector<uint32_t> v(n);
     v = geninput(inputtype, n);
-      measure(inputtype, "sp", [&] { spsort(&v[0], v.size()); });
+    //measure(inputtype, "sp", [&] { spsort(&v[0], v.size()); });
+    measure(inputtype, "magyar", [&] { MagyarSort::sort<uint32_t>(&v[0], v.size()); });
 
     for (auto r : results) printf("%9.3fs", r.second);
     puts("");
@@ -414,8 +415,8 @@
 }
 
 int main(void) {
-    int n = 100000000;
+  //int n = 100000000;
-    //int n = 10000000;
+  int n = 10000000;
   //int n = 100;
 
   // Uncomment this for profiling and alg!
@@ -457,16 +458,17 @@
     w = v;
     measure(inputtype, "4pasu", [&] { fourpassu(&w[0], w.size()); });
     assert(w == expected);
-      */
     w = v;
     measure(inputtype, "4rot", [&] { fourrots(&w[0], w.size()); });
     assert(w == expected);
     w = v;
-      /*measure(inputtype, "sp", [&] { spsort(&w[0], w.size()); });
+    measure(inputtype, "sp", [&] { spsort(&w[0], w.size()); });
     assert(w == expected);
     w = v;*/
     measure(inputtype, "gptbuck", [&] { gpt_bucket_sort(&w[0], w.size()); });
     assert(w == expected);
+    measure(inputtype, "mybuck", [&] { my_bucket_sort(&w[0], w.size()); });
+    assert(w == expected);
     /*
     w = v;
     measure(inputtype, "frewr", [&] { frewr(&w[0], w.size()); });