swab: fixed == to >= typo - now beats std::sort and is comparison sort fully!

schwab_sort.h

@@ -14,6 +14,11 @@
 #define SCHWAB_DELTA_THRESHOLD 32
 #endif /* SCHWAB_DELTA_THRESHOLD */
 
+/** Below this many elements we do insertion sort */
+#ifndef SCHWAB_INSERTION_THRESHOLD
+#define SCHWAB_INSERTION_THRESHOLD 64
+#endif /* SCHWAB_DELTA_THRESHOLD */
+
 typedef uint32_t sch_rand_state;
 
 /** Create rand state for schwab_sort using a seed - can give 0 if uninterested */
@@ -41,6 +46,22 @@ static inline void schwab_swap(uint32_t *a, uint32_t *b) {
 	*b = t;
 }
 
+/** Simple insertion sort for small cases */
+inline void sch_insertion_sort(uint32_t *arr, int low, int high) {
+    for (int i = low + 1; i <= high; ++i) {
+        uint32_t key = arr[i];
+        int j = i;
+
+        /* Move elements of arr[0..i-1] that are greater than key */
+        /* to one position ahead of their current position */
+        while (j > 0 && arr[j - 1] > key) {
+            arr[j] = arr[j - 1];
+            --j;
+        }
+        arr[j] = key;
+    }
+}
+
 /**
  * 3-way partitioning, in middle all the pivot elements.
  *
@@ -166,7 +187,7 @@ static inline int schwab_partition(
 
 		arr[b3] = arr[b2];
 		arr[b2] = (where == 2) ? arr[b2] : arr[b1];
-		arr[b1] = (where == 1) ? arr[b1] : arr[b0];
+		arr[b1] = (where >= 1) ? arr[b1] : arr[b0];
 
 		++b2;
 		b1 += (where < 2);
@@ -221,7 +242,9 @@ static inline void schwab_sort(
 		sch_rand_state *state) {
 
 	/* Loop handles longest sub-sort-task which ensused log tree depth */
+	/* Loop also handles start condition */
 	while(low < high) {
+		if(high - low > SCHWAB_INSERTION_THRESHOLD) {
 			int r0 = schwab_pick_pivot(state, (high + 1) - low) + low;
 			int r1 = schwab_pick_pivot(state, (high + 1) - low) + low;
 			uint32_t klo = array[r0];
@@ -273,6 +296,11 @@ static inline void schwab_sort(
 				/* low = low; */
 				high = plo - 1;
 			}
+		} else {
+			/* Just do an insertion sort instead */
+			sch_insertion_sort(array, low, high);
+			return;
+		}
 	}
 }