mormord ILP-variant "nearly sorting properly" but some values buggy

ypsu.cpp

@@ -265,7 +265,7 @@ static inline void mormord_sort_impl(uint32_t *a, int n) noexcept {
 	// Without it its data dependent like crazy...
 	uint32_t pivoti1 = 0;
 	uint32_t pivoti2 = boundz.second;
-	while((pivoti1 < boundz.first) && (pivoti2 < n)) { // FIXME: needs two more "finisher-loops" behind this!!!
+	while((pivoti1 < boundz.first) && (pivoti2 < n)) {
 
 		/* Pivot 1 */
 
@@ -301,6 +301,46 @@ static inline void mormord_sort_impl(uint32_t *a, int n) noexcept {
 		pivoti2 += !mask2;
 		radics2[radixval2] += !mask2; // undec index (!)
 	}
+	// Finish pivot1 if there are still elements..
+	while(pivoti1 < boundz.first) {
+
+		/* Pivot 1+ */
+
+		uint32_t radixval1 = morgrab<j>(a[pivoti1]);
+		uint32_t targeti1 = --radics1[radixval1]; // dec index (!)
+
+		// Bitmask: true -> 11.....1; false -> 00.....0
+		uint32_t mask1 = ~((targeti1 > pivoti1) - 1);
+
+		// Branchless swap (using bitmask)
+		uint32_t delta1 = (a[pivoti1] ^ a[targeti1]) & mask1;
+		a[pivoti1] = a[pivoti1] ^ delta1;
+		a[targeti1] = a[targeti1] ^ delta1;
+
+		// "else" branch
+		pivoti1 += !mask1;
+		radics1[radixval1] += !mask1; // undec index (!)
+	}
+	// Finish pivot2 if there are still elements..
+	while(pivoti2 < n) {
+
+		/* Pivot 2+ */
+
+		uint32_t radixval2 = morgrab<j>(a[pivoti2]);
+		uint32_t targeti2 = --radics2[radixval2]; // dec index (!)
+
+		// Bitmask: true -> 11.....1; false -> 00.....0
+		uint32_t mask2 = ~((targeti2 > pivoti2) - 1);
+
+		// Branchless swap (using bitmask)
+		uint32_t delta2 = (a[pivoti2] ^ a[targeti2]) & mask2;
+		a[pivoti2] = a[pivoti2] ^ delta2;
+		a[targeti2] = a[targeti2] ^ delta2;
+
+		// "else" branch
+		pivoti2 += !mask2;
+		radics2[radixval2] += !mask2; // undec index (!)
+	}
 
 	// Possible recursions
 	if constexpr (j != 0) {