added pre-randomized sorts (not so great so far - probably too much cache misses)

.gitmodules

@@ -4,3 +4,6 @@
 [submodule "FlameGraph"]
 	path = FlameGraph
 	url = https://github.com/brendangregg/FlameGraph
+[submodule "fastrand"]
+	path = fastrand
+	url = ssh://gitea@magosit.hu:8122/prenex/fastrand.git

cache_miss_flamegraph.sh

@@ -1,6 +1,9 @@
 #!/bin/sh
 
-rm perf.data; perf record -e L1-dcache-load-misses:u -c 1000 -g -- ./ypsu.out
+# time based sampling
+rm perf.data; perf record -e L1-dcache-load-misses:u -F 99 -g -- ./ypsu.out
+# counters only - might work or not
+#rm perf.data; perf record -e L1-dcache-load-misses:u -c 1000 -g -- ./ypsu.out
 perf script | FlameGraph/stackcollapse-perf.pl > out.perf-folded
 FlameGraph/flamegraph.pl out.perf-folded > perf.svg
 brave perf.svg

fastrand

@@ -0,0 +1 @@
+Subproject commit 42943f467831bfcc123214659f4b5b7f5a5e2c54

randominus.h

@@ -0,0 +1,47 @@
+#ifndef RANDOMINUS_H
+#define RANDOMINUS_H
+/* To randomize an array - hopefully as fast as possible */
+
+#include "fastrand/fastrand.h"
+
+/** swap */
+static void inline rd_swap(uint32_t *a, uint32_t *b) {
+	uint32_t tmp = *a;
+	*a = *b;
+	*b = tmp;
+}
+
+/** This is by no means "cryptographically correct" or stuff, but fast */
+static inline void randominus(uint32_t *a, int n, uint32_t seed) {
+
+	/** Initialized ILP random generator */
+	uint32_t ilp_seeds[8];
+	rand_ilp_state rsi;
+	rand_state rs = init_rand();
+	for(int i = 0; i < 8; ++i) {
+		uint32_t choice = rand_between(&rs, 0, n);
+		ilp_seeds[i] = choice;
+	}
+
+	/** Go over the array and randomly swap stuff - hand unrolled with ILP random get! */
+	for(int i = 0; i < (n - 8); i += 8) {
+		uint32_t to0 = fastmodlike(lcg_ilp(&rsi, A), n);
+		uint32_t to1 = fastmodlike(lcg_ilp(&rsi, B), n);
+		uint32_t to2 = fastmodlike(lcg_ilp(&rsi, C), n);
+		uint32_t to3 = fastmodlike(lcg_ilp(&rsi, D), n);
+		uint32_t to4 = fastmodlike(lcg_ilp(&rsi, E), n);
+		uint32_t to5 = fastmodlike(lcg_ilp(&rsi, F), n);
+		uint32_t to6 = fastmodlike(lcg_ilp(&rsi, G), n);
+		uint32_t to7 = fastmodlike(lcg_ilp(&rsi, H), n);
+		rd_swap(&a[i], &a[to0]);
+		rd_swap(&a[i + 1], &a[to1]);
+		rd_swap(&a[i + 2], &a[to2]);
+		rd_swap(&a[i + 3], &a[to3]);
+		rd_swap(&a[i + 4], &a[to4]);
+		rd_swap(&a[i + 5], &a[to5]);
+		rd_swap(&a[i + 6], &a[to6]);
+		rd_swap(&a[i + 7], &a[to7]);
+	}
+}
+
+#endif /* RANDOMINUS_H */

ypsu.cpp

@@ -23,6 +23,7 @@
 #include "qsort/chatgpt_qs.h"
 #include "threepass.h"
 #include "thiersort3.h"
+#include "randominus.h"
 
 // #define MAGYAR_SORT_DEFAULT_REUSE
 #include "magyarsort.h"
@@ -227,6 +228,15 @@ static inline void do_thier3(uint32_t *a, int n) noexcept {
 	thiersort3(a, &(tmp[0]), n);
 }
 
+/** rthier */
+static inline void do_rthier(uint32_t *a, int n) noexcept {
+	assert(n * uint32_t(sizeof(a[0])) <= INT_MAX);
+	uint32_t junk;
+	randominus(a, n, junk);
+	std::vector<uint32_t> tmp(n);
+	thiersort3(a, &(tmp[0]), n);
+}
+
 /** 3+1 pass bottom-up radix */
 static inline void do_threepass(uint32_t *a, int n) noexcept {
 	threepass(a, n);
@@ -895,8 +905,8 @@ int main(int argc, char **argv) {
   printf("Sorting %d elements:\n\n", n);
 
   // Uncomment this for profiling and alg!
-  //measure_single(n);
-  //return 0;
+  // measure_single(n);
+  // return 0;
 
   for (auto inputtype : inputtypes) {
     printf("%10s", inputtype.c_str());
@@ -921,7 +931,17 @@ int main(int argc, char **argv) {
     });
 
     w = v;
-    measure(inputtype, "magyar", [&] { MagyarSort::sort<uint32_t>(&w[0], w.size()); });
+    measure(inputtype, "magyar", [&] {
+			MagyarSort::sort<uint32_t>(&w[0], w.size());
+	});
+    assert(w == expected);
+
+    w = v;
+    measure(inputtype, "rmagyar", [&] {
+			uint32_t junk;
+			randominus(&w[0], w.size(), junk);
+			MagyarSort::sort<uint32_t>(&w[0], w.size());
+	});
     assert(w == expected);
 
     w = v;
@@ -1011,6 +1031,10 @@ int main(int argc, char **argv) {
     measure(inputtype, "thier3", [&] { do_thier3(&w[0], w.size()); });
     assert(w == expected);
 
+    w = v;
+    measure(inputtype, "rthier", [&] { do_rthier(&w[0], w.size()); });
+    assert(w == expected);
+
     w = v;
     measure(inputtype, "threep", [&] { do_threepass(&w[0], w.size()); });
     assert(w == expected);
@@ -1046,10 +1070,12 @@ int main(int argc, char **argv) {
     }
     assert(w == expected);
     */
-    /*
+
     w = v;
     measure(inputtype, "frewr", [&] { frewr(&w[0], w.size()); });
     assert(w == expected);
+
+    /*
     w = v;
     measure(inputtype, "vsort", [&] { vsort(&w[0], w.size()); });
     assert(w == expected);