prenex/magyarsort
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fixed thiersort2

Browse Source
This commit is contained in:
Richard Thier 2025-09-12 01:42:11 +02:00
parent a3643eba9b
commit 5a8f34efa0
3 changed files with 19 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
thiersort.h
View File

@ -531,6 +531,7 @@ static inline TSU8 ts_radixi(
/* Should never happen */
assert(false);
return 0;
}
/* Forward decl. */

21
thiersort2.h
View File

@ -46,24 +46,25 @@ static inline void thiersort2(uint32_t *arr, uint32_t *temparr, int n, sch_rand_
}
/* Count */
#pragma GCC unroll 4
#pragma GCC unroll 64
for(int i = 0; i < 256; ++i) {
bucket[i] = 0;
}
#pragma GCC unroll 4
#pragma GCC unroll 64
for(int i = 0; i < n; ++i) {
++bucket[witch_bucket(arr[i])];
}
/* Prefix sum (like in Magyarsort) */
uint32_t prev = 0;
#pragma GCC unroll 4
for (int i = 0; i < 256; i++) {
bucket[i] += prev;
prev = bucket[i];
}
/* Save end-offsets */
#pragma GCC unroll 4
#pragma GCC unroll 64
for(int i = 0; i < 256; ++i) {
bucket_end[i] = bucket[i];
}
@ -98,13 +99,13 @@ static inline void thiersort2(uint32_t *arr, uint32_t *temparr, int n, sch_rand_
uint32_t pivot = temparr[i];
#pragma GCC unroll 4
for(int j = begin + 1; j < end; ++j) {
if(temparr[j] == b) {
if(temparr[j] == pivot) {
/* swap to front partition */
++i;
uint32_t tmp = temparr[i];
temparr[i] = temparr[j];
temparr[j] = tmp;
} else if(temparr[j] < b) {
} else if(temparr[j] < pivot) {
/* copy to left */
arr[smalli++] = temparr[j];
} else {
@ -113,13 +114,15 @@ static inline void thiersort2(uint32_t *arr, uint32_t *temparr, int n, sch_rand_
}
}
/* Copy the mid elements back */
#pragma GCC unroll 4
int target = smalli;
#pragma GCC unroll 64
for(int j = begin; j < i + 1; ++j) {
arr[smalli++] = temparr[j];
arr[target++] = temparr[j];
}
/* Call schwabsort */
schwab_sort(arr, begin, end - 1, rstate);
/* Call schwabsort - only to [begin..smalli) and (biggie..end) */
schwab_sort(arr, begin, smalli - 1, rstate);
schwab_sort(arr, biggi + 1, end - 1, rstate);
}
}

10
ypsu.cpp
View File

@ -862,6 +862,7 @@ int main(void) {
//int n = 5000000;
int n = 1000000;
//int n = 100000;
//int n = 20001;
//int n = 20000;
//int n = 1000;
//int n = 200;
@ -889,6 +890,7 @@ int main(void) {
/*
w = v;
measure(inputtype, "ska", [&] { ska_sort(std::begin(w), std::end(w)); });
*/
w = v;
measure(inputtype, "ska_copy", [&] {
std::vector<uint32_t> buf(w.size());
@ -896,11 +898,14 @@ int main(void) {
w.swap(buf);
}
});
*/
w = v;
measure(inputtype, "magyar", [&] { MagyarSort::sort<uint32_t>(&w[0], w.size()); });
assert(w == expected);
w = v;
measure(inputtype, "gptbuck", [&] { gpt_bucket_sort(&w[0], w.size()); });
assert(w == expected);
/*
w = v;
measure(inputtype, "mormord", [&] { mormord_sort(&w[0], w.size()); });
@ -925,9 +930,6 @@ int main(void) {
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);
w = v;
measure(inputtype, "gpt_qsort", [&] { gpt_quicksort(w); });
assert(w == expected);
w = v;