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

more branchless mormord - slower

Browse Source
This commit is contained in:
Richard Thier 2024-04-11 19:53:58 +02:00
parent 2d2cad2c5a
commit f4ceffe6e2
1 changed files with 5 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
ypsu.cpp
View File

@ -185,18 +185,16 @@ static inline void mormord_sort_impl(uint32_t *a, int n) noexcept {
/* (because radix value 3 is not found in input) */
uint32_t prev = 0;
uint32_t reali = 0;
#pragma GCC unroll 16
#pragma GCC unroll 256
for(int i = 0; i < 256; ++i) {
radics[i] += prev;
if(radics[i] != 0) {
real_radics[reali] = prev;
real_radics[reali + 1] = radics[i];
reali += 2;
}
bool act =(radics[i] != 0);
real_radics[reali] = act ? prev : real_radics[reali];
real_radics[reali + 1] = act ? radics[i] : real_radics[reali + 1];
reali += 2 * act;
prev = radics[i];
}
// Inplace swap - own ideas + some ideas based-on "famous" ct-swap (for branchless / more ILP):
// void ct-swap(bool secret, uint64_t a[], uint64_t b[], size_t len) {
// uint64_t mask = ~((uint64_t)secret - 1); // 1->111....111; 0->000....000