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

schwab: goto-optimized - barely slower on my machine

Browse Source
This commit is contained in:
Richard Thier 2025-05-09 03:29:20 +02:00
parent 283783bf9b
commit b32c7540a2
1 changed files with 21 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
schwab_sort.h
View File

@ -158,11 +158,24 @@ static inline int schwab_partition(
/* TODO: should be copy of whole element when not just uint32s! */ /* TODO: should be copy of whole element when not just uint32s! */
uint32_t curr = arr[b3]; uint32_t curr = arr[b3];
/* TODO: We can do "ILP-memcpy"s here: /* Half-branchless and half-goto trickery */
* int where = (curr < klo) ? 0 :
* Key from b2->b3, value from b2->b3, key from b1->b2, value from b1... etc ((curr < kmid) ? 1 : 2);
* This is likely faster than calling a memcpy if we code this for not just uint32s! int target = (curr < klo) ? b0 :
*/ ((curr < kmid) ? b1 : b2);
arr[b3] = arr[b2];
if(where == 2) goto auss;
arr[b2] = arr[b1];
if(where == 1) goto auss;
arr[b1] = arr[b0];
auss:
++b2;
b1 += (where < 2);
b0 += (where < 1);
arr[target] = curr;
/* Same as this would have been:
if(curr < klo) { if(curr < klo) {
arr[b3] = arr[b2]; arr[b3] = arr[b2];
arr[b2] = arr[b1]; arr[b2] = arr[b1];
@ -176,12 +189,13 @@ static inline int schwab_partition(
arr[b3] = arr[b2]; arr[b3] = arr[b2];
arr[b2] = arr[b1]; arr[b2] = arr[b1];
arr[b1] = curr; arr[b1] = curr;
++b1; ++b2; ++b1;
} else { } else {
arr[b3] = arr[b2]; arr[b3] = arr[b2];
arr[b2] = curr; arr[b2] = curr;
++b2;
} }
++b2;
*/
} }
/* [*] Swap the chosen pivot to begin of last block */ /* [*] Swap the chosen pivot to begin of last block */