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

thier3: no unnecessary 4096 loops and storage because last commit makes it not necessary

Browse Source
This commit is contained in:
Richard Thier 2025-10-01 00:34:08 +02:00
parent 1f6ef0f2ea
commit 69d1432721
1 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
thiersort3.h
View File

@ -53,12 +53,12 @@ static inline uint32_t witch_bucket3(uint32_t key) {
* @param rstate Create with sch_rand_state rstate = schwab_rand_state(junk_uint32_t); * @param rstate Create with sch_rand_state rstate = schwab_rand_state(junk_uint32_t);
*/ */
static inline void thiersort3(uint32_t *arr, uint32_t *temparr, int n) { static inline void thiersort3(uint32_t *arr, uint32_t *temparr, int n) {
int bucket[4096]; /* Inclusive */ int bucket[256]; /* Inclusive */
int bucket_end[4096]; /* Not inclusive */ int bucket_end[256]; /* Not inclusive */
#ifndef NO_EXTRA_BIT #ifndef NO_EXTRA_BIT
int bucket_leftend[4096]; /* for extra 1bit split processing */ int bucket_leftend[256]; /* for extra 1bit split processing */
int bucket_left[4096]; /* for extra 1bit split processing */ int bucket_left[256]; /* for extra 1bit split processing */
#endif /* NO_EXTRA_BIT */ #endif /* NO_EXTRA_BIT */
/* Check if need to sort at all - needed for invariants later */ /* Check if need to sort at all - needed for invariants later */
@ -68,7 +68,7 @@ static inline void thiersort3(uint32_t *arr, uint32_t *temparr, int n) {
/* Count */ /* Count */
#pragma GCC unroll 64 #pragma GCC unroll 64
for(int i = 0; i < 4096; ++i) { for(int i = 0; i < 256; ++i) {
bucket[i] = 0; bucket[i] = 0;
} }
#pragma GCC unroll 64 #pragma GCC unroll 64
@ -79,14 +79,14 @@ static inline void thiersort3(uint32_t *arr, uint32_t *temparr, int n) {
/* Prefix sum (like in Magyarsort) */ /* Prefix sum (like in Magyarsort) */
uint32_t prev = 0; uint32_t prev = 0;
#pragma GCC unroll 4 #pragma GCC unroll 4
for (int i = 0; i < 4096; i++) { for (int i = 0; i < 256; i++) {
bucket[i] += prev; bucket[i] += prev;
prev = bucket[i]; prev = bucket[i];
} }
/* Save end-offsets */ /* Save end-offsets */
#pragma GCC unroll 64 #pragma GCC unroll 64
for(int i = 0; i < 4096; ++i) { for(int i = 0; i < 256; ++i) {
bucket_end[i] = bucket[i]; bucket_end[i] = bucket[i];
} }
@ -115,7 +115,7 @@ static inline void thiersort3(uint32_t *arr, uint32_t *temparr, int n) {
/* temparr -> arr each bucket and sort them in-place */ /* temparr -> arr each bucket and sort them in-place */
#pragma GCC unroll 64 #pragma GCC unroll 64
for(int b = 0; b < 4096; ++b) { for(int b = 0; b < 256; ++b) {
int begin = bucket[b]; int begin = bucket[b];
int end = bucket_end[b]; int end = bucket_end[b];
@ -149,7 +149,7 @@ static inline void thiersort3(uint32_t *arr, uint32_t *temparr, int n) {
/* temparr -> arr each bucket and sort them in-place */ /* temparr -> arr each bucket and sort them in-place */
#pragma GCC unroll 64 #pragma GCC unroll 64
for(int b = 0; b < 4096; ++b) { for(int b = 0; b < 256; ++b) {
assert(bucket_leftend[b] == bucket[b]); assert(bucket_leftend[b] == bucket[b]);
int lbegin = bucket_left[b]; int lbegin = bucket_left[b];
int lend = bucket[b]; /* non-inclusive */ int lend = bucket[b]; /* non-inclusive */