Queue take changed + benchmarks added

README.md

@@ -84,4 +84,18 @@ We want to generate 1M strings of length 12 with our own charset
 We want to generate 1M strings of length 12 with default charset ascii
 ```
 ./seqperm --random 1000000,12 --charset ascii
+```
+## Benchmarks (vs itertools)
+### Apple M1 with N_THREADS = 8
+Refer to total value for real execution time
+
+Trying to generate all permutations from length 1 to length 11 of `a b c d e f g h i l m`
+```
+seqperm 30,06s user 60,88s system 26,947 total
+itertools 72,94s user 0,06s system 1:13,10 total
+```
+Trying to generate all permutations from length 1 to length 8 of `a b c d e f g h i l m n o p q`
+```
+seqperm 91,48s user 219,63s system 1:21,62 total
+itertools 169,68s user 0,13s system 2:50,11 total
 ```

src/Makefile

@@ -2,7 +2,7 @@ DEBUG ?= 0
 ifeq ($(DEBUG), 1)
     CFLAGS = -g3
 else
-    CFLAGS = -Ofast
+    CFLAGS = -Ofast -march=native
 endif
 CC = cc
 TARGET = seqperm

src/main.c

@@ -552,6 +552,7 @@ int main(int argc, char **argv)
   }
   gen_bin_perms(bin, word_size, 0, max_len, 0, min_len);
   FREE_P(bin);
+  global_queue->head = global_queue->tail - 1;
   pthread_t tworker[N_THREAD];
   memset(&tworker, 0x0, sizeof(tworker));
   size_t pos = 0;

src/queue.c

@@ -37,8 +37,8 @@ void push_queue(Queue_t *Q, char **aux, size_t len)
 input_t *pop_queue(Queue_t *Q)
 {
   input_t *aux = NULL;
-  unsigned int idx = atomic_fetch_add(&Q->head, 1);
-  aux = idx < Q->maxsize ? Q->words[idx] : NULL;
+  long idx = atomic_fetch_sub(&Q->head, 1);
+  aux = idx >= 0 ? Q->words[idx] : NULL;
   return aux;
 }
 

src/queue.h

@@ -29,7 +29,7 @@ typedef struct input_aux_perm
 typedef struct Queue
 {
   input_t **words;
-  atomic_uint head;
+  atomic_long head;
   size_t tail;
   size_t maxsize;
 } Queue_t;