#1569291: speed up array.repeat() by making only O(log n) memcpy() calls; the code follows unicode_repeat.
diff --git a/Modules/arraymodule.c b/Modules/arraymodule.c
index 18e5e66..77a162c 100644
--- a/Modules/arraymodule.c
+++ b/Modules/arraymodule.c
@@ -674,11 +674,9 @@
static PyObject *
array_repeat(arrayobject *a, Py_ssize_t n)
{
- Py_ssize_t i;
Py_ssize_t size;
arrayobject *np;
- char *p;
- Py_ssize_t nbytes;
+ Py_ssize_t oldbytes, newbytes;
if (n < 0)
n = 0;
if ((Py_SIZE(a) != 0) && (n > PY_SSIZE_T_MAX / Py_SIZE(a))) {
@@ -688,13 +686,23 @@
np = (arrayobject *) newarrayobject(&Arraytype, size, a->ob_descr);
if (np == NULL)
return NULL;
- p = np->ob_item;
- nbytes = Py_SIZE(a) * a->ob_descr->itemsize;
- for (i = 0; i < n; i++) {
- memcpy(p, a->ob_item, nbytes);
- p += nbytes;
+ if (n == 0)
+ return (PyObject *)np;
+ oldbytes = Py_SIZE(a) * a->ob_descr->itemsize;
+ newbytes = oldbytes * n;
+ /* this follows the code in unicode_repeat */
+ if (oldbytes == 1) {
+ memset(np->ob_item, a->ob_item[0], newbytes);
+ } else {
+ Py_ssize_t done = oldbytes;
+ Py_MEMCPY(np->ob_item, a->ob_item, oldbytes);
+ while (done < newbytes) {
+ Py_ssize_t ncopy = (done <= newbytes-done) ? done : newbytes-done;
+ Py_MEMCPY(np->ob_item+done, np->ob_item, ncopy);
+ done += ncopy;
+ }
}
- return (PyObject *) np;
+ return (PyObject *)np;
}
static int