traverseda
/
cpython
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Issue #24221: Small optimizations for heapq. 

Replaces the PyList_GET_ITEM and PyList_SET_ITEM macros with normal array
accesses.  Replace the siftup unpredicatable branch with arithmetic.
Replace the rc == -1 tests with rc < 0.  Gives nicer looking assembly
with both Clang and GCC-4.9.  Also gives a small performance both for both.
This commit is contained in:
Raymond Hettinger 2015-05-22 00:41:57 -07:00
parent 35e24a50c5
commit 5cbd8331ff
2 changed files with 44 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Include/listobject.h
View File

@ -72,6 +72,7 @@ PyAPI_FUNC(void) _PyList_DebugMallocStats(FILE *out);
#define PyList_GET_ITEM(op, i) (((PyListObject *)(op))->ob_item[i]) #define PyList_GET_ITEM(op, i) (((PyListObject *)(op))->ob_item[i])
#define PyList_SET_ITEM(op, i, v) (((PyListObject *)(op))->ob_item[i] = (v)) #define PyList_SET_ITEM(op, i, v) (((PyListObject *)(op))->ob_item[i] = (v))
#define PyList_GET_SIZE(op) Py_SIZE(op) #define PyList_GET_SIZE(op) Py_SIZE(op)
#define _PyList_ITEMS(op) (((PyListObject *)(op))->ob_item)
#endif #endif
#ifdef __cplusplus #ifdef __cplusplus

80
Modules/_heapqmodule.c
View File

@ -11,7 +11,7 @@ annotated by François Pinard, and converted to C by Raymond Hettinger.
static int static int
siftdown(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos) siftdown(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
{ {
PyObject *newitem, *parent; PyObject *newitem, *parent, **arr;
Py_ssize_t parentpos, size; Py_ssize_t parentpos, size;
int cmp; int cmp;
@ -24,12 +24,13 @@ siftdown(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
/* Follow the path to the root, moving parents down until finding /* Follow the path to the root, moving parents down until finding
a place newitem fits. */ a place newitem fits. */
newitem = PyList_GET_ITEM(heap, pos); arr = _PyList_ITEMS(heap);
newitem = arr[pos];
while (pos > startpos) { while (pos > startpos) {
parentpos = (pos - 1) >> 1; parentpos = (pos - 1) >> 1;
parent = PyList_GET_ITEM(heap, parentpos); parent = arr[parentpos];
cmp = PyObject_RichCompareBool(newitem, parent, Py_LT); cmp = PyObject_RichCompareBool(newitem, parent, Py_LT);
if (cmp == -1) if (cmp < 0)
return -1; return -1;
if (size != PyList_GET_SIZE(heap)) { if (size != PyList_GET_SIZE(heap)) {
PyErr_SetString(PyExc_RuntimeError, PyErr_SetString(PyExc_RuntimeError,
@ -38,10 +39,11 @@ siftdown(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
} }
if (cmp == 0) if (cmp == 0)
break; break;
parent = PyList_GET_ITEM(heap, parentpos); arr = _PyList_ITEMS(heap);
newitem = PyList_GET_ITEM(heap, pos); parent = arr[parentpos];
PyList_SET_ITEM(heap, parentpos, newitem); newitem = arr[pos];
PyList_SET_ITEM(heap, pos, parent); arr[parentpos] = newitem;
arr[pos] = parent;
pos = parentpos; pos = parentpos;
} }
return 0; return 0;
@ -51,7 +53,7 @@ static int
siftup(PyListObject *heap, Py_ssize_t pos) siftup(PyListObject *heap, Py_ssize_t pos)
{ {
Py_ssize_t startpos, endpos, childpos, limit; Py_ssize_t startpos, endpos, childpos, limit;
PyObject *tmp1, *tmp2; PyObject *tmp1, *tmp2, **arr;
int cmp; int cmp;
assert(PyList_Check(heap)); assert(PyList_Check(heap));
@ -63,19 +65,19 @@ siftup(PyListObject *heap, Py_ssize_t pos)
} }
/* Bubble up the smaller child until hitting a leaf. */ /* Bubble up the smaller child until hitting a leaf. */
arr = _PyList_ITEMS(heap);
limit = endpos / 2; /* smallest pos that has no child */ limit = endpos / 2; /* smallest pos that has no child */
while (pos < limit) { while (pos < limit) {
/* Set childpos to index of smaller child. */ /* Set childpos to index of smaller child. */
childpos = 2*pos + 1; /* leftmost child position */ childpos = 2*pos + 1; /* leftmost child position */
if (childpos + 1 < endpos) { if (childpos + 1 < endpos) {
cmp = PyObject_RichCompareBool( cmp = PyObject_RichCompareBool(
PyList_GET_ITEM(heap, childpos), arr[childpos],
PyList_GET_ITEM(heap, childpos + 1), arr[childpos + 1],
Py_LT); Py_LT);
if (cmp == -1) if (cmp < 0)
return -1; return -1;
if (cmp == 0) childpos += ((unsigned)cmp ^ 1); /* increment when cmp==0 */
childpos++; /* rightmost child is smallest */
if (endpos != PyList_GET_SIZE(heap)) { if (endpos != PyList_GET_SIZE(heap)) {
PyErr_SetString(PyExc_RuntimeError, PyErr_SetString(PyExc_RuntimeError,
"list changed size during iteration"); "list changed size during iteration");
@ -83,10 +85,11 @@ siftup(PyListObject *heap, Py_ssize_t pos)
} }
} }
/* Move the smaller child up. */ /* Move the smaller child up. */
tmp1 = PyList_GET_ITEM(heap, childpos); arr = _PyList_ITEMS(heap);
tmp2 = PyList_GET_ITEM(heap, pos); tmp1 = arr[childpos];
PyList_SET_ITEM(heap, childpos, tmp2); tmp2 = arr[pos];
PyList_SET_ITEM(heap, pos, tmp1); arr[childpos] = tmp2;
arr[pos] = tmp1;
pos = childpos; pos = childpos;
} }
/* Bubble it up to its final resting place (by sifting its parents down). */ /* Bubble it up to its final resting place (by sifting its parents down). */
@ -227,7 +230,7 @@ heappushpop(PyObject *self, PyObject *args)
} }
cmp = PyObject_RichCompareBool(PyList_GET_ITEM(heap, 0), item, Py_LT); cmp = PyObject_RichCompareBool(PyList_GET_ITEM(heap, 0), item, Py_LT);
if (cmp == -1) if (cmp < 0)
return NULL; return NULL;
if (cmp == 0) { if (cmp == 0) {
Py_INCREF(item); Py_INCREF(item);
@ -362,7 +365,7 @@ PyDoc_STRVAR(heapify_doc,
static int static int
siftdown_max(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos) siftdown_max(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
{ {
PyObject *newitem, *parent; PyObject *newitem, *parent, **arr;
Py_ssize_t parentpos, size; Py_ssize_t parentpos, size;
int cmp; int cmp;
@ -375,12 +378,13 @@ siftdown_max(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
/* Follow the path to the root, moving parents down until finding /* Follow the path to the root, moving parents down until finding
a place newitem fits. */ a place newitem fits. */
newitem = PyList_GET_ITEM(heap, pos); arr = _PyList_ITEMS(heap);
newitem = arr[pos];
while (pos > startpos) { while (pos > startpos) {
parentpos = (pos - 1) >> 1; parentpos = (pos - 1) >> 1;
parent = PyList_GET_ITEM(heap, parentpos); parent = arr[parentpos];
cmp = PyObject_RichCompareBool(parent, newitem, Py_LT); cmp = PyObject_RichCompareBool(parent, newitem, Py_LT);
if (cmp == -1) if (cmp < 0)
return -1; return -1;
if (size != PyList_GET_SIZE(heap)) { if (size != PyList_GET_SIZE(heap)) {
PyErr_SetString(PyExc_RuntimeError, PyErr_SetString(PyExc_RuntimeError,
@ -389,10 +393,11 @@ siftdown_max(PyListObject *heap, Py_ssize_t startpos, Py_ssize_t pos)
} }
if (cmp == 0) if (cmp == 0)
break; break;
parent = PyList_GET_ITEM(heap, parentpos); arr = _PyList_ITEMS(heap);
newitem = PyList_GET_ITEM(heap, pos); parent = arr[parentpos];
PyList_SET_ITEM(heap, parentpos, newitem); newitem = arr[pos];
PyList_SET_ITEM(heap, pos, parent); arr[parentpos] = newitem;
arr[pos] = parent;
pos = parentpos; pos = parentpos;
} }
return 0; return 0;
@ -402,7 +407,7 @@ static int
siftup_max(PyListObject *heap, Py_ssize_t pos) siftup_max(PyListObject *heap, Py_ssize_t pos)
{ {
Py_ssize_t startpos, endpos, childpos, limit; Py_ssize_t startpos, endpos, childpos, limit;
PyObject *tmp1, *tmp2; PyObject *tmp1, *tmp2, **arr;
int cmp; int cmp;
assert(PyList_Check(heap)); assert(PyList_Check(heap));
@ -414,19 +419,19 @@ siftup_max(PyListObject *heap, Py_ssize_t pos)
} }
/* Bubble up the smaller child until hitting a leaf. */ /* Bubble up the smaller child until hitting a leaf. */
arr = _PyList_ITEMS(heap);
limit = endpos / 2; /* smallest pos that has no child */ limit = endpos / 2; /* smallest pos that has no child */
while (pos < limit) { while (pos < limit) {
/* Set childpos to index of smaller child. */ /* Set childpos to index of smaller child. */
childpos = 2*pos + 1; /* leftmost child position */ childpos = 2*pos + 1; /* leftmost child position */
if (childpos + 1 < endpos) { if (childpos + 1 < endpos) {
cmp = PyObject_RichCompareBool( cmp = PyObject_RichCompareBool(
PyList_GET_ITEM(heap, childpos + 1), arr[childpos + 1],
PyList_GET_ITEM(heap, childpos), arr[childpos],
Py_LT); Py_LT);
if (cmp == -1) if (cmp < 0)
return -1; return -1;
if (cmp == 0) childpos += ((unsigned)cmp ^ 1); /* increment when cmp==0 */
childpos++; /* rightmost child is smallest */
if (endpos != PyList_GET_SIZE(heap)) { if (endpos != PyList_GET_SIZE(heap)) {
PyErr_SetString(PyExc_RuntimeError, PyErr_SetString(PyExc_RuntimeError,
"list changed size during iteration"); "list changed size during iteration");
@ -434,10 +439,11 @@ siftup_max(PyListObject *heap, Py_ssize_t pos)
} }
} }
/* Move the smaller child up. */ /* Move the smaller child up. */
tmp1 = PyList_GET_ITEM(heap, childpos); arr = _PyList_ITEMS(heap);
tmp2 = PyList_GET_ITEM(heap, pos); tmp1 = arr[childpos];
PyList_SET_ITEM(heap, childpos, tmp2); tmp2 = arr[pos];
PyList_SET_ITEM(heap, pos, tmp1); arr[childpos] = tmp2;
arr[pos] = tmp1;
pos = childpos; pos = childpos;
} }
/* Bubble it up to its final resting place (by sifting its parents down). */ /* Bubble it up to its final resting place (by sifting its parents down). */