[Python-checkins] cpython: Issue #25349: Optimize bytes % int
victor.stinner
python-checkins at python.org
Fri Oct 9 17:04:44 EDT 2015
https://hg.python.org/cpython/rev/d9a89c9137d2
changeset: 98622:d9a89c9137d2
user: Victor Stinner <victor.stinner at gmail.com>
date: Fri Oct 09 22:43:24 2015 +0200
summary:
Issue #25349: Optimize bytes % int
Optimize bytes.__mod__(args) for integere formats: %d (%i, %u), %o, %x and %X.
_PyBytesWriter is now used to format directly the integer into the writer
buffer, instead of using a temporary bytes object.
Formatting is between 30% and 50% faster on a microbenchmark.
files:
Include/longobject.h | 7 +
Objects/bytesobject.c | 36 +++++++++
Objects/longobject.c | 120 +++++++++++++++++++++++------
3 files changed, 136 insertions(+), 27 deletions(-)
diff --git a/Include/longobject.h b/Include/longobject.h
--- a/Include/longobject.h
+++ b/Include/longobject.h
@@ -182,6 +182,13 @@
int base,
int alternate);
+PyAPI_FUNC(char*) _PyLong_FormatBytesWriter(
+ _PyBytesWriter *writer,
+ char *str,
+ PyObject *obj,
+ int base,
+ int alternate);
+
/* Format the object based on the format_spec, as defined in PEP 3101
(Advanced String Formatting). */
PyAPI_FUNC(int) _PyLong_FormatAdvancedWriter(
diff --git a/Objects/bytesobject.c b/Objects/bytesobject.c
--- a/Objects/bytesobject.c
+++ b/Objects/bytesobject.c
@@ -841,6 +841,42 @@
case 'o':
case 'x':
case 'X':
+ if (PyLong_CheckExact(v)
+ && width == -1 && prec == -1
+ && !(flags & (F_SIGN | F_BLANK))
+ && c != 'X')
+ {
+ /* Fast path */
+ int alternate = flags & F_ALT;
+ int base;
+
+ switch(c)
+ {
+ default:
+ assert(0 && "'type' not in [diuoxX]");
+ case 'd':
+ case 'i':
+ case 'u':
+ base = 10;
+ break;
+ case 'o':
+ base = 8;
+ break;
+ case 'x':
+ case 'X':
+ base = 16;
+ break;
+ }
+
+ /* Fast path */
+ writer.min_size -= 2; /* size preallocated by "%d" */
+ res = _PyLong_FormatBytesWriter(&writer, res,
+ v, base, alternate);
+ if (res == NULL)
+ goto error;
+ continue;
+ }
+
temp = formatlong(v, flags, prec, c);
if (!temp)
goto error;
diff --git a/Objects/longobject.c b/Objects/longobject.c
--- a/Objects/longobject.c
+++ b/Objects/longobject.c
@@ -1582,7 +1582,9 @@
static int
long_to_decimal_string_internal(PyObject *aa,
PyObject **p_output,
- _PyUnicodeWriter *writer)
+ _PyUnicodeWriter *writer,
+ _PyBytesWriter *bytes_writer,
+ char **bytes_str)
{
PyLongObject *scratch, *a;
PyObject *str;
@@ -1664,6 +1666,13 @@
kind = writer->kind;
str = NULL;
}
+ else if (bytes_writer) {
+ *bytes_str = _PyBytesWriter_Prepare(bytes_writer, *bytes_str, strlen);
+ if (*bytes_str == NULL) {
+ Py_DECREF(scratch);
+ return -1;
+ }
+ }
else {
str = PyUnicode_New(strlen, '9');
if (str == NULL) {
@@ -1673,13 +1682,8 @@
kind = PyUnicode_KIND(str);
}
-#define WRITE_DIGITS(TYPE) \
+#define WRITE_DIGITS(p) \
do { \
- if (writer) \
- p = (TYPE*)PyUnicode_DATA(writer->buffer) + writer->pos + strlen; \
- else \
- p = (TYPE*)PyUnicode_DATA(str) + strlen; \
- \
/* pout[0] through pout[size-2] contribute exactly \
_PyLong_DECIMAL_SHIFT digits each */ \
for (i=0; i < size - 1; i++) { \
@@ -1699,6 +1703,16 @@
/* and sign */ \
if (negative) \
*--p = '-'; \
+ } while (0)
+
+#define WRITE_UNICODE_DIGITS(TYPE) \
+ do { \
+ if (writer) \
+ p = (TYPE*)PyUnicode_DATA(writer->buffer) + writer->pos + strlen; \
+ else \
+ p = (TYPE*)PyUnicode_DATA(str) + strlen; \
+ \
+ WRITE_DIGITS(p); \
\
/* check we've counted correctly */ \
if (writer) \
@@ -1708,25 +1722,34 @@
} while (0)
/* fill the string right-to-left */
- if (kind == PyUnicode_1BYTE_KIND) {
+ if (bytes_writer) {
+ char *p = *bytes_str + strlen;
+ WRITE_DIGITS(p);
+ assert(p == *bytes_str);
+ }
+ else if (kind == PyUnicode_1BYTE_KIND) {
Py_UCS1 *p;
- WRITE_DIGITS(Py_UCS1);
+ WRITE_UNICODE_DIGITS(Py_UCS1);
}
else if (kind == PyUnicode_2BYTE_KIND) {
Py_UCS2 *p;
- WRITE_DIGITS(Py_UCS2);
+ WRITE_UNICODE_DIGITS(Py_UCS2);
}
else {
Py_UCS4 *p;
assert (kind == PyUnicode_4BYTE_KIND);
- WRITE_DIGITS(Py_UCS4);
+ WRITE_UNICODE_DIGITS(Py_UCS4);
}
#undef WRITE_DIGITS
+#undef WRITE_UNICODE_DIGITS
Py_DECREF(scratch);
if (writer) {
writer->pos += strlen;
}
+ else if (bytes_writer) {
+ (*bytes_str) += strlen;
+ }
else {
assert(_PyUnicode_CheckConsistency(str, 1));
*p_output = (PyObject *)str;
@@ -1738,7 +1761,7 @@
long_to_decimal_string(PyObject *aa)
{
PyObject *v;
- if (long_to_decimal_string_internal(aa, &v, NULL) == -1)
+ if (long_to_decimal_string_internal(aa, &v, NULL, NULL, NULL) == -1)
return NULL;
return v;
}
@@ -1750,7 +1773,8 @@
static int
long_format_binary(PyObject *aa, int base, int alternate,
- PyObject **p_output, _PyUnicodeWriter *writer)
+ PyObject **p_output, _PyUnicodeWriter *writer,
+ _PyBytesWriter *bytes_writer, char **bytes_str)
{
PyLongObject *a = (PyLongObject *)aa;
PyObject *v;
@@ -1812,6 +1836,11 @@
kind = writer->kind;
v = NULL;
}
+ else if (writer) {
+ *bytes_str = _PyBytesWriter_Prepare(bytes_writer, *bytes_str, sz);
+ if (*bytes_str == NULL)
+ return -1;
+ }
else {
v = PyUnicode_New(sz, 'x');
if (v == NULL)
@@ -1819,13 +1848,8 @@
kind = PyUnicode_KIND(v);
}
-#define WRITE_DIGITS(TYPE) \
+#define WRITE_DIGITS(p) \
do { \
- if (writer) \
- p = (TYPE*)PyUnicode_DATA(writer->buffer) + writer->pos + sz; \
- else \
- p = (TYPE*)PyUnicode_DATA(v) + sz; \
- \
if (size_a == 0) { \
*--p = '0'; \
} \
@@ -1860,30 +1884,50 @@
} \
if (negative) \
*--p = '-'; \
+ } while (0)
+
+#define WRITE_UNICODE_DIGITS(TYPE) \
+ do { \
+ if (writer) \
+ p = (TYPE*)PyUnicode_DATA(writer->buffer) + writer->pos + sz; \
+ else \
+ p = (TYPE*)PyUnicode_DATA(v) + sz; \
+ \
+ WRITE_DIGITS(p); \
+ \
if (writer) \
assert(p == ((TYPE*)PyUnicode_DATA(writer->buffer) + writer->pos)); \
else \
assert(p == (TYPE*)PyUnicode_DATA(v)); \
} while (0)
- if (kind == PyUnicode_1BYTE_KIND) {
+ if (bytes_writer) {
+ char *p = *bytes_str + sz;
+ WRITE_DIGITS(p);
+ assert(p == *bytes_str);
+ }
+ else if (kind == PyUnicode_1BYTE_KIND) {
Py_UCS1 *p;
- WRITE_DIGITS(Py_UCS1);
+ WRITE_UNICODE_DIGITS(Py_UCS1);
}
else if (kind == PyUnicode_2BYTE_KIND) {
Py_UCS2 *p;
- WRITE_DIGITS(Py_UCS2);
+ WRITE_UNICODE_DIGITS(Py_UCS2);
}
else {
Py_UCS4 *p;
assert (kind == PyUnicode_4BYTE_KIND);
- WRITE_DIGITS(Py_UCS4);
+ WRITE_UNICODE_DIGITS(Py_UCS4);
}
#undef WRITE_DIGITS
+#undef WRITE_UNICODE_DIGITS
if (writer) {
writer->pos += sz;
}
+ else if (bytes_writer) {
+ (*bytes_str) += sz;
+ }
else {
assert(_PyUnicode_CheckConsistency(v, 1));
*p_output = v;
@@ -1897,9 +1941,9 @@
PyObject *str;
int err;
if (base == 10)
- err = long_to_decimal_string_internal(obj, &str, NULL);
+ err = long_to_decimal_string_internal(obj, &str, NULL, NULL, NULL);
else
- err = long_format_binary(obj, base, 1, &str, NULL);
+ err = long_format_binary(obj, base, 1, &str, NULL, NULL, NULL);
if (err == -1)
return NULL;
return str;
@@ -1911,9 +1955,31 @@
int base, int alternate)
{
if (base == 10)
- return long_to_decimal_string_internal(obj, NULL, writer);
+ return long_to_decimal_string_internal(obj, NULL, writer,
+ NULL, NULL);
else
- return long_format_binary(obj, base, alternate, NULL, writer);
+ return long_format_binary(obj, base, alternate, NULL, writer,
+ NULL, NULL);
+}
+
+char*
+_PyLong_FormatBytesWriter(_PyBytesWriter *writer, char *str,
+ PyObject *obj,
+ int base, int alternate)
+{
+ char *str2;
+ int res;
+ str2 = str;
+ if (base == 10)
+ res = long_to_decimal_string_internal(obj, NULL, NULL,
+ writer, &str2);
+ else
+ res = long_format_binary(obj, base, alternate, NULL, NULL,
+ writer, &str2);
+ if (res < 0)
+ return NULL;
+ assert(str2 != NULL);
+ return str2;
}
/* Table of digit values for 8-bit string -> integer conversion.
--
Repository URL: https://hg.python.org/cpython
More information about the Python-checkins
mailing list