Rooba
/
cpython
mirror of https://github.com/python/cpython.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Issue #13624: Write a specialized UTF-8 encoder to allow more optimization 

The main bottleneck was the PyUnicode_READ() macro.
This commit is contained in:
Victor Stinner 2011-12-18 14:22:26 +01:00
parent b66dcb66ba
commit 6099a03202
3 changed files with 212 additions and 150 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Doc/whatsnew/3.3.rst
View File

@ -712,7 +712,9 @@ Major performance enhancements have been added:
* the memory footprint is divided by 2 to 4 depending on the text
* encode an ASCII string to UTF-8 doesn't need to encode characters anymore,
the UTF-8 representation is shared with the ASCII representation
* getting a substring of a latin1 strings is 4 times faster
* the UTF-8 encoder has been optimized
* repeating a single ASCII letter and getting a substring of a ASCII strings
is 4 times faster
Build and C API Changes

197
Objects/stringlib/codecs.h
View File

@ -153,4 +153,201 @@ STRINGLIB(utf8_try_decode)(const char *start, const char *end,
#undef LONG_PTR_MASK
#undef ASCII_CHAR_MASK
/* UTF-8 encoder specialized for a Unicode kind to avoid the slow
PyUnicode_READ() macro. Delete some parts of the code depending on the kind:
UCS-1 strings don't need to handle surrogates for example. */
Py_LOCAL_INLINE(PyObject *)
STRINGLIB(utf8_encoder)(PyObject *unicode,
STRINGLIB_CHAR *data,
Py_ssize_t size,
const char *errors)
{
#define MAX_SHORT_UNICHARS 300 /* largest size we'll do on the stack */
Py_ssize_t i; /* index into s of next input byte */
PyObject *result; /* result string object */
char *p; /* next free byte in output buffer */
Py_ssize_t nallocated; /* number of result bytes allocated */
Py_ssize_t nneeded; /* number of result bytes needed */
#if STRINGLIB_SIZEOF_CHAR > 1
PyObject *errorHandler = NULL;
PyObject *exc = NULL;
PyObject *rep = NULL;
#endif
#if STRINGLIB_SIZEOF_CHAR == 1
const Py_ssize_t max_char_size = 2;
char stackbuf[MAX_SHORT_UNICHARS * 2];
#elif STRINGLIB_SIZEOF_CHAR == 2
const Py_ssize_t max_char_size = 3;
char stackbuf[MAX_SHORT_UNICHARS * 3];
#else /* STRINGLIB_SIZEOF_CHAR == 4 */
const Py_ssize_t max_char_size = 4;
char stackbuf[MAX_SHORT_UNICHARS * 4];
#endif
assert(size >= 0);
if (size <= MAX_SHORT_UNICHARS) {
/* Write into the stack buffer; nallocated can't overflow.
* At the end, we'll allocate exactly as much heap space as it
* turns out we need.
*/
nallocated = Py_SAFE_DOWNCAST(sizeof(stackbuf), size_t, int);
result = NULL; /* will allocate after we're done */
p = stackbuf;
}
else {
if (size > PY_SSIZE_T_MAX / max_char_size) {
/* integer overflow */
return PyErr_NoMemory();
}
/* Overallocate on the heap, and give the excess back at the end. */
nallocated = size * max_char_size;
result = PyBytes_FromStringAndSize(NULL, nallocated);
if (result == NULL)
return NULL;
p = PyBytes_AS_STRING(result);
}
for (i = 0; i < size;) {
Py_UCS4 ch = data[i++];
if (ch < 0x80) {
/* Encode ASCII */
*p++ = (char) ch;
}
else
#if STRINGLIB_SIZEOF_CHAR > 1
if (ch < 0x0800)
#endif
{
/* Encode Latin-1 */
*p++ = (char)(0xc0 | (ch >> 6));
*p++ = (char)(0x80 | (ch & 0x3f));
}
#if STRINGLIB_SIZEOF_CHAR > 1
else if (Py_UNICODE_IS_SURROGATE(ch)) {
Py_ssize_t newpos;
Py_ssize_t repsize, k, startpos;
startpos = i-1;
rep = unicode_encode_call_errorhandler(
errors, &errorHandler, "utf-8", "surrogates not allowed",
unicode, &exc, startpos, startpos+1, &newpos);
if (!rep)
goto error;
if (PyBytes_Check(rep))
repsize = PyBytes_GET_SIZE(rep);
else
repsize = PyUnicode_GET_LENGTH(rep);
if (repsize > max_char_size) {
Py_ssize_t offset;
if (result == NULL)
offset = p - stackbuf;
else
offset = p - PyBytes_AS_STRING(result);
if (nallocated > PY_SSIZE_T_MAX - repsize + max_char_size) {
/* integer overflow */
PyErr_NoMemory();
goto error;
}
nallocated += repsize - max_char_size;
if (result != NULL) {
if (_PyBytes_Resize(&result, nallocated) < 0)
goto error;
} else {
result = PyBytes_FromStringAndSize(NULL, nallocated);
if (result == NULL)
goto error;
Py_MEMCPY(PyBytes_AS_STRING(result), stackbuf, offset);
}
p = PyBytes_AS_STRING(result) + offset;
}
if (PyBytes_Check(rep)) {
char *prep = PyBytes_AS_STRING(rep);
for(k = repsize; k > 0; k--)
*p++ = *prep++;
} else /* rep is unicode */ {
enum PyUnicode_Kind repkind;
void *repdata;
if (PyUnicode_READY(rep) < 0)
goto error;
repkind = PyUnicode_KIND(rep);
repdata = PyUnicode_DATA(rep);
for(k=0; k<repsize; k++) {
Py_UCS4 c = PyUnicode_READ(repkind, repdata, k);
if (0x80 <= c) {
raise_encode_exception(&exc, "utf-8",
unicode,
i-1, i,
"surrogates not allowed");
goto error;
}
*p++ = (char)c;
}
}
Py_CLEAR(rep);
}
else
#if STRINGLIB_SIZEOF_CHAR > 2
if (ch < 0x10000)
#endif
{
*p++ = (char)(0xe0 | (ch >> 12));
*p++ = (char)(0x80 | ((ch >> 6) & 0x3f));
*p++ = (char)(0x80 | (ch & 0x3f));
}
#if STRINGLIB_SIZEOF_CHAR > 2
else /* ch >= 0x10000 */
{
assert(ch <= MAX_UNICODE);
/* Encode UCS4 Unicode ordinals */
*p++ = (char)(0xf0 | (ch >> 18));
*p++ = (char)(0x80 | ((ch >> 12) & 0x3f));
*p++ = (char)(0x80 | ((ch >> 6) & 0x3f));
*p++ = (char)(0x80 | (ch & 0x3f));
}
#endif /* STRINGLIB_SIZEOF_CHAR > 2 */
#endif /* STRINGLIB_SIZEOF_CHAR > 1 */
}
if (result == NULL) {
/* This was stack allocated. */
nneeded = p - stackbuf;
assert(nneeded <= nallocated);
result = PyBytes_FromStringAndSize(stackbuf, nneeded);
}
else {
/* Cut back to size actually needed. */
nneeded = p - PyBytes_AS_STRING(result);
assert(nneeded <= nallocated);
_PyBytes_Resize(&result, nneeded);
}
#if STRINGLIB_SIZEOF_CHAR > 1
Py_XDECREF(errorHandler);
Py_XDECREF(exc);
#endif
return result;
#if STRINGLIB_SIZEOF_CHAR > 1
error:
Py_XDECREF(rep);
Py_XDECREF(errorHandler);
Py_XDECREF(exc);
Py_XDECREF(result);
return NULL;
#endif
#undef MAX_SHORT_UNICHARS
}
#endif /* STRINGLIB_IS_UNICODE */

161
Objects/unicodeobject.c
View File

@ -4987,20 +4987,9 @@ _Py_DecodeUTF8_surrogateescape(const char *s, Py_ssize_t size)
PyObject *
_PyUnicode_AsUTF8String(PyObject *unicode, const char *errors)
{
#define MAX_SHORT_UNICHARS 300 /* largest size we'll do on the stack */
Py_ssize_t i; /* index into s of next input byte */
PyObject *result; /* result string object */
char *p; /* next free byte in output buffer */
Py_ssize_t nallocated; /* number of result bytes allocated */
Py_ssize_t nneeded; /* number of result bytes needed */
char stackbuf[MAX_SHORT_UNICHARS * 4];
PyObject *errorHandler = NULL;
PyObject *exc = NULL;
int kind;
enum PyUnicode_Kind kind;
void *data;
Py_ssize_t size;
PyObject *rep = NULL;
if (!PyUnicode_Check(unicode)) {
PyErr_BadArgument();
@ -5018,144 +5007,18 @@ _PyUnicode_AsUTF8String(PyObject *unicode, const char *errors)
data = PyUnicode_DATA(unicode);
size = PyUnicode_GET_LENGTH(unicode);
assert(size >= 0);
if (size <= MAX_SHORT_UNICHARS) {
/* Write into the stack buffer; nallocated can't overflow.
* At the end, we'll allocate exactly as much heap space as it
* turns out we need.
*/
nallocated = Py_SAFE_DOWNCAST(sizeof(stackbuf), size_t, int);
result = NULL; /* will allocate after we're done */
p = stackbuf;
switch(kind) {
default:
assert(0);
case PyUnicode_1BYTE_KIND:
/* the string cannot be ASCII, or PyUnicode_UTF8() would be set */
assert(!PyUnicode_IS_ASCII(unicode));
return ucs1lib_utf8_encoder(unicode, data, size, errors);
case PyUnicode_2BYTE_KIND:
return ucs2lib_utf8_encoder(unicode, data, size, errors);
case PyUnicode_4BYTE_KIND:
return ucs4lib_utf8_encoder(unicode, data, size, errors);
}
else {
/* Overallocate on the heap, and give the excess back at the end. */
nallocated = size * 4;
if (nallocated / 4 != size) /* overflow! */
return PyErr_NoMemory();
result = PyBytes_FromStringAndSize(NULL, nallocated);
if (result == NULL)
return NULL;
p = PyBytes_AS_STRING(result);
}
for (i = 0; i < size;) {
Py_UCS4 ch = PyUnicode_READ(kind, data, i++);
if (ch < 0x80)
/* Encode ASCII */
*p++ = (char) ch;
else if (ch < 0x0800) {
/* Encode Latin-1 */
*p++ = (char)(0xc0 | (ch >> 6));
*p++ = (char)(0x80 | (ch & 0x3f));
} else if (Py_UNICODE_IS_SURROGATE(ch)) {
Py_ssize_t newpos;
Py_ssize_t repsize, k, startpos;
startpos = i-1;
rep = unicode_encode_call_errorhandler(
errors, &errorHandler, "utf-8", "surrogates not allowed",
unicode, &exc, startpos, startpos+1, &newpos);
if (!rep)
goto error;
if (PyBytes_Check(rep))
repsize = PyBytes_GET_SIZE(rep);
else
repsize = PyUnicode_GET_LENGTH(rep);
if (repsize > 4) {
Py_ssize_t offset;
if (result == NULL)
offset = p - stackbuf;
else
offset = p - PyBytes_AS_STRING(result);
if (nallocated > PY_SSIZE_T_MAX - repsize + 4) {
/* integer overflow */
PyErr_NoMemory();
goto error;
}
nallocated += repsize - 4;
if (result != NULL) {
if (_PyBytes_Resize(&result, nallocated) < 0)
goto error;
} else {
result = PyBytes_FromStringAndSize(NULL, nallocated);
if (result == NULL)
goto error;
Py_MEMCPY(PyBytes_AS_STRING(result), stackbuf, offset);
}
p = PyBytes_AS_STRING(result) + offset;
}
if (PyBytes_Check(rep)) {
char *prep = PyBytes_AS_STRING(rep);
for(k = repsize; k > 0; k--)
*p++ = *prep++;
} else /* rep is unicode */ {
enum PyUnicode_Kind repkind;
void *repdata;
if (PyUnicode_READY(rep) < 0)
goto error;
repkind = PyUnicode_KIND(rep);
repdata = PyUnicode_DATA(rep);
for(k=0; k<repsize; k++) {
Py_UCS4 c = PyUnicode_READ(repkind, repdata, k);
if (0x80 <= c) {
raise_encode_exception(&exc, "utf-8",
unicode,
i-1, i,
"surrogates not allowed");
goto error;
}
*p++ = (char)c;
}
}
Py_CLEAR(rep);
} else if (ch < 0x10000) {
*p++ = (char)(0xe0 | (ch >> 12));
*p++ = (char)(0x80 | ((ch >> 6) & 0x3f));
*p++ = (char)(0x80 | (ch & 0x3f));
} else /* ch >= 0x10000 */ {
assert(ch <= MAX_UNICODE);
/* Encode UCS4 Unicode ordinals */
*p++ = (char)(0xf0 | (ch >> 18));
*p++ = (char)(0x80 | ((ch >> 12) & 0x3f));
*p++ = (char)(0x80 | ((ch >> 6) & 0x3f));
*p++ = (char)(0x80 | (ch & 0x3f));
}
}
if (result == NULL) {
/* This was stack allocated. */
nneeded = p - stackbuf;
assert(nneeded <= nallocated);
result = PyBytes_FromStringAndSize(stackbuf, nneeded);
}
else {
/* Cut back to size actually needed. */
nneeded = p - PyBytes_AS_STRING(result);
assert(nneeded <= nallocated);
_PyBytes_Resize(&result, nneeded);
}
Py_XDECREF(errorHandler);
Py_XDECREF(exc);
return result;
error:
Py_XDECREF(rep);
Py_XDECREF(errorHandler);
Py_XDECREF(exc);
Py_XDECREF(result);
return NULL;
#undef MAX_SHORT_UNICHARS
}
PyObject *