gnustep/libs-base
0
0
Fork 0
mirror of https://github.com/gnustep/libs-base.git synced 2025-04-22 16:33:29 +00:00
Code Issues Projects Releases Packages Wiki Activity

Restructuring and various optimisations to drastically improve appendFormat:

Browse source 
performance.


git-svn-id: svn+ssh://svn.gna.org/svn/gnustep/libs/base/trunk@19307 72102866-910b-0410-8b05-ffd578937521
This commit is contained in:
Richard Frith-Macdonald 2004-05-14 10:52:30 +00:00
parent 543b3df634
commit c6df2e3852
8 changed files with 677 additions and 761 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
ChangeLog
View file

@ -1,3 +1,19 @@
2004-05-14 Richard Frith-Macdonald <rfm@gnu.org>
* Source/GSFormat.h:
* Source/GSFormat.m:
* Source/GSPrivate.h:
* Source/GSString.m:
* Source/GSeq.h:
* Source/NSString.m:
Updates to allow GSFormat to write directly into an existing string
in either 16bit or 8bit format. Also tidyups and modifications to
use temporary buffers on stack rather than malloc/free as long as
the buffers are reasonably sized.
* Source/benchmark.m: Trivial string format benchmarks ...
Performance improvement of 10% for initWithFormat and 90% for
appendFormat operations.
2004-05-13 Richard Frith-Macdonald <rfm@gnu.org>
* Source/NSCalendarDate.m: ([initWithString:calendarFormat:locale:])

10
Source/GSFormat.h
View file

@ -26,18 +26,12 @@
#define __GSFormat_H_
#include <Foundation/NSZone.h>
#include "GSPrivate.h"
@class NSDictionary;
typedef struct {
unichar *buf;
size_t len;
size_t size;
NSZone *z;
} FormatBuf_t;
void
GSFormat(FormatBuf_t *fb, const unichar *fmt, va_list ap, NSDictionary *loc);
GSFormat(GSStr fb, const unichar *fmt, va_list ap, NSDictionary *loc);
#endif

52
Source/GSFormat.m
View file

@ -127,7 +127,7 @@ struct printf_info
};
/* Type of a printf specifier-handler function.
STREAM is the FormatBuf_t on which to write output.
STREAM is the GSStr on which to write output.
INFO gives information about the format specification.
ARGS is a vector of pointers to the argument data;
the number of pointers will be the number returned
@ -733,32 +733,8 @@ parse_one_spec (const unichar *format, size_t posn, struct printf_spec *spec,
}
#define outchar(Ch) \
do \
{ \
register const wint_t outc = (Ch); \
if (s->len+1 >= s->size) { \
s->size += s->size / 2; \
s->buf = NSZoneRealloc(s->z, s->buf, s->size*sizeof(s->buf[0])); \
} \
s->buf[s->len++] = outc; \
++done; \
} \
while (0)
#define outstring(String, Len) \
do \
{ \
unsigned i; \
\
if (s->len+((unsigned)(Len)) >= s->size) { \
s->size += s->size/2 > ((unsigned)(Len))? s->size/2: (unsigned)(Len); \
s->buf = NSZoneRealloc(s->z, s->buf, s->size*sizeof(s->buf[0])); \
} \
for (i=0; i < ((unsigned)(Len)); i++) s->buf[s->len++] = (String)[i]; \
done += (unsigned)(Len); \
} \
while (0)
#define outchar(Ch) GSStrAppendUnichar(s, Ch)
#define outstring(String, Len) GSStrAppendUnichars(s, String, Len)
/* For handling long_double and longlong we use the same flag. If
`long' and `long long' are effectively the same type define it to
@ -787,7 +763,7 @@ static const unichar null[] = {'(','n','u','l','l',')','\0'};
/* Handle unknown format specifier. */
static int printf_unknown (FormatBuf_t *, const struct printf_info *,
static int printf_unknown (GSStr, const struct printf_info *,
const void *const *);
/* Group digits of number string. */
@ -796,7 +772,7 @@ static unichar *group_number (unichar *, unichar *, const char *, NSString *);
/* The function itself. */
void
GSFormat (FormatBuf_t *s, const unichar *format, va_list ap,
GSFormat (GSStr s, const unichar *format, va_list ap,
NSDictionary *locale)
{
/* The character used as thousands separator. */
@ -866,12 +842,6 @@ NSDictionary *locale)
#define CHAR_CLASS(Ch) (jump_table[(wint_t) (Ch) - ' '])
# define JUMP_TABLE_TYPE const void *const
if (s->size == 0)
{
s->buf = NSZoneMalloc(s->z, 100*sizeof(unichar));
s->size = 100;
}
/* Initialize local variables. */
done = 0;
grouping = (const char *) -1;
@ -1768,7 +1738,7 @@ NSDictionary *locale)
/* Allocate dynamically an array which definitely is long
enough for the wide character version. */
if (len < 8192
|| ((string = (unichar *) NSZoneMalloc(s->z, len * sizeof (unichar)))
|| ((string = (unichar *) NSZoneMalloc(s->_zone, len * sizeof (unichar)))
== NULL))
string = (unichar *) alloca (len * sizeof (unichar));
else
@ -1795,7 +1765,7 @@ NSDictionary *locale)
/* Allocate dynamically an array which definitely is long
enough for the wide character version. */
if (len < 8192
|| ((string = (unichar *) NSZoneMalloc(s->z, len * sizeof (unichar)))
|| ((string = (unichar *) NSZoneMalloc(s->_zone, len * sizeof (unichar)))
== NULL))
string = (unichar *) alloca (len * sizeof (unichar));
else
@ -1819,7 +1789,7 @@ NSDictionary *locale)
PAD (' ');
}
if (string_malloced)
NSZoneFree(s->z, string);
NSZoneFree(s->_zone, string);
}
break;
@ -1851,7 +1821,7 @@ NSDictionary *locale)
/* Allocate dynamically an array which definitely is long
enough for the wide character version. */
if (len < 8192
|| ((string = (unichar *) NSZoneMalloc(s->z, len * sizeof (unichar)))
|| ((string = (unichar *) NSZoneMalloc(s->_zone, len * sizeof (unichar)))
== NULL))
string = (unichar *) alloca (len * sizeof (unichar));
else
@ -1875,7 +1845,7 @@ NSDictionary *locale)
PAD (' ');
}
if (string_malloced)
NSZoneFree(s->z, string);
NSZoneFree(s->_zone, string);
}
break;
@ -1919,7 +1889,7 @@ all_done:
/* Handle an unknown format specifier. This prints out a canonicalized
representation of the format spec itself. */
static int
printf_unknown (FormatBuf_t *s, const struct printf_info *info,
printf_unknown (GSStr s, const struct printf_info *info,
const void *const *args)
{

36
Source/GSPrivate.h
View file

@ -106,6 +106,42 @@ GS_EXPORT BOOL GSIsByteEncoding(NSStringEncoding encoding);
}
@end
/*
* GSMutableString - concrete mutable string, capable of changing its storage
* from holding 8-bit to 16-bit character set.
*/
@interface GSMutableString : NSMutableString
{
union {
unichar *u;
unsigned char *c;
} _contents;
unsigned int _count;
struct {
unsigned int wide: 1;
unsigned int free: 1;
unsigned int unused: 2;
unsigned int hash: 28;
} _flags;
NSZone *_zone;
unsigned int _capacity;
}
@end
/*
* Typedef for access to internals of concrete string objects.
*/
typedef struct {
@defs(GSMutableString)
} GSStr_t;
typedef GSStr_t *GSStr;
/*
* Functions to append to GSStr
*/
extern void GSStrAppendUnichar(GSStr s, unichar);
extern void GSStrAppendUnichars(GSStr s, const unichar *u, unsigned l);
/*
* Enumeration for MacOS-X compatibility user defaults settings.
* For efficiency, we save defaults information which is used by the

940
Source/GSString.m
View file

File diff suppressed because it is too large Load diff

8
Source/GSeq.h
View file

@ -290,14 +290,14 @@ static inline void GSeq_uppercase(GSeq seq)
* Set up macros for dealing with 'self' on the basis of GSQ_S
*/
#if GSEQ_S == GSEQ_US
#define GSEQ_ST ivars
#define GSEQ_ST GSStr
#define GSEQ_SLEN s->_count
#define GSEQ_SGETC(I) s->_contents.u[I]
#define GSEQ_SGETR(B,R) memcpy(B, &s->_contents.u[R.location], 2*(R).length)
#define GSEQ_SRANGE(I) (*srImp)((id)s, ranSel, I)
#else
#if GSEQ_S == GSEQ_CS
#define GSEQ_ST ivars
#define GSEQ_ST GSStr
#define GSEQ_SLEN s->_count
#define GSEQ_SGETC(I) (unichar)s->_contents.c[I]
#define GSEQ_SGETR(B,R) ( { \
@ -322,14 +322,14 @@ static inline void GSeq_uppercase(GSeq seq)
* Set up macros for dealing with 'other' string on the basis of GSQ_O
*/
#if GSEQ_O == GSEQ_US
#define GSEQ_OT ivars
#define GSEQ_OT GSStr
#define GSEQ_OLEN o->_count
#define GSEQ_OGETC(I) o->_contents.u[I]
#define GSEQ_OGETR(B,R) memcpy(B, &o->_contents.u[R.location], 2*(R).length)
#define GSEQ_ORANGE(I) (*orImp)((id)o, ranSel, I)
#else
#if GSEQ_O == GSEQ_CS
#define GSEQ_OT ivars
#define GSEQ_OT GSStr
#define GSEQ_OLEN o->_count
#define GSEQ_OGETC(I) (unichar)o->_contents.c[I]
#define GSEQ_OGETR(B,R) ( { \

356
Source/NSString.m
View file

@ -69,6 +69,7 @@
// For private method _decodePropertyListForKey:
#include "Foundation/NSKeyedArchiver.h"
#include "GNUstepBase/GSMime.h"
#include "GSPrivate.h"
#include "GSFormat.h"
#include <limits.h>
#include <sys/stat.h>
@ -1153,321 +1154,70 @@ handle_printf_atsign (FILE *stream,
locale: (NSDictionary*)locale
arguments: (va_list)argList
{
FormatBuf_t f;
unichar *fmt;
unsigned char buf[2048];
GSStr_t f;
unichar fbuf[1024];
unichar *fmt = fbuf;
size_t len;
/*
* First we provide an array of unichar characters containing the
* format string. For performance reasons we try to use an on-stack
* buffer if the format string is small enough ... it almost always
* will be.
*/
len = [format length];
fmt = objc_malloc((len+1)*sizeof(unichar));
if (len >= 1024)
{
fmt = objc_malloc((len+1)*sizeof(unichar));
}
[format getCharacters: fmt];
fmt[len] = '\0';
f.z = NSDefaultMallocZone();
f.buf = NSZoneMalloc(f.z, 100*sizeof(unichar));
f.len = 0;
f.size = 100;
/*
* Now set up 'f' as a GSMutableString object whose initial buffer is
* allocated on the stack. The GSFormat function can write into it.
*/
f.isa = GSMutableStringClass;
f._zone = NSDefaultMallocZone();
f._contents.c = buf;
f._capacity = sizeof(buf);
f._count = 0;
f._flags.wide = 0;
f._flags.free = 0;
GSFormat(&f, fmt, argList, locale);
objc_free(fmt);
// don't use noCopy because f.size > f.len!
self = [self initWithCharacters: f.buf length: f.len];
NSZoneFree(f.z, f.buf);
return self;
}
#if 0
/* xxx Change this when we have non-CString classes */
- (id) initWithFormat: (NSString*)format
locale: (NSDictionary*)locale
arguments: (va_list)argList
{
#if defined(HAVE_VSPRINTF) || defined(HAVE_VASPRINTF)
const char *format_cp = [format lossyCString];
int format_len = strlen (format_cp);
#ifdef HAVE_VASPRINTF
char *buf;
int printed_len = 0;
NSString *ret;
#ifndef HAVE_REGISTER_PRINTF_FUNCTION
NSZone *z = GSObjCZone(self);
/* If the available libc doesn't have `register_printf_function()', then
the `%@' printf directive isn't available with printf() and friends.
Here we make a feable attempt to handle it. */
{
/* We need a local copy since we change it. (Changing and undoing
the change doesn't work because some format strings are constant
strings, placed in a non-writable section of the executable, and
writing to them will cause a segfault.) */
char format_cp_copy[format_len+1];
char *atsign_pos; /* points to a location inside format_cp_copy */
char *format_to_go = format_cp_copy;
char *buf_l;
#define _PRINTF_BUF_LEN 256
int printed_local_len, avail_len = _PRINTF_BUF_LEN;
int cstring_len;
buf = NSZoneMalloc(z, _PRINTF_BUF_LEN);
strcpy (format_cp_copy, format_cp);
/* Loop once for each `%@' in the format string. */
while ((atsign_pos = strstr (format_to_go, "%@")))
{
const char *cstring;
char *formatter_pos; // Position for formatter.
/* If there is a "%%@", then do the right thing: print it literally. */
if ((*(atsign_pos-1) == '%')
&& atsign_pos != format_cp_copy)
continue;
/* Temporarily terminate the string before the `%@'. */
*atsign_pos = '\0';
/* Print the part before the '%@' */
printed_local_len = VASPRINTF_LENGTH (vasprintf (&buf_l,
format_to_go, argList));
if(buf_l)
{
if(avail_len < printed_local_len+1)
{
NS_DURING
{
buf = NSZoneRealloc(z, buf,
printed_len+printed_local_len+_PRINTF_BUF_LEN);
avail_len += _PRINTF_BUF_LEN;
}
NS_HANDLER
{
free(buf_l);
[localException raise];
}
NS_ENDHANDLER
}
memcpy(&buf[printed_len], buf_l, printed_local_len+1);
avail_len -= printed_local_len;
printed_len += printed_local_len;
free(buf_l);
}
else
{
[NSException raise: NSMallocException
format: @"No available memory"];
}
/* Skip arguments used in last vsprintf(). */
while ((formatter_pos = strchr(format_to_go, '%')))
{
char *spec_pos; // Position of conversion specifier.
if (*(formatter_pos+1) == '%')
{
format_to_go = formatter_pos+2;
continue;
}
spec_pos = strpbrk(formatter_pos+1, "dioxXucsfeEgGpn\0");
switch (*spec_pos)
{
#ifndef powerpc
/* FIXME: vsprintf on powerpc apparently advances the arg list
so this doesn't need to be done. Make a more general check
for this */
case 'd': case 'i': case 'o':
case 'x': case 'X': case 'u': case 'c':
va_arg(argList, int);
break;
case 's':
if (*(spec_pos - 1) == '*')
va_arg(argList, int*);
va_arg(argList, char*);
break;
case 'f': case 'e': case 'E': case 'g': case 'G':
va_arg(argList, double);
break;
case 'p':
va_arg(argList, void*);
break;
case 'n':
va_arg(argList, int*);
break;
#endif /* NOT powerpc */
case '\0':
spec_pos--;
break;
}
format_to_go = spec_pos+1;
}
/* Get a C-string (char*) from the String object, and print it. */
cstring = [[(id) va_arg (argList, id) description] lossyCString];
if (!cstring)
cstring = "<null string>";
cstring_len = strlen(cstring);
if(cstring_len)
{
if(avail_len < cstring_len+1)
{
buf = NSZoneRealloc(z, buf,
printed_len+cstring_len+_PRINTF_BUF_LEN);
avail_len += _PRINTF_BUF_LEN;
}
memcpy(&buf[printed_len], cstring, cstring_len+1);
avail_len -= cstring_len;
printed_len += cstring_len;
}
/* Skip over this `%@', and look for another one. */
format_to_go = atsign_pos + 2;
}
/* Print the rest of the string after the last `%@'. */
printed_local_len = VASPRINTF_LENGTH (vasprintf (&buf_l,
format_to_go, argList));
if(buf_l)
{
if(avail_len < printed_local_len+1)
{
NS_DURING
{
buf = NSZoneRealloc(z, buf,
printed_len+printed_local_len+_PRINTF_BUF_LEN);
avail_len += _PRINTF_BUF_LEN;
}
NS_HANDLER
{
free(buf_l);
[localException raise];
}
NS_ENDHANDLER
}
memcpy(&buf[printed_len], buf_l, printed_local_len+1);
avail_len -= printed_local_len;
printed_len += printed_local_len;
free(buf_l);
}
else
{
[NSException raise: NSMallocException
format: @"No available memory"];
}
}
#else /* HAVE_VSPRINTF */
/* The available libc has `register_printf_function()', so the `%@'
printf directive is handled by printf and friends. */
printed_len = VASPRINTF_LENGTH (vasprintf (&buf, format_cp, argList));
if(!buf)
if (fmt != fbuf)
{
[NSException raise: NSMallocException
format: @"No available memory"];
objc_free(fmt);
}
#endif /* !HAVE_REGISTER_PRINTF_FUNCTION */
ret = [self initWithCString: buf];
#ifndef HAVE_REGISTER_PRINTF_FUNCTION
NSZoneFree(z, buf);
#else
free(buf);
#endif
return ret;
#else
/* xxx horrible disgusting BUFFER_EXTRA arbitrary limit; fix this! */
#define BUFFER_EXTRA 1024*500
char buf[format_len + BUFFER_EXTRA];
int printed_len = 0;
/*
* Don't use noCopy because f._contents.u may be memory on the stack,
* and even if it wasn't f._capacity may be greater than f._count so
* we could be wasting quite a bit of space. Better to accept a
* performance hit due to copying data (and allocating/deallocating
* the temporary buffer) for large strings. For most strings, the
* on-stack memory will have been used, so we will get better performance.
*/
if (f._flags.wide == 1)
{
self = [self initWithCharacters: f._contents.u length: f._count];
}
else
{
self = [self initWithCString: f._contents.c length: f._count];
}
#ifndef HAVE_REGISTER_PRINTF_FUNCTION
/* If the available libc doesn't have `register_printf_function()', then
the `%@' printf directive isn't available with printf() and friends.
Here we make a feable attempt to handle it. */
{
/* We need a local copy since we change it. (Changing and undoing
the change doesn't work because some format strings are constant
strings, placed in a non-writable section of the executable, and
writing to them will cause a segfault.) */
char format_cp_copy[format_len+1];
char *atsign_pos; /* points to a location inside format_cp_copy */
char *format_to_go = format_cp_copy;
strcpy (format_cp_copy, format_cp);
/* Loop once for each `%@' in the format string. */
while ((atsign_pos = strstr (format_to_go, "%@")))
{
const char *cstring;
char *formatter_pos; // Position for formatter.
/* If there is a "%%@", then do the right thing: print it literally. */
if ((*(atsign_pos-1) == '%')
&& atsign_pos != format_cp_copy)
continue;
/* Temporarily terminate the string before the `%@'. */
*atsign_pos = '\0';
/* Print the part before the '%@' */
printed_len += VSPRINTF_LENGTH (vsprintf (buf+printed_len,
format_to_go, argList));
/* Skip arguments used in last vsprintf(). */
while ((formatter_pos = strchr(format_to_go, '%')))
{
char *spec_pos; // Position of conversion specifier.
if (*(formatter_pos+1) == '%')
{
format_to_go = formatter_pos+2;
continue;
}
spec_pos = strpbrk(formatter_pos+1, "dioxXucsfeEgGpn\0");
switch (*spec_pos)
{
#ifndef powerpc
/* FIXME: vsprintf on powerpc apparently advances the arg list
so this doesn't need to be done. Make a more general check
for this */
case 'd': case 'i': case 'o':
case 'x': case 'X': case 'u': case 'c':
(void)va_arg(argList, int);
break;
case 's':
if (*(spec_pos - 1) == '*')
(void)va_arg(argList, int*);
(void)va_arg(argList, char*);
break;
case 'f': case 'e': case 'E': case 'g': case 'G':
(void)va_arg(argList, double);
break;
case 'p':
(void)va_arg(argList, void*);
break;
case 'n':
(void)va_arg(argList, int*);
break;
#endif /* NOT powerpc */
case '\0':
spec_pos--;
break;
}
format_to_go = spec_pos+1;
}
/* Get a C-string (char*) from the String object, and print it. */
cstring = [[(id) va_arg (argList, id) description] lossyCString];
if (!cstring)
cstring = "<null string>";
strcat (buf+printed_len, cstring);
printed_len += strlen (cstring);
/* Skip over this `%@', and look for another one. */
format_to_go = atsign_pos + 2;
}
/* Print the rest of the string after the last `%@'. */
printed_len += VSPRINTF_LENGTH (vsprintf (buf+printed_len,
format_to_go, argList));
}
#else
/* The available libc has `register_printf_function()', so the `%@'
printf directive is handled by printf and friends. */
printed_len = VSPRINTF_LENGTH (vsprintf (buf, format_cp, argList));
#endif /* !HAVE_REGISTER_PRINTF_FUNCTION */
/* Raise an exception if we overran our buffer. */
NSParameterAssert (printed_len < format_len + BUFFER_EXTRA - 1);
return [self initWithCString: buf];
#endif /* HAVE_VASPRINTF */
#else /* HAVE_VSPRINTF || HAVE_VASPRINTF */
[self notImplemented: _cmd];
/*
* If the string had to grow beyond the initial buffer size, we must
* release any allocated memory.
*/
if (f._flags.free == 1)
{
NSZoneFree(f._zone, f._contents.c);
}
return self;
#endif
}
#endif
/**
* Initialises the receiver with the supplied data, using the

20
Testing/benchmark.m
View file

@ -371,6 +371,7 @@ bench_str()
{
int i;
NSString *str;
NSMutableString *ms;
id plist;
NSString *plstr;
Class arc = [NSArchiver class];
@ -408,6 +409,25 @@ bench_str()
plstr = [plist description];
printf("NSString\n");
START_TIMER;
for (i = 0; i < MAX_COUNT; i++)
{
str = [[stringClass alloc] initWithFormat: @"Hello %d", i];
RELEASE(str);
}
END_TIMER;
PRINT_TIMER("NSString (1 initWithFormat:) \t");
ms = [NSMutableString stringWithCapacity: 0];
START_TIMER;
for (i = 0; i < MAX_COUNT; i++)
{
[ms appendFormat: @"%d", i];
}
END_TIMER;
PRINT_TIMER("NSString (1 appendFormat:) \t\t");
START_TIMER;
for (i = 0; i < MAX_COUNT; i++)
{