Perf #1473
Perf #1473
Conversation
The arguments to our mbrtowc(3) wrapper should not alias, since they also must not for mbrtowc(3). Might help some compilers optimizing the code.
|
The reason for the
|
/proc/<pid>/stat is parsed for every displayed process on every iteration containing several numbers to parse. Use out own implementations since strto(u)l(3) came up during profiling.
| @@ -78,11 +78,68 @@ static FILE* fopenat(openat_arg_t openatArg, const char* pathname, const char* m | |||
| return fp; | |||
| } | |||
|
|
|||
| static long fast_strtol_dec(char** str, int maxlen) { | |||
There was a problem hiding this comment.
How about just calling fast_stroull_dec() and downcast the result for this one?
static long fast_strtol_dec(char** str, int maxlen) {
long long result = fast_strtoll_dec(str, maxlen);
assert(result <= LONG_MAX);
assert(result >= LONG_MIN);
return (long)result;
}
| result *= 10; | ||
| result += **str - '0'; | ||
| (*str)++; | ||
| } |
There was a problem hiding this comment.
This "while" loop can be converted to a fast_strtoull_dec(str, maxlen) and then cast the result from unsigned to signed.
| static inline uint64_t fast_strtoull_dec(char** str, int maxlen) { | ||
| register uint64_t result = 0; | ||
|
|
||
| if (!maxlen) | ||
| --maxlen; | ||
| maxlen = 20; // 18446744073709551615 |
There was a problem hiding this comment.
Documenting the previous default makes no sense for this line. If at all we want documentation it would explain, that this is "assume max reasonable length if none given". I'm for dropping this comment though …
There was a problem hiding this comment.
Just FYI, the standard strtoull() permits almost unlimited number of leading zeros in the string. While it won't certainly happen in parsing any Linux procfs file, having an arbitrary length limit in a strtoull() implementation would technically violate ISO C.
There was a problem hiding this comment.
The issue here is less about ISO C compliance, but more about squeeze out cycles. fast_strtoull is not meant to comply to ISO C, but to parse the data we actually encounter in realistic situations. Otherwise we'd have to support locales and a full load of other stuff (different radix prefixes) we don't need …
| } | ||
|
|
||
| if (!maxlen) | ||
| maxlen = 20; // −9223372036854775808 |
There was a problem hiding this comment.
If at all, document the why not the magic value …
And the why is basically "because C does not have default arguments" …
| } | ||
|
|
||
| if (!maxlen) | ||
| maxlen = 20; // −9223372036854775808 |
| unsigned long result = 0; | ||
|
|
||
| if (!maxlen) | ||
| maxlen = 20; // 18446744073709551615 |
| while (maxlen-- && **str >= '0' && **str <= '9') { | ||
| result *= 10; | ||
| result += **str - '0'; | ||
| (*str)++; | ||
| } |
There was a problem hiding this comment.
Same replacement of this while loop could be done here too …
Annotate mbstowcs_nonfatal() with restrict
The arguments to our mbrtowc(3) wrapper should not alias, since they also must not for mbrtowc(3).
Might help some compilers optimizing the code.
Linux: use fast integer parsing for /proc//stat
/proc//stat is parsed for every displayed process on every iteration containing several numbers to parse.
Use out own implementations since strto(u)l(3) came up during profiling.
@BenBE what was the reason for adding the parameter
maxlento those functions?