This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Exploiting knowing sizes of string.

On 04/06/15 15:23, OndÅej BÃlka wrote:
> On Thu, Jun 04, 2015 at 02:59:20PM +0200, Jakub Jelinek wrote:
>> On Thu, Jun 04, 2015 at 02:53:31PM +0200, OndÅej BÃlka wrote:
>>> On Thu, Jun 04, 2015 at 02:33:19PM +0200, Jakub Jelinek wrote:
>>>> On Thu, Jun 04, 2015 at 12:26:03PM +0000, Joseph Myers wrote:
>>>>>> Again is this worth a gcc pass?
>>>>> This isn't a matter of compiler passes; it's additional checks in existing 
>>>>> built-in function handling.  Maybe that built-in function handling should 
>>>>> move to the match-and-simplify infrastructure (some, for libm functions 
>>>>> and bswap, already has) to make this even simpler to implement.
>>>> GCC already has a pass that attempts to track known and earlier computed
>>>> lengths of strings, and do various transformations and optimizations based
>>>> on that, see the tree-ssa-strlen.c pass.  Most of that you really can't do
>>>> at the glibc headers level.
>>> Yes, I was writing down ideas that I have and this was one of these. I
>>> didn't knew it does transformations, just checked that it doesn't use
>>> length from stpcpy or in
>> It does use length from stpcpy in various cases, but stpcpy needs to be
>> prototyped.  See gcc/testsuite/gcc.dg/strlenopt* for what it does.
>>> int foo(char *s)
>>> {
>>>   int l = strlen (s);
>>>   char *p = strchr (s,'a');
>>>   return p+l;
>>> }
>> And what do you want to optimize here?  The length from strlen is different
>> from the difference between p and s.  Furthermore, p can return NULL.
>> 	Jakub
> Change that into
> int foo(char *s)
> {
>   int l = strlen (s);
>   char *p = memchr (s, 'a', l);
>   return p+l;
> }

Which is still meaningless if 'a' does not appear in s => when the
result is NULL + l.

In fact, unless 'a' is the first character the result is possibly
meaningless anyway, since you can't know that p+l doesn't point more
than one beyond the end of the object.

Perhaps you just meant to return 'p'?


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]