This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc 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: Contributing to the glibc Project



On 13/10/2016 14:03, Joseph Myers wrote:
> On Thu, 13 Oct 2016, Adhemerval Zanella wrote:
> 
>> For strlen we do have an armv6, armv6 thumb2, and the default one
>> which should work on minimum supported arm (armv5?).
> 
> Minimum is v4 (though anything less than v4t may not work well with the 
> EABI).
> 
>> Also note that for armv7+ we use ifunc to select of different
>> implementation for memcpy. One possible work I would suggest you
>> and use the memcpy* for amrv7 and create also a multiarch memmove
>> implementation based on current armv7 strategies, since armv7 
>> still uses the default memmove.S implementation.
> 
> Note that any string function patches motivated by performance need 
> thorough before-and-after benchmarking to justify that they are 
> improvements for the affected processors.  In addition to running the 
> glibc testsuite of course - and for processors such as ARM that support 
> both endiannesses, string tests need running for both endiannesses.
> 

Indeed, all new performance algorithms should follow the usual
rationale.  I suggested memmove only because ARM default one
(sysdeps/arm/memmove.S) could use the similar strategies as already
placed on specialized one (__memcpy_{vfp,neon}) for backwards
copy.


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