This is the mail archive of the 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: [PATCH] Improves __ieee754_exp() performance by greater than 5x on sparc/x86.

On Monday 11 December 2017 10:29 PM, Patrick McGehearty wrote:
> log, log10, pow, cbrt are on my short list of functions to investigate
> as these all show significant performance advantage (1.8x or greater)
> using the Solaris Studio libm functions vs the Linux libm functions in
> the preliminary testing I did months ago. I intend to take these one
> at a time, first with a trial port and extensive accuracy and perf tests.
> Assuming the performance advantage applies across multiple platforms
> and accuracy does not suffer to an unacceptable degree, I will
> propose each in turn for patching. "log" is my next target, likely
> with log10 following close behind. I'll also look at the 32 bit functions
> to see if they offer similar opportunities, but I haven't written/run those
> tests yet. I don't have plans to work on 80 bit or 128 bit (long double)
> functions at this time.
> As always, the above is not a formal commitment as my management
> may redirect efforts to meet other corporate goals.
> My background project right now is supporting Vladimir Mezentsev's
> work on improving accuracy and range of complex divide.
> By range, I mean the range of input values which currently cause
> overflow/underflow but don't need to with appropriate scaling.

Sure, I won't hold you to it :)


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