Date: Sun, 12 Dec 2021 18:22:23 -0800 From: Steve Kargl <sgk@troutmask.apl.washington.edu> To: Mark Murray <markm@freebsd.org> Cc: freebsd-hackers@freebsd.org, freebsd-current@freebsd.org Subject: Re: What to do about tgammal? Message-ID: <20211213022223.GA41440@troutmask.apl.washington.edu> In-Reply-To: <E5711C71-1095-4B6B-A33A-4CDFF123AB62@FreeBSD.org> References: <20211204185352.GA20452@troutmask.apl.washington.edu> <E5711C71-1095-4B6B-A33A-4CDFF123AB62@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Dec 04, 2021 at 11:48:13PM +0000, Mark Murray wrote:
>
>
> > On 4 Dec 2021, at 18:53, Steve Kargl <sgk@troutmask.apl.washington.edu> wrote:
> >
> >
> > So, is anyone interested in seeing a massive patch?
>
> Me, please!
>
So, I have the ld80 case working. I'll open a PR if
on does not exist. The pr will contain
1) Diff for nonfunctional changes to code for tgamma(x).
This is a re-organization of the code so I could reverse
it. Some cleanup of comments and documentations. A
move towards style(9)
2) Advice on what to do about tgammal(x) on ld128 hardware.
Essentially, need to do the git equivalent of 'svn mv
imprecise.c ld128/b_tgammal.c.
3) Diff with the new code for ld80 tgamma.
Interval tested for tgammal: [6,1755.1]
5000000 calls, 2.056068 secs, 0.41121 usecs/call
ulp <= 0.5: 90.233% 4511666 | 90.233% 4511666
0.5 < ulp < 0.6: 7.061% 353033 | 97.294% 4864699
0.6 < ulp < 0.7: 2.393% 119644 | 99.687% 4984343
0.7 < ulp < 0.8: 0.300% 14981 | 99.986% 4999324
0.8 < ulp < 0.9: 0.014% 676 | 100.000% 5000000
Max ulp: 0.873414 at 1.480588145237629047468e+03
Interval tested for tgammal: [1.0662,6]
5000000 calls, 0.748966 secs, 0.14979 usecs/call
ulp <= 0.5: 96.355% 4817737 | 96.355% 4817737
0.5 < ulp < 0.6: 3.291% 164534 | 99.645% 4982271
0.6 < ulp < 0.7: 0.328% 16403 | 99.973% 4998674
0.7 < ulp < 0.8: 0.026% 1297 | 99.999% 4999971
0.8 < ulp < 0.9: 0.001% 29 | 100.000% 5000000
Max ulp: 0.861508 at 1.999467927053585410537e+00
Interval tested for tgammal: [1.01e-17,1.0661]
5000000 calls, 0.904246 secs, 0.18085 usecs/call
ulp <= 0.5: 96.201% 4810043 | 96.201% 4810043
0.5 < ulp < 0.6: 3.297% 164875 | 99.498% 4974918
0.6 < ulp < 0.7: 0.412% 20581 | 99.910% 4995499
0.7 < ulp < 0.8: 0.082% 4108 | 99.992% 4999607
0.8 < ulp < 0.9: 0.008% 389 | 100.000% 4999996
0.9 < ulp < 1.0: 0.000% 4 | 100.000% 5000000
Max ulp: 0.938041 at 1.023286481537296307856e+00
Interval tested for tgammal: [-1.9999,-1.0001]
5000000 calls, 1.740820 secs, 0.34816 usecs/call
ulp <= 0.5: 56.596% 2829776 | 56.596% 2829776
0.5 < ulp < 0.6: 8.616% 430793 | 65.211% 3260569
0.6 < ulp < 0.7: 7.452% 372621 | 72.664% 3633190
0.7 < ulp < 0.8: 6.250% 312511 | 78.914% 3945701
0.8 < ulp < 0.9: 5.147% 257372 | 84.061% 4203073
0.9 < ulp < 1.0: 4.100% 205010 | 88.162% 4408083
1.0 < ulp < 1.5: 9.846% 492324 | 98.008% 4900407
1.5 < ulp < 2.0: 1.790% 89514 | 99.798% 4989921
2.0 < ulp < 3.0: 0.201% 10074 | 100.000% 4999995
3.0 < ulp < 0.0: 0.000% 5 | 100.000% 5000000
--
Steve
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20211213022223.GA41440>