This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] PPC64: First in the series of patches implementing POWER8 vector math.
- From: Joseph Myers <joseph at codesourcery dot com>
- To: GT <tnggil at protonmail dot com>
- Cc: "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>
- Date: Mon, 18 Feb 2019 18:38:11 +0000
- Subject: Re: [PATCH] PPC64: First in the series of patches implementing POWER8 vector math.
- References: <7z03WSXwPIOYABvW1ZYzHobMOv90wBWKo-e0YptRFovFUpRnrSEGN-zv08kOe_fPJMFnHx1khbcfcfooubTStEAtMX96Q2U4cRVZOm_TZl8=@protonmail.com> <alpine.DEB.2.21.1902142104290.4022@digraph.polyomino.org.uk> <70qZ5GQ6aTHy6AlzYLn1tWsF7DIOMCpZTuDyYPoSaggIEF1wtsXJMogvDb4UaIOAxbTobvPFwbL7cAKBhbIrGeF4XiuujSzthNZbPCc3Qr4=@protonmail.com>
On Mon, 18 Feb 2019, GT wrote:
> I have addressed all the issues you raised except for one. I haven't
> been able to locate documentation which explains derivations of
> corrections to some constants used in the x86_64 code. Corrections to PI
> when used in repeated FMA operations, and corrections to 1/factorial(n)
> coefficients in the approximation polynomial. I put the question of how
> to obtain these derivations on the mailing list, and am also still
> looking online for a solution. May I resubmit the patch or is the lack
> of clarity regarding these constants a show-stopper?
Well, you could copy the comments from the x86_64 code, if these are the
same constants.
I'd expect, for example, given the comments, that __dPI1_FMA + __dPI2_FMA
+ __dPI3_FMA (using the x86_64 names) is an approximation to pi to about
3*53 bits, which is something that should be easy to verify.
--
Joseph S. Myers
joseph@codesourcery.com