From mboxrd@z Thu Jan 1 00:00:00 1970 Content-Type: text/plain; charset=gb2312 Content-Transfer-Encoding: quoted-printable References: <71A3F6B7-CC61-468D-B8B2-3D46AB92483D@gmail.com> <45fced5cac8155366565f0195c0b47b9@lilly.quanstro.net> <68D3A127-CECA-4E8D-951F-756992D11710@gmail.com> <915a03201f12f2e4e5879a1f1a8b9e23@lilly.quanstro.net> From: Kenny Lasse Hoff Levinsen In-Reply-To: <915a03201f12f2e4e5879a1f1a8b9e23@lilly.quanstro.net> Message-Id: <220EB545-B402-4D69-BEA1-D2EAE06A8A6F@gmail.com> Date: Tue, 23 Feb 2016 18:14:10 +0100 To: Fans of the OS Plan 9 from Bell Labs <9fans@9fans.net> Mime-Version: 1.0 (1.0) Subject: Re: [9fans] Go: FP in note handler Topicbox-Message-UUID: 890cdb98-ead9-11e9-9d60-3106f5b1d025 A proper duffcopy/duffzero/memmove is also an option. Best regards, Kenny Levinsen > On 23. feb. 2016, at 18.02, erik quanstrom wrote: >=20 >> On Tue Feb 23 07:55:26 PST 2016, kennylevinsen@gmail.com wrote: >> A benchmark was supposedly made of the new duffcopy/duffzero which claime= d significant speedup for larger copies: https://github.com/golang/go/commit= /5cf281a9b791f0f10efd1574934cbb19ea1b33da >>=20 >> I have no clue whether this holds true or not. My intention to reenable d= uffcopy and continue to use duffzero is mostly to avoid differences and ensu= re that the note handlers are floating point free in the future. Whether the= duffcopy/duffzero=A1=AFs current form is an actual optimization or just a c= omplexity, I cannot say. A test was made in #cat-v out of annoyance where th= e result seemed to be that it was indeed faster to use MOVUPS, but I don=A1=AF= t remember the details. >=20 > that post is a speedup relative to the original asm, which might not be as= good as the best > non-sse versions, and it is also for amd64. >=20 > - erik >=20