From mboxrd@z Thu Jan 1 00:00:00 1970 From: erik quanstrom Date: Tue, 23 Feb 2016 09:02:27 -0800 To: 9fans@9fans.net Message-ID: <915a03201f12f2e4e5879a1f1a8b9e23@lilly.quanstro.net> In-Reply-To: <68D3A127-CECA-4E8D-951F-756992D11710@gmail.com> References: <71A3F6B7-CC61-468D-B8B2-3D46AB92483D@gmail.com> <45fced5cac8155366565f0195c0b47b9@lilly.quanstro.net> <68D3A127-CECA-4E8D-951F-756992D11710@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Subject: Re: [9fans] Go: FP in note handler Topicbox-Message-UUID: 88ff72d2-ead9-11e9-9d60-3106f5b1d025 On Tue Feb 23 07:55:26 PST 2016, kennylevinsen@gmail.com wrote: > A benchmark was supposedly made of the new duffcopy/duffzero which clai= med significant speedup for larger copies: https://github.com/golang/go/c= ommit/5cf281a9b791f0f10efd1574934cbb19ea1b33da >=20 > I have no clue whether this holds true or not. My intention to reenable= duffcopy and continue to use duffzero is mostly to avoid differences and= ensure that the note handlers are floating point free in the future. Whe= ther the duffcopy/duffzero=E2=80=99s current form is an actual optimizati= on or just a complexity, I cannot say. A test was made in #cat-v out of a= nnoyance where the result seemed to be that it was indeed faster to use M= OVUPS, but I don=E2=80=99t remember the details. that post is a speedup relative to the original asm, which might not be a= s good as the best non-sse versions, and it is also for amd64. - erik