Date: Tue, 23 Mar 2010 13:40:01 -0700 From: "Matthew Fleming" <matthew.fleming@isilon.com> To: "Kostik Belousov" <kostikbel@gmail.com>, <arch@freebsd.org>, <amd64@freebsd.org> Subject: RE: XMM register usage in the kernel Message-ID: <06D5F9F6F655AD4C92E28B662F7F853E0388A6A9@seaxch09.desktop.isilon.com> In-Reply-To: <20100322095701.GE2415@deviant.kiev.zoral.com.ua> References: <20100322095701.GE2415@deviant.kiev.zoral.com.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
> The KPI consists of two functions, fpu_kern_enter() and fpu_kern_leave(). > The pair should brace the region of code that intend to use coprocessor. > Caller shall provide the context save area, I expect that usually this > memory will be allocated as part of larger subsystem data structure. >=20 > fpu_kern_enter() allows the nesting, assuming fpu_kern_leave() is called > proper number of times to pop the context stack. Isilon would like it if something like this functionality ended up in CURRENT. We have some hacks that do something similar. We use SSE2 instructions to significantly speed up computing ... stuff. :-) Thanks, matthew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?06D5F9F6F655AD4C92E28B662F7F853E0388A6A9>