[Bf-committers] 64-bit builds with ray optimization
sergey.vfx at gmail.com
Mon Dec 12 08:45:44 CET 2011
Tomas, as i understand we're speaking about x86_64 platform, not x86 and
all x86_64 bit compatible processor are comng with sse2 at least (i'm not
sure if sse3 is used in out code). It wouldn't be a problem to enable
rayoptimization and compile with sse/sse2 support (which also improves
speed of tracking).
Main issue is about x86 platforms. Yes, we've got code which detects if
sse2 is supported and it's already used in Cycles and blender internal,
we're only missing such check for motion tracking which is compiling
without sse2 support on x86 platforms. The here issue is: Eigen library can
use sse2/sse3 instructions to improve speed of matrix operations, but Eigen
is totally header-based library and making runtime detection of sse support
simply wouldn't work here.
So i think we should compile 64bit builds of Blender with rayoptimization,
sse and sse2 optimizations. For 32bit keep global sse optimizations
disabled and enable only rayoptimization (it seems to be quite safe).
On Mon, Dec 12, 2011 at 10:59 AM, Thomas Dinges <blender at dingto.org> wrote:
> SSE3 was introduced in 2004 by Intel (Pentium 4) and in 2005 by AMD.
> Thats 6/7 years now already.
> If there is any code that can't detect SSE2/3 on runtime, we should just
> make extra builds. But making only 1 build (even for x86) would be stupid.
> It's okay to support 10 year old cpu's who don't have these extensions,
> but at the same time we should not disable these optimizations for
> people using, well "old" hardware too. 6/7 years is old. ;-)
> Am 12.12.2011 02:10, schrieb Antony Riakiotakis:
> > Hi everyone, it's release time soon.
> > There has been some debate on irc on the topic of enabling ray
> > optimization for release builds.
> > Pros are: faster program, cons of course are that we are less
> > compatible with older machines.
> > I see though no reason why 64-bit machines, that support sse2, should
> > not be built with ray optimization on by default.
> > If this is decided upon, there are some -sse3 flags in cycles that
> > should probably be taken care of.
> > _______________________________________________
> > Bf-committers mailing list
> > Bf-committers at blender.org
> > http://lists.blender.org/mailman/listinfo/bf-committers
> Thomas Dinges
> Blender Developer, Artist and Musician
> Bf-committers mailing list
> Bf-committers at blender.org
With best regards, Sergey Sharybin
More information about the Bf-committers