Re: [eigen] about changeset 6eb14e380

[ Thread Index | Date Index | More Archives ]

2010/8/18 joel falcou <joel.falcou@xxxxxx>
On 18/08/10 17:52, Benoit Jacob wrote:
if yes, I wonder, because rows/cols are not too relevant to matrix products and other blocked algorithms.
hm, are you saying that you are aligning each row/col?
table are aligned globally, matrix are aligned on each leading dimension so we are sure each leading block is occupiyng full cache line alone.
Don't forget we openMPify every _expression_ as long as our CT check told us it was useful and not only matrix stuff.

ah OK. This makes sense. matrix products and other blocked algos often dwarve other use cases, but indeed the other uses cases are not negligible.

....and for the record, the other big reason: we want to be friendly to linux distros wrt binary libraries using Eigen. if libfoo uses Eigen, and you're a distro packager for x86, you want to ship only 1 package for libfoo.. You decide whether you build it with SSE/AVX/nothing. Then  You're free to do so because ABI is independent on that. Then app developers/packagers can link against libfoo regardless of how libfoo was built and regardless of their own simd settings. This is basically a requirement in environments, like linux distros, which rely heavily on system-wide shared libraries (rather than app bundles).

Ah yeah of course. I'll have to investigate this behavior in our case (again different target public)

Yes :-)


Mail converted by MHonArc 2.6.19+