Raj Prakash wrote:
> Would a facility of the following nature be useful?
>
> Build two (or more) binaries that are [b]optimized [/b]for different
> processors - for example, one for ultra1 and one for ultra3. A tool combines
> the binaries into a single binary. The combined binary may be used on [b]any
> [/b]Solaris machine and the most appropriate optimized binary will be chosen
> and run. The combined binary can used to package, ship, etc just as your
> usual binaries.
You mean:
http://en.wikipedia.org/wiki/Fat_binary
The subject has come up once or twice, but questions regarding the
nature of the container (is in this new ELF type), and how multiple
objects (libraries) might be mapped efficiently from a container
into memory, resulted in the conversation not going very far.
We've tended to leverage off of filters and such to enable the
use of alternative objects.
Of course there's the other data point that many customers aren't
interested in producing multiple objects, at least not multiple
flavors of optimization. Each new object involves another addition
to their text matrix. They want their one "generic" binary to
run everywhere. If you gave customers a container technology, would
they use it? There hasn't been much "pull" for this technology so far.
Presumably your IPO world sees a use for such a container?
--
Rod.
