Where will D sit in the web service space?
via Digitalmars-d
digitalmars-d at puremagic.com
Sat Jul 25 08:14:59 PDT 2015
On Saturday, 25 July 2015 at 14:49:36 UTC, David Nadlinger wrote:
> Why would that be so? You can still e.g. keep vector
> instructions in the IR and specialize according to the actual
> hardware width when you generate the binary.
You have several rather peculiar SIMD instructions that reflect
the specifics of the CPU pipeline that can shave off quite a few
cycles for specific algorithms. Different CPUs have different
penalties for alignments, ARM can do rather well with
misalignments. Different CPUs have different primitives, e.g. on
ARM complex functions like sincos can be broken down into simpler
primitives, so you can do less exact calculations (fewer
iterations than on x86).
More information about the Digitalmars-d
mailing list