Hi,
In general, which type of arithmetic would be done faster on a modern
desktop pc where accuracy and the dynamic range of floating point were
not
important?
a)16 bit fixed point using 8 way SIMD
b)32 bit single precision floating point using 4 way SIMD
>From what I've read, b) would be as fast/faster on modern cpus, but
wouldn't being able to operate on twice the amount of data per
operation
make up for/or exceed that?
Thanks!
