Processor competition in the desktop market has been lackluster since that time. Intel has been sitting on their laurels mostly. The vectored primitives and bigger cache and core counts are nice but honestly you have been fooled by the branding and continual product lines.
Loop unrolling still helps even with brand spankin new CPUs. Some intent cannot be inferred.
> Processor competition in the desktop market has been lackluster since that time. Intel has been sitting on their laurels mostly. The vectored primitives and bigger cache and core counts are nice but honestly you have been fooled by the branding and continual product lines.
I might be inclined to agree with you... if I didn't know that Haswell:
* implemented the computed goto transformation (for interpreter loops) in hardware
* made unaligned cache hits crossing cache line boundaries have no latency penalty
Both of those things are things that I would have told you were impossible(/infeasible) to do in hardware before finding out that Intel did them.
There's actually quite a lot of amazing microarchitectural changes going on in Intel processors, you just don't hear a lot about them.
