Frequent thing I see here on HN and many other sites is the immediate dismissal of VLIW... This sort of archtiecture has had a bad rap since it was first talked about (over 30 years ago now), to the present day. This is really encouraging me to write a post on why VLIWs have been unjustly shit on, and why what most people think of VLIWs (E.g. Itanium) is not a true VLIW.
At least the linked article gives a mostly fair (but very brief) description of VLIW, but then inappropriately compares it to Intel/HP's EPIC, which while inspired by VLIW, went so far off the rails, which in my opinion, led it to be a failed architecture.
Please do, I would love to hear a piece with a pro-VLIW perspective.
I always found Transmeta's approach to VLIW very interesting. In a way, it did for VLIW what current GPUs are doing for SIMD: take the dispatching away from the compiler and use the more rich runtime information to fill in the slots.
On a side-note, how does Rex's architecture compare to other VLIW tiled mesh-based architectures from before (e.g. Tilera)?
Ok. Just link to examples where it works vs modern processors while avoiding failures of Itanium etc. Outside of embedded space or constraints: HPC or at least desktop-grade stuff. That should either strongly support your claim or settle the debate depending on quality of evidence.
At least the linked article gives a mostly fair (but very brief) description of VLIW, but then inappropriately compares it to Intel/HP's EPIC, which while inspired by VLIW, went so far off the rails, which in my opinion, led it to be a failed architecture.