There seems to be no hardware implementation right now, not even on FPGA (last message about it dating from 2014). I also couldn't find a GCC or LLVM backend for its architecture and no software simulator with or without cycle counters.
As much as I'd love to see something new in the general-purpose CPU arena (was Itanic the last attempt?) I will remain skeptical of such claims until they can be demonstrated by actual benchmarks.
The instruction-set doc is slightly out-of-date and we know that and we're regenerating it soon, but most ops have same latencies now as documented. We also know there is scant examples for those who don't want to hand-roll stuff; we are working on doing some prettily-rendered examples, initially without auto-vectorization and loop-pipelining (so not really unlocking the full potential of the Mill, but better than nothing).
> I also couldn't find a GCC or LLVM backend for its architecture and no software simulator with or without cycle counters.
They do have a compiler, and they do have a simulator. Or rather compilers and simulators: one for each Mill family member. The simulator, compiler and related intermediary tooling (and the specification they use to produce them) are the first “real” things they've written and gotten working, and the simulator defines the instruction set. They've done a live demo of editing the specification and producing a new simulator, so I'm inclined to believe there's no subterfuge here.
However, they're not public. Possibly because they haven't finished filing all their patents?
The Mill team seem to have a strong aim for commercial success. Talks and documentation are typically only released after they have secured their patents. I'm also skeptical as to if they will make a viable product at the end of the day. None the less these videos have been some of the most enjoyable technical talks I've come across.
But I sincerely hope they do find commercial success.
On top of this, they have gotten patents filed, and as they do go through the process, information gets released about what is now covered by patent.
As in, they are making good on the promise of actually proving it isn't vaporware, it is just a slow and painful process.
Anything that gets leaked ahead of time can be used to derail their patent process and depending on what it is, it could be the difference between being a profitable venture or not.
Everyone who makes certain types of processors (which includes Intel and AMD both, along with all the ARM licensees) have a vested interest in ruining this effort, no matter if Mill ends up being successful or not (as in, Mill holding certain patents even if their implementation was not successful, is still a highly profitable venture).
As a side note, a lot of people use Itanium as an excuse to shit on Mill: just remember, quite a few things engineered for Itanium have found their way back into today's x86 designs. As in, things Intel patented in Itanium are still in use and lived on. Even though the Mill is strange, if it doesn't succeed, parts of it will live on in other designs.
As much as I'd love to see something new in the general-purpose CPU arena (was Itanic the last attempt?) I will remain skeptical of such claims until they can be demonstrated by actual benchmarks.