With the way architectures have gone, I think you'd end up recreating VLIW. The thing holding back VLIW was compilers were too dumb and computers too slow to really take advantage of it. You ended up with a lot of "NOP"s as a result in the output. VLIW is essentially how modern GPUs operate.

The main benefit of VLIW is that it simplifies the processor design by moving the complicated tasks/circuitry into the compiler. Theoretically, the compiler has more information about the intent of the program which allows it to better optimize things.

It would also be somewhat of a security boon. VLIW moves the branch prediction (and rewinding) into the processor. With exploits like spectre, pulling that out would make it easier to integrate compiler hints on security sensitive code "hey, don't spec ex here".

> The thing holding back VLIW was compilers were too dumb

That’s not really the problem.

The real issue is that VLIW requires branches to be strongly biased, statically, so a compiler can exploit them.

But in fact branches are very dynamic but trivially predicted by branch predictors, so branch predictors win.

Not to mention that even vliw cores use branch predictors, because the branch resolution latency is too long to wait for the branch outcome to be known.