[kaba0]

I am way out of my depth here, but wouldn’t a machine code to machine code JIT compiler solve the problem of underutilization of itanium? (I remember reading a paper on x86->x86 jit compiler as well that could provide some speed up)

If so complex branch prediction and pipelining can be done in hardware alone, much more clever (and patchable!) optimizations can be done in software, or I would think so. So while mainstream languages may not be able to reuse Itanium’s architecture efficiently at compile time, a separate program could reorder instructions to make use of some instruction level parallelism, couldn’t it?