To reinvent the processor

The fundamental work of an instruction
A typical instruction in a typical instruction set performs something like the following:
Read the operands from the global register file. Out-of-order execution also introduces costs for instruction issue and wakeup: In a fully out-of-order processor, any write to the LRF could make any locally-queued instruction ready to issue, and any write the the GRF could make any instruction ready to issue from any BW. You can get most of the motivation needed to invent the Mill’s belt by noticing how much more headroom operand forwarding has than large global register files, primarily because of the comparatively miniscule number of sources.

Source: medium.com