Systolic Generation Optimization

[calebmkim]

Optimizes Systolic Array Generation:

• Data that is passed between registers are continuous assignments now, rather than being scheduled. As long as we pass the from memories into these registers in at the right time, and the PEs read from them at the write time, we are fine (i.e., they will pass garbage but the garbage will never be read)
• Also, simplified the schedule for invoking PEs by directly sending a signal into the mul_ready port.

[calebmkim]

Merging this; I'll address any comments later.

Initial experiments show this can reduce LUT usage.

[rachitnigam]

Awesome! Worth pointing out in the paper that this is only possible because of static operators (if you think this argument is true). Specifically, in a dynamic control program, we don't know when the registers will be read so we cannot do these continuous assignments because we might read garbage values.