The final comparison, in Table 2, shows synthesis results for UCLOCK (as reported in [8]) and MINIMALIST.
For a fair and interesting comparison, we plugged some of the MINIMALIST tools into the UCLOCK path, to isolate and highlight two differences: (i) machine implementation style (choice of fed-back vs. no fed-back outputs), and (ii) state minimization algorithms. Even though UCLOCK does not use any optimal state assignment algorithms, we attached CHASM and HFMIN as a back end, to isolate these front-end differences. We also limited MINIMALIST to the only logic minimization modes that are available in UCLOCK: the cost function is product cardinality, and the logic implementation style is multi-output.
Table 2 shows the experimental results. In both MINIMALIST and the ``improved'' UCLOCK, reported results are the best of several fixed-length trials at or near the minimum code length. The majority of the MINIMALIST results use the fed-back output machine implementation style.
Not surprisingly, many MINIMALIST and UCLOCK results are nearly identical, since the operating modes are very similar. However, MINIMALIST's use of fed-back outputs results in significant gains in several cases (e.g., dram-ctrl and scsi-isend-bm). In addition, MINIMALIST obtains synthesis results in several cases where UCLOCK failed to complete, again due in part to MINIMALIST's more capable state minimization method.
A performance-oriented comparison to UCLOCK (like the above comparison to 3D) is possible, but is omitted, due to space considerations.
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
