While creating nmigen designs for MachXO2 that require Diamond, I noticed that the synthesis was significantly slower than compared to the manual shell script that @gatecat created for prjtrellis. I talked to them today, and we determined that removing the -forceAll option from each step significantly increases the synthesis speed because synthesis steps are not run repeatedly. By extension, removing -forceAll significantly reduces the output log size from running build_*.{bat, sh} > log.txt. @gatecat doesn't believe -forceAll is necessary at this time, and looking at a test bitstream built without -forceAll in EPIC, I'm inclined to agree.

I have created an MVCE with all the files already generated to show the synthesis speeds/log file sizes. Run python3 mk_build.py > log.txt to regenerate the output log. log-noforceall was generated by removing -forceAll manually and running the build script directly. The size difference between the logs is significant; 800 vs 4700 lines (the latter doing completely redundant work?).

Files

MVCE: diamond-test.zip
Log of the above, with -forceAll manually removed from the generated .bat (should also apply to .sh): log-noforceall.txt