How do I enable retiming in yosys-abc?

[oharboe]

Simply adding "retime" to the abc script didn't seem to have any effect at all: #1781

I would have expected regressions somewhere.... Certainly on area vs. setup time.

Unless it worked and the change is always better :-)

This is curious... I see some big improvements https://jenkins.openroad.tools/blue/organizations/jenkins/OpenROAD-flow-scripts-Public%2Fpublic_tests_all-pr/detail/PR-1781-head/2/pipeline/1139

[INFO] synth__design__instance__area__stdcell pass test: 85242.672 <= 98630.16
[INFO] constraints__clocks__count pass test: 1.0 == 1.0
[INFO] placeopt__design__instance__area pass test: 103573.0 <= 122596.0
[INFO] placeopt__design__instance__count__stdcell pass test: 8519.0 <= 10094.0
[INFO] detailedplace__design__violations pass test: 0.0 == 0.0
[INFO] cts__design__instance__count__setup_buffer pass test: 0.0 <= 878.0
[INFO] cts__design__instance__count__hold_buffer pass test: 0.0 <= 878.0
[INFO] detailedroute__route__wirelength pass test: 358603.0 <= 402019.0
[INFO] detailedroute__route__drc_errors pass test: 0.0 <= 0.0
[INFO] finish__timing__setup__ws pass test: 0.407628 >= 0.0
[INFO] finish__design__instance__area pass test: 113390.0 <= 123114.0
[INFO] finish__timing__drv__setup_violation_count pass test: 0.0 <= 439.0
[INFO] finish__timing__drv__hold_violation_count pass test: 0.0 <= 100.0
[INFO] finish__timing__wns_percent_delay pass test: 7.185671 >= -10.0

[maliberty]

I doubt you can in synthesis as yosys doesn't pass any sequentials to abc afaik. It seems more of a function of standalone abc.

[oharboe]

@maliberty Latest master has the same improvements. Does this mean that the reference data is dated?

$ make DESIGN_CONFIG=designs/sky130hs/riscv32i/config.mk metadata
./logs/sky130hs/riscv32i/base
Log                       Elapsed seconds
1_1_yosys                         20
2_1_floorplan                      2
2_2_floorplan_io                   1
2_3_floorplan_tdms                 1
2_4_floorplan_macro                1
2_5_floorplan_tapcell              1
2_6_floorplan_pdn                  1
3_1_place_gp_skip_io               2
3_2_place_iop                      1
3_3_place_gp                      19
3_4_place_resized                  5
3_5_place_dp                       4
4_1_cts                           17
5_1_grt                           32
5_2_fillcell                       1
5_3_route                       1087
6_1_merge                          1
6_report                          13
Total                           1209
[WARN] Overwriting Tag finish__runtime__total
[WARN] Overwriting Tag finish__cpu__total
[WARN] Overwriting Tag finish__mem__peak
[INFO] synth__design__instance__area__stdcell pass test: 84954.96 <= 98630.16
[INFO] constraints__clocks__count pass test: 1.0 == 1.0
[INFO] placeopt__design__instance__area pass test: 102531.0 <= 122596.0
[INFO] placeopt__design__instance__count__stdcell pass test: 8543.0 <= 10094.0
[INFO] detailedplace__design__violations pass test: 0.0 == 0.0
[INFO] cts__design__instance__count__setup_buffer pass test: 0.0 <= 878.0
[INFO] cts__design__instance__count__hold_buffer pass test: 0.0 <= 878.0
[INFO] detailedroute__route__wirelength pass test: 378862.0 <= 402019.0
[INFO] detailedroute__route__drc_errors pass test: 0.0 <= 0.0
[INFO] finish__timing__setup__ws pass test: 0.137163 >= 0.0
[INFO] finish__design__instance__area pass test: 113576.0 <= 123114.0
[INFO] finish__timing__drv__setup_violation_count pass test: 0.0 <= 439.0
[INFO] finish__timing__drv__hold_violation_count pass test: 0.0 <= 100.0
[INFO] finish__timing__wns_percent_delay pass test: 2.308676 >= -10.0
All metadata rules passed (14 rules)

[maliberty]

The reference numbers are from when the metrics were last updated. It is better to do a direct comparison yourself on such a change. https://dashboard.theopenroadproject.org/ can be helpful