Typical timing closure strategies involve running multiple implementation strategies and selecting the best one for lab testing. Machine learning (ML) strategies provide an alternative by requiring only three strategies to achieve a similar QoR benefit. ML strategies analyze features from a post-route design to predict the performance of different strategies on the same design.
The best three strategies are stored in RQS files generated by
report_qor_suggestions (and write_qor_suggestions)
and can be applied in subsequent runs. This approach significantly reduces the server
resources required.
When the directive is set to RQS on implementation commands, the tool references the RQS file for both the directive and other tool command options. The flow is shown in the following figure:
Strategy Suggestions Flow
- Generate strategy suggestions by running
report_qor_suggestionson a fully routed design generated using eitherDefaultorExploredirectives. For detailed requirements, see ML Strategy Availability. - Write the strategy RQS files by using the following command to
generate the required RQS files in the specified directory:
write_qor_suggestions -strategy_dir <dir>. By default, three strategies are generated. Each strategy RQS file contains all suggestion objects and the strategy suggestions object. The RQS file created usingwrite_qor_suggestions -file <fn>.rqscan be discarded because the information is duplicated in the strategy RQS files.Note: To generate more strategies, increase the number with:report_qor_suggestions -max_strategies <n> - Load the strategy RQS file by reading the generated RQS file into the new implementation run.
- Run the RQS strategy flow by setting the directive to RQS. Include calls to
opt_design,place_design,phys_opt_designandroute_designin the script.