优化策略之PowerOptimization
Power_Opt_Design
1 Available Logic Optimizations
opt_design[-retarget][-propconst][-sweep][-bram_power_opt][-remap]
[-resynth_area][-resynth_seq_area][-directive<arg>][-muxf_remap]
[-hier_fanout_limit<arg>][-bufg_opt][-control_set_merge][-quiet][-verbose]
1.1Retargeting (Default)
Retargetingreplaces one cell type with another to ease optimization. For example, a MUXF7
replaced by aLUT3 can be combined with other LUTs. In addition, simple cells such as
inverters areabsorbed into downstream logic.
1.2Constant Propagation (Default)
ConstantPropagation propagates constant values through logic, which results in:
• Eliminatedlogic:
For example, anAND with a constant 0 input
• Reduced logic:
For example, A3-input AND with a constant 1 input is reduced to a 2-input AND.
• Redundantlogic:
For example, A2-input OR with a logic 0 input is reduced to a wire.
1.3Sweep (Default)
Sweep removescells that have no loads.
1.4Block RAM Power Optimization(Default)
Block RAM PowerOptimization enables power optimization on block RAM cells including:
• Changing the WRITE_MODE on unread ports of true dual-port RAMs to NO_CHANGE.
• Applyingintelligent clock gating to block RAM outputs.
1.5Remap
Remap combinesmultiple LUTs into a single LUT to reduce the depth of the logic.
1.6Resynth Area
Resynth Areaperforms re-synthesis in area mode to reduce the number of LUTs.
1.7Mux Optimization
Remaps MUXF7,MUXF8, and MUXF9 primitives to LUT3 to improve route-ability.
1.8Control Set Merging
Reduce thedrivers of logically-equivalent control signals to a single driver. This islike a reverse fanout replication, and results in nets that are better suitedfor module-based replication.
1.9Global Clock Buffer Insertion
Logicoptimization conservatively inserts global clock buffers on clock nets andhigh-fanout non-clock nets such as device-wide resets.
For 7 seriesdesigns, clock buffers are inserted as long as 12 total global clock buffersarenot exceeded.
For UltraScale designs,there is no limit for clock buffers inserted on clock nets.
For non-clocknets:
• Global clockbuffers are only inserted as long as 24 total clock buffers are not
exceeded, notincluding BUFG_GT buffers.
• The fanoutmust be above 25,000.
For fabric-drivenclock nets, the fanout must be 30 or greater.
1.10 Module-Based FanoutOptimization
Net drivers withfanout greater than the specified limit, provided as an argument with thisoption, will be replicated according to the logical hierarchy.
For each hierarchicalinstance driven by the high-fanout net, if the fanout within the hierarchy isgreater than the specified limit, then the net within the hierarchy is drivenby a replica of the driver of the high-fanout net.
2 Using Directives
2.1Explore
Runs multiplepasses of optimization.
进行多次优化
2.2ExploreArea
Runs multiplepasses of optimization with emphasis on reducing combinational logic.
进行多次优化，重点是减少组合逻辑
2.3AddRemap
Runs the defaultlogic optimization flow and includes LUT remapping to reduce logic levels.
将LUT重新优化到逻辑单元内部。
2.4ExploreSequentialArea
Runs multiplepasses of optimization with emphasis on reducing registers and related combinationallogic.
减少寄存器和组合逻辑
2.5RuntimeOptimized
Runs minimalpasses of optimization, trading design performance for faster run time.
更快通过优化来替代设计性能
2.6NoBramPowerOpt
Runs all thedefault opt_design optimizations except block RAMPower Optimization.
对BRAM不做功耗优化
2.7ExploreWithRemap
Same as the Explore directive but includes the Remap optimization.
与Remap类似，但是多了一个重新映射。