gg9":SSKrSSKJrJrJr \"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"S S\R S5\"S S\R S5\"S S\R S5\"S S\R S5\"S S\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5\"SS\R S5/rSrSrSr Sr S r S#S!jr S"r g)$N) MetricRequestRequestedMetricsParser Importancezdram__cycles_active.avgzdram__cycles_active.maxzdram__cycles_active.minzdram__cycles_active.sumzdram__cycles_elapsed.sumzl1tex__cycles_active.avgzl1tex__cycles_active.maxzl1tex__cycles_active.minzl1tex__cycles_active.sumzl1tex__cycles_elapsed.sumzlts__cycles_active.avgzlts__cycles_active.maxzlts__cycles_active.minzlts__cycles_active.sumzlts__cycles_elapsed.sumzsm__cycles_active.avgzsm__cycles_active.maxzsm__cycles_active.minzsm__cycles_active.sumzsm__cycles_elapsed.sumzsmsp__cycles_active.avgzsmsp__cycles_active.maxzsmsp__cycles_active.minzsmsp__cycles_active.sumzsmsp__cycles_elapsed.sumcg)NWorkloadImbalancer5nsight-compute-2025.1.1/sections/WorkloadImbalance.pyget_identifierr 9s r cg)NzWorkload Imbalancerrr r get_namer =s r cg)Nz]Analysis of workload distribution in active cycles of SM, SMP, SMSP, L1 & L2 caches, and DRAMrrr r get_descriptionrAs jr cg)NWorkloadDistributionrrr r get_section_identifierrEs !r cS/$)NComputerrr r get_parent_rules_identifiersrIs ;r c XS3R5S:XdXS3R5S:XagSXS3R5XS3R5- - S-nSXS3R5XS3R5- - S-nXRSS5S 3R5nXS 3R5n XiU- -n X:agXd:aTXt:aO[Xg- 5S :a>UR[R R S US US USSUSS3 US35n OXd:a@Xg:a;UR[R R S USUSSUSS3US35n OFXt:a@Xv:a;UR[R R S US USSUSS3US35n OgURU [R RU 5 URU XS3R5XS3R5[R RUbU5 gSUS35 g)Nz.maxrz.avgdz.minactiveelapsedz.sumz One or more zks have a much higher number of active cycles than the average number of active cycles. Additionally, other zps have a much lower number of active cycles than the average number of active cycles. Maximum instance value is z.2fz9% above the average, while the minimum instance value is z% below the average.zs Workload Imbalancezqs have a much higher number of active cycles than the average number of active cycles. Maximum instance value is z-Balancing the number of active cycles across z+s would result in a more optimized workload) valuereplaceabsmessageNvRules IFrontendMsgType_MSG_OPTIMIZATIONspeedupSpeedupType_GLOBAL focus_metricnameSeverity_SEVERITY_HIGH) femetricsmetric_base_nameid min_speeduprecommendationmax_distance_from_avgmin_distance_from_avg total_elapsed total_activer#msg_ids r analyze_imbalancer3Ls) $D)*002a77EWW[C\;];c;c;eij;j',>d*C"D"J"J"LwYkkoWpOqOwOwOy"yzBB',>d*C"D"J"J"LwYkkoWpOqOwOwOy"yzBB77)LMTRSYY[M/t45;;=L#m'CDG, ! 0 %= > BG--FFrd#--B3,GHABWX[A\\pq$*+ - .#;G--FFrd#--B3,GHABWX[A\\pq$*+- .#;G--FFrd#--B3,GHABWX[A\\pq$*+- JJvw((;;WEOO$D)*//1$D)*00200)5  >kkmjnoZ<[ r cD[R"U5nURS5RS5nUR 5n[ X5R [5n[X4SSSS9 [X4SSSS9 [X4SS SS9 [X4S S SS9 [X4S S SS9 g)Nrsm__cycles_activeSM)r(r)r*r+r,smsp__cycles_activeSMSPl1tex__cycles_activezL1 Slicelts__cycles_activezL2 Slicedram__cycles_activez DRAM Slice) r get_context range_by_idx action_by_idxfrontendrparserequested_metricsr3)handlectxactionr(r)s r applyrFs   f %C   a . .q 1F B$V4::;LMG?RW[ijk?TY_mno?UZdrst?SXbpqr?TYestur )N)r RequestedMetricsrrrOPTIONALrBr r rrrr3rFrr r rIsM2NN+T:3F3FJ+T:3F3FJ+T:3F3FJ+T:3F3FJ,dJ4G4GK,dJ4G4GK,dJ4G4GK,dJ4G4GK,dJ4G4GK-tZ5H5H!L*D*2E2EqI*D*2E2EqI*D*2E2EqI*D*2E2EqI+T:3F3FJ)41D1DaH)41D1DaH)41D1DaH)41D1DaH*D*2E2EqI+T:3F3FJ+T:3F3FJ+T:3F3FJ+T:3F3FJ,dJ4G4GK3: k"<| vr