ggxjSSKrSSKJrJr \"SS5\"SS5/rSrSrS rS rS r S r S r g)N) MetricRequestRequestedMetricsParserz2smsp__thread_inst_executed_per_inst_executed.ratiothread_inst_executedz:smsp__thread_inst_executed_pred_on_per_inst_executed.ratiothread_inst_executed_NPOcg)NThreadDivergencer 4nsight-compute-2025.1.1/sections/ThreadDivergence.pyget_identifierr "s r cg)NzThread Divergencer r r r get_namer%s r cg)Nz%Warp and thread control flow analysisr r r r get_descriptionr(s 2r cg)NWarpStateStatsr r r r get_section_identifierr+s r cS/$)NComputer r r r get_parent_rules_identifiersr.s ;r c[X5nSUS- - nSnXP;a'[RRnX@U-S-nXg4$[RRnUS-nXg4$)N compute_throughput_normalizedd)minNvRules IFrontendSpeedupType_GLOBALSpeedupType_LOCAL)parent_weightsrrnum_threads_usedimprovement_localcompute_throughput_name speedup_typeimprovement_percents r get_estimated_speedupr'2s/J-22=0((;; /AX2YY\__  ,,((:: /#5  ,,r c [R"U5nURS5RS5nUR 5n[ X5R [5nURS5nUSR5nUSR5n/nSn Xi:dXy:GaSRU5n URUSR5U[RRS45 Xv:aRU SRU5- n URUSR5U[RRS 45 UR![RR"U 5n [%XVU5upUR'XU 5 UH#nUR)XSUS US US 5 M% gg) NrrrraInstructions are executed in warps, which are groups of 32 threads. Optimal instruction throughput is achieved if all 32 threads of a warp execute the same instruction. The chosen launch configuration, early thread completion, and divergent flow control can significantly lower the number of active threads in a warp per cycle. This workload achieves an average of {0:.1f} threads being active per cycle.z9Increase the number of threads per instruction towards 32aO This is further reduced to {0:.1f} threads per warp due to predication. The compiler may use predication to avoid an actual branch. Instead, all instructions are scheduled, but a per-thread condition code or predicate controls which threads execute the instructions. Try to avoid different execution paths within a warp when possible.zGIncrease the number of predicated-on threads per instruction towards 32r)r get_context range_by_idx action_by_idxfrontendrparserequested_metricsreceive_dict_from_parentvalueformatappendnamerSeverity_SEVERITY_LOWSeverity_SEVERITY_HIGHmessageMsgType_MSG_OPTIMIZATIONr'speedup focus_metric)handlectxactionfemetricsr!rrfms thresholdr9msg_idr% speedup_valuefms r applyrGAs   f %C   a . .q 1F B$V4::;LMG00;N"#9:@@B&'ABHHJ CI'+C+OippqEF G2388:rIsS2BFH^_NPjk 3 -@r