-trace_opcode_latency_initiation_int 4,2
-trace_opcode_latency_initiation_sp 4,2
-trace_opcode_latency_initiation_dp 8,4
-trace_opcode_latency_initiation_sfu 20,8
-trace_opcode_latency_initiation_tensor 8,4

#execute branch insts on spec unit 1
#in Turing, there is a dedicated branch unit
#<enabled>,<num_units>,<max_latency>,<ID_OC_SPEC>,<OC_EX_SPEC>,<NAME>
-specialized_unit_1 1,4,4,4,4,BRA
#<latency>,<initiation>
-trace_opcode_latency_initiation_spec_op_1 4,4

#TEX unit, make fixed latency for all tex insts
-specialized_unit_2 1,1,200,4,4,TEX
-trace_opcode_latency_initiation_spec_op_2 200,4

#tensor unit
-specialized_unit_3 1,4,8,4,4,TENSOR
-trace_opcode_latency_initiation_spec_op_3 8,4

#UDP unit
#for more info about UDP, see https://www.hotchips.org/hc31/HC31_2.12_NVIDIA_final.pdf
-specialized_unit_4 1,4,4,4,4,UDP
-trace_opcode_latency_initiation_spec_op_4 4,2