Identifier: "ComputeWorkloadAnalysis"
DisplayName: "Compute Workload Analysis"
Description: "Detailed analysis of the compute resources of the streaming multiprocessors (SM), including the achieved instructions per clock (IPC) and the utilization of each available pipeline. Pipelines with very high utilization might limit the overall performance."
Order: 20
Sets {
  Identifier: "detailed"
}
Sets {
  Identifier: "full"
}
Header {
  Metrics {
    Label: "Executed Ipc Elapsed"
    Name: "sm__inst_executed.avg.per_cycle_elapsed"
  }
  Metrics {
    Label: "SM Busy"
    Name: "sm__instruction_throughput.avg.pct_of_peak_sustained_active"
  }
  Metrics {
    Label: "Executed Ipc Active"
    Name: "sm__inst_executed.avg.per_cycle_active"
  }
  Metrics {
    Label: "Issue Slots Busy"
    Name: "sm__inst_issued.avg.pct_of_peak_sustained_active"
  }
  Metrics {
    Label: "Issued Ipc Active"
    Name: "sm__inst_issued.avg.per_cycle_active"
  }
  Metrics {
    Name: ""
  }
}
Metrics {
  Metrics {
    Label: "Max. Issued Slots Busy"
    Name: "sm__inst_issued.max.pct_of_peak_sustained_active"
  }
}

Body {
  DisplayName: "Pipe Utilization"
  Items {
    HorizontalContainer {
      Items {
        BarChart {
          Label: "Pipe Utilization (% of active cycles)"
          Description: "Pipeline utilization based on the number of cycles the pipeline was active. This takes the rates of different instructions executing on the pipeline into account. For an instruction requiring 4 cycles to complete execution, the counter is increased by 1 for 4 cycles."
          SortKey: ByValue
          SortDirection: Descending
          ValueAxis {
            Label: "Utilization [%]"
            Range {
              Max: 100
            }
          }
          ValueAxisAlignment: ValueAxisAlignments_Both
          Metrics {
            Label: "ALU"
            Name: "sm__pipe_alu_cycles_active.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "FMA"
            Name: "sm__pipe_fma_cycles_active.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "FP64"
            Name: "sm__pipe_fp64_cycles_active.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "Tensor (All)"
            Name: "sm__pipe_tensor_cycles_active.avg.pct_of_peak_sustained_active"
            Options {
              Name: "sm__pipe_tensor_cycles_active_v2.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_89
                MaxArch: CC_89
              }
            }
          }
          Metrics {
            Label: "Tensor (DP)"
            Name: "sm__pipe_tensor_op_dmma_cycles_active.avg.pct_of_peak_sustained_active"
            Filter {
              Items {
                MinArch: CC_80
                MaxArch: CC_80
              }
              Items {
                MinArch: CC_90
                MaxArch: CC_90
              }
            }
          }
          Metrics {
            Label: "Tensor (FP)"
            Name: "sm__pipe_tensor_op_hmma_cycles_active.avg.pct_of_peak_sustained_active"
            Options {
              Name: "sm__pipe_tensor_op_hmma_cycles_active_v2.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_89
                MaxArch: CC_89
              }
            }
          }
          Metrics {
            Label: "Tensor (INT)"
            Name: "sm__pipe_tensor_op_imma_cycles_active.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_72
            }
            Options {
              Name: "sm__pipe_tensor_op_imma_cycles_active_v2.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_89
                MaxArch: CC_89
              }
            }
          }
          Metrics {
            Label: "Shared (FP64+FP16+Tensor)"
            Name: "sm__pipe_shared_cycles_active.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_70
              MaxArch: CC_70
            }
            Options {
              Label: "Shared (FP16+Tensor)"
              Name: "sm__pipe_shared_cycles_active.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_72
                MaxArch: CC_75
              }
            }
            Options {
              Label: "Shared (FP64+Tensor)"
              Name: "sm__pipe_shared_cycles_active.avg.pct_of_peak_sustained_active"
              Filter {
                Items {
                  MinArch: CC_80
                  MaxArch: CC_80
                }
                Items {
                  MinArch: CC_90
                  MaxArch: CC_90
                }
              }
            }
          }
          Metrics {
            Label: "TMA"
            Name: "sm__pipe_tma_cycles_active.avg.pct_of_peak_sustained_active"
            Filter {
              Items {
                MinArch: CC_90
                MaxArch: CC_90
              }
            }
          }
        }
      }
      Items {
        BarChart {
          Label: "Pipe Utilization (% of peak instructions executed)"
          Description: "Pipeline utilization based on the number of executed instructions. This does not account for any variation in instruction latencies for this pipeline. For an instruction requiring 4 cycles to complete execution, the counter is increased by 1 only."
          SortKey: ByValue
          SortDirection: Descending
          ValueAxis {
            Label: "Utilization [%]"
            Range {
              Max: 100
            }
          }
          ValueAxisAlignment: ValueAxisAlignments_Both
          Metrics {
            Label: "ADU"
            Name: "sm__inst_executed_pipe_adu.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "ALU"
            Name: "sm__inst_executed_pipe_alu.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "CBU"
            Name: "sm__inst_executed_pipe_cbu.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "FMA"
            Name: "sm__inst_executed_pipe_fma.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "FP16"
            Name: "sm__inst_executed_pipe_fp16.avg.pct_of_peak_sustained_active"
            Filter {
              MaxArch: CC_80
            }
            Options {
              Label: "FMA (FP16)"
              Name: "sm__inst_executed_pipe_fma_type_fp16.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_86
              }
            }
          }
          Metrics {
            Label: "FP64 (DMMA)"
            Name: "sm__inst_executed_pipe_fp64_op_dmma.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_86
              MaxArch: CC_89
            }
          }
          Metrics {
            Label: "FP64 (FP64)"
            Name: "sm__inst_executed_pipe_fp64_op_fp64.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_86
              MaxArch: CC_89
            }
          }

          Metrics {
            Label: "FP64"
            Name: "sm__inst_executed_pipe_fp64.avg.pct_of_peak_sustained_active"
            Filter {
              Items {
                MaxArch: CC_75
              }
              Items {
                MinArch: CC_80
                MaxArch: CC_80
              }
              Items {
                MinArch: CC_90
              }
            }
          }
          Metrics {
            Label: "LSU"
            Name: "sm__inst_executed_pipe_lsu.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "Tensor (DP)"
            Name: "sm__inst_executed_pipe_tensor_op_dmma.avg.pct_of_peak_sustained_active"
            Filter {
              Items {
                MinArch: CC_80
                MaxArch: CC_80
              }
              Items {
                MinArch: CC_90
                MaxArch: CC_90
              }
            }
          }
          Metrics {
            Label: "Tensor (Warp Group)"
            Name: "sm__inst_executed_pipe_tensor_op_gmma.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_90
              MaxArch: CC_90
            }
          }
          Metrics {
            Label: "Tensor (FP)"
            Name: "sm__inst_executed_pipe_tensor_op_hmma.avg.pct_of_peak_sustained_active"
            Options {
              Name: "sm__inst_executed_pipe_tensor_op_hmma_v2.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_89
                MaxArch: CC_89
              }
            }
          }
          Metrics {
            Label: "Tensor (INT)"
            Name: "sm__inst_executed_pipe_tensor_op_imma.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_72
            }
            Options {
              Name: "sm__inst_executed_pipe_tensor_op_imma_v2.avg.pct_of_peak_sustained_active"
              Filter {
                MinArch: CC_89
                MaxArch: CC_89
              }
            }
          }
          Metrics {
            Label: "TEX"
            Name: "sm__inst_executed_pipe_tex.avg.pct_of_peak_sustained_active"
          }
          Metrics {
            Label: "TMA"
            Name: "sm__inst_executed_pipe_tma.avg.pct_of_peak_sustained_active"
            Filter {
              Items {
                MinArch: CC_90
                MaxArch: CC_90
              }
            }
          }
          Metrics {
            Label: "Uniform"
            Name: "sm__inst_executed_pipe_uniform.avg.pct_of_peak_sustained_active"
            Filter {
              MinArch: CC_75
            }
          }
          Metrics {
            Label: "XU"
            Name: "sm__inst_executed_pipe_xu.avg.pct_of_peak_sustained_active"
          }
        }
      }
    }
  }
}