Buffer and Arithmetic Levels
----------------------------
Level 0
-------
=== mac ===

    SPECS
    -----
    Word bits             : 16
    Instances             : 131072 (512*256)
    Compute energy        : 1.41 pJ

    STATS
    -----
    Utilized instances      : 5120
    Computes (total)        : 5120
    Cycles                  : 1
    Energy (total)          : 7208.96 pJ
    Area (total)            : 129184.56 um^2

Level 1
-------
=== scratchpad ===

    SPECS
    -----
        Technology                  : SRAM
        Size                        : 1
        Word bits                   : 16
        Block size                  : 1
        Cluster size                : 1
        Instances                   : 131072 (512*256)
        Shared bandwidth            : -
        Read bandwidth              : -
        Write bandwidth             : -
        Multiple buffering          : 1.00
        Effective size              : 1
        Min utilization             : 0.00
        Vector access energy        : 0.00 pJ
        Vector access energy source : ERT
        Area                        : 0.00 um^2

    MAPPING
    -------
    Loop nest:

    STATS
    -----
    Cycles               : 1
    Bandwidth throttling : 1.00
    Weights:
        Partition size                           : 1
        Utilized capacity                        : 1
        Utilized instances (max)                 : 5120
        Utilized clusters (max)                  : 5120
        Scalar reads (per-instance)              : 1
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 0
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 1
        Energy (per-scalar-access)               : 0.00 pJ
        Energy (per-instance)                    : 0.00 pJ
        Energy (total)                           : 0.00 pJ
        Temporal Reduction Energy (per-instance) : 0.00 pJ
        Temporal Reduction Energy (total)        : 0.00 pJ
        Address Generation Energy (per-cluster)  : 0.00 pJ
        Address Generation Energy (total)        : 0.00 pJ
        Shared Bandwidth (per-instance)          : 1.00 words/cycle
        Shared Bandwidth (total)                 : 5120.00 words/cycle
        Read Bandwidth (per-instance)            : 1.00 words/cycle
        Read Bandwidth (total)                   : 5120.00 words/cycle
        Write Bandwidth (per-instance)           : 0.00 words/cycle
        Write Bandwidth (total)                  : 0.00 words/cycle

Level 2
-------
=== dummy_buffer ===

    SPECS
    -----
        Technology                  : SRAM
        Size                        : 0
        Word bits                   : 16
        Block size                  : 1
        Cluster size                : 1
        Instances                   : 8 (4*2)
        Shared bandwidth            : -
        Read bandwidth              : -
        Write bandwidth             : -
        Multiple buffering          : 1.00
        Effective size              : 0
        Min utilization             : 0.00
        Vector access energy        : 0.00 pJ
        Vector access energy source : ERT
        Area                        : 0.00 um^2

    MAPPING
    -------
    Loop nest:

    STATS
    -----
    Cycles               : 1
    Bandwidth throttling : 1.00

Level 3
-------
=== input_buffer ===

    SPECS
    -----
        Technology                  : SRAM
        Size                        : 12288
        Word bits                   : 16
        Block size                  : 1
        Cluster size                : 1
        Instances                   : 8 (4*2)
        Shared bandwidth            : -
        Read bandwidth              : -
        Write bandwidth             : -
        Multiple buffering          : 1.00
        Effective size              : 12288
        Min utilization             : 0.00
        Vector access energy        : 4.97 pJ
        Vector access energy source : ERT
        Area                        : 41746.70 um^2

    MAPPING
    -------
    Loop nest:

    STATS
    -----
    Cycles               : 1
    Bandwidth throttling : 1.00
    Inputs:
        Partition size                           : 128
        Utilized capacity                        : 128
        Utilized instances (max)                 : 8
        Utilized clusters (max)                  : 8
        Scalar reads (per-instance)              : 128
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 128
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 256
        Energy (per-scalar-access)               : 3.97 pJ
        Energy (per-instance)                    : 1016.49 pJ
        Energy (total)                           : 8131.90 pJ
        Temporal Reduction Energy (per-instance) : 0.00 pJ
        Temporal Reduction Energy (total)        : 0.00 pJ
        Address Generation Energy (per-cluster)  : 0.00 pJ
        Address Generation Energy (total)        : 0.00 pJ
        Shared Bandwidth (per-instance)          : 256.00 words/cycle
        Shared Bandwidth (total)                 : 2048.00 words/cycle
        Read Bandwidth (per-instance)            : 128.00 words/cycle
        Read Bandwidth (total)                   : 1024.00 words/cycle
        Write Bandwidth (per-instance)           : 128.00 words/cycle
        Write Bandwidth (total)                  : 1024.00 words/cycle

Level 4
-------
=== shared_glb ===

    SPECS
    -----
        Technology                  : SRAM
        Size                        : 65536
        Word bits                   : 16
        Block size                  : 4
        Cluster size                : 1
        Instances                   : 1 (1*1)
        Shared bandwidth            : -
        Read bandwidth              : 16.00
        Write bandwidth             : 16.00
        Multiple buffering          : 1.00
        Effective size              : 65536
        Min utilization             : 0.00
        Vector access energy        : 41.71 pJ
        Vector access energy source : ERT
        Area                        : 590548.00 um^2

    MAPPING
    -------
    Loop nest:

    STATS
    -----
    Cycles               : 34
    Bandwidth throttling : 0.03
    Inputs:
        Partition size                           : 512
        Utilized capacity                        : 512
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 512
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 512
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 1024
        Energy (per-scalar-access)               : 10.36 pJ
        Energy (per-instance)                    : 10612.58 pJ
        Energy (total)                           : 10612.58 pJ
        Temporal Reduction Energy (per-instance) : 0.00 pJ
        Temporal Reduction Energy (total)        : 0.00 pJ
        Address Generation Energy (per-cluster)  : 0.00 pJ
        Address Generation Energy (total)        : 0.00 pJ
        Shared Bandwidth (per-instance)          : 30.80 words/cycle
        Shared Bandwidth (total)                 : 30.80 words/cycle
        Read Bandwidth (per-instance)            : 15.40 words/cycle
        Read Bandwidth (total)                   : 15.40 words/cycle
        Write Bandwidth (per-instance)           : 15.40 words/cycle
        Write Bandwidth (total)                  : 15.40 words/cycle
    Outputs:
        Partition size                           : 10
        Utilized capacity                        : 10
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 0
        Scalar updates (per-instance)            : 10
        Scalar fills (per-instance)              : 10
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 20
        Energy (per-scalar-access)               : 12.51 pJ
        Energy (per-instance)                    : 250.25 pJ
        Energy (total)                           : 250.25 pJ
        Temporal Reduction Energy (per-instance) : 0.00 pJ
        Temporal Reduction Energy (total)        : 0.00 pJ
        Address Generation Energy (per-cluster)  : 0.00 pJ
        Address Generation Energy (total)        : 0.00 pJ
        Shared Bandwidth (per-instance)          : 0.60 words/cycle
        Shared Bandwidth (total)                 : 0.60 words/cycle
        Read Bandwidth (per-instance)            : 0.00 words/cycle
        Read Bandwidth (total)                   : 0.00 words/cycle
        Write Bandwidth (per-instance)           : 0.60 words/cycle
        Write Bandwidth (total)                  : 0.60 words/cycle

Level 5
-------
=== DRAM ===

    SPECS
    -----
        Technology                  : DRAM
        Size                        : -
        Word bits                   : 16
        Block size                  : 4
        Cluster size                : 1
        Instances                   : 1 (1*1)
        Shared bandwidth            : -
        Read bandwidth              : -
        Write bandwidth             : -
        Multiple buffering          : 1.00
        Effective size              : -
        Min utilization             : 0.00
        Vector access energy        : 512.00 pJ
        Vector access energy source : ERT
        Area                        : 0.00 um^2

    MAPPING
    -------
    Loop nest:

    STATS
    -----
    Cycles               : 1
    Bandwidth throttling : 1.00
    Inputs:
        Partition size                           : 512
        Utilized capacity                        : 512
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 512
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 0
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 512
        Energy (per-scalar-access)               : 128.00 pJ
        Energy (per-instance)                    : 65536.00 pJ
        Energy (total)                           : 65536.00 pJ
        Temporal Reduction Energy (per-instance) : 0.00 pJ
        Temporal Reduction Energy (total)        : 0.00 pJ
        Address Generation Energy (per-cluster)  : 0.00 pJ
        Address Generation Energy (total)        : 0.00 pJ
        Shared Bandwidth (per-instance)          : 512.00 words/cycle
        Shared Bandwidth (total)                 : 512.00 words/cycle
        Read Bandwidth (per-instance)            : 512.00 words/cycle
        Read Bandwidth (total)                   : 512.00 words/cycle
        Write Bandwidth (per-instance)           : 0.00 words/cycle
        Write Bandwidth (total)                  : 0.00 words/cycle
    Outputs:
        Partition size                           : 10
        Utilized capacity                        : 10
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 0
        Scalar updates (per-instance)            : 10
        Scalar fills (per-instance)              : 0
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 10
        Energy (per-scalar-access)               : 153.60 pJ
        Energy (per-instance)                    : 1536.00 pJ
        Energy (total)                           : 1536.00 pJ
        Temporal Reduction Energy (per-instance) : 0.00 pJ
        Temporal Reduction Energy (total)        : 0.00 pJ
        Address Generation Energy (per-cluster)  : 0.00 pJ
        Address Generation Energy (total)        : 0.00 pJ
        Shared Bandwidth (per-instance)          : 10.00 words/cycle
        Shared Bandwidth (total)                 : 10.00 words/cycle
        Read Bandwidth (per-instance)            : 0.00 words/cycle
        Read Bandwidth (total)                   : 0.00 words/cycle
        Write Bandwidth (per-instance)           : 10.00 words/cycle
        Write Bandwidth (total)                  : 10.00 words/cycle

Networks
--------
Network 0
---------
A2D_NoC

    SPECS
    -----
        Type            : SimpleMulticast
        ConnectionType  : 2
        Word bits       : 16
        Action Name       : transfer

    STATS
    -----
    Weights:
        Fanout                                  : 0
        Multicast factor                        : 0
        Ingresses                               : 0.00
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
    Inputs:
        Fanout                                  : 1
        Multicast factor                        : 1
        Ingresses                               : 128.00
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
    Outputs:
        Fanout                                  : 1
        Multicast factor                        : 1
        Ingresses                               : 5.00
        Energy (per-instance)                   : 1605.84 pJ
        Energy (total)                          : 12846.69 pJ

Network 1
---------
D2A_NoC

    SPECS
    -----
        Type            : SimpleMulticast
        ConnectionType  : 1
        Word bits       : 16
        Action Name       : transfer

    STATS
    -----
    Weights:
        Fanout                                  : 0
        Multicast factor                        : 0
        Ingresses                               : 0.00
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
    Inputs:
        Fanout                                  : 1
        Multicast factor                        : 1
        Ingresses                               : 128.00
        Energy (per-instance)                   : 195.95 pJ
        Energy (total)                          : 1567.62 pJ
    Outputs:
        Fanout                                  : 1
        Multicast factor                        : 1
        Ingresses                               : 5.00
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ

Network 2
---------
DRAM <==> shared_glb

    SPECS
    -----
        Type            : Legacy
        Legacy sub-type : 
        ConnectionType  : 3
        Word bits       : 16
        Router energy   : - pJ
        Wire energy     : - pJ/b/mm
        Fill latency     : 0
        Drain latency     : 0

    STATS
    -----
    Weights:
        Fanout                                  : 0
        Fanout (distributed)                    : 0
        Multicast factor                        : 0
        Ingresses                               : 0.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.00
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Inputs:
        Fanout                                  : 1
        Fanout (distributed)                    : 0
        Multicast factor                        : 1
        Ingresses                               : 512.00
            @multicast 1 @scatter 1: 512.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.50
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Outputs:
        Fanout                                  : 1
        Fanout (distributed)                    : 0
        Multicast factor                        : 1
        Ingresses                               : 10.00
            @multicast 1 @scatter 1: 10.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.50
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ

Network 3
---------
dummy_buffer <==> scratchpad

    SPECS
    -----
        Type            : Legacy
        Legacy sub-type : 
        ConnectionType  : 3
        Word bits       : 16
        Router energy   : - pJ
        Wire energy     : - pJ/b/mm
        Fill latency     : 0
        Drain latency     : 0

    STATS
    -----
    Weights:
        Fanout                                  : 0
        Fanout (distributed)                    : 0
        Multicast factor                        : 0
        Ingresses                               : 0.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.00
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Inputs:
        Fanout                                  : 640
        Fanout (distributed)                    : 0
        Multicast factor                        : 5
        Ingresses                               : 128.00
            @multicast 5 @scatter 128: 128.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 57.29
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Outputs:
        Fanout                                  : 640
        Fanout (distributed)                    : 0
        Multicast factor                        : 128
        Ingresses                               : 5.00
            @multicast 128 @scatter 5: 5.00
        Link transfers                          : 0
        Spatial reductions                      : 635
        Average number of hops                  : 291.47
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ

Network 4
---------
scratchpad <==> mac

    SPECS
    -----
        Type            : Legacy
        Legacy sub-type : 
        ConnectionType  : 3
        Word bits       : 16
        Router energy   : - pJ
        Wire energy     : - pJ/b/mm
        Fill latency     : 0
        Drain latency     : 0

    STATS
    -----
    Weights:
        Fanout                                  : 1
        Fanout (distributed)                    : 0
        Multicast factor                        : 1
        Ingresses                               : 1.00
            @multicast 1 @scatter 1: 1.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.50
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Inputs:
        Fanout                                  : 1
        Fanout (distributed)                    : 0
        Multicast factor                        : 1
        Ingresses                               : 1.00
            @multicast 1 @scatter 1: 1.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.50
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Outputs:
        Fanout                                  : 1
        Fanout (distributed)                    : 0
        Multicast factor                        : 1
        Ingresses                               : 1.00
            @multicast 1 @scatter 1: 1.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.50
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ

Network 5
---------
shared_glb <==> input_buffer

    SPECS
    -----
        Type            : Legacy
        Legacy sub-type : 
        ConnectionType  : 3
        Word bits       : 16
        Router energy   : - pJ
        Wire energy     : - pJ/b/mm
        Fill latency     : 0
        Drain latency     : 0

    STATS
    -----
    Weights:
        Fanout                                  : 0
        Fanout (distributed)                    : 0
        Multicast factor                        : 0
        Ingresses                               : 0.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 0.00
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Inputs:
        Fanout                                  : 8
        Fanout (distributed)                    : 0
        Multicast factor                        : 2
        Ingresses                               : 512.00
            @multicast 2 @scatter 4: 512.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 3.21
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ
    Outputs:
        Fanout                                  : 8
        Fanout (distributed)                    : 0
        Multicast factor                        : 4
        Ingresses                               : 10.00
            @multicast 4 @scatter 2: 10.00
        Link transfers                          : 0
        Spatial reductions                      : 30
        Average number of hops                  : 5.16
        Energy (per-hop)                        : 0.00 fJ
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
        Link transfer energy (per-instance)     : 0.00 pJ
        Link transfer energy (total)            : 0.00 pJ
        Spatial Reduction Energy (per-instance) : 0.00 pJ
        Spatial Reduction Energy (total)        : 0.00 pJ


Operational Intensity Stats
---------------------------
    Total elementwise ops                   : 5120
    Total reduction ops                     : 5110
    Total ops                               : 10230
    Total memory accesses required          : 523
    Optimal Op per Byte                     : 9.78

=== scratchpad ===
    Total scalar accesses                   : 5120
    Op per Byte                             : 1.00
=== dummy_buffer ===
=== input_buffer ===
    Total scalar accesses                   : 2048
    Op per Byte                             : 2.50
=== shared_glb ===
    Total scalar accesses                   : 1044
    Op per Byte                             : 4.90
=== DRAM ===
    Total scalar accesses                   : 522
    Op per Byte                             : 9.80


Summary Stats
-------------
GFLOPs (@1GHz): 300.88
Utilization: 0.00
Cycles: 34
Energy: 0.11 uJ
EDP(J*cycle): 3.66e-06
Area: 0.00 mm^2

Computes = 5120
pJ/Compute
    mac                          = 1.41
    scratchpad                   = 0.00
    dummy_buffer                 = 0.00
    input_buffer                 = 1.59
    shared_glb                   = 2.12
    DRAM                         = 13.10
    A2D_NoC                      = 2.51
    D2A_NoC                      = 0.31
    DRAM <==> shared_glb         = 0.00
    dummy_buffer <==> scratchpad = 0.00
    scratchpad <==> mac          = 0.00
    shared_glb <==> input_buffer = 0.00
    Total                        = 21.03

