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

    SPECS
    -----
    Word bits             : 16
    Instances             : 589824 (2304*256)
    Compute energy        : 1.41 pJ

    STATS
    -----
    Utilized instances      : 589824
    Computes (total)        : 37748736
    Cycles                  : 64
    Energy (total)          : 53150220.29 pJ
    Area (total)            : 581330.53 um^2

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

    SPECS
    -----
        Technology                  : SRAM
        Size                        : 1
        Word bits                   : 16
        Block size                  : 1
        Cluster size                : 1
        Instances                   : 589824 (2304*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               : 64
    Bandwidth throttling : 1.00
    Weights:
        Partition size                           : 1
        Utilized capacity                        : 1
        Utilized instances (max)                 : 589824
        Utilized clusters (max)                  : 589824
        Scalar reads (per-instance)              : 64
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 0
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 64
        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)                 : 589824.00 words/cycle
        Read Bandwidth (per-instance)            : 1.00 words/cycle
        Read Bandwidth (total)                   : 589824.00 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                   : 36 (18*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               : 64
    Bandwidth throttling : 1.00

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

    SPECS
    -----
        Technology                  : SRAM
        Size                        : 12288
        Word bits                   : 16
        Block size                  : 1
        Cluster size                : 1
        Instances                   : 36 (18*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               : 64
    Bandwidth throttling : 1.00
    Inputs:
        Partition size                           : 3413
        Utilized capacity                        : 1024
        Utilized instances (max)                 : 36
        Utilized clusters (max)                  : 36
        Scalar reads (per-instance)              : 8192
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 8192
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 16384
        Energy (per-scalar-access)               : 3.97 pJ
        Energy (per-instance)                    : 65055.21 pJ
        Energy (total)                           : 2341987.61 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)                 : 9216.00 words/cycle
        Read Bandwidth (per-instance)            : 128.00 words/cycle
        Read Bandwidth (total)                   : 4608.00 words/cycle
        Write Bandwidth (per-instance)           : 128.00 words/cycle
        Write Bandwidth (total)                  : 4608.00 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               : 9216
    Bandwidth throttling : 0.01
    Inputs:
        Partition size                           : 25600
        Utilized capacity                        : 7680
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 147456
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 25600
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 173056
        Energy (per-scalar-access)               : 10.32 pJ
        Energy (per-instance)                    : 1785809.89 pJ
        Energy (total)                           : 1785809.89 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)          : 18.78 words/cycle
        Shared Bandwidth (total)                 : 18.78 words/cycle
        Read Bandwidth (per-instance)            : 16.00 words/cycle
        Read Bandwidth (total)                   : 16.00 words/cycle
        Write Bandwidth (per-instance)           : 2.78 words/cycle
        Write Bandwidth (total)                  : 2.78 words/cycle
    Outputs:
        Partition size                           : 16384
        Utilized capacity                        : 2048
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 0
        Scalar updates (per-instance)            : 16384
        Scalar fills (per-instance)              : 16384
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 32768
        Energy (per-scalar-access)               : 10.43 pJ
        Energy (per-instance)                    : 341677.67 pJ
        Energy (total)                           : 341677.67 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)          : 3.56 words/cycle
        Shared Bandwidth (total)                 : 3.56 words/cycle
        Read Bandwidth (per-instance)            : 0.00 words/cycle
        Read Bandwidth (total)                   : 0.00 words/cycle
        Write Bandwidth (per-instance)           : 3.56 words/cycle
        Write Bandwidth (total)                  : 3.56 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               : 64
    Bandwidth throttling : 1.00
    Inputs:
        Partition size                           : 25600
        Utilized capacity                        : 25600
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 25600
        Scalar updates (per-instance)            : 0
        Scalar fills (per-instance)              : 0
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 25600
        Energy (per-scalar-access)               : 128.00 pJ
        Energy (per-instance)                    : 3276800.00 pJ
        Energy (total)                           : 3276800.00 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)          : 400.00 words/cycle
        Shared Bandwidth (total)                 : 400.00 words/cycle
        Read Bandwidth (per-instance)            : 400.00 words/cycle
        Read Bandwidth (total)                   : 400.00 words/cycle
        Write Bandwidth (per-instance)           : 0.00 words/cycle
        Write Bandwidth (total)                  : 0.00 words/cycle
    Outputs:
        Partition size                           : 16384
        Utilized capacity                        : 16384
        Utilized instances (max)                 : 1
        Utilized clusters (max)                  : 1
        Scalar reads (per-instance)              : 0
        Scalar updates (per-instance)            : 16384
        Scalar fills (per-instance)              : 0
        Temporal reductions (per-instance)       : 0
        Address generations (per-cluster)        : 16384
        Energy (per-scalar-access)               : 128.00 pJ
        Energy (per-instance)                    : 2097152.00 pJ
        Energy (total)                           : 2097152.00 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)          : 256.00 words/cycle
        Shared Bandwidth (total)                 : 256.00 words/cycle
        Read Bandwidth (per-instance)            : 0.00 words/cycle
        Read Bandwidth (total)                   : 0.00 words/cycle
        Write Bandwidth (per-instance)           : 256.00 words/cycle
        Write Bandwidth (total)                  : 256.00 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                               : 8192.00
        Energy (per-instance)                   : 0.00 pJ
        Energy (total)                          : 0.00 pJ
    Outputs:
        Fanout                                  : 1
        Multicast factor                        : 1
        Ingresses                               : 8192.00
        Energy (per-instance)                   : 2631003.01 pJ
        Energy (total)                          : 94716108.47 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                               : 8192.00
        Energy (per-instance)                   : 12540.97 pJ
        Energy (total)                          : 451474.88 pJ
    Outputs:
        Fanout                                  : 1
        Multicast factor                        : 1
        Ingresses                               : 8192.00
        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                               : 25600.00
            @multicast 1 @scatter 1: 25600.00
        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                               : 16384.00
            @multicast 1 @scatter 1: 16384.00
        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                                  : 16384
        Fanout (distributed)                    : 0
        Multicast factor                        : 128
        Ingresses                               : 8192.00
            @multicast 128 @scatter 128: 8192.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 1495.68
        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                                  : 16384
        Fanout (distributed)                    : 0
        Multicast factor                        : 128
        Ingresses                               : 8192.00
            @multicast 128 @scatter 128: 8192.00
        Link transfers                          : 0
        Spatial reductions                      : 1040384
        Average number of hops                  : 1495.68
        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                               : 64.00
            @multicast 1 @scatter 1: 64.00
        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                               : 64.00
            @multicast 1 @scatter 1: 64.00
        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                               : 64.00
            @multicast 1 @scatter 1: 64.00
        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                                  : 36
        Fanout (distributed)                    : 0
        Multicast factor                        : 2
        Ingresses                               : 147456.00
            @multicast 2 @scatter 18: 147456.00
        Link transfers                          : 0
        Spatial reductions                      : 0
        Average number of hops                  : 7.04
        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                                  : 36
        Fanout (distributed)                    : 0
        Multicast factor                        : 18
        Ingresses                               : 16384.00
            @multicast 18 @scatter 2: 16384.00
        Link transfers                          : 0
        Spatial reductions                      : 278528
        Average number of hops                  : 25.42
        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                   : 37748736
    Total reduction ops                     : 37732352
    Total ops                               : 75481088
    Total memory accesses required          : 41985
    Optimal Op per Byte                     : 898.91

=== scratchpad ===
    Total scalar accesses                   : 37748736
    Op per Byte                             : 1.00
=== dummy_buffer ===
=== input_buffer ===
    Total scalar accesses                   : 589824
    Op per Byte                             : 63.99
=== shared_glb ===
    Total scalar accesses                   : 205824
    Op per Byte                             : 183.36
=== DRAM ===
    Total scalar accesses                   : 41984
    Op per Byte                             : 898.93


Summary Stats
-------------
GFLOPs (@1GHz): 8190.22
Utilization: 0.01
Cycles: 9216
Energy: 158.16 uJ
EDP(J*cycle): 1.46e+00
Area: 0.00 mm^2

Computes = 37748736
pJ/Compute
    mac                          = 1.41
    scratchpad                   = 0.00
    dummy_buffer                 = 0.00
    input_buffer                 = 0.06
    shared_glb                   = 0.06
    DRAM                         = 0.14
    A2D_NoC                      = 2.51
    D2A_NoC                      = 0.01
    DRAM <==> shared_glb         = 0.00
    dummy_buffer <==> scratchpad = 0.00
    scratchpad <==> mac          = 0.00
    shared_glb <==> input_buffer = 0.00
    Total                        = 4.19

