==PROF== Connected to process 3054214 (/accel-sim/accel-sim-framework-dev/gpu-app-collection/src/cuda-test/Tango-master/GPU/AlexNet/single_kernel_execute3DconvolutionCuda)
==PROF== Profiling "execute3DconvolutionCuda" - 0: 0%....50%....100% - 9 passes
execute3DconvolutionCuda done.
conv_out[0] = 1.396400
conv_out[1] = 1.533199
conv_out[2] = 1.536999
conv_out[3] = 1.543399
conv_out[4] = 1.538799
conv_out[5] = 1.536399
conv_out[6] = 1.544999
conv_out[7] = 1.542599
conv_out[8] = 1.542399
conv_out[9] = 1.542199
==PROF== Disconnected from process 3054214
[3054214] single_kernel_execute3DconvolutionCuda@127.0.0.1
  execute3DconvolutionCuda(float *, float *, float *, float *, int, int, int, int, int, int, int, int) (128, 1, 1)x(27, 27, 1), Context 1, Stream 7, Device 0, CC 12.0
    Section: GPU Speed Of Light Throughput
    ----------------------- ----------- ------------
    Metric Name             Metric Unit Metric Value
    ----------------------- ----------- ------------
    DRAM Frequency                  Ghz        14.78
    SM Frequency                    Ghz         2.27
    Elapsed Cycles                cycle       420149
    Memory Throughput                 %        40.73
    DRAM Throughput                   %         0.44
    Duration                         us       184.67
    L1/TEX Cache Throughput           %        51.27
    L2 Cache Throughput               %         3.43
    SM Active Cycles              cycle    333231.37
    Compute (SM) Throughput           %        40.73
    ----------------------- ----------- ------------

    OPT   This kernel grid is too small to fill the available resources on this device, resulting in only 0.8 full      
          waves across all SMs. Look at Launch Statistics for more details.                                             

    Section: Launch Statistics
    -------------------------------- --------------- ---------------
    Metric Name                          Metric Unit    Metric Value
    -------------------------------- --------------- ---------------
    Block Size                                                   729
    Cluster Scheduling Policy                           PolicySpread
    Cluster Size                                                   0
    Function Cache Configuration                     CachePreferNone
    Grid Size                                                    128
    Registers Per Thread             register/thread              39
    Shared Memory Configuration Size           Kbyte            8.19
    Driver Shared Memory Per Block       Kbyte/block            1.02
    Dynamic Shared Memory Per Block       byte/block               0
    Static Shared Memory Per Block        byte/block               0
    # SMs                                         SM              84
    Stack Size                                                  1024
    Threads                                   thread           93312
    # TPCs                                                        42
    Enabled TPC IDs                                              all
    Uses Green Context                                             0
    Waves Per SM                                                0.76
    -------------------------------- --------------- ---------------

    OPT   Est. Speedup: 0.9511%                                                                                         
          Threads are executed in groups of 32 threads called warps. This kernel launch is configured to execute 729    
          threads per block. Consequently, some threads in a warp are masked off and those hardware resources are       
          unused. Try changing the number of threads per block to be a multiple of 32 threads. Between 128 and 256      
          threads per block is a good initial range for experimentation. Use smaller thread blocks rather than one      
          large thread block per multiprocessor if latency affects performance.  This is particularly beneficial to     
          kernels that frequently call __syncthreads(). See the Hardware Model                                          
          (https://docs.nvidia.com/nsight-compute/ProfilingGuide/index.html#metrics-hw-model) description for more      
          details on launch configurations.                                                                             
    ----- --------------------------------------------------------------------------------------------------------------
    OPT   If you execute __syncthreads() to synchronize the threads of a block, it is recommended to have at least two  
          blocks per multiprocessor (compared to the currently executed 1.5 blocks) This way, blocks that aren't        
          waiting for __syncthreads() can keep the hardware busy.                                                       

    Section: Occupancy
    ------------------------------- ----------- ------------
    Metric Name                     Metric Unit Metric Value
    ------------------------------- ----------- ------------
    Max Active Clusters                 cluster            0
    Max Cluster Size                      block            8
    Overall GPU Occupancy                     %            0
    Cluster Occupancy                         %            0
    Block Limit Barriers                  block           24
    Block Limit SM                        block           24
    Block Limit Registers                 block            2
    Block Limit Shared Mem                block            8
    Block Limit Warps                     block            2
    Theoretical Active Warps per SM        warp           46
    Theoretical Occupancy                     %        95.83
    Achieved Occupancy                        %        66.50
    Achieved Active Warps Per SM           warp        31.92
    ------------------------------- ----------- ------------

    OPT   Est. Local Speedup: 30.61%                                                                                    
          The difference between calculated theoretical (95.8%) and measured achieved occupancy (66.5%) can be the      
          result of warp scheduling overheads or workload imbalances during the kernel execution. Load imbalances can   
          occur between warps within a block as well as across blocks of the same kernel. See the CUDA Best Practices   
          Guide (https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/index.html#occupancy) for more details on     
          optimizing occupancy.                                                                                         

    Section: GPU and Memory Workload Distribution
    -------------------------- ----------- ------------
    Metric Name                Metric Unit Metric Value
    -------------------------- ----------- ------------
    Average DRAM Active Cycles       cycle        11900
    Total DRAM Elapsed Cycles        cycle     21841920
    Average L1 Active Cycles         cycle    333231.37
    Total L1 Elapsed Cycles          cycle     35236512
    Average L2 Active Cycles         cycle    207448.75
    Total L2 Elapsed Cycles          cycle     12185184
    Average SM Active Cycles         cycle    333231.37
    Total SM Elapsed Cycles          cycle     35236512
    Average SMSP Active Cycles       cycle    321221.11
    Total SMSP Elapsed Cycles        cycle    140946048
    -------------------------- ----------- ------------

    OPT   Est. Speedup: 15.87%                                                                                          
          One or more SMs have a much lower number of active cycles than the average number of active cycles. Maximum   
          instance value is 19.98% above the average, while the minimum instance value is 27.02% below the average.     
    ----- --------------------------------------------------------------------------------------------------------------
    OPT   Est. Speedup: 17.34%                                                                                          
          One or more SMSPs have a much higher number of active cycles than the average number of active cycles.        
          Additionally, other SMSPs have a much lower number of active cycles than the average number of active         
          cycles. Maximum instance value is 22.65% above the average, while the minimum instance value is 24.57% below  
          the average.                                                                                                  
    ----- --------------------------------------------------------------------------------------------------------------
    OPT   Est. Speedup: 15.87%                                                                                          
          One or more L1 Slices have a much lower number of active cycles than the average number of active cycles.     
          Maximum instance value is 19.98% above the average, while the minimum instance value is 27.02% below the      
          average.                                                                                                      

