Platform: ARM Platform
  Device: Mali-T628
    Driver version  : 1.2 (Linux ARM)
    Compute units   : 4
    Clock frequency : 600 MHz

    Global memory bandwidth (GBPS)
      float   : 5.22
      float2  : 6.91
      float4  : 7.02
      float8  : 6.92
      float16 : 6.53

    Single-precision compute (GFLOPS)
      float   : 19.00
      float2  : 37.62
      float4  : 35.07
      float8  : 34.81
      float16 : 25.82

    Half-precision compute (GFLOPS)
      half   : 38.18
      half2  : 75.81
      half4  : 73.85
      half8  : 68.67
      half16 : 66.50

    Double-precision compute (GFLOPS)
      double   : 9.14
      double2  : 17.32
      double4  : 17.26
      double8  : 13.18
      double16 : 5.17

    Integer compute (GIOPS)
      int   : 18.96
      int2  : 37.61
      int4  : 35.26
      int8  : 34.79
      int16 : 25.40

    Integer compute Fast 24bit (GIOPS)
      int   : 18.96
      int2  : 37.61
      int4  : 35.26
      int8  : 34.78
      int16 : 25.40

    Transfer bandwidth (GBPS)
      enqueueWriteBuffer              : 0.76
      enqueueReadBuffer               : 0.76
      enqueueWriteBuffer non-blocking : 0.76
      enqueueReadBuffer non-blocking  : 0.76
      enqueueMapBuffer(for read)      : 2.29
        memcpy from mapped ptr        : 2.25
      enqueueUnmap(after write)       : 2.34
        memcpy to mapped ptr          : 2.27

    Kernel launch latency : 168.80 us

  Device: Mali-T628
    Driver version  : 1.2 (Linux ARM)
    Compute units   : 2
    Clock frequency : 600 MHz

    Global memory bandwidth (GBPS)
      float   : 2.89
      float2  : 4.66
      float4  : 6.49
      float8  : 6.46
      float16 : 6.04

    Single-precision compute (GFLOPS)
      float   : 9.50
      float2  : 18.82
      float4  : 17.55
      float8  : 17.40
      float16 : 12.91

    Half-precision compute (GFLOPS)
      half   : 19.09
      half2  : 37.90
      half4  : 36.92
      half8  : 34.33
      half16 : 33.24

    Double-precision compute (GFLOPS)
      double   : 4.57
      double2  : 8.67
      double4  : 8.63
      double8  : 6.60
      double16 : 2.61

    Integer compute (GIOPS)
      int   : 9.48
      int2  : 18.80
      int4  : 17.65
      int8  : 17.38
      int16 : 12.71

    Integer compute Fast 24bit (GIOPS)
      int   : 9.48
      int2  : 18.80
      int4  : 17.66
      int8  : 17.38
      int16 : 12.71

    Transfer bandwidth (GBPS)
      enqueueWriteBuffer              : 0.70
      enqueueReadBuffer               : 0.70
      enqueueWriteBuffer non-blocking : 0.70
      enqueueReadBuffer non-blocking  : 0.70
      enqueueMapBuffer(for read)      : 2.04
        memcpy from mapped ptr        : 2.26
      enqueueUnmap(after write)       : 2.08
        memcpy to mapped ptr          : 2.24

    Kernel launch latency : 130.93 us