dave_user
207
edits
Changes
no edit summary
Copyright @2020 Xilinx Inc. All Rights Reserved.
'''Dsight''' per DNNDK, coarse grain profiling
<pre>
---------------------------------------------------------------
red_apple_1.jpg
[Time] LoadImage 0.0128658 ms
[Time] PreprocessImage 0.00510821 ms
[Time] SetInputImageInHWCFP32 0.00115788 ms
[DPU Time] dpuSetInputTensorInHWCFP32 1.70365 ms
[DPU Time] dpuRunTask 1.96536 ms
[DPU Time] dpuGetOutputTensorInHWCFP32 0.00884 ms
[DPU Time] dpuRunSoftmax 0.00472 ms
[DPU tot time] 3.68257 ms
[DPU throughput] 271.549 FPS
[Time] CpuArgmax<float> 1.2e-07 ms
[Time] CpuSoftmax 9.6e-07 ms
[Time] RunCustomCNN 0.0108261 ms
[Time] TopK 8.17e-06 ms
1) red_apple : 0.999664
2) orange : 0.00033535
3) avocado : 1.85477e-07
4) hand : 1.52249e-08
5) banana : 1.13961e-12
6) green_apple : 3.13808e-17
_______________________________________________________________
</pre>
---------------------------------------------------------------
red_apple_1.jpg
[Time] LoadImage 0.012864 ms
[Time] PreprocessImage 0.0051113 ms
[Time] SetInputImageInHWCFP32 0.00120038 ms
[DNNDK] Performance profile - DPU Kernel "pruned_dbg_custom_cnn_0" DPU Task "pruned_dbg_custom_cnn_0-1"
=====================================================================================================
ID NodeName Workload(MOP) Mem(MB) RunTime(ms) Perf(GOPS) Utilization MB/S
1 conv2d_1_Conv2D 85.163 0.53 0.362 235.3 19.1% 1457.7
2 conv2d_2_Conv2D 218.991 0.48 0.206 1063.1 86.5% 2318.8
3 conv2d_3_Conv2D 99.680 0.15 0.109 914.5 74.4% 1383.7
4 conv2d_4_Conv2D 84.935 0.13 0.090 943.7 76.8% 1470.7
5 dense_1_MatMul 9.437 4.52 1.101 8.6 0.7% 4108.1
6 predictions_MatMul 0.003 0.00 0.012 0.3 0.0% 158.0
Total Nodes In Avg:
All 498.209 6.10 1.880 265.0 21.6% 3242.4
=====================================================================================================
[Time] CpuArgmax<float> 1.3e-07 ms
[Time] CpuSoftmax 1.08e-06 ms
[Time] RunCustomCNN 0.0108636 ms
[Time] TopK 7.83e-06 ms
1) red_apple : 0.999664
2) orange : 0.00033535
3) avocado : 1.85477e-07
4) hand : 1.52249e-08
5) banana : 1.13961e-12
6) green_apple : 3.13808e-17
_______________________________________________________________
<pre>
</pre>