TIOVX User Guide
J784S4 Linux Performance Report
SoC J784S4
Test EVM J784S4 TI EVM
HOST ARM A72 running Linux at default frequency
Target Standalone C7x DSP running Free RTOS at default frequency, L1/L2 Cache Enabled (Cache-only based performance.)



TIOVX Kernel Performance
Index Kernel Variant Frame Size (Pixels) Graph Performance (msec) Node Performance (msec)
1 Accumulate 640x480 (307200) 1.359000 1.286000
2 AccumulateSquare 640x480 (307200) 1.951000 1.868000
3 AccumulateWeighted 640x480 (307200) 0.952000 0.880000
4 Addition U8+U8=U8 640x480 (307200) 0.767000 0.694000
5 Addition U8+U8=S16 640x480 (307200) 0.806000 0.733000
6 Addition U8+S16=S16 640x480 (307200) 1.300000 1.227000
7 Addition S16+S16=S16 640x480 (307200) 1.419000 1.345000
8 Subtraction U8-U8=U8 640x480 (307200) 0.815000 0.745000
9 Subtraction U8-U8=S16 640x480 (307200) 0.803000 0.731000
10 Subtraction S16-U8=S16 640x480 (307200) 1.866000 1.794000
11 Subtraction S16-S16=S16 640x480 (307200) 1.405000 1.332000
12 Not Operation U8 640x480 (307200) 0.622000 0.554000
13 And Operation U8 640x480 (307200) 0.758000 0.691000
14 OR Operation U8 640x480 (307200) 0.729000 0.661000
15 XOR Operation U8 640x480 (307200) 0.740000 0.672000
16 Abs Diff U8 640x480 (307200) 0.786000 0.719000
17 Abs Diff S16 640x480 (307200) 1.612000 1.538000
18 Box 3x3 640x480 (307200) 1.545000 1.470000
19 Dilate 3x3 640x480 (307200) 0.634000 0.561000
20 Erode 3x3 640x480 (307200) 0.629000 0.555000
21 Fast Corners No NMS 640x480 (307200) 13.224000 13.135000
22 Canny 640x480 (307200) 7.625000 7.541000
23 Channel Combine 3 to 1 (RGB) 640x480 (307200) 1.449000 1.328000
24 Channel Combine 4 to 1 (RGBX) 640x480 (307200) 1.325000 1.247000
25 Channel Combine YUYV 640x480 (307200) 0.928000 0.851000
26 Channel Combine 2 to 1 (NV12) 640x480 (307200) 0.770000 0.698000
27 Channel Extract 1 of 3 (RGB) 640x480 (307200) 1.558000 1.483000
28 Channel Extract 1 of 4 (RGBX) 640x480 (307200) 2.130000 2.055000
29 Color Convert RGB=>RGBX 640x480 (307200) 3.541000 3.459000
30 Color Convert RGB=>NV12 640x480 (307200) 2.902000 2.822000
31 Color Convert RGB=>IYUV 640x480 (307200) 2.891000 2.805000
32 Color Convert RGB=>YUV4 640x480 (307200) 3.026000 2.944000
33 Color Convert RGBX=>RGB 640x480 (307200) 3.816000 3.732000
34 Color Convert RGBX=>NV12 640x480 (307200) 3.487000 3.407000
35 Color Convert RGBX=>IYUV 640x480 (307200) 3.430000 3.354000
36 Color Convert RGBX=>YUV4 640x480 (307200) 3.512000 3.430000
37 Color Convert NVXX=>RGB 640x480 (307200) 2.807000 2.729000
38 Color Convert NVXX=>RGBX 640x480 (307200) 3.671000 3.593000
39 Color Convert NVXX=>IYUV 640x480 (307200) 0.767000 0.694000
40 Color Convert NVXX=>YUV4 640x480 (307200) 0.993000 0.918000
41 Color Convert YUVX=>RGB 640x480 (307200) 3.213000 3.131000
42 Color Convert YUVX=>RGBX 640x480 (307200) 2.839000 2.760000
43 Color Convert YUVX=>NV12 640x480 (307200) 1.674000 1.599000
44 Color Convert YUVX=>IYUV 640x480 (307200) 1.665000 1.589000
45 Color Convert IYUV=>RGB 640x480 (307200) 2.910000 2.832000
46 Color Convert IYUV=>RGBX 640x480 (307200) 3.120000 3.041000
47 Color Convert IYUV=>NV12 640x480 (307200) 0.783000 0.709000
48 Color Convert IYUV=>YUV4 640x480 (307200) 0.867000 0.790000
49 Convert Depth U8 to S16 640x480 (307200) 0.543000 0.468000
50 Convert Depth S16 to U8 640x480 (307200) 0.995000 0.924000
51 Convolution 3x3, O: U8 640x480 (307200) 1.164000 1.092000
52 Convolution 3x3, O: S16 640x480 (307200) 1.098000 1.024000
53 Convolution 9x3, O: U8 640x480 (307200) 47.197000 47.120000
54 Convolution 9x3, O: S16 640x480 (307200) 48.137000 48.059000
55 Convolution 3x9, O: U8 640x480 (307200) 62.892000 62.813000
56 Convolution 3x9, O: S16 640x480 (307200) 63.424000 63.349000
57 Convolution 5x5, O: U8 640x480 (307200) 2.209000 2.136000
58 Convolution 5x5, O: S16 640x480 (307200) 2.220000 2.148000
59 Convolution 7x7, O: U8 640x480 (307200) 3.618000 3.545000
60 Convolution 7x7, O: S16 640x480 (307200) 4.000000 3.922000
61 Convolution 9x9, O: U8 640x480 (307200) 121.855000 121.773000
62 Convolution 9x9, O: S16 640x480 (307200) 122.774000 122.695000
63 Equalize Histogram 640x480 (307200) 2.002000 1.929000
64 Gaussian 3x3 640x480 (307200) 0.791000 0.723000
65 Gaussian Pyramid 640x480 (307200) 6.896000 6.812000
66 Half Scale Gaussian Pyramid 640x480 (307200) 0.586000 0.515000
67 Harris Corners 640x480 (307200) 16.685000 16.593000
68 Histogram U8 640x480 (307200) 1.125000 1.051000
69 Integral Image 640x480 (307200) 1.113000 1.036000
70 Laplacian Pyramid U8; Levels = 6 640x480 (307200) 3.165000 3.085000
71 Laplacian Reconstruct U8 640x480 (307200) 4.257000 4.176000
72 LookUpTable U8 640x480 (307200) 1.659000 1.585000
73 LookUpTable U8 640x480 (307200) 0.750000 0.678000
74 Magnitude S16 640x480 (307200) 3.790000 3.712000
75 Mean/Standard Deviation U8 640x480 (307200) 0.516000 0.444000
76 Median 3x3 640x480 (307200) 1.295000 1.224000
77 Min Max Loc U8 640x480 (307200) 0.941000 0.870000
78 Min Max Loc S16 640x480 (307200) 1.836000 1.761000
79 Multiply U8 x U8 = U8 640x480 (307200) 0.843000 0.775000
80 Multiply U8 x U8 = S16 640x480 (307200) 0.872000 0.796000
81 Multiply U8 x S16 = S16 640x480 (307200) 2.163000 2.089000
82 Multiply S16 x S16 = S16 640x480 (307200) 2.728000 2.652000
83 NonLinear Filter mask=5x5; Erosion 640x480 (307200) 2.677000 2.603000
84 NonLinear Filter mask=5x5; Dilation 640x480 (307200) 2.673000 2.596000
85 NonLinear Filter mask=5x5; Median 640x480 (307200) 60.111000 60.033000
86 Optical Flow 5x5 640x480 (307200) 5.009000 4.923000
87 Phase 640x480 (307200) 4.643000 4.566000
88 Remap Bilinear Interpolation *640x480 (307200) 9.022000 8.941000
89 Remap Nearest Neighbor *640x480 (307200) 7.622000 7.537000
90 Scale Image 1 to 2; Nearest Neighbor *640x480 (307200) 1.388000 1.315000
91 Scale Image 1 to 2; Bilinear Interpolation *640x480 (307200) 5.161000 5.088000
92 Sobel 3x3 640x480 (307200) 1.440000 1.356000
93 Threshold Binary 640x480 (307200) 0.548000 0.480000
94 Warp Affine Nearest Neighbor *640x480 (307200) 0.323000 0.250000
95 Warp Affine Bilinear Interpolation *640x480 (307200) 1.075000 1.005000
96 Warp Pespective Nearest Neighbor *640x480 (307200) 0.514000 0.442000
97 Warp Pespective Bilinear Interpolation *640x480 (307200) 1.661000 1.588000

Footnote 1: All optional parameters for relevant kernels are used

Footnote 2: All image sizes are for input parameters unless noted with an asterisk (*) in which case these are the output parameter sizes