linux

~~Images 23.08.04-rpardini-369~~
vim3-ubuntu-22.04-gnome-linux-5.15-fenix-1.6.3-240112.img.xz
vim3l-ubuntu-22.04-gnome-linux-5.15-fenix-1.6.3-240112.img.xz
- 重大利好，最新系统原装支持opencl！！！

utils

~~dd if=/dev/zero of=/dev/mmcblk1 conv=fsync,notrunc bs=512 count=8~~

这个方案现在已经无法恢复emmc中的android系统了。。。。

activate the Cooling Fan in VIM3

use oowow, enter shell:

enable fan

i2cset -y 0 0x18 0x88 3 b # max speed
i2cset -y 0 0x18 0x88 2 b # mid
i2cset -y 0 0x18 0x88 1 b # low

disable fan

i2cset -y 0 0x18 0x88 0 b

Max out the Vim3 performance (Optional)

alias performance="echo performance | sudo tee /sys/bus/cpu/devices/cpu[02]/cpufreq/scaling_governor /sys/class/devfreq/ffe40000.gpu/governor"
alias ondemand="echo ondemand | sudo tee /sys/bus/cpu/devices/cpu[02]/cpufreq/scaling_governor && echo simple_ondemand | sudo tee /sys/class/devfreq/ffe40000.gpu/governor"
alias powersave="echo powersave | sudo tee /sys/bus/cpu/devices/cpu[046]/cpufreq/scaling_governor /sys/class/devfreq/ffe40000.gpu/governor"

CPU

Cortex-a73

single thread @2.4GHz

$ ./mnn_perf                            
Creating MNN Interpreter: efficientformerv2_s0
The device support i8sdot:0, support fp16:0, support i8mm: 0
(index: 985,  score: 11.719070), (index: 644,  score: 4.942233), (index: 309,  score: 3.837214), 
[197 iters] min = 100.88ms max = 105.35ms median = 101.07ms mean = 101.46ms mean = 101.63ms
Creating MNN Interpreter: efficientformerv2_s1
(index: 985,  score: 13.267050), (index: 984,  score: 4.345687), (index: 308,  score: 4.289116), 
[128 iters] min = 155.78ms max = 161.50ms median = 155.91ms mean = 156.58ms mean = 156.76ms
Creating MNN Interpreter: efficientformerv2_s2
(index: 985,  score: 12.623688), (index: 22,  score: 3.934839), (index: 309,  score: 3.621716), 
[73 iters] min = 274.74ms max = 290.52ms median = 275.22ms mean = 276.12ms mean = 276.30ms
Creating MNN Interpreter: SwiftFormer_XS
(index: 985,  score: 11.777540), (index: 883,  score: 4.872621), (index: 309,  score: 4.720006), 
[149 iters] min = 133.34ms max = 141.05ms median = 133.43ms mean = 134.15ms mean = 134.32ms
Creating MNN Interpreter: SwiftFormer_S
(index: 985,  score: 13.017500), (index: 720,  score: 4.264023), (index: 89,  score: 4.236666), 
[101 iters] min = 197.81ms max = 204.70ms median = 198.18ms mean = 198.94ms mean = 199.11ms
Creating MNN Interpreter: SwiftFormer_L1
(index: 985,  score: 13.600887), (index: 310,  score: 4.213078), (index: 309,  score: 4.000207), 
[66 iters] min = 302.61ms max = 311.13ms median = 302.61ms mean = 304.57ms mean = 304.73ms
Creating MNN Interpreter: EMO_1M
(index: 985,  score: 9.835707), (index: 309,  score: 4.373645), (index: 310,  score: 3.886256), 
[237 iters] min =  83.73ms max =  90.55ms median =  84.04ms mean =  84.45ms mean =  84.62ms
Creating MNN Interpreter: EMO_2M
(index: 985,  score: 9.492629), (index: 309,  score: 3.385969), (index: 308,  score: 3.221591), 
[157 iters] min = 126.91ms max = 134.41ms median = 132.72ms mean = 127.95ms mean = 128.12ms
Creating MNN Interpreter: EMO_5M
(index: 985,  score: 9.188762), (index: 883,  score: 2.813849), (index: 872,  score: 2.550485), 
[89 iters] min = 224.29ms max = 230.90ms median = 230.06ms mean = 225.60ms mean = 225.76ms
Creating MNN Interpreter: EMO_6M
(index: 985,  score: 9.295821), (index: 309,  score: 2.286871), (index: 308,  score: 2.279376), 
[84 iters] min = 237.03ms max = 245.71ms median = 242.57ms mean = 238.62ms mean = 238.78ms
Creating MNN Interpreter: edgenext_xx_small
(index: 985,  score: 10.561434), (index: 309,  score: 5.250806), (index: 310,  score: 4.911592), 
[238 iters] min =  82.57ms max =  89.85ms median =  83.49ms mean =  83.97ms mean =  84.21ms
Creating MNN Interpreter: edgenext_x_small
(index: 985,  score: 9.692822), (index: 309,  score: 4.415436), (index: 308,  score: 3.541662), 
[135 iters] min = 145.23ms max = 156.45ms median = 148.20ms mean = 148.34ms mean = 148.58ms
Creating MNN Interpreter: edgenext_small
(index: 985,  score: 12.133100), (index: 309,  score: 4.457827), (index: 308,  score: 3.973910), 
[58 iters] min = 345.45ms max = 354.89ms median = 346.19ms mean = 348.27ms mean = 348.51ms
Creating MNN Interpreter: mobilevitv2_050
(index: 985,  score: 8.414091), (index: 309,  score: 2.655013), (index: 89,  score: 2.475482), 
[155 iters] min = 127.86ms max = 134.29ms median = 133.47ms mean = 128.76ms mean = 129.03ms
Creating MNN Interpreter: mobilevitv2_075
(index: 985,  score: 8.284222), (index: 309,  score: 2.720599), (index: 308,  score: 2.143365), 
[83 iters] min = 239.62ms max = 246.08ms median = 240.63ms mean = 241.10ms mean = 241.33ms
Creating MNN Interpreter: mobilevitv2_100
(index: 985,  score: 8.258598), (index: 557,  score: 2.323085), (index: 309,  score: 2.103235), 
[51 iters] min = 394.67ms max = 401.57ms median = 395.00ms mean = 396.69ms mean = 396.92ms
Creating MNN Interpreter: mobilevitv2_125
(index: 985,  score: 8.478318), (index: 309,  score: 2.083009), (index: 113,  score: 1.427779), 
[35 iters] min = 574.81ms max = 586.25ms median = 576.52ms mean = 578.52ms mean = 578.77ms
Creating MNN Interpreter: mobilevitv2_150
(index: 985,  score: 9.081230), (index: 308,  score: 2.288944), (index: 301,  score: 2.262866), 
[26 iters] min = 774.75ms max = 783.18ms median = 775.54ms mean = 778.24ms mean = 778.47ms
Creating MNN Interpreter: mobilevitv2_175
(index: 985,  score: 8.934436), (index: 494,  score: 2.101688), (index: 309,  score: 1.884150), 
[20 iters] min =1006.96ms max =1016.74ms median =1013.49ms mean =1011.13ms mean =1011.38ms
Creating MNN Interpreter: mobilevitv2_200
(index: 985,  score: 8.605928), (index: 309,  score: 2.242715), (index: 308,  score: 2.195836), 
[16 iters] min =1291.35ms max =1301.74ms median =1296.40ms mean =1295.61ms mean =1295.85ms
Creating MNN Interpreter: mobilevit_xx_small
(index: 985,  score: 12.430065), (index: 309,  score: 6.491374), (index: 308,  score: 6.247639), 
[167 iters] min = 118.25ms max = 126.02ms median = 119.44ms mean = 119.88ms mean = 120.13ms
Creating MNN Interpreter: mobilevit_x_small
(index: 985,  score: 13.046449), (index: 89,  score: 6.823447), (index: 309,  score: 5.870105), 
[74 iters] min = 269.55ms max = 280.27ms median = 272.20ms mean = 271.86ms mean = 272.11ms
Creating MNN Interpreter: mobilevit_small
(index: 985,  score: 10.438556), (index: 309,  score: 3.712978), (index: 838,  score: 3.705502), 
[46 iters] min = 431.53ms max = 441.50ms median = 436.69ms mean = 434.63ms mean = 434.88ms
Creating MNN Interpreter: LeViT_128S
(index: 985,  score: 11.709265), (index: 308,  score: 3.568025), (index: 309,  score: 3.375850), 
[332 iters] min =  59.66ms max =  66.22ms median =  60.03ms mean =  60.25ms mean =  60.42ms
Creating MNN Interpreter: LeViT_128
(index: 985,  score: 11.346601), (index: 309,  score: 3.408505), (index: 113,  score: 3.297331), 
[241 iters] min =  82.18ms max =  88.99ms median =  82.77ms mean =  82.95ms mean =  83.12ms
Creating MNN Interpreter: LeViT_192
(index: 985,  score: 11.811327), (index: 324,  score: 3.396981), (index: 326,  score: 3.303845), 
[167 iters] min = 119.10ms max = 128.50ms median = 119.51ms mean = 120.12ms mean = 120.28ms
Creating MNN Interpreter: LeViT_256
(index: 985,  score: 11.188658), (index: 108,  score: 3.035140), (index: 309,  score: 2.935834), 
[102 iters] min = 196.32ms max = 204.04ms median = 196.73ms mean = 197.78ms mean = 197.96ms
Creating MNN Interpreter: resnet50
(index: 985,  score: 7.986690), (index: 113,  score: -5.246479), (index: 310,  score: -5.445749), 
[41 iters] min = 493.34ms max = 506.56ms median = 493.92ms mean = 496.23ms mean = 496.41ms
Creating MNN Interpreter: mobilenetv3_large_100
(index: 985,  score: 9.726588), (index: 310,  score: 2.717167), (index: 308,  score: 2.388681), 
[406 iters] min =  48.78ms max =  54.35ms median =  49.58ms mean =  49.22ms mean =  49.37ms
Creating MNN Interpreter: tf_efficientnetv2_b0
(index: 985,  score: 9.735632), (index: 309,  score: 2.588478), (index: 310,  score: 2.398535), 
[156 iters] min = 126.92ms max = 136.60ms median = 127.48ms mean = 128.08ms mean = 128.23ms
Creating MNN Interpreter: tf_efficientnetv2_b1
(index: 985,  score: 9.685658), (index: 309,  score: 2.281629), (index: 310,  score: 2.219452), 
[98 iters] min = 202.73ms max = 215.40ms median = 203.00ms mean = 204.64ms mean = 204.85ms
Creating MNN Interpreter: tf_efficientnetv2_b2
(index: 985,  score: 10.036093), (index: 883,  score: 2.635378), (index: 309,  score: 2.177715), 
[66 iters] min = 301.23ms max = 311.59ms median = 301.77ms mean = 303.69ms mean = 303.90ms
Creating MNN Interpreter: tf_efficientnetv2_b3
(index: 985,  score: 9.172971), (index: 955,  score: 2.843595), (index: 310,  score: 2.219716), 
[39 iters] min = 518.76ms max = 534.56ms median = 520.05ms mean = 523.93ms mean = 524.26ms

Mali-G52

The Panfrost driver on Mali-G52 is conformant for OpenGL ES 3.1. However, OpenGL ES on other hardware is non-conformant, as per the official documentation. OpenGL 3.1 support is work-in-progress and non-conformant.

Both the Panfrost and Lima drivers are included in Mesa and should work out-of-the-box after installing the relevant packages (which are, in practice, libglx-mesa0 and libgl1-mesa-dri).

OpenCL support is not implemented yet. Hardware video acceleration is not within the scope of the Panfrost or Lima drivers.

$ sudo dmesg | grep panfrost
[sudo] password for albert: 
[    7.787988] panfrost ffe40000.gpu: clock rate = 24000000
[    7.788061] panfrost ffe40000.gpu: error -ENODEV: _opp_set_regulators: no regulator (mali) found
[    7.800299] panfrost ffe40000.gpu: mali-g52 id 0x7212 major 0x0 minor 0x0 status 0x0
[    7.809936] panfrost ffe40000.gpu: features: 00000000,00000cf7, issues: 00000000,00000400
[    7.825795] panfrost ffe40000.gpu: Features: L2:0x07110206 Shader:0x00000000 Tiler:0x00000809 Mem:0x1 MMU:0x00002830 AS:0xff JS:0x7
[    7.862762] panfrost ffe40000.gpu: shader_present=0x3 l2_present=0x1
[    7.878129] [drm] Initialized panfrost 1.2.0 20180908 for ffe40000.gpu on minor 1

first performace

$ glmark2
=======================================================
    glmark2 2023.01
=======================================================
    OpenGL Information
    GL_VENDOR:      Panfrost
    GL_RENDERER:    Mali-G52 (Panfrost)
    GL_VERSION:     3.1 Mesa 22.3.6
    Surface Config: buf=32 r=8 g=8 b=8 a=8 depth=24 stencil=0 samples=0
    Surface Size:   800x600 windowed
=======================================================
[build] use-vbo=false: FPS: 802 FrameTime: 1.248 ms
[build] use-vbo=true: FPS: 897 FrameTime: 1.115 ms
[texture] texture-filter=nearest: FPS: 812 FrameTime: 1.232 ms
[texture] texture-filter=linear: FPS: 817 FrameTime: 1.224 ms
[texture] texture-filter=mipmap: FPS: 818 FrameTime: 1.223 ms
[shading] shading=gouraud: FPS: 738 FrameTime: 1.355 ms
[shading] shading=blinn-phong-inf: FPS: 742 FrameTime: 1.348 ms
[shading] shading=phong: FPS: 600 FrameTime: 1.669 ms
[shading] shading=cel: FPS: 664 FrameTime: 1.507 ms
[bump] bump-render=high-poly: FPS: 455 FrameTime: 2.201 ms
[bump] bump-render=normals: FPS: 1047 FrameTime: 0.955 ms
[bump] bump-render=height: FPS: 1015 FrameTime: 0.986 ms
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 580 FrameTime: 1.726 ms
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 306 FrameTime: 3.276 ms
[pulsar] light=false:quads=5:texture=false: FPS: 704 FrameTime: 1.421 ms
[desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: FPS: 283 FrameTime: 3.537 ms
[desktop] effect=shadow:windows=4: FPS: 753 FrameTime: 1.329 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 255 FrameTime: 3.931 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: FPS: 260 FrameTime: 3.851 ms
[buffer] columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 311 FrameTime: 3.220 ms
[ideas] speed=duration: FPS: 385 FrameTime: 2.604 ms
[jellyfish] <default>: FPS: 492 FrameTime: 2.033 ms
[terrain] <default>: FPS: 59 FrameTime: 17.003 ms
[shadow] <default>: FPS: 451 FrameTime: 2.217 ms
[refract] <default>: FPS: 112 FrameTime: 8.938 ms
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 640 FrameTime: 1.563 ms
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 633 FrameTime: 1.580 ms
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 640 FrameTime: 1.564 ms
[function] fragment-complexity=low:fragment-steps=5: FPS: 623 FrameTime: 1.606 ms
[function] fragment-complexity=medium:fragment-steps=5: FPS: 633 FrameTime: 1.582 ms
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 635 FrameTime: 1.577 ms
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 612 FrameTime: 1.636 ms
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 630 FrameTime: 1.588 ms
=======================================================
                                  glmark2 Score: 587 
=======================================================

$ glmark2-es2
=======================================================
    glmark2 2023.01
=======================================================
    OpenGL Information
    GL_VENDOR:      Panfrost
    GL_RENDERER:    Mali-G52 (Panfrost)
    GL_VERSION:     OpenGL ES 3.1 Mesa 22.3.6
    Surface Config: buf=32 r=8 g=8 b=8 a=8 depth=24 stencil=0 samples=0
    Surface Size:   800x600 windowed
=======================================================
[build] use-vbo=false: FPS: 832 FrameTime: 1.202 ms
[build] use-vbo=true: FPS: 941 FrameTime: 1.063 ms
[texture] texture-filter=nearest: FPS: 891 FrameTime: 1.122 ms
[texture] texture-filter=linear: FPS: 901 FrameTime: 1.110 ms
[texture] texture-filter=mipmap: FPS: 912 FrameTime: 1.097 ms
[shading] shading=gouraud: FPS: 795 FrameTime: 1.259 ms
[shading] shading=blinn-phong-inf: FPS: 779 FrameTime: 1.285 ms
[shading] shading=phong: FPS: 649 FrameTime: 1.543 ms
[shading] shading=cel: FPS: 712 FrameTime: 1.406 ms
[bump] bump-render=high-poly: FPS: 474 FrameTime: 2.112 ms
[bump] bump-render=normals: FPS: 1111 FrameTime: 0.901 ms
[bump] bump-render=height: FPS: 1078 FrameTime: 0.928 ms
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 579 FrameTime: 1.730 ms
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 306 FrameTime: 3.276 ms
[pulsar] light=false:quads=5:texture=false: FPS: 786 FrameTime: 1.273 ms
[desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: FPS: 286 FrameTime: 3.505 ms
[desktop] effect=shadow:windows=4: FPS: 761 FrameTime: 1.315 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 271 FrameTime: 3.691 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: FPS: 271 FrameTime: 3.692 ms
[buffer] columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 321 FrameTime: 3.117 ms
[ideas] speed=duration: FPS: 390 FrameTime: 2.566 ms
[jellyfish] <default>: FPS: 503 FrameTime: 1.989 ms
[terrain] <default>: FPS: 60 FrameTime: 16.862 ms
[shadow] <default>: FPS: 466 FrameTime: 2.147 ms
[refract] <default>: FPS: 120 FrameTime: 8.389 ms
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 733 FrameTime: 1.365 ms
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 722 FrameTime: 1.387 ms
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 728 FrameTime: 1.374 ms
[function] fragment-complexity=low:fragment-steps=5: FPS: 734 FrameTime: 1.364 ms
[function] fragment-complexity=medium:fragment-steps=5: FPS: 647 FrameTime: 1.548 ms
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 730 FrameTime: 1.370 ms
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 728 FrameTime: 1.374 ms
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 635 FrameTime: 1.576 ms
=======================================================
                                  glmark2 Score: 630 
=======================================================

Vulkan

performance

PAN_I_WANT_A_BROKEN_VULKAN_DRIVER=1 ./ncnn_perf --backend=v
WARNING: panvk is not a conformant Vulkan implementation, testing use only.
[0 Mali-G52 (Panfrost)]  queueC=0[1]  queueG=0[1]  queueT=0[1]
[0 Mali-G52 (Panfrost)]  bugsbn1=0  bugbilz=0  bugcopc=0  bugihfa=0
[0 Mali-G52 (Panfrost)]  fp16-p/s/a=1/0/0  int8-p/s/a=1/0/0
[0 Mali-G52 (Panfrost)]  subgroup=0  basic=0  vote=0  ballot=0  shuffle=0
Creating ncnn net: resnet50
(index: 985,  score: 7.964844), (index: 113,  score: -5.253906), (index: 310,  score: -5.449219), 
[35 iters] min = 579.76ms max = 593.46ms median = 587.83ms mean = 587.69ms
Creating ncnn net: efficientformerv2_s0
(index: 985,  score: 11.734375), (index: 644,  score: 4.976562), (index: 954,  score: 3.843750), 
[79 iters] min = 254.07ms max = 263.65ms median = 255.50ms mean = 255.95ms
Creating ncnn net: efficientformerv2_s1
(index: 985,  score: 13.265625), (index: 984,  score: 4.414062), (index: 308,  score: 4.296875), 
[55 iters] min = 362.87ms max = 369.85ms median = 364.51ms mean = 364.91ms
Creating ncnn net: efficientformerv2_s2
(index: 985,  score: 12.656250), (index: 22,  score: 3.953125), (index: 309,  score: 3.607422), 
[35 iters] min = 578.56ms max = 585.85ms median = 580.88ms mean = 581.46ms
Creating ncnn net: mobilevitv2_050
(index: 985,  score: 8.414062), (index: 309,  score: 2.660156), (index: 89,  score: 2.474609), 
[165 iters] min = 120.82ms max = 128.99ms median = 121.33ms mean = 121.87ms
Creating ncnn net: mobilevitv2_075
(index: 985,  score: 8.265625), (index: 309,  score: 2.697266), (index: 308,  score: 2.125000), 
[92 iters] min = 215.95ms max = 223.61ms median = 216.56ms mean = 217.44ms
Creating ncnn net: mobilevitv2_100
(index: 985,  score: 8.250000), (index: 557,  score: 2.314453), (index: 309,  score: 2.097656), 
[59 iters] min = 338.54ms max = 346.08ms median = 339.24ms mean = 340.52ms
Creating ncnn net: mobilevitv2_125
(index: 985,  score: 8.468750), (index: 309,  score: 2.072266), (index: 113,  score: 1.417969), 
[40 iters] min = 508.52ms max = 516.53ms median = 509.54ms mean = 510.92ms
Creating ncnn net: mobilevitv2_150
(index: 985,  score: 9.039062), (index: 308,  score: 2.265625), (index: 301,  score: 2.246094), 
[29 iters] min = 705.92ms max = 714.68ms median = 707.95ms mean = 709.52ms
Creating ncnn net: mobilevitv2_175
(index: 985,  score: 8.921875), (index: 494,  score: 2.097656), (index: 309,  score: 1.884766), 
[21 iters] min = 952.38ms max = 958.87ms median = 957.45ms mean = 956.40ms
Creating ncnn net: mobilevitv2_200
(index: 985,  score: 8.585938), (index: 309,  score: 2.228516), (index: 308,  score: 2.189453), 
[17 iters] min =1222.97ms max =1229.14ms median =1227.83ms mean =1227.13ms
Creating ncnn net: mobilenetv3_large_100
(index: 985,  score: 9.734375), (index: 310,  score: 2.720703), (index: 308,  score: 2.394531), 
[248 iters] min =  80.22ms max =  85.90ms median =  80.67ms mean =  80.95ms
Creating ncnn net: tf_efficientnetv2_b0
(index: 985,  score: 9.718750), (index: 309,  score: 2.591797), (index: 310,  score: 2.396484), 
[91 iters] min = 216.68ms max = 226.54ms median = 219.52ms mean = 220.10ms
Creating ncnn net: tf_efficientnetv2_b1
(index: 985,  score: 9.679688), (index: 309,  score: 2.285156), (index: 310,  score: 2.218750), 
[55 iters] min = 353.31ms max = 378.73ms median = 363.68ms mean = 364.05ms
Creating ncnn net: tf_efficientnetv2_b2
(index: 985,  score: 10.023438), (index: 883,  score: 2.626953), (index: 309,  score: 2.171875), 
[47 iters] min = 422.02ms max = 446.40ms median = 435.65ms mean = 433.58ms
Creating ncnn net: tf_efficientnetv2_b3
(index: 985,  score: 9.179688), (index: 955,  score: 2.847656), (index: 310,  score: 2.222656), 
[30 iters] min = 660.96ms max = 685.13ms median = 671.26ms mean = 671.43ms

MISC

panvk: Drop support for Midgard
mali gpu wikipedia The Lima and Panfrost FOSS drivers

See also: Free and open-source device drivers: graphics § ARM

On January 21, 2012, Phoronix reported that Luc Verhaegen was driving a reverse-engineering attempt aimed at the Mali series of GPUs, specifically the Mali 200 and Mali 400 versions. The project was known as Lima and targeted support for OpenGL ES 2.0.[117] The reverse-engineering project was presented at FOSDEM, February 4, 2012,[118][119] followed by the opening of a website[120] demonstrating some renders. On February 2, 2013, Verhaegen demonstrated Quake III Arena in timedemo mode, running on top of the Lima driver.[121] In May 2018, a Lima developer posted the driver for inclusion in the Linux kernel.[122] In May 2019, the Lima driver became part of the mainline Linux kernel.[123] The Mesa userspace counterpart was merged at the same time. It currently supports OpenGL ES 1.1, 2.0 and parts of Desktop OpenGL 2.1, and the fallback emulation in MESA provides full support for graphical desktop environments.[124]

Panfrost is a reverse-engineered driver effort for Mali Txxx (Midgard) and Gxx (Bifrost) GPUs. Introducing Panfrost[125] talk was presented at X.Org Developer's Conference 2018. As of May 2019, the Panfrost driver is part of the mainline Linux kernel.[126] and MESA. Panfrost supports OpenGL ES 2.0, 3.0 and 3.1, as well as OpenGL 3.1.[127]

andorid

Cortex-a73

single thread @1.7GHz

Creating MNN Interpreter: efficientformerv2_s0
parsed /proc/cpuinfo Hardware = "Amlogic"
The device support i8sdot:0, support fp16:0, support i8mm: 0
(index: 985,  score: 11.719067), (index: 644,  score: 4.942239), (index: 309,  score: 3.837213), 
[144 iters] min = 138.44ms max = 142.42ms median = 138.99ms mean = 139.19ms mean = 139.36ms
Creating MNN Interpreter: efficientformerv2_s1
(index: 985,  score: 13.267044), (index: 984,  score: 4.345684), (index: 308,  score: 4.289115), 
[93 iters] min = 214.72ms max = 218.59ms median = 215.16ms mean = 215.56ms mean = 215.72ms
Creating MNN Interpreter: efficientformerv2_s2
(index: 985,  score: 12.623688), (index: 22,  score: 3.934839), (index: 309,  score: 3.621716), 
[53 iters] min = 375.51ms max = 382.97ms median = 375.73ms mean = 378.02ms mean = 378.18ms
Creating MNN Interpreter: SwiftFormer_XS
(index: 985,  score: 11.777542), (index: 883,  score: 4.872621), (index: 309,  score: 4.720006), 
[110 iters] min = 181.22ms max = 186.88ms median = 182.49ms mean = 182.67ms mean = 182.84ms
Creating MNN Interpreter: SwiftFormer_S
(index: 985,  score: 13.017501), (index: 720,  score: 4.264025), (index: 89,  score: 4.236668), 
[74 iters] min = 269.78ms max = 277.71ms median = 273.01ms mean = 271.47ms mean = 271.63ms
Creating MNN Interpreter: SwiftFormer_L1
(index: 985,  score: 13.600889), (index: 310,  score: 4.213076), (index: 309,  score: 4.000207), 
[49 iters] min = 408.87ms max = 418.33ms median = 414.24ms mean = 411.00ms mean = 411.17ms
Creating MNN Interpreter: EMO_1M
(index: 985,  score: 9.835767), (index: 309,  score: 4.373718), (index: 310,  score: 3.886150), 
[175 iters] min = 113.17ms max = 119.55ms median = 113.45ms mean = 114.23ms mean = 114.40ms
Creating MNN Interpreter: EMO_2M
(index: 985,  score: 9.492679), (index: 309,  score: 3.385825), (index: 308,  score: 3.221593), 
[116 iters] min = 171.11ms max = 178.53ms median = 171.95ms mean = 172.63ms mean = 172.79ms
Creating MNN Interpreter: EMO_5M
(index: 985,  score: 9.188854), (index: 883,  score: 2.813967), (index: 872,  score: 2.550658), 
[66 iters] min = 302.18ms max = 310.56ms median = 303.59ms mean = 305.07ms mean = 305.23ms
Creating MNN Interpreter: EMO_6M
(index: 985,  score: 9.295814), (index: 309,  score: 2.286917), (index: 308,  score: 2.279395), 
[63 iters] min = 320.19ms max = 327.41ms median = 321.01ms mean = 322.46ms mean = 322.62ms
Creating MNN Interpreter: edgenext_xx_small
(index: 985,  score: 10.561436), (index: 309,  score: 5.250808), (index: 310,  score: 4.911592), 
[200 iters] min =  97.44ms max = 106.35ms median =  99.19ms mean = 100.07ms mean = 100.30ms
Creating MNN Interpreter: edgenext_x_small
(index: 985,  score: 9.692823), (index: 309,  score: 4.415437), (index: 308,  score: 3.541663), 
[98 iters] min = 200.59ms max = 214.70ms median = 200.95ms mean = 204.32ms mean = 204.58ms
Creating MNN Interpreter: edgenext_small
(index: 985,  score: 12.133101), (index: 309,  score: 4.457826), (index: 308,  score: 3.973911), 
[42 iters] min = 477.80ms max = 490.18ms median = 478.43ms mean = 482.13ms mean = 482.35ms
Creating MNN Interpreter: mobilevitv2_050
(index: 985,  score: 8.414263), (index: 309,  score: 2.655451), (index: 89,  score: 2.475034), 
[116 iters] min = 171.08ms max = 175.89ms median = 175.14ms mean = 172.45ms mean = 172.71ms
Creating MNN Interpreter: mobilevitv2_075
(index: 985,  score: 8.283899), (index: 309,  score: 2.720386), (index: 308,  score: 2.143100), 
[63 iters] min = 319.38ms max = 324.26ms median = 324.26ms mean = 321.24ms mean = 321.48ms
Creating MNN Interpreter: mobilevitv2_100
(index: 985,  score: 8.259032), (index: 557,  score: 2.323329), (index: 309,  score: 2.103631), 
[37 iters] min = 543.37ms max = 551.02ms median = 546.47ms mean = 546.45ms mean = 546.71ms
Creating MNN Interpreter: mobilevitv2_125
(index: 985,  score: 8.478145), (index: 309,  score: 2.082687), (index: 113,  score: 1.427791), 
[26 iters] min = 766.73ms max = 775.93ms median = 775.93ms mean = 771.00ms mean = 771.26ms
Creating MNN Interpreter: mobilevitv2_150
(index: 985,  score: 9.081184), (index: 308,  score: 2.288956), (index: 301,  score: 2.262746), 
[20 iters] min =1035.52ms max =1041.57ms median =1037.95ms mean =1037.98ms mean =1038.23ms
Creating MNN Interpreter: mobilevitv2_175
(index: 985,  score: 8.934433), (index: 494,  score: 2.101466), (index: 309,  score: 1.884507), 
[15 iters] min =1347.30ms max =1356.08ms median =1347.30ms mean =1351.55ms mean =1351.82ms
Creating MNN Interpreter: mobilevitv2_200
(index: 985,  score: 8.606405), (index: 309,  score: 2.243204), (index: 308,  score: 2.195781), 
[12 iters] min =1791.63ms max =1799.40ms median =1796.55ms mean =1795.51ms mean =1795.79ms
Creating MNN Interpreter: mobilevit_xx_small
(index: 985,  score: 12.430826), (index: 309,  score: 6.491026), (index: 308,  score: 6.248070), 
[121 iters] min = 163.69ms max = 170.59ms median = 169.24ms mean = 165.23ms mean = 165.49ms
Creating MNN Interpreter: mobilevit_x_small
(index: 985,  score: 13.045725), (index: 89,  score: 6.823128), (index: 309,  score: 5.870834), 
[54 iters] min = 367.78ms max = 377.10ms median = 370.68ms mean = 370.28ms mean = 370.54ms
Creating MNN Interpreter: mobilevit_small
(index: 985,  score: 10.438497), (index: 309,  score: 3.711923), (index: 838,  score: 3.707543), 
[35 iters] min = 579.29ms max = 589.61ms median = 585.42ms mean = 582.55ms mean = 582.80ms
Creating MNN Interpreter: LeViT_128S
(index: 985,  score: 11.709261), (index: 308,  score: 3.568023), (index: 309,  score: 3.375848), 
[242 iters] min =  81.73ms max =  87.88ms median =  82.17ms mean =  82.67ms mean =  82.84ms
Creating MNN Interpreter: LeViT_128
(index: 985,  score: 11.346600), (index: 309,  score: 3.408505), (index: 113,  score: 3.297330), 
[176 iters] min = 112.68ms max = 117.54ms median = 114.08ms mean = 113.95ms mean = 114.12ms
Creating MNN Interpreter: LeViT_192
(index: 985,  score: 11.811327), (index: 324,  score: 3.396981), (index: 326,  score: 3.303845), 
[121 iters] min = 163.69ms max = 169.00ms median = 165.47ms mean = 165.18ms mean = 165.34ms
Creating MNN Interpreter: LeViT_256
(index: 985,  score: 11.188659), (index: 108,  score: 3.035138), (index: 309,  score: 2.935833), 
[68 iters] min = 294.23ms max = 299.54ms median = 298.18ms mean = 295.53ms mean = 295.70ms
Creating MNN Interpreter: resnet50
(index: 985,  score: 7.986822), (index: 113,  score: -5.246409), (index: 310,  score: -5.445825), 
[30 iters] min = 675.37ms max = 681.39ms median = 675.37ms mean = 677.97ms mean = 678.15ms
Creating MNN Interpreter: mobilenetv3_large_100
(index: 985,  score: 9.726589), (index: 310,  score: 2.717164), (index: 308,  score: 2.388678), 
[301 iters] min =  65.67ms max =  69.82ms median =  66.29ms mean =  66.43ms mean =  66.59ms
Creating MNN Interpreter: tf_efficientnetv2_b0
(index: 985,  score: 9.735810), (index: 309,  score: 2.588191), (index: 310,  score: 2.398264), 
[114 iters] min = 174.75ms max = 181.15ms median = 175.56ms mean = 176.20ms mean = 176.37ms
Creating MNN Interpreter: tf_efficientnetv2_b1
(index: 985,  score: 9.687206), (index: 309,  score: 2.282777), (index: 310,  score: 2.219707), 
[72 iters] min = 275.88ms max = 282.41ms median = 276.86ms mean = 277.58ms mean = 277.79ms
Creating MNN Interpreter: tf_efficientnetv2_b2
(index: 985,  score: 10.035254), (index: 883,  score: 2.634550), (index: 309,  score: 2.177393), 
[50 iters] min = 397.71ms max = 406.87ms median = 405.86ms mean = 401.07ms mean = 401.28ms
Creating MNN Interpreter: tf_efficientnetv2_b3
(index: 985,  score: 9.174591), (index: 955,  score: 2.843929), (index: 310,  score: 2.220167), 
[30 iters] min = 678.75ms max = 687.81ms median = 680.27ms mean = 683.10ms mean = 683.43ms

vulkan

MNN

Creating MNN Interpreter: tf_efficientnetv2_b0
parsed /proc/cpuinfo Hardware = "Amlogic"
The device support i8sdot:0, support fp16:0, support i8mm: 0
(index: 985,  score: 9.679688), (index: 309,  score: 2.583984), (index: 310,  score: 2.398438), 
[203 iters] min =  94.67ms max = 185.23ms median =  96.66ms mean =  98.41ms mean =  98.74ms
Creating MNN Interpreter: tf_efficientnetv2_b1
(index: 985,  score: 9.656250), (index: 309,  score: 2.291016), (index: 310,  score: 2.214844), 
[144 iters] min = 132.80ms max = 233.61ms median = 136.19ms mean = 138.95ms mean = 139.42ms
Creating MNN Interpreter: tf_efficientnetv2_b2
(index: 985,  score: 10.031250), (index: 883,  score: 2.625000), (index: 309,  score: 2.160156), 
[101 iters] min = 189.62ms max = 212.26ms median = 194.26ms mean = 198.03ms mean = 198.76ms
Creating MNN Interpreter: tf_efficientnetv2_b3
(index: 985,  score: 9.179688), (index: 955,  score: 2.820312), (index: 310,  score: 2.199219), 
[63 iters] min = 308.76ms max = 331.63ms median = 314.95ms mean = 316.58ms mean = 317.51ms

devboard: Khadas Vim3&Vim3l - YingkunZhou/EdgeTransformerBench GitHub Wiki

linux

utils

CPU

Cortex-a73

Mali-G52

Vulkan

performance

MISC

andorid

Cortex-a73

vulkan

MNN

⚠️ GitHub.com Fallback ⚠️

devboard: Khadas Vim3&Vim3l - YingkunZhou/EdgeTransformerBench GitHub Wiki

linux

utils

CPU

Cortex-a73

Mali-G52

Vulkan

performance

MISC

andorid

Cortex-a73

vulkan

MNN

⚠️ **GitHub.com Fallback** ⚠️

⚠️ GitHub.com Fallback ⚠️