How to run with OpenGL backend on RPi4

[dimkastan]

Dear all, first of all thanks for sharing your implementation.
I am trying to run a Yolov5 on a RPI4 using OpenGL backend an I am getting the following error:

Can't open file:/sys/devices/system/cpu/cpufreq/ondemand/affected_cpus
CPU Group: [ 0  1  2  3 ], 600000 - 1600000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
not support half 
Segmentation fault

More specifically:

On a headless RPI4, I compiled the MNN with OpenGL support and tried to run a Yolo network (f32 bit)

When running my program I get:

Can't open file:/sys/devices/system/cpu/cpufreq/ondemand/affected_cpus
CPU Group: [ 0  1  2  3 ], 600000 - 1600000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
not support half 
Segmentation fault

To enable the GL backend in my code I used the following commands:

    MNN::ScheduleConfig config;
    config.type = MNN_FORWARD_OPENGL;  // Use OpenGL backend
    int numThread = 4;  // Default number of threads

    if (argc > 3) {  // Check if the third argument (argv[3]) exists
        numThread = atoi(argv[3]);
        if (numThread <= 0) {
            std::cerr << "Invalid number of threads specified. Using default: 4\n";
            numThread = 4;
        }
    }
    config.numThread = numThread;
    // Force full precision (FP32)
    MNN::BackendConfig backendConfig;
    backendConfig.precision = MNN::BackendConfig::Precision_Normal;
    config.backendConfig = &backendConfig;
    auto session = interpreter->createSession(config);

and I compiled using:

 g++ -std=c++11 -Wl,--no-as-needed -o mnn_test mnn_model_test.cpp -I/home/pi/mytest/MNN-3.0.4/include -L/home/pi/mytest/MNN-3.0.4/build/source/backend/opengl -lMNN_GL -L/home/pi/mytest/MNN-3.0.4/build -lMNN -lGLEW -lGLESv2 -lEGL

and exported the following env variable:

 export EGL_PLATFORM=surfaceless

the LDD command returns:

	linux-vdso.so.1 (0x0000007fa41a8000)
	libMNN_GL.so => /home/pi/mytest/MNN-3.0.4/build/source/backend/opengl/libMNN_GL.so (0x0000007fa4128000)
	libMNN.so => /home/pi/mytest/MNN-3.0.4/build/libMNN.so (0x0000007fa3ee4000)
	libGLEW.so.2.1 => /lib/aarch64-linux-gnu/libGLEW.so.2.1 (0x0000007fa3e0b000)
	libGLESv2.so.2 => /lib/aarch64-linux-gnu/libGLESv2.so.2 (0x0000007fa3dd6000)
	libEGL.so.1 => /lib/aarch64-linux-gnu/libEGL.so.1 (0x0000007fa3db1000)
	libstdc++.so.6 => /lib/aarch64-linux-gnu/libstdc++.so.6 (0x0000007fa3bd9000)
	libm.so.6 => /lib/aarch64-linux-gnu/libm.so.6 (0x0000007fa3b2e000)
	libgcc_s.so.1 => /lib/aarch64-linux-gnu/libgcc_s.so.1 (0x0000007fa3b0a000)
	libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6 (0x0000007fa3996000)
	libpthread.so.0 => /lib/aarch64-linux-gnu/libpthread.so.0 (0x0000007fa3965000)
	libGL.so.1 => /lib/aarch64-linux-gnu/libGL.so.1 (0x0000007fa386e000)
	libX11.so.6 => /lib/aarch64-linux-gnu/libX11.so.6 (0x0000007fa371f000)
	libGLdispatch.so.0 => /lib/aarch64-linux-gnu/libGLdispatch.so.0 (0x0000007fa3594000)
	libdl.so.2 => /lib/aarch64-linux-gnu/libdl.so.2 (0x0000007fa3580000)
	/lib/ld-linux-aarch64.so.1 (0x0000007fa4178000)
	libGLX.so.0 => /lib/aarch64-linux-gnu/libGLX.so.0 (0x0000007fa353d000)
	libxcb.so.1 => /lib/aarch64-linux-gnu/libxcb.so.1 (0x0000007fa3505000)
	libXau.so.6 => /lib/aarch64-linux-gnu/libXau.so.6 (0x0000007fa34f1000)
	libXdmcp.so.6 => /lib/aarch64-linux-gnu/libXdmcp.so.6 (0x0000007fa34db000)
	libbsd.so.0 => /lib/aarch64-linux-gnu/libbsd.so.0 (0x0000007fa34b6000)
	libmd.so.0 => /lib/aarch64-linux-gnu/libmd.so.0 (0x0000007fa349a000)

My model was extracted using the ultralytics python SDK

When running GetMNNInfo I get:

Can't open file:/sys/devices/system/cpu/cpufreq/ondemand/affected_cpus
CPU Group: [ 0  1  2  3 ], 600000 - 1600000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
Model default dimensionFormat is NCHW
Model Inputs:
[ images ]: dimensionFormat: NCHW, size: [ 1,3,640,640 ], type is float
Model Outputs:
[ onnx::Sigmoid_361 ]
[ onnx::Sigmoid_405 ]
[ output0 ]
Model Version: 3.0.4 
Model bizCode: {"description": "Ultralytics 0p8GFLOPS-92p5 model ", "author": "Ultralytics", "date": "2025-02-12T08:38:39.595401", "version": "8.3.74", "license": "AGPL-3.0 License (https://ultralytics.com/license)", "docs": "https://docs.ultralytics.com", "stride": 32, "task": "detect", "batch": 1, "imgsz": [640, 640], "names": {"0": "person"}, "args": {"batch": 1, "half": false, "int8": false}}

Please note that with CPU backend the network runs properly.

Any help/support will be highly appreciated!

[dimkastan]

Dear all,

I converted the model using the MNN converter and got the following error when running with OpenGL backend.

./mnn_test model_fp32.mnn
Can't open file:/sys/devices/system/cpu/cpufreq/ondemand/affected_cpus
CPU Group: [ 0 1 2 3 ], 600000 - 1600000
The device supports: i8sdot:0, fp16:0, i8mm: 0, sve2: 0
eglChooseConfig error !!!
mContext error !!!
terminate called after throwing an instance of 'std::logic_error'
what(): basic_string::_M_construct null not valid
Aborted

Any suggestion or comment is welcome!

[jxt1234]

OpenGL Backend is deprecated now. You can use Vulkan or OpenCL Backend instead to enable GPU compute.