The reason why the ncnn converted model on raspberry pie 4B always crashes when called

Recently I got a raspberry pie 4B Do experiments with your board , Compile ncnn after , Convert the model into bin and param When calling after a file, it is always because Segment Fault collapse , Compiled debug Version of ncnn After use gdb Check it out , I found that I always collapsed in convolution_3x3.h in :

#0  0x000dccb0 in ncnn::_ZN4ncnnL14conv3x3s1_neonERKNS_3MatERS0_S2_S2_RKNS_6OptionE._omp_fn.13(void) () at /root/ncnn/src/layer/arm/convolution_3x3.h:423
#1  0xb6eabc64 in ?? () from /usr/lib/arm-linux-gnueabihf/libgomp.so.1

The code at the crash is :

sum0 += r0[0] * k0[0];

r0 It is the present. layer Input characteristic data ,k0 Convolution kernel data , There is no apparent problem , Then I checked from the beginning , I found that the board burned 32 A system of (armv7l), No 64 position (aarch64) Of , So compile ncnn Pay attention to add

     -DCMAKE_TOOLCHAIN_FILE=../toolchains/pi3.toolchain.cmake

This option , Otherwise, the converted model will crash every time it is called , Recompile with this option ncnn, And then use onnx2ncn The converted model can be used normally .