I am trying to compile a Mxnet model to my PYNQ-Z2 board.
My images version is 2.3.
LLVM version is 4.0.1.
However, it occurs this error.
Here is the code of compilation.
Here is the code of compilation.
I think I’ll need a little more context about your example; it’s odd that you’re using GCC toolchain for android; can you use vanilla GCC instead? perhaps you’re adding binaries to your PATH that shouldn’t be there?
I used to do some tests on Android.
So, I set export TVM_NDK_CC=/opt/androidndk16b-toolchain-armv7/bin/arm-linux-androideabi-g++.
How can I change that?
Yes, I think that’s the issue. I would use whichever GCC ships with your linux distro. What does which gcc give you when you run it in a clean environment?
My target is target = tvm.target.create(‘llvm -device=arm_cpu -target=arm-linux-gnueabihf’).
try the following, llvm -target=armv7-none-linux-gnueabihf
This is where the host target is set: https://github.com/dmlc/tvm/blob/master/vta/python/vta/environment.py#L245
Ok, I will try.
Is it any problem when I set export TVM_NDK_CC=/bin/gcc?
And I want to know why I have to do it?
And what is vanilla GCC?
I don’t think that’s the issue; we shouldn’t be using NDK for the Zynq target. I believe it’s cross compiling for ARM but ARM has different ISA variants; Zynq uses armv7 which is worth specifying.
However, I’m thinking there could be a different setup issue in your build. Can you try just running the simple example on the pynq z1 unmodified? does that work?
