Using wamrc for target thumbv7

[schmalzer]

Hello,

I wanted to try out the AOT performance of wamr for a simple nbody benchmark that runs fine in interpreter mode on Zephyr. I am running the benchmark on a STM32WB55RG that has a cortex-m4 implementing the thumbv7 instruction set.

To compile the wasm binary file to AOT I followed the instructions to build wamrc and used the following command to compile the working benchmark (the wasm binary file) to an AOT file:
./wamrc -o nbody.c.aot --target=thumbv7 --cpu=cortex-m4 nbody.c.wasm

Then I use: xxd -i nbody.c.aot > nbody.c.aot.h to get the header file, which I can use to load the AOT file in memory which I did exactly like loading the wasm binary file:

#include "../benchmarks_zephyr/nbody/build/nbody.c.aot.h"
...
...
wasm_file_buf = (uint8 *)___build_nbody_c_aot;
wasm_file_size = ___build_nbody_c_aot_len;

if (!(wasm_module = wasm_runtime_load(wasm_file_buf, wasm_file_size, error_buf, sizeof(error_buf)))) {
    printf("%s\n", error_buf);
    goto fail1;
}
...
...

Then I use west build -p always -b nucleo_wb55rg C:\dev\zephyr\simple -t ram_report and west flash

However after the program is started I get the following output on my serial port:

[00:00:00.014,000] <err> os: ***** USAGE FAULT *****
[00:00:00.014,000] <err> os:   No coprocessor instructions
[00:00:00.014,000] <err> os: r0/a1:  0x200047b8  r1/a2:  0x20023480  r2/a3:  0x20023b64
[00:00:00.014,000] <err> os: r3/a4:  0x20023cfc r12/ip:  0x200035b1 r14/lr:  0x2000358b
[00:00:00.014,000] <err> os:  xpsr:  0x210b0000
[00:00:00.014,000] <err> os: Faulting instruction address (r15/pc): 0x2000322a
[00:00:00.014,000] <err> os: >>> ZEPHYR FATAL ERROR 33: Unknown error on CPU 0
[00:00:00.014,000] <err> os: Current thread: 0x20001b78 (unknown)
[00:00:00.080,000] <err> os: Halting system

The error seems to happen in wasm_application_execute_main(module_inst, app_argc, app_argv);

Also, if I remove everything from the benchmark code and just run return 0; from the main, this error does not happen and the program terminates successfully. However, I am not sure if the code just gets optimized away when only returning 0. As soon as I add simple math operations the error is back.
I think some pitfalls are that the STM32WB55RG has a coprocessor and a FPU.

Any idea, how to tackle this problem? Thanks for any hints!