In the same spirit as #903, it would be nice to have an example that shows how to place the tensor arena in external RAM.
If I understand correctly, this should be much easier to do than moving the weights to RAM. Looking at the two_models_one_arena example, it should be fairly simple to just add -DSHARED_TENSOR_ARENA to the APP_FLAGS in the Makefile, and add declare the array as follows
// Place in BSS - zero-initialise in bootstrap - do not occupy space in the flash image
__attribute__ ((section(".ExtMem.bss")))
uint8_t tensor_arena[LARGEST_TENSOR_ARENA_SIZE] ALIGN(8);
I'm not exactly sure if there's anything else to be done (I haven't tried this yet), but I think this would be a very valuable example to have.
In the same spirit as #903, it would be nice to have an example that shows how to place the tensor arena in external RAM.
If I understand correctly, this should be much easier to do than moving the weights to RAM. Looking at the two_models_one_arena example, it should be fairly simple to just add
-DSHARED_TENSOR_ARENA
to theAPP_FLAGS
in the Makefile, and add declare the array as followsI'm not exactly sure if there's anything else to be done (I haven't tried this yet), but I think this would be a very valuable example to have.