Issue after updating STM32CubeAI Studio to v1.2: No longer generates raw files

[pawatJoy]

Hello,

I have just updated STM32CubeAI Studio to version 1.2.

In the previous version, the AI Studio generated:

• Middleware
• Application code
• Raw files (e.g., network_atonbuf.xSPI2.raw)

My workflow was to add the middleware and application code to my project, then convert the raw file (network_atonbuf.xSPI2.raw) to a hex file. Everything worked fine until I updated the studio.

In version 1.2, instead of generating the raw file, it now generates a C source file: network_atonbuf.xSPI2.c.

I am unsure what to do with this new .c file. I placed it in my project, but nothing works as expected.

Could someone please advise on how to proceed or explain the change in workflow?

 

[pawatJoy]

Thank you for your useful information.

Can I ask another question? If I want to use that C file, how should I edit my Linker script (.ld)?

[hamitiya]

Hello  @pawatJoy  

It depends how you want to integrate your network, but if you want to map your weights in 0x7100_0000, you will need to do the following. It is only a proposal, and can be fine-tuned based on  your board and what you want to achieve:

 

- Create a section (memory definition) in your linker script, starting from the address you want:

 

```

/* Memories definition */

MEMORY

{

RAM (xrw) : ORIGIN = 0x34000000, LENGTH = 0x100000

ROM (xrw) : ORIGIN = 0x70100400, LENGTH = 0x6ffc00

EXTFLASH (rx) : ORIGIN = 0x70180000, LENGTH = 0x71000000 - 0x70180000

NETWORK (rx) : ORIGIN = 0x71000000, LENGTH = 0x40000

}

```

 

in Sections:

 

.AI_NETWORK :
{
. = ALIGN(32);
KEEP(*(.AI_NETWORK))
. = ALIGN(32);
} >NETWORK

 

 

then, in your C Code, you can define this pragma:

 

#define AI_NETWORK __attribute__((section(".AI_NETWORK")))

 

and in your network file in AppliAIAppnetwork_atonbuf.xSPI2.c, add the corresponding pragma to your weights allocation:

 

 

 

For more information, you can refer to the LD documentation here, especially the Memory Layout section:

Using LD, the GNU linker - Command Language

 

If you want to put your weights in the same section as "ROM", you can integrate your C file as-is. No modification needed. However, your ROM should have enough space in order to compile and not have an overflow. 

 

Best regards,

Yanis

 

 

[hamitiya]

Hello  @pawatJoy  

with this new update, STM32Cube AI Studio uses the ability to generate weights as C code instead of binary file.

If you want to reuse the same behavior as before, you can go to:

Target > Memory Pool

and, at the far bottom, under Weights Mapping:

 

 

 

Usage of C file is easier to integrate in your application since you can ship one unique .elf / .bin file containing the whole app. However, it means you need to edit your linker script and add a macro to map the weights where you want.

 

Best regards,

Yanis