# Dual Core

## Setup Environment

Goal: Ensure a simple program can run on the dual core mcu successfully.

- Setup cubeIDE for dual core
https://www.youtube.com/watch?v=jI1k6p-fduE
- Debugging for dual core
- https://www.youtube.com/watch?v=k3mXhPZSasw (Official tutorial)
- Download program without dubugging on STM32CubeIDE
(Need to download ST-LINK Utility or STM32CubeProgrammer)
- https://www.youtube.com/watch?v=BdNq7bfXtjs
- [Discovery kit with STM32H747XI MCU - User manual](https://www.st.com/resource/en/user_manual/dm00504240-discovery-kit-with-stm32h747xi-mcu-stmicroelectronics.pdf)
- Development environment choices
- STM32cubeIDE (including STM32cubeMX)
- STM32cubeMX + Keil MDK
- Communicate between CM4 and CM7
https://community.st.com/s/article/stm32h7-dual-core-how-to-communicate-between-cm4-cm7
https://www.st.com/resource/en/reference_manual/rm0399-stm32h745755-and-stm32h747757-advanced-armbased-32bit-mcus-stmicroelectronics.pdf (p.554)
(HSEM register map: p.566)

## Issues

- HSEM: (HS)
- survey FreeRTOS
- Ensure tasks order - between different CPU

- Resorce Sharing (YF)
- 64KB SRAM are accessible for CM4 and CM7

- Scheduling
- Pipeline
- split two cores for each op
- grouping layers

- High resolution solution (CC)

- frame diff

- port model on F767 (one core) (CC and YC)
- mobilevit with mbedOS
- mobilevit with stm32cubeIDE - cubeAI
- pipeline
- profiler
- 找出model中所有cut vertices
- 
- profile程序
- 用cut vertex把model分成兩部分
- 大core拿前半段，小core拿後半段
- 跑一遍看各自花多少時間
- report 所有cut vertices情形的latency數據
- profile一遍找最好的分配方式
- 最好的定義lat_c0最接近lat_c1
- parallelize op 內 計算
- 拆tile分給兩個core
- statically
- compile time 決定，定義在model file的op參數裡
- dynamically
- 用特殊的方式算出比例
- 什麼方式要再想
- 或用producer-consumer架構，讓core自己去搶tile計算
- 需要之道拆成多小的tile才不會對compiler優化有影響
- 特殊方式分配計算
- 什麼方式要再想
- 必須考量到compiler好不好優化

Reference:
https://www.st.com/resource/en/application_note/an5617-stm32h745755-and-stm32h747757-lines-interprocessor-communications-stmicroelectronics.pdf

## Run a model

- Bring tflm to dual core
https://mirzafahad.github.io/2020-06-26-tflite-stm32-part3/
https://www.youtube.com/watch?v=gDFWCxrJruQ

- Replace tflm with tinyTS

- Schedule a model on CM4 and CM7