Our #BL602 TensorFlow Build hitting some Math errors ... Let's fix them

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

#BL602 Makefile for TensorFlow Lite needs this ... To fix the Math build errors

https://github.com/lupyuen/tflite-bl602/blob/main/bouffalo.mk#L21-L27

#BL602 TensorFlow Lite also needs the "ruy" library ... For matrix multiplication

https://github.com/lupyuen/tflite-bl602/blob/main/bouffalo.mk#L71-L82

Our #BL602 Build of TensorFlow Lite ... Fixing a few more missing C++ Functions

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

Source Folders for TensorFlow Lite are defined in the #BL602 Makefiles "bouffalo.mk" and "component.mk"

https://github.com/lupyuen/tflite-bl602/blob/main/bouffalo.mk

Our #BL602 Build of TensorFlow Lite calls "new" and "delete" to Allocate and Deallocate C++ Objects in Heap Memory ... But it's supposed to use Static Memory, not Heap Memory ... Let's find out why

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

Setting this flag will tell TensorFlow Lite to use Static Memory (instead of Heap Memory) on #BL602 ... Neat!

https://github.com/lupyuen/tflite-bl602/blob/main/bouffalo.mk#L46-L49

Final few Missing Functions for TensorFlow Lite on #BL602 ... Let's find out what they are

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

Just realised that "CPPFLAGS" works for building #BL602 C++ Libraries, but not for BL602 C++ Apps ... Dang

https://github.com/lupyuen/bl_iot_sdk/blob/tflite/customer_app/sdk_app_tflite/Makefile#L12-L20

Aha "CPPFLAGS" should be defined in #BL602 "bouffalo.mk" ... Not "Makefile" 🙄

https://github.com/lupyuen/bl_iot_sdk/blob/tflite/customer_app/sdk_app_tflite/sdk_app_tflite/bouffalo.mk#L28-L41

We disable Thread-Safe Initialisation by setting "-fno-threadsafe-statics" ... This fixes the Missing Functions "__cxa_guard_acquire" and "__cxa_guard_release" in #BL602 TensorFlow

https://github.com/lupyuen/bl_iot_sdk/blob/tflite/customer_app/sdk_app_tflite/sdk_app_tflite/bouffalo.mk#L37-L41

We set the #BL602 Global Destructor "__dso_handle" to null ... Since we're not destroying the Static TensorFlow Variables

https://github.com/lupyuen/bl_iot_sdk/blob/tflite/customer_app/sdk_app_tflite/sdk_app_tflite/demo.c#L20-L22

#TensorFlow Lite builds OK on #BL602 yay! Let's flash and run the firmware! 🤩

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

#TensorFlow Lite runs OK on #BL602 yay! 🎉 This pre-trained TensorFlow Model is supposed to infer "y = sin(x)" ... Let's verify the output! 🥧

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

#BL602 runs #TensorFlow Lite Model that infers "y = sin(x)" ... Output looks reasonably accurate!

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

#TensorFlow Lite on #BL602 ... Now with a proper Command-Line Interface

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

How we might teach #BL602 ... To light up the LED 😂

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

We have a #TensorFlow Lite Model that generates a Sine Wave ... Let's use it to light up the #BL602 LED ... In a fun and interesting way!

https://github.com/lupyuen/bl_iot_sdk/tree/tflite/customer_app/sdk_app_tflite

@lupyuen
Lupy, does pine cone have enough number crunching power to run bigger ai models? Or do You use tf here just to train models on pc rather than running them?

@lupyuen
If youd get me a hw i could do some sw devel for you. Anyway. Youre no longer as responsive as you used to be. Making up my mind about unfollowing. Cheers.

@lupyuen
What are possible uses of TF on #BL602 ?