AM2634-Q1: Fixed-point math lib and detailed benchmark comparison with C2000 MCU

Will Gao

Part Number: AM2634-Q1

Hi experts,

In my customer side, the control algorithm running on AM263x consumes more cycles than running on C2000. I proposed that allocating data and codes into TCM which decreases the executing cycles a lot.

But the customer still has concerns about the efficiency of AM263x, so they request for fixed-point math lib and detailed benchmark comparison with C2000 MCU. Please help on this requests.

Thanks & Regards,

Will

over 3 years ago

+1 Sri Vidya Gunturi over 3 years ago

TI__Expert 7586 points

Hi Will

Control algorithm running on AM263x consumes more cycles than running on C2000. Because, C2K has a customized C28 core from TI, along with TMU and Control Law Accelerator to compute faster calculations.

You can do the following to reduce execution time:

Always Build code in Release mode with compiler optimizations set to z or fast. This is very important as debug mode with no optimizations creates a lot of overhead in the code and increases execution time.
Enable link time optimization in your compiler options. You need to add this -flto flag in both Arm Compiler and Arm Linker Options of Build Settings. As shown below:
Link time optimizations help reduce good amount of time.
Your control Loop code is present in ISR, so the whole ISR, stack, irq stack needs to be placed in the TCM.
1. You can refer this link link to achieve maximum optimization through TCM.
2. https://e2e.ti.com/support/microcontrollers/arm-based-microcontrollers-group/arm-based-microcontrollers/f/arm-based-microcontrollers-forum/1165683/faq-mcu-plus-sdk-am263x-how-to-add-code-into-tcm-using-linker-cmd-file
3. A linker file is attached as reference in that FAQ, you can use the same linker file for your application for better results.
Do you have any trigonometric functions being used in your Control Loop?, you can replace those functions with the mathTrig Library functions for faster execution: https://www.ti.com/lit/an/sprad27a/sprad27a.pdf?ts=1668059872384&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FAM2634

I will work with internal team to see if we have any benchmark comparisons to C2000.

But there is one application note on the AM263 benchmarks, refer to this section - AM26x FOC Loop Benchmarks:

https://www.ti.com/lit/an/sprad32/sprad32.pdf?ts=1668059715060&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FAM2634

The execution time for different instances in the Control loop is given.

Thanks & regards

Sri Vidya

0 Will Gao over 3 years ago in reply to Sri Vidya Gunturi

TI__Intellectual 2335 points

Hi Sri Vidya,

Thanks for your instructions. Please help check with internal team if we have any benchmark comparisons to C2000.

Thanks & Regards,

Will

0 Sri Vidya Gunturi over 3 years ago in reply to Will Gao

TI__Expert 7586 points

Hi Will

Currently, there is no official comparison to C2000. There is no plan currently to implement any benchmark comparisons from C2000 to AM263. If anything comes up in future will do let you know.

Otherwise, if there is any piece of code that you could share that you worked on C2000. We will help you port with AM263 that will give you best results.

Thanks & Regards

Sri Vidya

Arm-based microcontrollers

Arm-based microcontrollers forum

AM2634-Q1: Fixed-point math lib and detailed benchmark comparison with C2000 MCU