TMS570 DCAN Performance Question

JR Simma

I've been having a problem with slow reading times from a DCAN on a TMS570 running at 160MHz. VCLK and VCLKA are both at 80MHz.

Two 32-bit reads from the IF3DATA and IF3DATB registers is taking 350ns. I'm fairly sure it is not a pipeline issue because for testing purposes I'm doing 10 32-bit reads and increased it to 12 32-bit reads. Going from 10 to 12 reads increased the time by 350ns.

Can anyone tell me why this might be?

Thanks,

JR Simma

Simma Software, Inc.

The J1939 Experts

over 14 years ago

0 JR Simma over 14 years ago

Intellectual 660 points

I also want to mention that I did the same block of 10 reads from the GIO section and they are taking exactly the same amount of time (1.9us). So whatever is going on here probably isn't related to just the DCAN.

0 Pratip Kumar over 14 years ago in reply to JR Simma

TI__Intellectual 2285 points

Hi Simma,

Are you reading these sections and writing it to RAM ? How would you define this read time , in case of DCAN and GIO ?

- Pratip

0 JR Simma over 14 years ago in reply to Pratip Kumar

Intellectual 660 points

Yes I am reading directly from IF3DATA and IF3DATB DCAN registers and storing into onchip RAM. We are using an o-scope to measure the time.

0 Sunil Oak over 14 years ago in reply to JR Simma

TI__Mastermind 49120 points

Hello,

The Cortex-R4F CPU is optimized for maximizing the bandwidth for accesses to tightly-coupled memories such as the CPU RAM. Accesses to peripherals take several cycles, which is essentially what you are measuring. You could offload the CPU by using the DMA module to move data from the DCAN to the CPU data RAM.

Regards, Sunil

0 JR Simma over 14 years ago in reply to Sunil Oak

Intellectual 660 points

Thank you for the information. I agree that we will need to do DMA then. Does the 570 group have any DMA examples? One for the DCAN would be great.

In case the 570 team would like to know, I did the same test on a C2000 running at 150MHz. The 2 reads for a C2000 takes 88ns with a 570 at 160MHz taking 350ns.

0 KGreb over 14 years ago in reply to JR Simma

TI__Mastermind 23000 points

As a further suggestion, you may also want to evaluate the memory ordering you have set for the DCAN peripheral. By default the peripheral region is strongly ordered, which means buffered accesses to peripherals are not allowed. If you set an MPU region on the DCAN, you could change the memory ordering to device, which allows buffered writes. This can improve your write access performance to the peripheral.

0 JR Simma over 14 years ago in reply to KGreb

Intellectual 660 points

Thank you for the suggestion. We plan to do use DMA because of the read times. If you guys have an example of using DMA on the 570, that would be great.

Thanks,

Arm-based microcontrollers

Arm-based microcontrollers forum

TMS570 DCAN Performance Question