imx8m mini with FLEXSPI

取消
显示结果 
显示  仅  | 搜索替代 
您的意思是: 
已解决

imx8m mini with FLEXSPI

跳至解决方案
2,659 次查看
Abdellahkafi
Contributor I

Hi,

I want to use FlexSPI on the imx8m mini. It seems working but when i write bock of data ,i have noticed that writing data to FlexSPI  TX IP FIFO is really slow, in my case writing 32 bytes needs around 4 us, also when i try to toggle GPIO by directly writing to the dedicated register takes long time (150 ns , the assembly code looks fine with only 3 instructions) which seems too slow, i tried also to fire a timer interrupt to be sure that clock/pll configurations are fine. 

All the tests are made on the M4 core and the code is based on the SDK_2.8.2_EVKB-IMXRT1050; so i am wondering what is the time needed  to access peripherals registers.

 

 

标签 (1)
0 项奖励
回复
1 解答
2,645 次查看
igorpadykov
NXP Employee
NXP Employee

Hi Abdellahkafi

 

regarding "time needed  to access peripherals registers" - general answer

this is caused by bus arbiters, like NIC301 in i.MX6 processors, similar modules

are used in i.MX8 processors. Good latency description can be found in 

sect.3.4 NIC transfer latency AN4947 "Understanding Vybrid Architecture"

https://www.nxp.com/docs/en/application-note/AN4947.pdf

So 150 ns for toggling GPIO by directly writing to the dedicated register may be

considered as normal.

 

Best regards
igor

在原帖中查看解决方案

0 项奖励
回复
2 回复数
2,646 次查看
igorpadykov
NXP Employee
NXP Employee

Hi Abdellahkafi

 

regarding "time needed  to access peripherals registers" - general answer

this is caused by bus arbiters, like NIC301 in i.MX6 processors, similar modules

are used in i.MX8 processors. Good latency description can be found in 

sect.3.4 NIC transfer latency AN4947 "Understanding Vybrid Architecture"

https://www.nxp.com/docs/en/application-note/AN4947.pdf

So 150 ns for toggling GPIO by directly writing to the dedicated register may be

considered as normal.

 

Best regards
igor

0 项奖励
回复
2,639 次查看
Abdellahkafi
Contributor I

Thanks for your fast answer

Concerning FlexSPI, writing 32 bytes (eight 32-bits words ) to IP TX-FIFO, I changed the watermark level to 3 in my case and the filling time is around 2.5 us, is it expected and is there any way to make it faster without using DMA?

Best regards 

0 项奖励
回复