S32K142 & S32K144 & S32K146运算速度

cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

S32K142 & S32K144 & S32K146运算速度

Jump to solution
5,510 Views
zhengjianfei1
Contributor III

Hi community,   

   最近发现一个比较奇怪的问题:S32K142运算速度比S32K144和S32K146慢。同样主频配置条件下,运行同样的代码,S32K142速度会慢一些。测试代码如下。测试结果是S32K142执行时间3.78ms,S32K144和S32K146执行时间是3.55ms。从芯片上看,都是M4内核,主频也设置来一样的,应该运行速度一样才对的,为什么会出现这样的差异的?

pastedImage_1.png

pastedImage_2.png

0 Kudos
1 Solution
5,246 Views
Robin_Shen
NXP TechSupport
NXP TechSupport

Hi jianfei,

Yes, you are right, this can be caused by the P-Flash port width difference between S32K142 and S32K144:

36.1.2 Flash memory sizes

S32K144 has two 256KB P-Flash interleave block and form a 128-bit port width for data access. while S32K142 has a non-interleave 256KB P-Flash block and its access port width is 64-bit. This feature is very useful for Instruction and Data Speculation & Prefech implemented on S32K1xx Flash controller(detailed configurations can be refered to OCMDR0~OCMDR2 register of MSCM module), and it is enabled by default.

34.4.2.18 On-Chip Memory Descriptor Register (OCMDR0).png

Best Regards,

Robin

 

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

View solution in original post

6 Replies
5,246 Views
Robin_Shen
NXP TechSupport
NXP TechSupport

Hi jianfei,

是否使用CLKOUT功能引出内部HCLK或者BUS_CLK,使用示波器观察其频率是否一致?

如果使用IO口反转方式来测代码执行时间其实是不准确的,因为从MCU执行反转IO代码到实际IO口反转电平还需要一段时间。(当然对你这个测试目的来说是没有问题的,因为如果频率配置相同IO口翻转消耗的时间应该也是相同的)

这篇文章Cycle Counting on ARM Cortex-M with DWT提到的DWT模块可以用来测代码执行时间。(你也可以使用SysTick观察test函数前后的counter差值)

Best Regards,

Robin

 

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

0 Kudos
5,246 Views
zhengjianfei1
Contributor III

Hi Robin,

      按照您的方法测试,CLOCK OUT将bus CLK输出,S32K142和S32K146的结果是:S32K142 23.99MHZ,S32K146 23.98MHZ,两者几乎无差异。

     此外,代码里面增加SYSTICK部分代码,通过寄存器查看每次执行后SYSTICK的差值,代码如下。

pastedImage_1.png

测试情况是:S32K142执行test函数SYSTICK差值为:90475。S32K146执行test函数的SYSTICK差值为85388。可以看出,S32K146确实是比S32K142快一些,不知道为什么?

0 Kudos
5,246 Views
Robin_Shen
NXP TechSupport
NXP TechSupport

抱歉让你久等了,我刚得到AE的答复:

Please double check and confirm the S32K142 and S32K144/6 project using the same C compiler optimization level setting. You can find the detailed compilation configurations for a certain C file from its .arg file located in the same location under the project build target folder(here is Debug_FLASH, which is generated after build the project) as below for example:

pastedImage_1.png

In general, -o1 runs faster than -o0 due to less instructions generated for the same function in C code.

 

Tips: the S32DS example project is default set to -o1 optimization level, while a new created project has -o0 optimization level

Best Regards,

Robin

 

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

0 Kudos
5,246 Views
zhengjianfei1
Contributor III

Hi Robin,

      我这边是新建的工程,确认了是没有开优化的,O0等级。如下。

pastedImage_1.png

0 Kudos
5,246 Views
zhengjianfei1
Contributor III

对了,还有一个信息。当我开启了指令缓存后,S32K142和S32K144两者速度就一样了,从SYSTICK差值可以看出,完全是一样的。从示波器的测试结果也一样的。我觉得会不会是S32K142和S32K144的flash读取速率有差异呢?

0 Kudos
5,247 Views
Robin_Shen
NXP TechSupport
NXP TechSupport

Hi jianfei,

Yes, you are right, this can be caused by the P-Flash port width difference between S32K142 and S32K144:

36.1.2 Flash memory sizes

S32K144 has two 256KB P-Flash interleave block and form a 128-bit port width for data access. while S32K142 has a non-interleave 256KB P-Flash block and its access port width is 64-bit. This feature is very useful for Instruction and Data Speculation & Prefech implemented on S32K1xx Flash controller(detailed configurations can be refered to OCMDR0~OCMDR2 register of MSCM module), and it is enabled by default.

34.4.2.18 On-Chip Memory Descriptor Register (OCMDR0).png

Best Regards,

Robin

 

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------