how to calculating the MIPS for iMx

[kumar_ugesh003]

I looking some support to calculating the MIPS for iMx solo and Dual

[kumar_ugesh004]

Hi Jimmy,

 Thanks for the document reference but I have to calculate the MIPS for the algorithm which I have written to measure the performance. To measure the performance , I was looking for some sort of formula.

Could you please provide if you know some formula? which consider all active threads, sockets, pipeline instruction etc.

Regards.

Munirajulu Ugesh

[jimmychan]

We don't have the formula. Because the performance is affected by many factors.  In the i.MX6, there are different IP modules such as GPU for 3D/2D, VPU for video encode/decode and IPU for image processing. The MIPS only can show you the Cortex A9 performance. It cannot show you the whole system performance. Different board configuration and hardware design also has different performance. e.g. DDR, BSP..etc.

You can search "Benchmark i.mx6" in google. There are some performance reports from different board design.

For your reference:

1. Benchmark tests - iMX6 Rex Projects 

2. Nitrogen8M - Android Benchmark - Boundary Devices 

3. https://liliputing.com/2012/10/freescale-i-mx6-quad-core-cpu-benchmarks-ampe-a10-tablet.html 

4. Freescale i.MX6 Quad-Core Benchmark Demo 

[kumar_ugesh004]

Hi Jim,

I am completely agree with you but my case is like below
I have a process A, B, C, D and would like to calculate the MIPS consider for each of the process to evaluate the performance system. Do we have something to find the MIPS taken for each process?

Example: Lets say we have a process A,B,C,D and system is running for 25sec with 800MhZ then what is the MIPS consider for each of the process?

[jimmychan]

I think you need the tools for that. You can consult our partners Green Hills and Lauterbach.

NXP - Partner Profile Information (Green Hills)

NXP - Partner Profile Information  (Lauterbach)

[joanxie]

I got answer from expert team:

"The DMIPS for i.MX6SL/6DL/7D is 2500/5000/3800. You can check it on Arm website directly."

[kumar_ugesh004]

if the below is the formula

                Processor Clock Frequency (MCLK)             cycles / second      million instructions
   MIPS = ------------------------------------------------ = -------------------- = --------------------
          Average Cycles per Instruction (CPI) * 1,000,000   cycles / instruction          second

How to get the Average Cycles per Instruction ?

[jimmychan]

You can find the information from here: (at page 35)

https://cache.freescale.com/files/training/doc/dwf/DWF13_AMF_CON_T0060.pdf

MIPS calculation:  For Cortex A9, it is 2.5 DMIPS/MHz per core

Details please refer to List of ARM microarchitectures - Wikipedia 

[kumar_ugesh004]

There is a lack of information regarding the MIPS calculations. Do you have specific link for this?
I am using iMX6 Dual core A9.