EVK-MIMXRT685 eIQ Glow for RT685 Lab CIFAR-10 example build error

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

EVK-MIMXRT685 eIQ Glow for RT685 Lab CIFAR-10 example build error

Jump to solution
887 Views
tugrulyatagan
Contributor I

Hi,

I am following https://community.nxp.com/t5/eIQ-Machine-Learning-Software/eIQ-Glow-Lab-for-i-MX-RT/ta-p/1123119 eIQ Glow for RT685 Lab.pdf file step-by-step. I have installed Python 3.8 and Tensorflow 2.4. I have imported the evkmimxrt685_dsp_cifar10_demo_cm33 example and I have created the mnist.h, mnist.o, and mnist.weights.txt glow_bundle files as explained in the lab notes. However generated mnist.weights.txt size is too big (~2.5 MB). Thus, project cannot be compiled at step "5 Run Glow on the RT685" because generated mnist model doesn't fit to the device. I did what the document said and here is the result. Did I miss something?

Memory region         Used Size  Region Size  %age Used
            SRAM:     2373800 B      1536 KB    150.92%
c:/nxp/mcuxpressoide_11.3.0_5222/ide/plugins/com.nxp.mcuxpresso.tools.win32_11.3.0.202008311133/tools/bin/../lib/gcc/arm-none-eabi/9.3.1/../../../../arm-none-eabi/bin/ld.exe: evkmimxrt685_dsp_cifar10_demo_cm33_org.axf section `.dspcode' will not fit in region `SRAM'
        SRAM_ROM:          0 GB       128 KB      0.00%
c:/nxp/mcuxpressoide_11.3.0_5222/ide/plugins/com.nxp.mcuxpresso.tools.win32_11.3.0.202008311133/tools/bin/../lib/gcc/arm-none-eabi/9.3.1/../../../../arm-none-eabi/bin/ld.exe: region `SRAM' overflowed by 800936 bytes
     SRAM_SHARED:          0 GB       384 KB      0.00%
collect2.exe: error: ld returned 1 exit status
        SRAM_DSP:          0 GB      2560 KB      0.00%
         USB_RAM:          0 GB        16 KB      0.00%
make: *** [makefile:51: evkmimxrt685_dsp_cifar10_demo_cm33_org.axf] Error 1
"make -r -j4 all" terminated with exit code 2. Build might be incomplete.

Thanks,

Tugrul

Labels (1)
0 Kudos
1 Solution
840 Views
kerryzhou
NXP TechSupport
NXP TechSupport

Hi @tugrulyatagan 

Please check my attached project, it contains the flash, but my project is imported from the new SDK2.10.0, you can test it on your side, and add the code again, whether you still have the RAM overflow issues or not?

kerryzhou_0-1625558323085.png

 

Wish it helps you!

Best Regards,

kerry

View solution in original post

3 Replies
845 Views
tugrulyatagan
Contributor I

Hi @kerryzhou,

I am using RT685 SDK 2.9.0. I was following https://community.nxp.com/t5/eIQ-Machine-Learning-Software/eIQ-Glow-Lab-for-i-MX-RT/ta-p/1123119 eIQ Glow for RT685 Lab.pdf since I am using RT685, not RT1060. In this lab hand out dsp_cifar10_demo_cm33 example is used for demo.

tugrulyatagan_1-1625490280289.png

 

However in the RT1060 eIQ video you sent glow_lenet_mnist example is used. But there isn't glow_lenet_mnist example in RT685 SDK. There is dsp_lenet_demo example and it is completely different.

tugrulyatagan_2-1625490372352.png

 

I check the dsp_cifar10_demo_cm33 project settings and there isn't any FLASH option.

tugrulyatagan_0-1625490188659.png

Thanks,

Tugrul

0 Kudos
841 Views
kerryzhou
NXP TechSupport
NXP TechSupport

Hi @tugrulyatagan 

Please check my attached project, it contains the flash, but my project is imported from the new SDK2.10.0, you can test it on your side, and add the code again, whether you still have the RAM overflow issues or not?

kerryzhou_0-1625558323085.png

 

Wish it helps you!

Best Regards,

kerry

870 Views
kerryzhou
NXP TechSupport
NXP TechSupport

Hi @tugrulyatagan 

  Please also tell me which SDK version for RT685 you are using?

  I will test it on my side.

 BTW, do you watch the video?

https://www.nxp.com/design/training/getting-started-with-eiq-ml-software-for-glow-neural-network-com...

From your build result, seems the file memory location issues, all put in the SRAM, do you select the project build in the RAM directly? Do you enable the flash?

kerryzhou_0-1624441962896.png

But your code didn't have the flash memory.

So, please check the configuration again.

Best Regards,

kerry

kerryzhou_1-1624441989245.png

 

I check the video, the build result is:

kerryzhou_2-1624442004706.png

You can find it also use the SDRAM and the flash.

So, please check the configuration again.

Best Regards,

Kerry

 

 

0 Kudos