Assistance with Yocto Project and i.MX93 EVK Queries

[kumarchaitanyab]

I am currently working with the i.MX93 EVK (11x11mm board) and am new to the Yocto project environment. I have a few questions that I hope you can assist with:

1. I successfully compiled the imx-image-core build with TensorFlow and ran the tensorflow-lite-2.16.2 image classification example. I was able to observe the logs as described in the UG10166 user guide. However, I am unable to locate the source code for the application. Could you please guide me on where to find it?
2. I am interested in measuring power consumption during application execution. Is it possible to measure power using the pca9451a device? If so, could you provide guidance on how to achieve this?
3. According to the RM00285 reference manual, the uPower API is specific to the i.MX 8ULP. Could you confirm whether this API is also compatible with the i.MX93 EVK?

[kumarchaitanyab]

Hi @Chavira ,

I am unable to locate the source code for the M33 binaries (e.g., imx93-11x11-evk_m33_TCM_rpmsg_lite_pingpong_rtos_linux_remote.bin or imx93-11x11-evk_m33_TCM_low_power_wakeword.bin) generated during the build process using either "bitbake imx-image-core" or "bitbake imx-image-full". I would greatly appreciate it if you could guide me to the source code for these binaries.

Additionally, after loading the M33 image using the command:
load mmc 0:1 0x80000000 imx93-11x11-evk_m33_TCM_rpmsg_lite_pingpong_rtos_linux_remote.bin
and then executing:
bootaux 0x80000000
from the U-Boot console, I observe the message:
Starting auxiliary core addr = 0x80000000.

Following this, U-Boot execution resumes, and Linux starts successfully on COM3 (USB3). However, I do not see any output on COM4 (USB4).

Could you confirm whether 0x80000000 is a valid address for this operation?

For reference, the boot mode is set to SD card (0010).

Thank you for your assistance!

[Chavira]

Hi @kumarchaitanyab!

Thank you for contacting NXP Support!

1 >> The Machine learning packages are only available on full image.

2 >> You can follow the AN13917 to do Power Consumption Measurement in iMX93-EVK.

3 >> Yes, the uPoweris only for iMX8ULP .

Best Regards!

Alejandro

[kumarchaitanyab]

  Hi @Chavira 

Actually, I have added below in local.conf file to include ML application and compiled the build

IMAGE_INSTALL:append = " packagegroup-imx-ml"
TOOLCHAIN_TARGET_TASK:append = " tensorflow-lite-dev onnxruntime-dev"
PACKAGECONFIG:append:pn-opencv_mx8 = " tests tests-imx"

and I am able to run the example ML application. It would be good if you can point me to the source code for referece?

[Chavira]

Hi @kumarchaitanyab!

All our source code can be found in our github.

Also you can refer to our IMX MACHINE LEARNING User Guide 

[kumarchaitanyab]

Hi @Chavira, thanks for the reply. Apart from the BCU tool, is there a way that we can measure the power of the i.MX93 EVK, more specifically through the APIs within the code?

[kumarchaitanyab]

Hi @Chavira ,

Is it possible to call APIs from within the application to read the power consumption data from either PCA9451A or PAC1934 device ? 

[Chavira]

Hi @kumarchaitanyab!

For iMX93 EVK you only can do the measurements only using BCU tool.

[kumarchaitanyab]

Hi @Chavira 

The BCU provides power values at a 10ms resolution, the inference in the label_image sample application completes in approximately 3ms when using Vela. How can we ensure that the power consumption measured by the BCU accurately reflects the actual usage? We did try to give the refresh rate parameter (-hz=0.001 or 0.0001) with different values to speed up the values but we see the samples are at 10ms rate only. Is there a better way to capture the samples at a faster rate to know the actual power consumption?

[Chavira]

Hi @kumarchaitanyab!

For that specific case you can do a rework to your board and connect directly to an external ammeter.

You can download the schematic of the evk board in this link.