TDA4VH-Q1: Using OpenCL to call the GPU in TDA4VH

tian liang

Part Number: TDA4VH-Q1
Other Parts Discussed in Thread: TDA4VH

Tool/software:

Dear TI Experts,

Hello! Our team is currently using the TDA4VH-Q1 device, with SDK version ti-processor-sdk-rtos-j784s4-evm-09_00_01_01. We would like to deploy our surround-view stitching algorithm onto the device’s GPU. The algorithm has already been developed based on OpenCL. However, we are unsure how to use OpenCL to call the GPU on the TDA4VH-Q1 and deploy our algorithm onto it. Could you please provide specific example code to help us with this process?

We look forward to your reply!

Best regards,

3 months ago

0 Sarabesh Srinivasan 3 months ago

TI__Genius 17500 points

Hello,

I believe we should have some example documentation to help guide you on using OpenCL. I will discuss with the team and get back to you.

Thanks,
Sarabesh S.

0 Sarabesh Srinivasan 2 months ago in reply to Sarabesh Srinivasan

TI__Genius 17500 points

Hello,

In SDK 9.0, OpenCL is not enabled. Could you deploy your SRV application on 11.0 SDK? 11.0 is when we've enabled this so it would be a significantly easier effort to run there.

Thank you,
Sarabesh S.

0 tian liang 2 months ago in reply to Sarabesh Srinivasan

Prodigy 100 points

Dear TI experts,

Hello! Our team is currently using the TDA4VH device with SDK version 11.0. We would like to deploy a surround‑view stitching algorithm onto the device’s GPU. The algorithm has been developed using OpenCL, but we are not sure how to use OpenCL to target/invoke the TDA4VH GPU. Could you provide concrete sample code to help us with this process, and also tell us how to compile it?

We look forward to your reply.

0 tian liang 2 months ago in reply to Sarabesh Srinivasan

Prodigy 100 points

Dear TI experts,

We look forward to your reply.

0 tian liang 2 months ago in reply to Sarabesh Srinivasan

Prodigy 100 points

Hello,

Our team is currently using the TDA4VH device with SDK version 11.0. We would like to deploy a surround‑view stitching algorithm onto the device’s GPU. The algorithm has been developed using OpenCL, but we are not sure how to use OpenCL to target/invoke the TDA4VH GPU. Could you provide concrete sample code to help us with this process, and also tell us how to compile it?

We look forward to your reply.

0 Sarabesh Srinivasan 2 months ago in reply to tian liang

TI__Genius 17500 points

Hello,

Let me consult with the development team to see if we can provide any examples. This should be fairly standard.

Thanks,
Sarabesh S.

0 tian liang 2 months ago in reply to Sarabesh Srinivasan

Prodigy 100 points

Hello,

May I ask if there has been any progress so far?

Thanks,

tian L.

0 Sarabesh Srinivasan 2 months ago in reply to tian liang

TI__Genius 17500 points

Hi Tian,

You can simply use gcc to compile your OpenCL code on the target. If you're needing OpenCL dev headers you would need to pull them from your yocto-sysroots since we don't ship dev-headers in the SDK release. Additionally, the target only has the Imagination GPU ICD so CL_DEVICE_TYPE_GPU will only return one GPU device.

Thank you,
Sarabesh S.

0 tian liang 2 months ago in reply to Sarabesh Srinivasan

Prodigy 100 points

Hi Sarabesh,

Do you have any sample code? We have been trying to implement it continuously but have encountered many problems. If you could provide sample code, it would be very helpful for our work.

Thanks,

tian L.

0 Sarabesh Srinivasan 2 months ago in reply to tian liang

TI__Genius 17500 points

Hi Tian,

I don't have any examples immediately to validate and share but I will look into it. Expect some delays to verify this due to other ongoing tasks.

Thanks,
Sarabesh S.

0 tian liang 2 months ago in reply to Sarabesh Srinivasan

Prodigy 100 points

Hi Sarabesh,

How can I check the real-time GPU utilization on TDA4VH SDK 11.0?

Thanks,

tian L.

0 Sarabesh Srinivasan 2 months ago in reply to tian liang

TI__Genius 17500 points

Hi Tian,

There's a couple of ways you can verify realtime GPU utilization. For one you can use Imaginations PVRTune server and client software. This will give you a very in depth variety of GPU metrics (https://developer.imaginationtech.com/solutions/pvrtune/).

The second way to very simply see the GPU utilization would be to cat the file located in /sys/kernel/debug/pvr/status.

Thank you,
Sarabesh S.

Processors

Processors forum

TDA4VH-Q1: Using OpenCL to call the GPU in TDA4VH