automatic deployment of a G diagram directly to a GPU target

[GoofyWires]

The upcoming GPU Analysis Toolkit that calls CUDA

https://decibel.ni.com/content/docs/DOC-7707

https://decibel.ni.com/content/docs/DOC-6064

and the CLoo4Labview that calls OpenCL http://lavag.org/topic/13342-get-a-cloo4labview-opencl-in-labview/page__hl__cuda__fromsearch__1  are great but only a first step.

I would like LV not only to let me deploy into GPU as they deploy into FPGA, I would like LV compiler to automatically find the available resources and make the best use of them.

Moving out some parallel computations from the CPU to the GPU should be done automatically behind the scene.

[JCC_(SK)]

I would prefere OpenCL instead CUDA

[Intaris]

Absolutely, OpenCL is the target here.

[GoofyWires]

Yep, OpenCL is cross platform and hardware indifferent, thus, more suiting.

If you like the idea why didn't you give it a Kudo?

Once LV's parallelism go GPGPU the performance will be x10 at least for parallel calculations and in LV there are lots of such even fundamental actions behind the scene.

[wanghh]

I thought you are able to realize automatic deployment of a G diagram directly to a GPU target.......

[Intaris]

Well the point of OpenCL is that creating an executable which is open-CL enabled does NOT rely on specific hardware, but will use it if present.

The compiler does NOT need to know anything about the hardware present on the compile machine, it auto-adapts depending on the available OpenCL drivers installed on the executing system.  Thus there would be no "deployment" as such.  It would be far more transparent that that.

As such the idea is flawed and I can't support it.

[Darren]

Any idea that has received less than 4 kudos within 4 years after posting will be automatically declined.