Does the TK1 support concurrent kernels?

Running the CUDA Samples ‘Concurrent Kernels’ sample implies that concurrent execution of kernels is either not supported or not working.

What does the deviceQuery sample indicate (assuming that sample is available in the tegra cuda version)?

It’s not reported in the Runtime API deviceQuery sample but is in the Driver API deviceQueryDrv example … except Driver API samples don’t compile on the TK1.

I modified deviceQuery to report deviceProp.concurrentKernels and, yes, concurrent kernels should be supported:

But the ‘Concurrent Kernels’ sample isn’t working.

This is still broken in the L4T 19.3 Driver Package + CUDA 6.0 Toolkit for L4T Rel 19.2.

This bug was reported (1526164) and is marked as fixed.

Any ETA when we’ll see this on the TK1? It’s kind of an important feature.

The bug is fixed in 6.5.
I don’t have an ETA for when it will be available for Jetson.

As an aside, if one comments out the check for libcuda.so in Makefile, then deviceQueryDrv will compile and run:

CUDALIB ?= $(shell find $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so)

ifeq (“$(CUDALIB)”,“”)

$(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<)

EXEC=@echo “[@]”

endif

LIBRARIES += -lcuda

Concurrent kernels are reported as supported in L4T 21.1 but still appear to be broken:

Help?

excuse me first,for my task i wish to replace the cueernce kernel on the board while i’d to save the config file ,here comes the question that where can i find it?

Linux kernel?

@cedar_luo, I think you’re asking about rebuilding the Linux kernel? It’s covered elsewhere in this forum. For example:

https://devtalk.nvidia.com/default/topic/751707/?comment=4264351

This thread is about a bug in CUDA concurrent GPU compute kernels and not building the Linux kernel. :)

first thank for your answers.i just want to get the .config file on the board but can not find it,can you please tell me which directory it may be?

@cedar_lou
It’s in /proc/config.gz

You can do something like:

$ zcat /proc/config.gz > /kernel/.config

to extract it.

thanks a lot!

There is a new 21.2 L4T and CUDA Toolkit and the driver still reports that concurrent kernels are supported when the multi-stream concurrent kernel CUDA Sample fails to run kernels concurrently.

It would be great if this bug were fixed.

@allanmac

any progress on this ? I am using 21.4 and I still can’t see the kernels running concurrently.