Is it possible to cross-compile a 64-bit CUDA application to the Jetson TX1. My TX1 has the latest 64-bit Jetpack and CUDA libraries. My host has the latest CUDA 7.5 release, but my host nvcc still gives the error “Cannot compile in the 64-bit mode when the host compiler targets arm”. This seems like an unnecessary limitation now that TX1 CUDA 7.0 is 64-bit. For reference, I’m using the Linaro compiler for targeting ARM on my host computer.
What version of L4T are you running? 23.x doesn’t support 64-bit applications.
I’m running L4T 24.1
The error is coming from my host machine’s installed NVCC.
I’ve not cross compiled CUDA…however, this makes me suspicious:
"Cannot compile in the 64-bit mode when the host compiler targets arm"
Perhaps it is just a lack of detail in the error message, but just “arm” tends to be 32-bit ARMv7. The correct target for 64-bit user space would be some variation on “arm64”, “aarch64”, or “ARMv8-a”.
If the reference is to the Linaro compiler, then definitely “arm” or “armhf” would be incorrect for 64-bit.