Appropriate flags for scons while building Arm Compute Library

[saswat0]

I was trying to build ARM Compute Library on my Khadas VIM 3 board and was confused regarding the flags that I need to pass to scons. The guide shows the following

scons arch=arm64-v8a neon=1 extra_cxx_flags="-fPIC" benchmark_tests=0 validation_tests=0 

But I'm not sure which one to use. Or if my hardware has neon. Is there any way to figure out the correct flags for my board?

Thanks in advance

[MikeJKelly]

Hi @saswat0

from looking at the specs that board has a quad core ARM Cortex-A73 and dual core Cortex-A53, those CPUs have an arm64-v8a architecture with the neon extensions.

https://developer.arm.com/ip-products/processors/cortex-a/cortex-a73 https://developer.arm.com/ip-products/processors/cortex-a/cortex-a53

It also has an ARM Mali-G52 GPU so if you want you can enable the GPU acceleration too:

https://developer.arm.com/ip-products/graphics-and-multimedia/mali-gpus/mali-g52-gpu

If you only want to use the CPU then the following should work:

scons arch=arm64-v8a neon=1 extra_cxx_flags="-fPIC" benchmark_tests=0 validation_tests=0

If you want to use the GPU too then you can use this:

scons arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" benchmark_tests=0 validation_tests=0

Best regards, Mike

[saswat0]

Thanks a ton @MikeJKelly. This pretty much solves my case

I just had one more question. What's the difference between building a stand-alone TfLite Delegate and Delegate with Arm NN. Will they yield different performances or so?