Meson option build fixes

[piotrrak]

Misc fixes for some of the options disabled by configuration options.
This is step in direction of fixing builds with various configuration options configurations.

Self evaluation:

1. Build test: [ ]Passed [ ]Failed [*]Skipped
2. Run test: [ ]Passed [ ]Failed [*]Skipped

How to evaluate:

1. Describe how to evaluate in order to be reproduced by reviewer(s).

meson setup --native-file=disables.ini

disables.ini with contents:

[project options]

enable-blas=false 
enable-tflite-backbone=false 
enable-memory-swap=false 
enable-nnstreamer-backbone=false 
enable-tflite-interpreter=false 
enable-app=false 
ml-api-support='disabled'
enable-test=true
werror=false

[baek2sm]

This script contains the code if get_option('enable-nnstreamer-backbone') and get_option('platform') != 'android' extra_defines += '-DENABLE_NNSTREAMER_BACKBONE=1' endif.

How about merging this part as well? It seems like it would be fine to move the part extra_defines += '-DENABLE_NNSTREAMER_BACKBONE=1' under this condition, too!

[djeong20]

FYI) line 407-409

[piotrrak]

@baek2sm @djeong20 thanks for suggestion, change updated/rebased.
I've dediced to move host_machine.system() != 'windows' there too, that probably won't be necessary in near future.

[piotrrak]

I've reverted back to not moving subdir('nnstreamer') to the earlier stage of meson.

[djeong20]

LGTM!

[skykongkong8]

According to discussion made from #2930, I believe enable-blas=true should set by default for both x86 and arm situation. Please share your opinion if you find this fix inappropriate!

[piotrrak]

According to discussion made from #2930, I believe enable-blas=true should set by default for both x86 and arm situation. Please share your opinion if you find this fix inappropriate!

The configuration I've picked here is non-default. This configuration was written to make very minimal thing, with as much things as disabled. Is "should" = "we never really shouldn't build with BLAS disabled on x86 or arm"?

Maybe we should consider removing this enable-blas option and make proper choice based on host machine/platform if non-default shouldn't or can't really work - if that is the case.

[skykongkong8]

Is "should" = "we never really shouldn't build with BLAS disabled on x86 or arm"?

Yes, I believe that would be more appropriate for the current status since we rely all single-precision computations on cblas (for both x86 and arm). And yes, I think we should really consider removing enable-blas option, but it actually needs more discussion because we might have to introduce multiple backend options per target architecture.

[jijoongmoon]

LGTM

[skykongkong8]

LGTM