fix: Allow building SSE kernels regardless of SSE 4.2 being (in)active

[atteneder]

I tried building KTX-Software for Android x86_64 and for some reason the Android NDK's compiler sets the define __SSE4_2__ even though the SSE version is specified via -msse4.1:

$ /path/to/android-ndk-r21d/toolchains/llvm/prebuilt/linux-x86_64/bin/clang++ --target=x86_64-none-linux-android24 -msse4.1 -dM -E - < /dev/null | egrep "SSE|AVX" | sort
#define __SSE__ 1
#define __SSE2__ 1
#define __SSE2_MATH__ 1
#define __SSE3__ 1
#define __SSE4_1__ 1
#define __SSE4_2__ 1
#define __SSE_MATH__ 1
#define __SSSE3__ 1

Removing paramter --target=x86_64-none-linux-android24 would solve it, but this is added by the CMake Android toolchain and I'm afraid beyond my control.

This brings me to my question: Does this absolutely have to fail if __SSE4_2__ is set? With the suggested change both basisu and KTX-Software build just fine, although I didn't run it yet.

Thanks for considering!

[richgel999]

I think this change would be fine. I'll queue this PR for testing.

[zeux]

Won't the build start failing after this when -msse4.1 is specified on other compilers? I would suggest removing the 4_2 check altogether, as the code doesn't use any 4.2 intrinsics atm.

[atteneder]

Won't the build start failing after this when -msse4.1 is specified on other compilers?

I didn't check, but I think you have a point.

I would suggest removing the 4_2 check altogether, as the code doesn't use any 4.2 intrinsics atm.

Agreed, probably the better solution.

[atteneder]

Changed it (and tested on Ubuntu x86_64)

[richgel999]

Yea, it doesn't use SSE 4.2. I'll get this in.

[richgel999]

Got it - thank you very much!