GDALVersionInfo: provide information about floating-point features

[dbaston]

Feature description

I'm comparing differences in gdalwarp output between two machines that are using the same version of GDAL 3.13 but with different architectures. I wondered if the differences could be related to differences in the use of intrinsics, but I don't see a way to get this information from gdal.VersionInfo('BUILD_INFO').

I experimented with a patch to GDALVersionInfo along the following lines:

        osBuildInfo += "FLOATING_POINT_OPTIONS="
#ifdef HAVE_SSE_AT_COMPILE_TIME
        "SSE: YES"
#else
        "SSE: NO"
#endif
#ifdef HAVE_SSSE3_AT_COMPILE_TIME
        ", SSSE3: YES"
#else
        ", SSSE3: NO"
#endif
#ifdef HAVE_AVX_AT_COMPILE_TIME
        ", AVX: YES"
#else
        ", AVX: NO"
#endif

But the various capabilities don't seem to be consistently exposed as definitions that can be picked up in this way. For example, CMake sets a variable GDAL_AVX2_FLAG if AVX2 support is detected, but there is no corresponding definition available to the preprocessor.

Additional context

No response

[rouault]

HAVE_AVX_AT_COMPILE_TIME doesn't presume that AVX will be effectively used. It just tells that the compiler knows about the AVX intrinsincs. GDAL can then use that to build some object files with -mavx and decide at runtime whether to use them if the CPU supports AVX

For warping, we only use SSE2, and the SSE2-Neon optimization on Arm, that should (I believe) yield exact same results.