I’m doing a cross-build for my armv7l platform with Gcc 4.9 + uclibc
As I am not using hardfloat feature, I tried disabling neon support with “ENABLE_NEON=OFF”
"
– CPU/HW features:
– Baseline: NEON FP16
– requested: DETECT
– disabled: VFPV3 NEON
"
But it seems the Neon is getting auto-detected and getting enabled,
If I pass CPU_BASELINE=FP16 in addition I could see the build gets success, but __m128 seems to be with SSE for x86 32/64 bit arch, doesn’t it?
kindly give me clarity on this please, why “ENABLE_NEON=OFF” doesn’t help in disabling it?
‘’’
uclibcgnueabi/OpenCV-4.9.0/opencv-4.9.0/modules/core/include/opencv2/core/cvdef.h:904:9: error: ‘__m128’ was not declared in this scope
__m128 v = _mm_load_ss(&x);
^
uclibcgnueabi/OpenCV-4.9.0/opencv-4.9.0/modules/core/include/opencv2/core/cvdef.h:905:52: error: ‘v’ was not declared in this scope
w = (ushort)_mm_cvtsi128_si32(_mm_cvtps_ph(v, 0));
‘’’
‘’’
– Performing Test HAVE_CPU_NEON_SUPPORT (check file: cmake/checks/cpu_neon.cpp)
– Performing Test HAVE_CPU_NEON_SUPPORT - Success
– Performing Test HAVE_CPU_FP16_SUPPORT (check file: cmake/checks/cpu_fp16.cpp)
– Performing Test HAVE_CPU_FP16_SUPPORT - Success
– Performing Test HAVE_CPU_BASELINE_FLAGS
– Performing Test HAVE_CPU_BASELINE_FLAGS - Success
‘’’