It seems that when building with option(NEON "Neon acceleration" ON) and option(AUTOVECTORIZE "Prefer autovectorizable code to one using C intrinsics" ON), the arm_vec_exponent_f32 function is not defined, which will cause error in arm_svm_polynomial_predict_f32.c line 407.
@Q1anWan You're right. It looks like the ARM_MATH_AUTOVECTORIZE test is not applied coherently to all Neon versions. The arm_svm_polynomial_predict_f32 is not disabled with ARM_MATH_AUTOVECTORIZE but it should.
It seems that when building with option(NEON "Neon acceleration" ON) and option(AUTOVECTORIZE "Prefer autovectorizable code to one using C intrinsics" ON), the arm_vec_exponent_f32 function is not defined, which will cause error in arm_svm_polynomial_predict_f32.c line 407.