Hi, I don't know if this is already known. I found the ARM NEON vcvt instruction mentioned in some fix in the mailing list archive, but not what has been fixed.
However, I found a bug in the implementation of that instruction. I observe that the type operands are emulated in the wrong order. Example: "vcvt.s32.f32 q0,q0" should convert a float in q0 into an integer, which works correctly on the CPU. With Qemu I have to change the order of the operands to get the correct result (i.e. "vcvt.f32.s32 q0,q0"). I did not test all combinations of conversions, just integer and float. cheers, André
