[zibba]

Hi

I'm trying to convert some Intel SSE code to ARM NEON and I can't find the equivalent NEON instructions of the SSE single-precision instructions _mm_mul_ss, _m_add_ss. Please help.

thanks

[isogen74]

I don't believe NEON has any direct equivalent - you'll either have to:

* multiply out everything and merge two result registers to get what you want
* do the scalar multiply on the ARM core using normal FPU instructions, and then merge that in. You could use single-lane load store rather than register mangling in this case, which _may_ be faster.

[zibba]

isogen74, on 15 May 2012 - 07:09 AM, said:

I don't believe NEON has any direct equivalent - you'll either have to:

* multiply out everything and merge two result registers to get what you want
* do the scalar multiply on the ARM core using normal FPU instructions, and then merge that in. You could use single-lane load store rather than register mangling in this case, which _may_ be faster.

Thanks for the quick reply. At least that explains why I couldn't find those instructions. I think I'll use the first version to avoid doing to and from the FPU.