1

My CPU doesn't support AVX512 so unfortunately I can't use the function _mm256_loadu_epi32().

It looks like I can use "_mm256_set_epi32()", but I'm not sure if it's hopelessly slower than "_mm256_loadu_XXXXXX()". Any idea? what's the best way to do this?

Noob
  • 103
  • 9
  • It suggested using _mm256_loadu_si256() which "Load 256-bits of integer data". I'm not sure about if 256-bits of integer data will be treated like 8 signed int32, guess I will test it. – Noob Jan 14 '20 at 16:13
  • What is the epi32 version adding that the AVX `_mm256_loadu_si256 ` doesn't do? I see it is `vmovdqu32 ymm, m256` vs `vmovdqu ymm, m256` but the documentation looks basically the same. – Fire Lancer Jan 14 '20 at 16:14
  • @rustyx Thanks for pointing to it, it's definitely helpful! solved now – Noob Jan 14 '20 at 16:19
  • @FireLancer Yeah I understand it now. It was just a bit unintuitive to switch from the function to loading float/double to loading integer. – Noob Jan 14 '20 at 16:21

0 Answers0