Search code examples
armsimdneon

How to convert uint32x4_t to uint8x16_t in Neon?

Then what I mean is that I want to split each 32-bit unsigned int into four 8-bit. The corresponding bit not change.

If the uint32x4_t is:

01000101001111100000001000010000 | 01000101001111100000001000010000 | 01000101001111100000001000010000 | 01000101001111100000001000010000

I want to get:

 01000101 | 00111110 | 00000010 | 00010000 | 01000101 | 00111110 | 00000010 | 00010000 | 01000101 | 00111110 | 00000010 | 00010000 | 01000101 | 00111110 | 00000010 | 00010000 |

How can I do this?

Solution

  • With vreinterpretq_u8_u32. The prototype is:

    uint8x16_t vreinterpretq_u8_u32 (uint32x4_t a);

    Edit: as @EOF pointed out in the comment below, you may need an endian swap (vrev32q_u8) too. GCC and clang define __BYTE_ORDER__ (to __ORDER_LITTLE_ENDIAN__ for little-endian and __ORDER_BIG_ENDIAN__ for big-endian), or if you need something a bit more portable see https://github.com/nemequ/portable-snippets/tree/master/endian