How to perform the inverse of _mm256_movemask_epi8 (VPMOVMSKB)?
avx, avx2, c, simd, x86
Solution
Here is an alternative to LUT or `pdep` instructions that might be more efficient:
- Copy your 32-bit mask to both low bytes of some `ymm` register and bytes 16..19 of the same register. You could use temporary array and `_mm256_load_si256`. Or you could move single copy of 32-bit mask to low bytes of some `ymm` register, then broadcast it with `VPBROADCASTD (_mm_broadcastd_epi32)` or other broadcast/shuffle instructions.
- Rearrange bytes of the register so that low 8 bytes (each) contain low 8 bits of your mask, next 8 bytes - next 8 bits, etc. This could be done with `VPSHUFB (_mm256_shuffle_epi8)` with control register containing '0' in low 8 bytes, '1' in next 8 bytes, etc.
- Select proper bit for each byte with `VPOR (_mm256_or_si256)` or `VPAND (_mm256_and_si256)`.
- Set MSB of appropriate bytes with `VPCMPEQB (_mm256_cmpeq_epi8)`. Compare each byte to `0xFF`. If you want each bit of the mask toggled, use `VPAND` on previous step and compare to zero.
Additional flexibility of this approach is that you could choose different control register for step #2 and different mask for step #3 to shuffle bits of your bit mask (for example you could copy this mask to `ymm` register in reversed order).
Problem
The intrinsic: ``` int mask = _mm256_movemask_epi8(__m256i s1) ``` creates a mask, with its `32` bits corresponding to the most significant bit of each byte of `s1`. After manipulating the mask using bit operations (`BMI2` for example) I would like to perform the inverse of `_mm256_movemask_epi8`, i.e., create a `__m256i` vector with the most significant bit of each byte containing the corresponding bit of the `uint32_t mask`. What is the best way to do this? Edit: I need to perform the inverse because the intrinsic `_mm256_blendv_epi8` accepts only `__m256i` type mask instead of `uint32_t`. As such, in the resulting `__m256i` mask, I can ignore the bits other than the MSB of each byte.