Best way to mask a single bit in AVX2? - c

Question

For example, with an input ymm vector x and bit index i I want an output vector with only the ith bit kept and everything else zeroed.
With AVX512 k registers, I could write the following, but AVX2 and below doesn't have k registers, so what do you think is the best way to do it?
__m512i m512i_maskBit(__m512i x, unsigned i) {
__mmask8 m = _cvtu32_mask8(1u << i / 64);
__m512i vm = _mm512_maskz_set1_epi64(m, 1ull << i % 64);
return _mm512_and_si512(x, vm);
}

Answer 1

Here is an approach using variable shifts (just creating the mask):
__m256i create_mask(unsigned i) {
__m256i ii = _mm256_set1_epi32(i);
ii = _mm256_sub_epi32(ii,_mm256_setr_epi32(0,32,64,96,128,160,192,224));
__m256i mask = _mm256_sllv_epi32(_mm256_set1_epi32(1), ii);
return mask;
}
_mm256_sllv_epi32 (vpsllvd) was introduced by AVX2 and it shifts each 32 bit element by a variable amount of bits. If the (unsigned) shift-amount is bigger than 31 (i.e., also for signed negative numbers), the corresponding result is 0.
Godbolt link with small test code: https://godbolt.org/z/a5xfqTcGs

Answer 2

How about the simplest approach:
__m256i m256i_create_mask(unsigned i) {
// Get the required bit in every byte of the vector
__m256i vm = _mm256_broadcastb_epi8(_mm_cvtsi32_si128(1u << (i & 7u)));
// Mask off the bytes that are outside the index
__m256i vi = _mm256_broadcastb_epi8(_mm_cvtsi32_si128(i >> 3u));
__m256i vm1 = _mm256_cmpeq_epi8(vi,
_mm256_setr_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31));
return _mm256_and_si256(vm, vm1);
}

Answer 3

Here’s another approach. Not sure it’s necessarily better, it depends on CPU model and surrounding code, but it might be.
// A buffer to load vectors with a single bit set in one lane
alignas( 64 ) static const std::array<int, 16> s_oneBuffer =
{
0, 0, 0, 0, 0, 0, 0, 0,
1, 0, 0, 0, 0, 0, 0, 0
};
__m256i maskSingleBit( __m256i x, uint32_t bitIndex )
{
// Load `1` into a single 32-bit lane of the vector
// The buffer aligned by 64 bytes, contained in a single cache line, no unaligned load penalty.
__m256i one = _mm256_loadu_si256( ( const __m256i* )( ( s_oneBuffer.data() + 8 ) - ( bitIndex / 32 ) ) );
// Left shift to move the `1` into the correct location
__m128i shift = _mm_cvtsi32_si128( bitIndex % 32 );
__m256i bit = _mm256_sll_epi32( one, shift );
// Bitwise AND with the value
return _mm256_and_si256( x, bit );
}