PCMPGTW - Packed CoMPare Greater Than Word

PCMPGTW xmm1, xmm2/m128    (S2
__m128i _mm_cmpgt_epi16(__m128i a, __m128i b)

For each signed WORD, if (1) > (2) set 1, else set 0, to all bits in the corresponding WORD of (3).
VPCMPGTW xmm1, xmm2, xmm3/m128    (V1
__m128i _mm_cmpgt_epi16(__m128i a, __m128i b)

For each signed WORD, if (1) > (2) set 1, else set 0, to all bits in the corresponding WORD of (3).
VPCMPGTW ymm1, ymm2, ymm3/m256    (V2
__m256i _mm256_cmpgt_epi16(__m256i a, __m256i b)

For each signed WORD, if (1) > (2) set 1, else set 0, to all bits in the corresponding WORD of (3).
VPCMPGTW k1{k2}, xmm2, xmm3/m128    (V5+BW+VL
__mmask8 _mm_cmpgt_epi16_mask(__m128i a, __m128i b)
__mmask8 _mm_mask_cmpgt_epi16_mask(__mmask8 k2, __m128i a, __m128i b)

For each signed WORD, if (1) > (2) set 1, else set 0, to the corresponding bit of (3).
If k2 bit is 0, the comparison is not done and the corresponding bit of (3) is set to zero. Upper bits of (3) are zero cleared.
VPCMPGTW k1{k2}, ymm2, ymm3/m256    (V5+BW+VL
__mmask16 _mm256_cmpgt_epi16_mask(__m256i a, __m256i b)
__mmask16 _mm256_mask_cmpgt_epi16_mask(__mmask16 k2, __m256i a, __m256i b)

For each signed WORD, if (1) > (2) set 1, else set 0, to the corresponding bit of (3).
If k2 bit is 0, the comparison is not done and the corresponding bit of (3) is set to zero. Upper bits of (3) are zero cleared.
VPCMPGTW k1{k2}, zmm2, zmm3/m512    (V5+BW
__mmask32 _mm512_cmpgt_epi16_mask(__m512i a, __m512i b)
__mmask32 _mm512_mask_cmpgt_epi16_mask(__mmask32 k2, __m512i a, __m512i b)

For each signed WORD, if (1) > (2) set 1, else set 0, to the corresponding bit of (3).
If k2 bit is 0, the comparison is not done and the corresponding bit of (3) is set to zero. Upper bits of (3) are zero cleared.

x86/x64 SIMD Instruction List  Feedback