VRCPPH - ReCiProcal Packed Half
VRCPPH xmm1{k1}{z}, xmm2/m128/m16bcst (V5+FP16+VL
__m128h _mm_rcp_ph(__m128h a)
__m128h _mm_mask_rcp_ph(__m128h s, __mmask8 k, __m128h a)
__m128h _mm_maskz_rcp_ph(__mmask8 k, __m128h a)
For each FP16, calculate approximate reciprocal of (1) and store the result in (2). (relative error < 2-11 + 2-14)
VRCPPH ymm1{k1}{z}, ymm2/m256/m16bcst (V5+FP16+VL
__m256h _mm256_rcp_ph(__m256h a)
__m256h _mm256_mask_rcp_ph(__m256h s, __mmask16 k, __m256h a)
__m256h _mm256_maskz_rcp_ph(__mmask16 k, __m256h a)
For each FP16, calculate approximate reciprocal of (1) and store the result in (2). (relative error < 2-11 + 2-14)
VRCPPH zmm1{k1}{z}, zmm2/m512/m16bcst (V5+FP16
__m512h _mm512_rcp_ph(__m512h a)
__m512h _mm512_mask_rcp_ph(__m512h s, __mmask32 k, __m512h a)
__m512h _mm512_maskz_rcp_ph(__mmask32 k, __m512h a)
For each FP16, calculate approximate reciprocal of (1) and store the result in (2). (relative error < 2-11 + 2-14)
x86/x64 SIMD Instruction List
Feedback