VRNDSCALEPH - RouND SCALE Packed Half
VRNDSCALEPH xmm1{k1}{z}, xmm2/m128/m16bcst, imm8 (V5+FP16+VL
__m128h _mm_roundscale_ph(__m128h a, int imm8)
__m128h _mm_mask_roundscale_ph(__m128h s, __mmask8 k, __m128h a, int imm8)
__m128h _mm_maskz_roundscale_ph(__mmask8 k, __m128h a, int imm8)
For each FP16, round (1) to n-bit below the decimal point and store the resulting FP16 in (2).
VRNDSCALEPH ymm1{k1}{z}, ymm2/m256/m16bcst, imm8 (V5+FP16+VL
__m256h _mm256_roundscale_ph(__m256h a, int imm8)
__m256h _mm256_mask_roundscale_ph(__m256h s, __mmask16 k, __m256h a, int imm8)
__m256h _mm256_maskz_roundscale_ph(__mmask16 k, __m256h a, int imm8)
For each FP16, round (1) to n-bit below the decimal point and store the resulting FP16 in (2).
VRNDSCALEPH zmm1{k1}{z}, zmm2/m512/m16bcst{sae}, imm8 (V5+FP16
__m512h _mm512_roundscale_ph(__m512h a, int imm8)
__m512h _mm512_mask_roundscale_ph(__m512h s, __mmask32 k, __m512h a, int imm8)
__m512h _mm512_maskz_roundscale_ph(__mmask32 k, __m512h a, int imm8)
__m512h _mm512_roundscale_round_ph(__m512h a, int imm8, int sae)
__m512h _mm512_mask_roundscale_round_ph(__m512h s, __mmask32 k, __m512h a, int imm8, int sae)
__m512h _mm512_maskz_roundscale_round_ph(__mmask32 k, __m512h a, int imm8, int sae)
For each FP16, round (1) to n-bit below the decimal point and store the resulting FP16 in (2).
imm8
| bit |
|
| 7:4 |
Bits to preserve below decimal point (0 to 15).
Set 0 in this field to round to decimal point (like ROUNDPD / ROUNDPS).
|
| 3 |
0: precision exception mask is specified by MXCSR
1: precision exception is masked
|
| 2 |
0: rounding mode is specified by bit 1:0
1: rounding mode is specified by MXCSR
|
| 1:0 |
00: round to nearest or even
01: round toward negative infinity
10: round toward positive inifinity
11: round toward zero
|
x86/x64 SIMD Instruction List
Feedback