vfmadd213ph
Fused Multiply-Add (213) Packed FP16
VFMADD213PH zmm1 {k1}, zmm2, zmm3/m512
Computes (Src1 * Dest) + Src2 in half-precision.
Details
Fused multiply-add instruction computing (zmm2 × zmm1) + zmm3/m512 in FP16 precision with a single rounding operation. The operand order (213) indicates zmm2 is the first multiplier, zmm1 is the second, and zmm3/m512 is the addend. Requires AVX-512-FP16; no arithmetic flags are modified; results are written to zmm1 under EVEX mask control.
Pseudocode Operation
for i = 0 to 31:
if (k1[i] == 1):
zmm1[16*i:16*i+15] ← fma_fp16(zmm2[16*i:16*i+15], zmm1[16*i:16*i+15], zmm3/m512[16*i:16*i+15])
else if (EVEX.z == 1):
zmm1[16*i:16*i+15] ← 0
// else zmm1[16*i:16*i+15] unchanged
Example
VFMADD213PH zmm1, zmm2, zmm3/m512
Encoding
Binary Layout
EVEX
+0
A8
+4
Operands
-
dest
512-bit ZMM AVX-512 register -
src1
512-bit ZMM AVX-512 register -
src2
512-bit ZMM AVX-512 register or Memory operand
Reference (Intel® SDM)
Instruction Forms
| Opcode | Instruction | Op/En | 64/32-bit Mode | CPUID | Description |
|---|---|---|---|---|---|
| EVEX.128.66.MAP6.W0 98 /r | VFMADD132PH xmm1{k1}{z}, xmm2, xmm3/m128/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from xmm1 and xmm3/m128/m16bcst, add to xmm2, and store the result in xmm1. |
| EVEX.256.66.MAP6.W0 98 /r | VFMADD132PH ymm1{k1}{z}, ymm2, ymm3/m256/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from ymm1 and ymm3/m256/m16bcst, add to ymm2, and store the result in ymm1. |
| EVEX.512.66.MAP6.W0 98 /r | VFMADD132PH zmm1{k1}{z}, zmm2, zmm3/m512/m16bcst {er} | A | V/V | AVX512_FP16 OR AVX10.1 | Multiply packed FP16 values from zmm1 and zmm3/m512/m16bcst, add to zmm2, and store the result in zmm1. |
| EVEX.128.66.MAP6.W0 A8 /r | VFMADD213PH xmm1{k1}{z}, xmm2, xmm3/m128/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from xmm1 and xmm2, add to xmm3/m128/m16bcst, and store the result in xmm1. |
| EVEX.256.66.MAP6.W0 A8 /r | VFMADD213PH ymm1{k1}{z}, ymm2, ymm3/m256/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from ymm1 and ymm2, add to ymm3/m256/m16bcst, and store the result in ymm1. |
| EVEX.512.66.MAP6.W0 A8 /r | VFMADD213PH zmm1{k1}{z}, zmm2, zmm3/m512/m16bcst {er} | A | V/V | AVX512_FP16 OR AVX10.1 | Multiply packed FP16 values from zmm1 and zmm2, add to zmm3/m512/m16bcst, and store the result in zmm1. |
| EVEX.128.66.MAP6.W0 B8 /r | VFMADD231PH xmm1{k1}{z}, xmm2, xmm3/m128/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from xmm2 and xmm3/m128/m16bcst, add to xmm1, and store the result in xmm1. |
| EVEX.256.66.MAP6.W0 B8 /r | VFMADD231PH ymm1{k1}{z}, ymm2, ymm3/m256/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from ymm2 and ymm3/m256/m16bcst, add to ymm1, and store the result in ymm1. |
| EVEX.512.66.MAP6.W0 B8 /r | VFMADD231PH zmm1{k1}{z}, zmm2, zmm3/m512/m16bcst {er} | A | V/V | AVX512_FP16 OR AVX10.1 | Multiply packed FP16 values from zmm2 and zmm3/m512/m16bcst, add to zmm1, and store the result in zmm1. |
| EVEX.128.66.MAP6.W0 9C /r | VFNMADD132PH xmm1{k1}{z}, xmm2, xmm3/m128/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from xmm1 and xmm3/m128/m16bcst, and negate the value. Add this value to xmm2, and store the result in xmm1. |
| EVEX.256.66.MAP6.W0 9C /r | VFNMADD132PH ymm1{k1}{z}, ymm2, ymm3/m256/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from ymm1 and ymm3/m256/m16bcst, and negate the value. Add this value to ymm2, and store the result in ymm1. |
| EVEX.512.66.MAP6.W0 9C /r | VFNMADD132PH zmm1{k1}{z}, zmm2, zmm3/m512/m16bcst {er} | A | V/V | AVX512_FP16 OR AVX10.1 | Multiply packed FP16 values from zmm1 and zmm3/m512/m16bcst, and negate the value. Add this value to zmm2, and store the result in zmm1. |
| EVEX.128.66.MAP6.W0 AC /r | VFNMADD213PH xmm1{k1}{z}, xmm2, xmm3/m128/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from xmm1 and xmm2, and negate the value. Add this value to xmm3/m128/m16bcst, and store the result in xmm1. |
| EVEX.256.66.MAP6.W0 AC /r | VFNMADD213PH ymm1{k1}{z}, ymm2, ymm3/m256/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from ymm1 and ymm2, and negate the value. Add this value to ymm3/m256/m16bcst, and store the result in ymm1. VF[,N]MADD[132,213,231]PH—Fused Multiply-Add of Packed FP16 Values Vol. 2C 5-214 Opcode/ Op/ 64/32 CPUID Feature Description Instruction En Bit Mode Flag Support |
| EVEX.512.66.MAP6.W0 AC /r | VFNMADD213PH zmm1{k1}{z}, zmm2, zmm3/m512/m16bcst {er} | A | V/V | AVX512_FP16 OR AVX10.1 | Multiply packed FP16 values from zmm1 and zmm2, and negate the value. Add this value to zmm3/m512/m16bcst, and store the result in zmm1. |
| EVEX.128.66.MAP6.W0 BC /r | VFNMADD231PH xmm1{k1}{z}, xmm2, xmm3/m128/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from xmm2 and xmm3/m128/m16bcst, and negate the value. Add this value to xmm1, and store the result in xmm1. |
| EVEX.256.66.MAP6.W0 BC /r | VFNMADD231PH ymm1{k1}{z}, ymm2, ymm3/m256/m16bcst | A | V/V | (AVX512_FP16 AND AVX512VL) OR AVX10.1 | Multiply packed FP16 values from ymm2 and ymm3/m256/m16bcst, and negate the value. Add this value to ymm1, and store the result in ymm1. |
| EVEX.512.66.MAP6.W0 BC /r | VFNMADD231PH zmm1{k1}{z}, zmm2, zmm3/m512/m16bcst {er} | A | V/V | AVX512_FP16 OR AVX10.1 | Multiply packed FP16 values from zmm2 and zmm3/m512/m16bcst, and negate the value. Add this value to zmm1, and store the result in zmm1. |
Instruction Operand Encoding
| Op/En | Tuple Type | Operand 1 | Operand 2 | Operand 3 | Operand 4 |
|---|---|---|---|---|---|
| A | Full | ModRM:reg (r, w) | VEX.vvvv (r) | ModRM:r/m (r) | N/A |
Description
This instruction performs a packed multiply-add or negated multiply-add computation on FP16 values using three source operands and writes the results in the destination operand. The destination operand is also the first source operand. The “N” (negated) forms of this instruction add the negated infinite precision intermediate product to the corresponding remaining operand. The notation’ “132”, “213” and “231” indicate the use of the operands in ±A * B
+ C, where each digit corresponds to the operand number, with the destination being operand 1; see Table 5-5.
The destination elements are updated according to the writemask.
Table 5-5. VF[,N]MADD[132,213,231]PH Notation for Operands
Notation Operands
132 dest = ± dest*src3+src2
231 dest = ± src2*src3+dest
213 dest = ± src2*dest+src3
VF[,N]MADD[132,213,231]PH—Fused Multiply-Add of Packed FP16 Values Vol. 2C 5-215
Operation
VF[,N]MADD132PH DEST, SRC2, SRC3 (EVEX encoded versions) when src3 operand is a register VL = 128, 256 or 512 KL := VL/16 IF (VL = 512) AND (EVEX.b = 1): SET_RM(EVEX.RC) ELSE SET_RM(MXCSR.RC) FOR j := 0 TO KL-1: IF k1[j] OR *no writemask*: IF *negative form*: DEST.fp16[j] := RoundFPControl(-DEST.fp16[j]*SRC3.fp16[j] + SRC2.fp16[j]) ELSE: DEST.fp16[j] := RoundFPControl(DEST.fp16[j]*SRC3.fp16[j] + SRC2.fp16[j]) ELSE IF *zeroing*: DEST.fp16[j] := 0 // else dest.fp16[j] remains unchanged DEST[MAXVL-1:VL] := 0 VF[,N]MADD132PH DEST, SRC2, SRC3 (EVEX encoded versions) when src3 operand is a memory source VL = 128, 256 or 512 KL := VL/16 FOR j := 0 TO KL-1: IF k1[j] OR *no writemask*: IF EVEX.b = 1: t3 := SRC3.fp16[0] ELSE: t3 := SRC3.fp16[j] IF *negative form*: DEST.fp16[j] := RoundFPControl(-DEST.fp16[j] * t3 + SRC2.fp16[j]) ELSE: DEST.fp16[j] := RoundFPControl(DEST.fp16[j] * t3 + SRC2.fp16[j]) ELSE IF *zeroing*: DEST.fp16[j] := 0 // else dest.fp16[j] remains unchanged DEST[MAXVL-1:VL] := 0 VF[,N]MADD[132,213,231]PH—Fused Multiply-Add of Packed FP16 Values Vol. 2C 5-216 VF[,N]MADD213PH DEST, SRC2, SRC3 (EVEX encoded versions) when src3 operand is a register VL = 128, 256 or 512 KL := VL/16 IF (VL = 512) AND (EVEX.b = 1): SET_RM(EVEX.RC) ELSE SET_RM(MXCSR.RC) FOR j := 0 TO KL-1: IF k1[j] OR *no writemask*: IF *negative form*: DEST.fp16[j] := RoundFPControl(-SRC2.fp16[j]*DEST.fp16[j] + SRC3.fp16[j]) ELSE DEST.fp16[j] := RoundFPControl(SRC2.fp16[j]*DEST.fp16[j] + SRC3.fp16[j]) ELSE IF *zeroing*: DEST.fp16[j] := 0 // else dest.fp16[j] remains unchanged DEST[MAXVL-1:VL] := 0 VF[,N]MADD213PH DEST, SRC2, SRC3 (EVEX encoded versions) when src3 operand is a memory source VL = 128, 256 or 512 KL := VL/16 FOR j := 0 TO KL-1: IF k1[j] OR *no writemask*: IF EVEX.b = 1: t3 := SRC3.fp16[0] ELSE: t3 := SRC3.fp16[j] IF *negative form*: DEST.fp16[j] := RoundFPControl(-SRC2.fp16[j] * DEST.fp16[j] + t3 ) ELSE: DEST.fp16[j] := RoundFPControl(SRC2.fp16[j] * DEST.fp16[j] + t3 ) ELSE IF *zeroing*: DEST.fp16[j] := 0 // else dest.fp16[j] remains unchanged DEST[MAXVL-1:VL] := 0 VF[,N]MADD[132,213,231]PH—Fused Multiply-Add of Packed FP16 Values Vol. 2C 5-217 VF[,N]MADD231PH DEST, SRC2, SRC3 (EVEX encoded versions) when src3 operand is a register VL = 128, 256 or 512 KL := VL/16 IF (VL = 512) AND (EVEX.b = 1): SET_RM(EVEX.RC) ELSE SET_RM(MXCSR.RC) FOR j := 0 TO KL-1: IF k1[j] OR *no writemask*: IF *negative form: DEST.fp16[j] := RoundFPControl(-SRC2.fp16[j]*SRC3.fp16[j] + DEST.fp16[j]) ELSE: DEST.fp16[j] := RoundFPControl(SRC2.fp16[j]*SRC3.fp16[j] + DEST.fp16[j]) ELSE IF *zeroing*: DEST.fp16[j] := 0 // else dest.fp16[j] remains unchanged DEST[MAXVL-1:VL] := 0 VF[,N]MADD231PH DEST, SRC2, SRC3 (EVEX encoded versions) when src3 operand is a memory source VL = 128, 256 or 512 KL := VL/16 FOR j := 0 TO KL-1: IF k1[j] OR *no writemask*: IF EVEX.b = 1: t3 := SRC3.fp16[0] ELSE: t3 := SRC3.fp16[j] IF *negative form*: DEST.fp16[j] := RoundFPControl(-SRC2.fp16[j] * t3 + DEST.fp16[j] ) ELSE: DEST.fp16[j] := RoundFPControl(SRC2.fp16[j] * t3 + DEST.fp16[j] ) ELSE IF *zeroing*: DEST.fp16[j] := 0 // else dest.fp16[j] remains unchanged DEST[MAXVL-1:VL] := 0 VF[,N]MADD[132,213,231]PH—Fused Multiply-Add of Packed FP16 Values Vol. 2C 5-218
Intel C/C++ Compiler Intrinsic Equivalent
VFMADD132PH, VFMADD213PH , and VFMADD231PH: __m128h _mm_fmadd_ph (__m128h a, __m128h b, __m128h c); __m128h _mm_mask_fmadd_ph (__m128h a, __mmask8 k, __m128h b, __m128h c); __m128h _mm_mask3_fmadd_ph (__m128h a, __m128h b, __m128h c, __mmask8 k); __m128h _mm_maskz_fmadd_ph (__mmask8 k, __m128h a, __m128h b, __m128h c); __m256h _mm256_fmadd_ph (__m256h a, __m256h b, __m256h c); __m256h _mm256_mask_fmadd_ph (__m256h a, __mmask16 k, __m256h b, __m256h c); __m256h _mm256_mask3_fmadd_ph (__m256h a, __m256h b, __m256h c, __mmask16 k); __m256h _mm256_maskz_fmadd_ph (__mmask16 k, __m256h a, __m256h b, __m256h c); __m512h _mm512_fmadd_ph (__m512h a, __m512h b, __m512h c); __m512h _mm512_mask_fmadd_ph (__m512h a, __mmask32 k, __m512h b, __m512h c); __m512h _mm512_mask3_fmadd_ph (__m512h a, __m512h b, __m512h c, __mmask32 k); __m512h _mm512_maskz_fmadd_ph (__mmask32 k, __m512h a, __m512h b, __m512h c); __m512h _mm512_fmadd_round_ph (__m512h a, __m512h b, __m512h c, const int rounding); __m512h _mm512_mask_fmadd_round_ph (__m512h a, __mmask32 k, __m512h b, __m512h c, const int rounding); __m512h _mm512_mask3_fmadd_round_ph (__m512h a, __m512h b, __m512h c, __mmask32 k, const int rounding); __m512h _mm512_maskz_fmadd_round_ph (__mmask32 k, __m512h a, __m512h b, __m512h c, const int rounding); VFNMADD132PH, VFNMADD213PH, and VFNMADD231PH: __m128h _mm_fnmadd_ph (__m128h a, __m128h b, __m128h c); __m128h _mm_mask_fnmadd_ph (__m128h a, __mmask8 k, __m128h b, __m128h c); __m128h _mm_mask3_fnmadd_ph (__m128h a, __m128h b, __m128h c, __mmask8 k); __m128h _mm_maskz_fnmadd_ph (__mmask8 k, __m128h a, __m128h b, __m128h c); __m256h _mm256_fnmadd_ph (__m256h a, __m256h b, __m256h c); __m256h _mm256_mask_fnmadd_ph (__m256h a, __mmask16 k, __m256h b, __m256h c); __m256h _mm256_mask3_fnmadd_ph (__m256h a, __m256h b, __m256h c, __mmask16 k); __m256h _mm256_maskz_fnmadd_ph (__mmask16 k, __m256h a, __m256h b, __m256h c); __m512h _mm512_fnmadd_ph (__m512h a, __m512h b, __m512h c); __m512h _mm512_mask_fnmadd_ph (__m512h a, __mmask32 k, __m512h b, __m512h c); __m512h _mm512_mask3_fnmadd_ph (__m512h a, __m512h b, __m512h c, __mmask32 k); __m512h _mm512_maskz_fnmadd_ph (__mmask32 k, __m512h a, __m512h b, __m512h c); __m512h _mm512_fnmadd_round_ph (__m512h a, __m512h b, __m512h c, const int rounding); __m512h _mm512_mask_fnmadd_round_ph (__m512h a, __mmask32 k, __m512h b, __m512h c, const int rounding); __m512h _mm512_mask3_fnmadd_round_ph (__m512h a, __m512h b, __m512h c, __mmask32 k, const int rounding); __m512h _mm512_maskz_fnmadd_round_ph (__mmask32 k, __m512h a, __m512h b, __m512h c, const int rounding);
Exceptions
SIMD Floating-Point Exceptions
Invalid, Underflow, Overflow, Precision, Denormal
Other Exceptions
EVEX-encoded instructions, see Table 2-48, “Type E2 Class Exception Conditions.”
VF[,N]MADD[132,213,231]PH—Fused Multiply-Add of Packed FP16 Values Vol. 2C 5-219