-
Notifications
You must be signed in to change notification settings - Fork 97
PSHUFD
PSHUFD — Shuffle Packed Doublewords
| Opcode/ Instruction | Op/ En | 64/32 bit Mode Support | CPUID Feature Flag | Description |
| 66 0F 70 /r ib PSHUFD xmm1, xmm2/m128, imm8 | A | V/V | SSE2 | Shuffle the doublewords in xmm2/m128 based on the encoding in imm8 and store the result in xmm1. |
| VEX.128.66.0F.WIG 70 /r ib VPSHUFD xmm1, xmm2/m128, imm8 | A | V/V | AVX | Shuffle the doublewords in xmm2/m128 based on the encoding in imm8 and store the result in xmm1. |
| VEX.256.66.0F.WIG 70 /r ib VPSHUFD ymm1, ymm2/m256, imm8 | A | V/V | AVX2 | Shuffle the doublewords in ymm2/m256 based on the encoding in imm8 and store the result in ymm1. |
| EVEX.128.66.0F.W0 70 /r ib VPSHUFD xmm1 {k1}{z}, xmm2/m128/m32bcst, imm8 | B | V/V | AVX512VL AVX512F | Shuffle the doublewords in xmm2/m128/m32bcst based on the encoding in imm8 and store the result in xmm1 using writemask k1. |
| EVEX.256.66.0F.W0 70 /r ib VPSHUFD ymm1 {k1}{z}, ymm2/m256/m32bcst, imm8 | B | V/V | AVX512VL AVX512F | Shuffle the doublewords in ymm2/m256/m32bcst based on the encoding in imm8 and store the result in ymm1 using writemask k1. |
| EVEX.512.66.0F.W0 70 /r ib VPSHUFD zmm1 {k1}{z}, zmm2/m512/m32bcst, imm8 | B | V/V | AVX512F | Shuffle the doublewords in zmm2/m512/m32bcst based on the encoding in imm8 and store the result in zmm1 using writemask k1. |
| Op/En | Tuple Type | Operand 1 | Operand 2 | Operand 3 | Operand 4 |
| A | NA | ModRM:reg (w) | ModRM:r/m (r) | imm8 | NA |
| B | Full Vector | ModRM:reg (w) | ModRM:r/m (r) | Imm8 | NA |
Copies doublewords from source operand (second operand) and inserts them in the destination operand (first operand) at the locations selected with the order operand (third operand). Figure 4-16 shows the operation of the 256-bit VPSHUFD instruction and the encoding of the order operand. Each 2-bit field in the order operand selects the contents of one doubleword location within a 128-bit lane and copy to the target element in the destination operand. For example, bits 0 and 1 of the order operand targets the first doubleword element in the low and high 128-bit lane of the destination operand for 256-bit VPSHUFD. The encoded value of bits 1:0 of the order operand (see the field encoding in Figure 4-16) determines which doubleword element (from the respective 128-bit lane) of the source operand will be copied to doubleword 0 of the destination operand.
For 128-bit operation, only the low 128-bit lane are operative. The source operand can be an XMM register or a 128-bit memory location. The destination operand is an XMM register. The order operand is an 8-bit immediate. Note that this instruction permits a doubleword in the source operand to be copied to more than one doubleword location in the destination operand.
| X7 | X6 | X5 | X4 | X3 | X2 | X1 | X0 |
| Y7 | Y6 | Y5 | Y4 | Y3 | Y2 | Y1 | Y0 |
11B - X3 Operand
Figure 4-16. 256-bit VPSHUFD Instruction Operation
The source operand can be an XMM register or a 128-bit memory location. The destination operand is an XMM register. The order operand is an 8-bit immediate. Note that this instruction permits a doubleword in the source operand to be copied to more than one doubleword location in the destination operand.
In 64-bit mode and not encoded in VEX/EVEX, using REX.R permits this instruction to access XMM8-XMM15.
128-bit Legacy SSE version: Bits (MAXVL-1:128) of the corresponding YMM destination register remain unchanged.
VEX.128 encoded version: The source operand can be an XMM register or a 128-bit memory location. The destina- tion operand is an XMM register. Bits (MAXVL-1:128) of the corresponding ZMM register are zeroed.
VEX.256 encoded version: The source operand can be an YMM register or a 256-bit memory location. The destina- tion operand is an YMM register. Bits (MAXVL-1:256) of the corresponding ZMM register are zeroed. Bits (255- 1:128) of the destination stores the shuffled results of the upper 16 bytes of the source operand using the imme- diate byte as the order operand.
EVEX encoded version: The source operand can be an ZMM/YMM/XMM register, a 512/256/128-bit memory loca- tion, or a 512/256/128-bit vector broadcasted from a 32-bit memory location. The destination operand is a ZMM/YMM/XMM register updated according to the writemask.
Each 128-bit lane of the destination stores the shuffled results of the respective lane of the source operand using the immediate byte as the order operand.
Note: EVEX.vvvv and VEX.vvvv are reserved and must be 1111b otherwise instructions will #UD.
DEST[31:0] ← (SRC >> (ORDER[1:0] * 32))[31:0];
DEST[63:32] ← (SRC >> (ORDER[3:2] * 32))[31:0];
DEST[95:64] ← (SRC >> (ORDER[5:4] * 32))[31:0];
DEST[127:96] ← (SRC >> (ORDER[7:6] * 32))[31:0];
DEST[MAXVL-1:128] (Unmodified)DEST[31:0] ← (SRC >> (ORDER[1:0] * 32))[31:0];
DEST[63:32] ← (SRC >> (ORDER[3:2] * 32))[31:0];
DEST[95:64] ← (SRC >> (ORDER[5:4] * 32))[31:0];
DEST[127:96] ← (SRC >> (ORDER[7:6] * 32))[31:0];
DEST[MAXVL-1:128] ← 0DEST[31:0] ← (SRC[127:0] >> (ORDER[1:0] * 32))[31:0];
DEST[63:32] ← (SRC[127:0] >> (ORDER[3:2] * 32))[31:0];
DEST[95:64] ← (SRC[127:0] >> (ORDER[5:4] * 32))[31:0];
DEST[127:96] ← (SRC[127:0] >> (ORDER[7:6] * 32))[31:0];
DEST[159:128] ← (SRC[255:128] >> (ORDER[1:0] * 32))[31:0];
DEST[191:160] ← (SRC[255:128] >> (ORDER[3:2] * 32))[31:0];
DEST[223:192] ← (SRC[255:128] >> (ORDER[5:4] * 32))[31:0];
DEST[255:224] ← (SRC[255:128] >> (ORDER[7:6] * 32))[31:0];
DEST[MAXVL-1:256] ← 0(KL, VL) = (4, 128), (8, 256), (16, 512)
FOR j ← 0 TO KL-1
i ← j * 32
IF (EVEX.b = 1) AND (SRC *is memory*)
THEN TMP_SRC[i+31:i] ← SRC[31:0]
ELSE TMP_SRC[i+31:i] ← SRC[i+31:i]
FI;
ENDFOR;
IF VL >= 128
TMP_DEST[31:0] ← (TMP_SRC[127:0] >> (ORDER[1:0] * 32))[31:0];
TMP_DEST[63:32] ← (TMP_SRC[127:0] >> (ORDER[3:2] * 32))[31:0];
TMP_DEST[95:64] ← (TMP_SRC[127:0] >> (ORDER[5:4] * 32))[31:0];
TMP_DEST[127:96] ← (TMP_SRC[127:0] >> (ORDER[7:6] * 32))[31:0];
FI;
IF VL >= 256
TMP_DEST[159:128] ← (TMP_SRC[255:128] >> (ORDER[1:0] * 32))[31:0];
TMP_DEST[191:160] ← (TMP_SRC[255:128] >> (ORDER[3:2] * 32))[31:0];
TMP_DEST[223:192] ← (TMP_SRC[255:128] >> (ORDER[5:4] * 32))[31:0];
TMP_DEST[255:224] ← (TMP_SRC[255:128] >> (ORDER[7:6] * 32))[31:0];
FI;
IF VL >= 512
TMP_DEST[287:256] ← (TMP_SRC[383:256] >> (ORDER[1:0] * 32))[31:0];
TMP_DEST[319:288] ← (TMP_SRC[383:256] >> (ORDER[3:2] * 32))[31:0];
TMP_DEST[351:320] ← (TMP_SRC[383:256] >> (ORDER[5:4] * 32))[31:0];
TMP_DEST[383:352] ← (TMP_SRC[383:256] >> (ORDER[7:6] * 32))[31:0];
TMP_DEST[415:384] ← (TMP_SRC[511:384] >> (ORDER[1:0] * 32))[31:0];
TMP_DEST[447:416] ← (TMP_SRC[511:384] >> (ORDER[3:2] * 32))[31:0];
TMP_DEST[479:448] ←(TMP_SRC[511:384] >> (ORDER[5:4] * 32))[31:0];
TMP_DEST[511:480] ← (TMP_SRC[511:384] >> (ORDER[7:6] * 32))[31:0];
FI;
FOR j ← 0 TO KL-1
i ← j * 32
IF k1[j] OR *no writemask*
THEN DEST[i+31:i] ← TMP_DEST[i+31:i]
ELSE
IF *merging-masking*
; merging-masking
THEN *DEST[i+31:i] remains unchanged*
ELSE *zeroing-masking*
; zeroing-masking
DEST[i+31:i] ← 0
FI
FI;
ENDFOR
DEST[MAXVL-1:VL] ← 0 VPSHUFD __m512i _mm512_shuffle_epi32(__m512i a, int n );
VPSHUFD __m512i _mm512_mask_shuffle_epi32(__m512i s, __mmask16 k, __m512i a, int n );
VPSHUFD __m512i _mm512_maskz_shuffle_epi32( __mmask16 k, __m512i a, int n );
VPSHUFD __m256i _mm256_mask_shuffle_epi32(__m256i s, __mmask8 k, __m256i a, int n );
VPSHUFD __m256i _mm256_maskz_shuffle_epi32( __mmask8 k, __m256i a, int n );
VPSHUFD __m128i _mm_mask_shuffle_epi32(__m128i s, __mmask8 k, __m128i a, int n );
VPSHUFD __m128i _mm_maskz_shuffle_epi32( __mmask8 k, __m128i a, int n );
(V)PSHUFD:__m128i _mm_shuffle_epi32(__m128i a, int n)
VPSHUFD:__m256i _mm256_shuffle_epi32(__m256i a, const int n)None.
None.
Non-EVEX-encoded instruction, see Exceptions Type 4. EVEX-encoded instruction, see Exceptions Type E4NF.
\#UD If VEX.vvvv ≠ 1111B or EVEX.vvvv ≠ 1111B.
Source: Intel Architecture Software Developer's Manual (July 2017)