vpshrdd

Packed Shift Right Double Concatenate

VPSHRDD zmm1 {k1}, zmm2, zmm3/m512, imm8

Funnel shift right of doublewords.

Details

Performs a funnel right-shift (double concatenate) on 32-bit doublewords: each output doubleword is the logical right-shift result of concatenating src1 and src2, shifted right by the immediate count. Bits shifted out on the right are discarded; bits from src1 fill in on the left. No arithmetic flags are modified; masking via k1 applies to the result.

Pseudocode Operation

for i = 0 to 15:  // 16 doublewords
  concatenated = (zmm2[32*i+31:32*i] << 32) | zmm3[32*i+31:32*i]
  shift_amount = cnt & 0x1F
  zmm1[32*i+31:32*i] ← (concatenated >> shift_amount) & 0xFFFFFFFF
  zmm1[32*i+31:32*i] ← zmm1[32*i+31:32*i] & k1_mask[i]

Example

VPSHRDD zmm1, zmm2, zmm3/m512, 3

Encoding

Binary Layout

EVEX

Format EVEX

Opcode EVEX.512.66.0F3A.W0 73 /r /ib

Extension AVX-512-VBMI2

Operands

dest
ZMM
src1
ZMM
src2
ZMM/Mem
cnt
Imm

Reference (Intel® SDM)

Instruction Forms

Opcode	Instruction	Op/En	64/32-bit Mode	CPUID	Description
EVEX.128.66.0F3A.W1 72 /r /ib	VPSHRDW xmm1{k1}{z}, xmm2, xmm3/m128, imm8	A	V/V	(AVX512_VBMI2 AND AVX512VL) OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into xmm1.
EVEX.256.66.0F3A.W1 72 /r /ib	VPSHRDW ymm1{k1}{z}, ymm2, ymm3/m256, imm8	A	V/V	(AVX512_VBMI2 AND AVX512VL) OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into ymm1.
EVEX.512.66.0F3A.W1 72 /r /ib	VPSHRDW zmm1{k1}{z}, zmm2, zmm3/m512, imm8	A	V/V	AVX512_VBMI2 OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into zmm1.
EVEX.128.66.0F3A.W0 73 /r /ib	VPSHRDD xmm1{k1}{z}, xmm2, xmm3/m128/m32bcst, imm8	B	V/V	(AVX512_VBMI2 AND AVX512VL) OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into xmm1.
EVEX.256.66.0F3A.W0 73 /r /ib	VPSHRDD ymm1{k1}{z}, ymm2, ymm3/m256/m32bcst, imm8	B	V/V	(AVX512_VBMI2 AND AVX512VL) OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into ymm1.
EVEX.512.66.0F3A.W0 73 /r /ib	VPSHRDD zmm1{k1}{z}, zmm2, zmm3/m512/m32bcst, imm8	B	V/V	AVX512_VBMI2 OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into zmm1.
EVEX.128.66.0F3A.W1 73 /r /ib	VPSHRDQ xmm1{k1}{z}, xmm2, xmm3/m128/m64bcst, imm8	B	V/V	(AVX512_VBMI2 AND AVX512VL) OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into xmm1.
EVEX.256.66.0F3A.W1 73 /r /ib	VPSHRDQ ymm1{k1}{z}, ymm2, ymm3/m256/m64bcst, imm8	B	V/V	(AVX512_VBMI2 AND AVX512VL) OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into ymm1.
EVEX.512.66.0F3A.W1 73 /r /ib	VPSHRDQ zmm1{k1}{z}, zmm2, zmm3/m512/m64bcst, imm8	B	V/V	AVX512_VBMI2 OR AVX10.1	Concatenate destination and source operands, extract result shifted to the right by constant value in imm8 into zmm1.

Instruction Operand Encoding

Op/En	Tuple Type	Operand 1	Operand 2	Operand 3	Operand 4
A	Full Mem	ModRM:reg (w)	EVEX.vvvv (r)	ModRM:r/m (r)	imm8 (r)
B	Full	ModRM:reg (w)	EVEX.vvvv (r)	ModRM:r/m (r)	imm8 (r)

Description

Concatenate packed data, extract result shifted to the right by constant value. This instruction supports memory fault suppression. VPSHRD—Concatenate and Shift Packed Data Right Logical Vol. 2C 5-628

Operation

VPSHRDW DEST, SRC2, SRC3, imm8
(KL, VL) = (8, 128), (16, 256), (32, 512)
FOR j := 0 TO KL-1:
IF MaskBit(j) OR *no writemask*:
DEST.word[j] := concat(SRC3.word[j], SRC2.word[j]) >> (imm8 & 15)
ELSE IF *zeroing*:
DEST.word[j] := 0
*ELSE DEST.word[j] remains unchanged*
DEST[MAX_VL-1:VL] := 0

VPSHRDD DEST, SRC2, SRC3, imm8
(KL, VL) = (4, 128), (8, 256), (16, 512)
FOR j := 0 TO KL-1:
IF SRC3 is broadcast memop:
tsrc3 := SRC3.dword[0]
ELSE:
tsrc3 := SRC3.dword[j]
IF MaskBit(j) OR *no writemask*:
DEST.dword[j] := concat(tsrc3, SRC2.dword[j]) >> (imm8 & 31)
ELSE IF *zeroing*:
DEST.dword[j] := 0
*ELSE DEST.dword[j] remains unchanged*
DEST[MAX_VL-1:VL] := 0

VPSHRDQ DEST, SRC2, SRC3, imm8
(KL, VL) = (2, 128), (4, 256), (8, 512)
FOR j := 0 TO KL-1:
IF SRC3 is broadcast memop:
tsrc3 := SRC3.qword[0]
ELSE:
tsrc3 := SRC3.qword[j]
IF MaskBit(j) OR *no writemask*:
DEST.qword[j] := concat(tsrc3, SRC2.qword[j]) >> (imm8 & 63)
ELSE IF *zeroing*:
DEST.qword[j] := 0
*ELSE DEST.qword[j] remains unchanged*
DEST[MAX_VL-1:VL] := 0





VPSHRD—Concatenate and Shift Packed Data Right Logical                                                                                     Vol. 2C 5-629

Intel C/C++ Compiler Intrinsic Equivalent

VPSHRDQ __m128i _mm_shrdi_epi64(__m128i, __m128i, int);
VPSHRDQ __m128i _mm_mask_shrdi_epi64(__m128i, __mmask8, __m128i, __m128i, int);
VPSHRDQ __m128i _mm_maskz_shrdi_epi64(__mmask8, __m128i, __m128i, int);
VPSHRDQ __m256i _mm256_shrdi_epi64(__m256i, __m256i, int);
VPSHRDQ __m256i _mm256_mask_shrdi_epi64(__m256i, __mmask8, __m256i, __m256i, int);
VPSHRDQ __m256i _mm256_maskz_shrdi_epi64(__mmask8, __m256i, __m256i, int);
VPSHRDQ __m512i _mm512_shrdi_epi64(__m512i, __m512i, int);
VPSHRDQ __m512i _mm512_mask_shrdi_epi64(__m512i, __mmask8, __m512i, __m512i, int);
VPSHRDQ __m512i _mm512_maskz_shrdi_epi64(__mmask8, __m512i, __m512i, int);
VPSHRDD __m128i _mm_shrdi_epi32(__m128i, __m128i, int);
VPSHRDD __m128i _mm_mask_shrdi_epi32(__m128i, __mmask8, __m128i, __m128i, int);
VPSHRDD __m128i _mm_maskz_shrdi_epi32(__mmask8, __m128i, __m128i, int);
VPSHRDD __m256i _mm256_shrdi_epi32(__m256i, __m256i, int);
VPSHRDD __m256i _mm256_mask_shrdi_epi32(__m256i, __mmask8, __m256i, __m256i, int);
VPSHRDD __m256i _mm256_maskz_shrdi_epi32(__mmask8, __m256i, __m256i, int);
VPSHRDD __m512i _mm512_shrdi_epi32(__m512i, __m512i, int);
VPSHRDD __m512i _mm512_mask_shrdi_epi32(__m512i, __mmask16, __m512i, __m512i, int);
VPSHRDD __m512i _mm512_maskz_shrdi_epi32(__mmask16, __m512i, __m512i, int);
VPSHRDW __m128i _mm_shrdi_epi16(__m128i, __m128i, int);
VPSHRDW __m128i _mm_mask_shrdi_epi16(__m128i, __mmask8, __m128i, __m128i, int);
VPSHRDW __m128i _mm_maskz_shrdi_epi16(__mmask8, __m128i, __m128i, int);
VPSHRDW __m256i _mm256_shrdi_epi16(__m256i, __m256i, int);
VPSHRDW __m256i _mm256_mask_shrdi_epi16(__m256i, __mmask16, __m256i, __m256i, int);
VPSHRDW __m256i _mm256_maskz_shrdi_epi16(__mmask16, __m256i, __m256i, int);
VPSHRDW __m512i _mm512_shrdi_epi16(__m512i, __m512i, int);
VPSHRDW __m512i _mm512_mask_shrdi_epi16(__m512i, __mmask32, __m512i, __m512i, int);
VPSHRDW __m512i _mm512_maskz_shrdi_epi16(__mmask32, __m512i, __m512i, int);

Exceptions

SIMD Floating-Point Exceptions

None.

Other Exceptions

See Table 2-51, “Type E4 Class Exception Conditions.” VPSHRD—Concatenate and Shift Packed Data Right Logical Vol. 2C 5-630