uzp1

SVE Unzip 1

UZP1 <Zd>.<T>, <Zn>.<T>, <Zm>.<T>

Selects even elements from concatenated vectors.

Details

SVE Unzip 1 deinterlaces the concatenation of Zn and Zm by selecting even-indexed elements (0, 2, 4, ...) and packing them into Zd. This is the inverse of a zip/transpose operation. This is an unpredicated operation. No condition flags are affected. This is an AArch64-only SVE instruction requiring SVE support.

Pseudocode Operation

integer esize = 8 << UInt(sz);
integer elements = VL / esize;
for e = 0 to elements-1
  integer src_index = 2 * e;
  if src_index < elements then
    Zd[e * esize +: esize] = Zn[src_index * esize +: esize];
  else
    Zd[e * esize +: esize] = Zm[(src_index - elements) * esize +: esize];

Example

UZP1 z0.s.T, z1.s.T, z2.s.T

Encoding

Binary Layout

00000101

size

011

Format SVE Permute

Opcode 0x05206800

Extension SVE

Operands

Zd
Destination scalable vector register (SVE)
Zn
First source scalable vector register (SVE)
Zm
Second source scalable vector register (SVE)

Reference (Arm A64 ISA)

Instruction Forms

Encoding	Instruction	ISA	Bit pattern
0x0E001800	UZP1 <Vd>.<T>, <Vn>.<T>, <Vm>.<T>	A64	0 \| Q \| 001110 \| size \| 0 \| Rm \| 0 \| 0 \| 0110 \| Rn \| Rd
0x05204800	UZP1 <Pd>.<T>, <Pn>.<T>, <Pm>.<T>	A64	00000101 \| size \| 10 \| Pm \| 010 \| 0 \| 1 \| 0 \| 0 \| Pn \| 0 \| Pd
0x05206800	UZP1 <Zd>.<T>, <Zn>.<T>, <Zm>.<T>	A64	00000101 \| size \| 1 \| Zm \| 011 \| 01 \| 0 \| Zn \| Zd
0x05A00800	UZP1 <Zd>.Q, <Zn>.Q, <Zm>.Q	A64	00000101101 \| Zm \| 000 \| 0 \| 1 \| 0 \| Zn \| Zd

Description

Concatenate adjacent even or odd-numbered elements from the first and second source vectors and place in elements of the destination vector. This instruction is unpredicated. Note: UZP1 is equivalent to truncating and packing each element from two source vectors into a single destination vector with elements of half the size. The 128-bit element variant requires that the Effective SVE vector length is at least 256 bits. ID_AA64ZFR0_EL1.F64MM indicates whether the 128-bit element variant is implemented. The 128-bit element variant is illegal when executed in Streaming SVE mode, unless FEAT_SME_FA64 is implemented and enabled.

Operation

if esize < 128 then CheckSVEEnabled(); else CheckNonStreamingSVEEnabled();
constant integer VL = CurrentVL;
if VL < esize * 2 then UNDEFINED;
constant integer pairs = VL DIV (esize * 2);
bits(VL) operand1 = Z[n, VL];
bits(VL) operand2 = Z[m, VL];
bits(VL) result = Zeros(VL);

for p = 0 to pairs - 1
    Elem[result, p, esize] = Elem[operand1, 2*p+part, esize];

for p = 0 to pairs - 1
    Elem[result, pairs+p, esize] = Elem[operand2, 2*p+part, esize];

Z[d, VL] = result;