Move prefix (predicated).
The predicated MOVPRFX instruction is a hint to hardware that the instruction may be combined with the destructive instruction which follows it in program order to create a single constructive operation. Since it is a hint it is also permitted to be implemented as a discrete vector copy, and the result of executing the pair of instructions with or without combining is identical. The choice of combined versus discrete operation may vary dynamically.
Unless the combination of a constructive operation with merging predication is specifically required, it is strongly recommended that for performance reasons software should prefer to use the zeroing form of predicated MOVPRFX or the unpredicated MOVPRFX instruction.
Although the operation of the instruction is defined as a simple predicated vector copy, it is required that the prefixed instruction at PC+4 must be an SVE destructive binary or ternary instruction encoding, or a unary operation with merging predication, but excluding other MOVPRFX instructions. The prefixed instruction must specify the same predicate register, and have the same maximum element size (ignoring a fixed 64-bit "wide vector" operand), and the same destination vector as the MOVPRFX instruction. The prefixed instruction must not use the destination register in any other operand position, even if they have different names but refer to the same architectural register state. Any other use is UNPREDICTABLE.
if !HaveSVE() then UNDEFINED; integer esize = 8 << UInt(size); integer g = UInt(Pg); integer n = UInt(Zn); integer d = UInt(Zd); boolean merging = (M == '1');
Is the name of the destination scalable vector register, encoded in the "Zd" field.
Is the size specifier,
Is the name of the governing scalable predicate register P0-P7, encoded in the "Pg" field.
Is the predication qualifier,
Is the name of the source scalable vector register, encoded in the "Zn" field.
CheckSVEEnabled(); integer elements = VL DIV esize; bits(PL) mask = P[g]; bits(VL) operand1 = Z[n]; bits(VL) dest = Z[d]; bits(VL) result; for e = 0 to elements-1 bits(esize) element = Elem[operand1, e, esize]; if ElemP[mask, e, esize] == '1' then Elem[result, e, esize] = element; elsif merging then Elem[result, e, esize] = Elem[dest, e, esize]; else Elem[result, e, esize] = Zeros(); Z[d] = result;