fcmla

SVE Floating-Point Complex Multiply-Add

FCMLA <Zda>.<T>, <Pg>/M, <Zn>.<T>, <Zm>.<T>, #<rot>

Performs complex multiply-accumulate.

Details

SVE floating-point complex multiply-add: performs a predicate-masked complex multiply-accumulate operation where Zn and Zm are multiplied with rotation, and the result is added to Zda. For each rotation value (0°, 90°, 180°, 270°), a different complex multiplication result is accumulated. Elements where the predicate is false leave Zda unchanged. Floating-point exceptions are signaled per IEEE 754. This is an SVE-only instruction.

Pseudocode Operation

for i = 0 to VL-1
  if Pg[i] == '1' then
    product ← FPComplexMultiply(Zn[i], Zm[i], rot)
    Zda[i] ← FPAdd(Zda[i], product)
  // else Zda[i] unchanged

Example

FCMLA z0.s.T, p0/m/M, z1.s.T, z2.s.T, #rot

Encoding

Binary Layout
01100100
size
0
Zm
0
rot
Pg
Zn
Zda
 
Format SVE FP Complex
Opcode 0x64000000
Extension SVE

Operands

  • Zda
    Accumulator scalable vector register (SVE)
  • Pg
    Mask
  • Zn
    First source scalable vector register (SVE)
  • Zm
    Second source scalable vector register (SVE)
  • rot
    Rot

Reference (Arm A64 ISA)

Instruction Forms

Encoding Instruction ISA Bit pattern
0x2F001000 FCMLA <Vd>.<T>, <Vn>.<T>, <Vm>.<Ts>[<index>], #<rotate> A64 0 | Q | 1 | 01111 | size | L | M | Rm | 0 | rot | 1 | H | 0 | Rn | Rd
0x2E00C400 FCMLA <Vd>.<T>, <Vn>.<T>, <Vm>.<T>, #<rotate> A64 0 | Q | 1 | 01110 | size | 0 | Rm | 110 | rot | 1 | Rn | Rd
0x64000000 FCMLA <Zda>.<T>, <Pg>/M, <Zn>.<T>, <Zm>.<T>, <const> A64 01100100 | size | 0 | Zm | 0 | rot | Pg | Zn | Zda
0x64A01000 FCMLA <Zda>.H, <Zn>.H, <Zm>.H[<imm>], <const> A64 01100100 | 1 | 0 | 1 | i2 | Zm | 0001 | rot | Zn | Zda
0x64E01000 FCMLA <Zda>.S, <Zn>.S, <Zm>.S[<imm>], <const> A64 01100100 | 1 | 1 | 1 | i1 | Zm | 0001 | rot | Zn | Zda

Description

Multiply the duplicated real components for rotations 0 and 180, or imaginary components for rotations 90 and 270, of the floating-point complex numbers in the first source vector by the corresponding complex number in the second source vector rotated by 0, 90, 180 or 270 degrees in the direction from the positive real axis towards the positive imaginary axis, when considered in polar representation. Then destructively add the products to the corresponding components of the complex numbers in the addend and destination vector, without intermediate rounding. These transformations permit the creation of a variety of multiply-add and multiply-subtract operations on complex numbers by combining two of these instructions with the same vector operands but with rotations that are 90 degrees apart. Each complex number is represented in a vector register as an even/odd pair of elements with the real part in the even-numbered element and the imaginary part in the odd-numbered element. Inactive elements in the destination vector register remain unmodified.

Operation

CheckSVEEnabled();
constant integer VL = CurrentVL;
constant integer PL = VL DIV 8;
constant integer pairs = VL DIV (2 * esize);
bits(PL) mask = P[g, PL];
bits(VL) operand1 = if AnyActiveElement(mask, esize) then Z[n, VL] else Zeros(VL);
bits(VL) operand2 = if AnyActiveElement(mask, esize) then Z[m, VL] else Zeros(VL);
bits(VL) operand3 = Z[da, VL];
bits(VL) result;

for p = 0 to pairs-1
    addend_r = Elem[operand3, 2 * p + 0, esize];
    addend_i = Elem[operand3, 2 * p + 1, esize];
    if ActivePredicateElement(mask, 2 * p + 0, esize) then
        bits(esize) elt1_a = Elem[operand1, 2 * p + sel_a, esize];
        bits(esize) elt2_a = Elem[operand2, 2 * p + sel_a, esize];
        if neg_r then elt2_a = FPNeg(elt2_a, FPCR);
        addend_r = FPMulAdd(addend_r, elt1_a, elt2_a, FPCR);
    if ActivePredicateElement(mask, 2 * p + 1, esize) then
        bits(esize) elt1_a = Elem[operand1, 2 * p + sel_a, esize];
        bits(esize) elt2_b = Elem[operand2, 2 * p + sel_b, esize];
        if neg_i then elt2_b = FPNeg(elt2_b, FPCR);
        addend_i = FPMulAdd(addend_i, elt1_a, elt2_b, FPCR);
    Elem[result, 2 * p + 0, esize] = addend_r;
    Elem[result, 2 * p + 1, esize] = addend_i;

Z[da, VL] = result;