Page 321 - ARM 64 Bit Assembly Language

310 Chapter 9

9.7.4 Multiply and accumulate

These instructions are used to multiply and accumulate the results. A fused multiply accu-
mulate operation does not perform rounding between the multiply and add operations. The
two operations are fused into one. Using the fused multiply accumulate can speed up and im-
prove the accuracy of many computations that involve the accumulation of products. NEON
provides the following fused multiply accumulate instructions.:
fmadd Fused Multiply Accumulate
fmsub Fused Multiply Subtract
fnmadd Fused Multiply Accumulate and Negate
fnmsub Fused Multiply Subtract and Negate

9.7.4.1 Syntax

f<op> Fd, Fn, Fm, Fa

• Fd, Fn, Fm,and Fa must specify either four single precision registers or four double preci-
sion registers.
• <op> is one of madd, msub, nmadd,or nmsub.

9.7.4.2 Operations

Name Effect Description
fmadd Fd ← Fa + Fn × Fm Multiply the values in the operand
registers, and add the value to the
destination register.
fmsub Fd ← Fa − Fn × Fm Multiply the values in the operand
registers, and subtract the value from
the destination register.
fnmadd Fd ←−(Fa + Fn × Fm) Multiply the values in the operand
registers, and add the value to the
destination register, then negate the
final result and store it in the destina-
tion register
fnmsub Fd ←−(Fa − Fn × Fm) Multiply the values in the operand
registers, and subtract the value from
the destination register, then negate
the final result and store it in the des-
tination register