[Bug target/82582] not quite optimal code for -2*x*y - 3*z: could use one less LEA for smaller code without increasing critical path latency for any input

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82582

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
For AARCH64 the correct thing is able to be produced:
foo32:
        sub     w2, w2, w2, lsl 2 ; w2 = z - z<<2; (rather -z*3)
        neg     w0, w0, lsl 1     ; w0 = x * -2
        madd    w0, w0, w1, w2    ; res = w0 * y + w2
        ret