[Bug target/70341] [5/6/7 Regression] cost model for addresses is incorrect, slsr is using reg + reg + CST for arm

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70341

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Fredrik Hederstierna from comment #8)
> Could it be something in tree-ssa-forwprop pass?
> 
> This pass is executed 4 times in -Os, and starting with GCC-4.9 it seems
> like statements that seems to generate instructions that are hard to
> eliminate later in compilation could be propagated?

No that is correct for this pass, this was done on purpose to allow for better
optimizations and better aliasing info.

Anyways if we change case 4: to default: for GCC 7, we get even better code on
aarch64 than previous versions (even at -O2):
        add     x0, x0, x1, sxtw 4
        ldp     w2, w0, [x0, 4]
        cmp     w2, 2
        beq     .L3
        cmp     w2, 3
        beq     .L4
        cmp     w2, 1
        beq     .L8
        b       handle_case_4
        .p2align 3
.L8:
        b       handle_case_1
        .p2align 3
.L4:
        b       handle_case_3
        .p2align 3
.L3:
        b       handle_case_2