[AArch64][GlobalISel] Improve i128 mul generation

[davemgreen]

i128 multiplies under GlobalISel could be producing more madds if it reassociated the add.
https://godbolt.org/z/Wr1r5ez1G

SDAG

        umulh   x8, x0, x2
        madd    x8, x0, x3, x8
        mul     x0, x0, x2
        madd    x1, x1, x2, x8
        ret

GISel

        mul     x9, x0, x3
        mul     x8, x0, x2
        umulh   x10, x0, x2
        madd    x9, x1, x2, x9
        mov     x0, x8
        add     x1, x9, x10
        ret

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: David Green (davemgreen)

i128 multiplies under GlobalISel could be producing more madds if it reassociated the add. https://godbolt.org/z/Wr1r5ez1G

SDAG

        umulh   x8, x0, x2
        madd    x8, x0, x3, x8
        mul     x0, x0, x2
        madd    x1, x1, x2, x8
        ret

GISel

        mul     x9, x0, x3
        mul     x8, x0, x2
        umulh   x10, x0, x2
        madd    x9, x1, x2, x9
        mov     x0, x8
        add     x1, x9, x10
        ret

[arsenm]

I think the multiply splitting needs work, I started #97194 but don't have time to get back to it

[tschuett]

; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=aarch64-apple-ios -global-isel -stop-before=instruction-select %s -o - | FileCheck %s --check-prefix=PRESELECTION

define i128 @ti128(i128 %a, i128 %b) {
  ; PRESELECTION-LABEL: name: ti128
  ; PRESELECTION: bb.1 (%ir-block.0):
  ; PRESELECTION-NEXT:   liveins: $x0, $x1, $x2, $x3
  ; PRESELECTION-NEXT: {{  $}}
  ; PRESELECTION-NEXT:   [[COPY:%[0-9]+]]:gpr(s64) = COPY $x0
  ; PRESELECTION-NEXT:   [[COPY1:%[0-9]+]]:gpr(s64) = COPY $x1
  ; PRESELECTION-NEXT:   [[COPY2:%[0-9]+]]:gpr(s64) = COPY $x2
  ; PRESELECTION-NEXT:   [[COPY3:%[0-9]+]]:gpr(s64) = COPY $x3
  ; PRESELECTION-NEXT:   [[MUL:%[0-9]+]]:gpr(s64) = G_MUL [[COPY]], [[COPY2]]
  ; PRESELECTION-NEXT:   [[MUL1:%[0-9]+]]:gpr(s64) = G_MUL [[COPY1]], [[COPY2]]
  ; PRESELECTION-NEXT:   [[MUL2:%[0-9]+]]:gpr(s64) = G_MUL [[COPY]], [[COPY3]]
  ; PRESELECTION-NEXT:   [[UMULH:%[0-9]+]]:gpr(s64) = G_UMULH [[COPY]], [[COPY2]]
  ; PRESELECTION-NEXT:   [[ADD:%[0-9]+]]:gpr(s64) = G_ADD [[MUL1]], [[MUL2]]
  ; PRESELECTION-NEXT:   [[ADD1:%[0-9]+]]:gpr(s64) = G_ADD [[ADD]], [[UMULH]]
  ; PRESELECTION-NEXT:   $x0 = COPY [[MUL]](s64)
  ; PRESELECTION-NEXT:   $x1 = COPY [[ADD1]](s64)
  ; PRESELECTION-NEXT:   RET_ReallyLR implicit $x0, implicit $x1
    %c = mul i128 %a, %b
    ret i128 %c
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; PRESELECTION: {{.*}}