[PATCH] D81992: [AArch64][GlobalISel] Make G_SEXT_INREG legal and add selection support.

[John Brawn via Phabricator via llvm-commits]

john.brawn added a comment.

It looks like this patch is causing incorrect code generation for

  signed int x = -0x8000;
  int fn(int *ptr) {
      return ptr[(signed short)(unsigned short)(x - 0x8000)];
  }

we have initially

  %6:_(s32) = nsw G_SUB %3:_, %5:_
  %7:_(s16) = G_TRUNC %6:_(s32)
  %8:_(s64) = G_SEXT %7:_(s16)

which then gets converted to

  %6:_(s32) = nsw G_SUB %3:_, %5:_
  %16:_(s64) = G_ANYEXT %6:_(s32)
  %8:_(s64) = G_SEXT_INREG %16:_, 16

The truncation has been lost, and we eventually end up with

  subs    w9, w9, #8, lsl #12
  mov     w0, w9
  ldr     w0, [x8, w0, uxtw #2]




Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81992/new/

https://reviews.llvm.org/D81992