[PATCH] D96881: Avoid Bitcast-GEP fusion for void pointers

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

In D96881#2615088 <https://reviews.llvm.org/D96881#2615088>, @hjooybar2 wrote:

> <...>

(unless i'm not aware of a RFC newer than https://lists.llvm.org/pipermail/llvm-dev/2019-December/137684.html)
I think my main point is being ignored:
This fix is temporary. After the pointer types are gone from IR,
the problem will reappear, because the GEP's will have pointer offsets,
they won't know anything about structs,
see http://lists.llvm.org/pipermail/llvm-dev/attachments/20191218/761e8893/attachment.obj:

  define i8* @gep(p0 %ptr) {
    %res = getelementptr { i8, i32 }, p0 %ptr, i32 5, i32 0
    ret i8* %res
  }

So it would be most future-proof to work towards making that transition as easy as possible,
by enhancing whatever passes aren't happy about this existing transformation.

All that being said, i understand that it is a *much* bigger endeavor than just adjusting InstCombine.



================
Comment at: llvm/test/Transforms/InstCombine/getelementptr.ll:1275
 }
 
 !0 = !{!"branch_weights", i32 2, i32 10}
----------------
For my own reference, the current result is: https://godbolt.org/z/TTE3nM


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D96881/new/

https://reviews.llvm.org/D96881