[PATCH] D67307: [MemorySSA] Do not create memoryaccesses for debug info intrinsics.

[Alina Sbirlea via llvm-commits]

Ack, I will update with the test from https://reviews.llvm.org/D67296 on
Monday.

On Fri, Sep 6, 2019, 11:08 PM George Burgess IV via Phabricator <
reviews at reviews.llvm.org> wrote:

> george.burgess.iv added a comment.
>
> Thanks for this!
>
> +1 for a simple -disable-basicaa test please.
>
>
>
> ================
> Comment at: lib/Analysis/MemorySSA.cpp:287
>      case Intrinsic::assume:
> +    case Intrinsic::dbg_addr:
> +    case Intrinsic::dbg_declare:
> ----------------
> this should be dynamically dead code, no? `II` is sourced from `MD`, so
> there has to be an existing `Def` for these. if we're special-casing
> elsewhere to ensure these `Def`s are never created, ...
>
> (if you'd rather make these new cases lead to `assert(false && "debuginfo
> shouldn't have associated defs!");` or similar, i'm equally content with
> that)
>
>
> ================
> Comment at: lib/Analysis/MemorySSA.cpp:1734
>    // FIXME: Replace this special casing with a more accurate modelling of
>    // assume's control dependency.
>    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I))
> ----------------
> nit: please include a note saying something like debuginfo intrinsics may
> be considered clobbers when we have a nonstandard AA pipeline, so we
> special-case them
>
>
> Repository:
>   rL LLVM
>
> CHANGES SINCE LAST ACTION
>   https://reviews.llvm.org/D67307/new/
>
> https://reviews.llvm.org/D67307
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190906/5b870ac1/attachment.html>