[llvm-dev] Normalize a SCEV Expression

37 views
Skip to first unread message

Stefanos Baziotis via llvm-dev

unread,
Jul 30, 2020, 4:03:27 PM7/30/20
to llvm-dev
Hi,

I have a SCEV like this: {16,+,8}. Say that this came from a loop like:
int64_t *p;
for (int64_t i = 0; i < ...; ++i)
  p[i+2]...

And assuming that I'm on a 64-bit machine. What I would like to do is normalize it
like that, basically this: {2,+,1} i.e. map it to the index.

Now, I tried to get the underlying element size of the pointer, then getUDivExpr(OriginalSCEV, ElementSize); But I don't get the desired SCEV back. Instead I'm getting:
{16,+,8} \u 8, i.e. just adding udiv in the original expression.

Why is this not simplified?

Thanks,
Stefanos Baziotis

Florian Hahn via llvm-dev

unread,
Jul 31, 2020, 5:37:05 AM7/31/20
to Stefanos Baziotis, llvm-dev


I guess the problem here is that the AddRec is missing overflow flags, which means the expression may overflow and if that happens, the folded expression may not be equivalent to the original one in all cases.

If the AddRec has the overflow flags set, it looks something like `{16,+,8}<nuw><nsw><%loop>`. I’d check if the original IR had nuw/nsw flags on add instruction for the AddRec. The way those flags are managed in SCEV are sometimes a bit surprising, because the expressions are not tied to a specific location and the flags need to be valid for the whole scope the expression can be evaluated in.

Cheers,
Florian

_______________________________________________
LLVM Developers mailing list
llvm...@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

Stefanos Baziotis via llvm-dev

unread,
Jul 31, 2020, 7:35:19 AM7/31/20
to Florian Hahn, llvm-dev
Indeed you're right, thanks! I need a nuw (which I have to invent, with a run-time check, since it doesn't exist in the original but anyway).

Best,
Stefanos

Florian Hahn via llvm-dev

unread,
Jul 31, 2020, 7:45:00 AM7/31/20
to Stefanos Baziotis, llvm-dev

> On Jul 31, 2020, at 12:34, Stefanos Baziotis <stefanos...@gmail.com> wrote:
>
> Indeed you're right, thanks! I need a nuw (which I have to invent, with a run-time check, since it doesn't exist in the original but anyway).
>

Right, I guess in your example, the flags on the add depend on the exact bound of the loop, which is omitted.

Stefanos Baziotis via llvm-dev

unread,
Jul 31, 2020, 8:17:15 AM7/31/20
to Florian Hahn, llvm-dev
Exactly.
Reply all
Reply to author
Forward
0 new messages