[PATCH] D14699: [GlobalOpt] Allow constant globals to be SRA'd

[Daniel Berlin via llvm-commits]

On Mon, Nov 16, 2015 at 8:01 AM, Mehdi AMINI via llvm-commits <
llvm-commits at lists.llvm.org> wrote:

> joker.eph added inline comments.
>
> ================
> Comment at: test/Transforms/GlobalOpt/globalsra.ll:31
> @@ +30,3 @@
> +define void @onlystore2() {
> +        store i32 123, i32* getelementptr ({ i32, float, { double } }, {
> i32, float, { double } }* @G2, i32 0, i32 0)
> +        ret void
> ----------------
> It is strange that we can store to a constant?
>
>

It's strange in the sense that it probably doesn't do anything sensible,
and is likely not well defined behavior ;-)

But you can't really disallow it in the IR, because if you didn't, you'd
have to do things like make sure optimizations never propagated constants
to certain uses.

IE you couldn't write:

Discover X is constant
for each use of X
  replace use with constant

you'd have to write

Discover X is constant
for each use of X
   if user of use is not a store
     replace use with constant

This is more expensive, and doesn't make the behavior more well defined.
It's still storing to the same place (or else you wouldn't have been able
to prove it constant!)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151116/c95ab94f/attachment.html>