[PATCH] D23824: [ADCE] Add handling of PHI nodes when removing control flow

[Daniel Berlin via llvm-commits]

Yes, that is why i said you may need to make the phi non-constant (ie make
a new parameter):

It is propagating the constant into the phi:
    %j.0 = phi i32 [ 0, %0 ], [ 1, %2 ]

Thus, you never end up marking anything interesting live there.





On Tue, Sep 6, 2016 at 3:30 PM, David Callahan <dcallahan at fb.com> wrote:

> david2050 added a comment.
>
> I ran this test:
>
>   int foo(int a, int b, int N) {
>     int j = 0;
>     if (b) {
>       int i;
>       for (i = 0; i < N; i++)
>         ;
>       j = 1;
>     }
>     return j;
>   }
>
> compiled to bc and then processed:  opt -sroa -adce -adce-remove-loops
> and it did successfully remove the loop.  I also tried this variant
>
>   int foo(int b, int N) {
>     int j = 0;
>     if (b) {
>       int i;
>       j = 1;
>       for (i = 0; i < N; i++)
>         ;
>     }
>     return j;
>   }
>
> The final output for that one looked like:
>
>   define i32 @foo(i32 %b, i32 %N) #0 {
>     %1 = icmp ne i32 %b, 0
>     br i1 %1, label %2, label %3
>
>   ; <label>:2:                                      ; preds = %0
>     br label %3
>
>   ; <label>:3:                                      ; preds = %2, %0
>     %j.0 = phi i32 [ 0, %0 ], [ 1, %2 ]
>     ret i32 %j.0
>   }
>
>
> https://reviews.llvm.org/D23824
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160906/11f27918/attachment.html>