[Polly][Refactor] IslAst [6/7]

Tobias Grosser tobias at grosser.es
Tue Jul 29 10:31:35 PDT 2014


On 29/07/2014 18:05, Johannes Doerfert wrote:
> I figured that [Refactor] would indicate that I do not change the behavior.
> However, I can make it more explicit.
>
> To be explicit here:
>    I do not intend to change the behavior (and I'm pretty sure I did not).
>    What I did is:
>       - I removed unnecessary computation.
>       - Made the variable names consistent.
>       - Got rid of the "mentioning" of OpenMP.
>       - Use a different kind of flags to express what we expressed before.

OK. Thanks for clarifying.

I was assuming this, but I am still a little confused by the code (also 
by the old one). Specifically, I try to understand if there is a 
difference between an innermost-parallel loop and an innermost and 
parallel loop and wonder which has been implemented before, what is 
implemented today.

I am specifically interested to try a couple more test cases that show 
this difference. As this was not here in the original code, it is 
obviously not your fault, but I would use this opportunity to make this
stupidly obvious.

I was just trying to compile and run this patch and it seems to not 
apply cleanly. Specifically, the part of the diff that tries to remove
a line 'assert(Info && "Post order visit assumes annotated for nodes")'
does not apply. Do I need any other patch to be able to apply and test 
this code? Do you happen to have a patch that applies cleanly?

Cheers,
Tobias



More information about the llvm-commits mailing list