[LLVMdev] Support for per-loop pragma

Junjie Gu jgu222 at gmail.com
Wed Jun 2 11:08:02 PDT 2010

Thanks for comments/suggestions.

Yes, attaching metadata to instructions will be good choices for many
cases. But for loops,
attaching metadata to back-edges requires that the front end to build
loops, which is an
additional task for the front end. And this task is really a backend's
job, not the front end's.

If the only concern is that it is hard for pragma intrinsics to stay
with their associated code,
another choice is to process pragma intrinsic at the beginning of
optimizer and associates
them to instructions by either attaching metadata to instructions and
some kind maps (between
instructions/BB to metadata, etc).


On Wed, Jun 2, 2010 at 9:48 AM, Devang Patel <devang.patel at gmail.com> wrote:
> On Tue, Jun 1, 2010 at 5:42 PM, Junjie Gu <jgu222 at gmail.com> wrote:
>> Any suggestions/ideas ?
> IIUC, Chris suggested something like following ...
> header:
>  br i1 %x, label %then, %label endif
> then:
>  ...
>  br i1 %y, label %loop_exit, label %header, !loop_pragma !1
> endif:
>  ...
>  br i1 %z, label %loop_exit, label %header, !loop_pragma !2
> loop_exit:
>  ret i32 1
> Where !1 and !2 are the metadata for your loop pragma.
> -
> Devang

More information about the llvm-dev mailing list