[LLVMdev] Shrink Wrap for ARM architecture?

Yin Ma

unread,

Oct 30, 2013, 4:47:48 PM10/30/13

to llvmdev Dev

Hi

Does anyone know if the shrink wrap pass works for ARM architecture?

I tried it seems not working. Which architecture shrink warp was

originally designed for?

Thanks,

Yin

mcro...@codeaurora.org

unread,

Oct 30, 2013, 5:10:22 PM10/30/13

to Yin Ma, llvmdev Dev

I know nothing about this pass, but here are a few comments.

> Does anyone know if the shrink wrap pass works for ARM architecture?

First, how are you invoking it?

Looks to be a command line option, so -mllvm -shrink-wrap should enable
it. Assuming you have a Debug build, you can see the debug output using
-debug-only=shrink-wrap.

> I tried it seems not working. Which architecture shrink warp was
> originally designed for?

AFAICT, the implementation is target independent.

Chad

_______________________________________________
LLVM Developers mailing list
LLV...@cs.uiuc.edu http://llvm.cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

Evan Cheng

unread,

Oct 30, 2013, 8:08:47 PM10/30/13

to mcro...@codeaurora.org, llvmdev Dev

The current implementation in LLVM is experimental at best. It probably should be ripped out.

Evan

Rafael Espíndola

unread,

Oct 31, 2013, 1:34:29 AM10/31/13

to Evan Cheng, llvmdev Dev

On 30 October 2013 20:08, Evan Cheng <evan....@apple.com> wrote:
> The current implementation in LLVM is experimental at best. It probably should be ripped out.

Patch attached :-)

Cheers,
Rafael

t.patch

Renato Golin

unread,

Oct 31, 2013, 2:34:06 AM10/31/13

to Rafael Espíndola, llvmdev Dev

Thanks Rafael! It's a shame to see a feature go away that easily, but I also don't have man-power to maintain this. At least it was well isolated so that anyone can start over from that when needed.

For my education, how bad was it? Was the problem in the implementation or in the edge case performance? Was it something wrong with the code or just under-tested?

Maybe, if it's not too bad (just need more testing), than I could possibly convince someone to look at it?

cheers,

--renato

Jakob Stoklund Olesen

unread,

Oct 31, 2013, 3:06:22 AM10/31/13

to Rafael Espíndola, llvmdev Dev

LGTM.

Was that the last use of SparseBitVector?

Thanks,
/jakob

> <t.patch>

Rafael Espíndola

unread,

Oct 31, 2013, 10:13:15 AM10/31/13

to Renato Golin, llvmdev Dev

> Thanks Rafael! It's a shame to see a feature go away that easily, but I also
> don't have man-power to maintain this. At least it was well isolated so that
> anyone can start over from that when needed.
>
> For my education, how bad was it? Was the problem in the implementation or
> in the edge case performance? Was it something wrong with the code or just
> under-tested?

Even can probably comment more, but two things I noticed:

* zero tests
* an arbitrary limit (500 basic blocks) where it would give up

> Maybe, if it's not too bad (just need more testing), than I could possibly
> convince someone to look at it?
>
> cheers,
> --renato

Cheers,
Rafael

Rafael Espíndola

unread,

Oct 31, 2013, 10:15:13 AM10/31/13

to Jakob Stoklund Olesen, llvmdev Dev

On 31 October 2013 03:06, Jakob Stoklund Olesen <stok...@2pi.dk> wrote:
> LGTM.
>
> Was that the last use of SparseBitVector?

No, AliveBlocks in LiveVariables uses it.

Cheers,
Rafael

Reply all

Reply to author

Forward