[llvm-commits] [PATCH] Take into account minimize size attribute in the inliner

[Eric Christopher]

Looks good. I hope no one needs to unroll loops quickly in C++ at -Oz
though :)

-eric


On Wed, Dec 12, 2012 at 4:04 PM, Quentin Colombet <qcolombet at apple.com>wrote:

> Hi Eric,
>
> Here is the patch with the test case in the backend parts.
> Moreover I've attached some performance numbers of runs with Oz with and
> without my patch (compile time, execution time, and code size).
>
> Cheers,
>
> -Quentin
>
>
>
>
>
> On Dec 12, 2012, at 11:36 AM, Quentin Colombet <qcolombet at apple.com>
> wrote:
>
>
> On Dec 12, 2012, at 11:32 AM, Eric Christopher <echristo at gmail.com> wrote:
>
>
>
>>
>> I had a phasing issue as llvm output was already inlined. That's why I
>> did a clang test since I used clang as run command.
>> I am not super familiar to llvm phasing, thus how should I produce an
>> input for a backend test in that case?
>>
>>
> Probably the easiest way would be to use clang -O0 to generate a .ll file
> out of your testcase, clean it up a bit to test only what you want and then
> run opt -inline to see what is and isn't being inlined before and after
> your test.
>
> Make sense?
>
> Perfect sense, thanks!
> Actually I've just seen that.
>
> -Quentin
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20121212/c336fedf/attachment.html>