llvm dev - Mar 2013 - [LLVMdev] Different results using -emit-llvm and llc together than just using -S?

Why would the generated assembly results for a piece of code be different
if I did this:

clang -S -emit-llvm sample.c -o sample.ll
llc sample.ll -o sample.s

then if I did this:

clang -S sample.c -o sample.s

Thanks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130312/5fc1a52a/attachment.html>

llc and clang have different set of defaults.

E.g. llc defaults to frame pointer elimination, etc.

On Wed, Mar 13, 2013 at 4:13 AM, Ryan Taylor <ryta1203 at gmail.com>
wrote:
> Why would the generated assembly results for a piece of code be different
if
> I did this:
>
> clang -S -emit-llvm sample.c -o sample.ll
> llc sample.ll -o sample.s
>
> then if I did this:
>
> clang -S sample.c -o sample.s
>
> Thanks.
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>


--
With best regards, Anton Korobeynikov
Faculty of Mathematics and Mechanics, Saint Petersburg State University

Hi Ryan,

On 13/03/13 01:13, Ryan Taylor wrote:
> Why would the generated assembly results for a piece of code be different
if I
> did this:
>
> clang -S -emit-llvm sample.c -o sample.ll
> llc sample.ll -o sample.s
>
> then if I did this:
>
> clang -S sample.c -o sample.s
because clang passes all kinds of options to llc that are not the llc defaults.
For example, by default llc targets your local processor while I don't think
clang does.

Ciao, Duncan.

Ryan Taylor wrote:
> Why would the generated assembly results for a piece of code be
> different if I did this:
>
> clang -S -emit-llvm sample.c -o sample.ll
> llc sample.ll -o sample.s
>
> then if I did this:
>
> clang -S sample.c -o sample.s
For one example, 'llc' defaults to -O2 but clang defaults to -O0. In 
general, clang sets up the equivalent of an opt|llc run with a slew of 
flags passed in. Also, note that clang -O0 is not the same as clang -O2 
-mllvm -disable-llvm-optzns, as clang chooses to emit more code which 
may prove useful to the optimizer, but would only waste time at -O0.

Nick

Understood; however, I see the same change/difference when calling llc -O0?

Are there any other differences between calling these separately? Or does
llc not take -O levels?

On Wed, Mar 13, 2013 at 3:53 AM, Nick Lewycky <nicholas at mxc.ca> wrote:
> Ryan Taylor wrote:
>
>> Why would the generated assembly results for a piece of code be
>> different if I did this:
>>
>> clang -S -emit-llvm sample.c -o sample.ll
>> llc sample.ll -o sample.s
>>
>> then if I did this:
>>
>> clang -S sample.c -o sample.s
>>
>
> For one example, 'llc' defaults to -O2 but clang defaults to -O0.
In
> general, clang sets up the equivalent of an opt|llc run with a slew of
> flags passed in. Also, note that clang -O0 is not the same as clang -O2
> -mllvm -disable-llvm-optzns, as clang chooses to emit more code which may
> prove useful to the optimizer, but would only waste time at -O0.
>
> Nick
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130313/2702bb53/attachment.html>

Anton,

  Is this true even at -O0?

Thanks.

On Wed, Mar 13, 2013 at 3:28 AM, Anton Korobeynikov <anton at
korobeynikov.info
> wrote:
> llc and clang have different set of defaults.
>
> E.g. llc defaults to frame pointer elimination, etc.
>
> On Wed, Mar 13, 2013 at 4:13 AM, Ryan Taylor <ryta1203 at gmail.com>
wrote:
> > Why would the generated assembly results for a piece of code be
> different if
> > I did this:
> >
> > clang -S -emit-llvm sample.c -o sample.ll
> > llc sample.ll -o sample.s
> >
> > then if I did this:
> >
> > clang -S sample.c -o sample.s
> >
> > Thanks.
> >
> > _______________________________________________
> > LLVM Developers mailing list
> > LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> >
>
>
>
> --
> With best regards, Anton Korobeynikov
> Faculty of Mathematics and Mechanics, Saint Petersburg State University
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130313/2a0b8cfd/attachment.html>