FYI, thanks for all the discussion and feedback!<div><br></div><div>It sounds like folks are generally OK with the old stuff being cleared out, so I plan to go ahead and do that. Specific concerns and the strategy for them being addressed:</div>
<div><br></div><div>- Alastair's GSoC project is moving to the new branch probability infrastructure and specifically branch weight metadata. It sounds like nuking the profile stuff won't really even disrupt this work, so no need to worry there.</div>
<div><br></div><div>- The llvm-prof, <a href="http://profile.pl">profile.pl</a>, related profiling tools may work for some, but not others. The only user is out-of-tree, and is already regularly broken by updates. My suggestion would be to not update, and/or contribute analogous tools for branch weight metadata representation. I'm happy to give advice or help out here a little if I can. As we're already breaking these, I don't think we should actually keep them in a half-dead zombie state. If they're re-written, they should ideally be in C++ and better tested.</div>
<div><br></div><div>- I'll watch the lists and try to help out anyone impacted by this.</div><div><br></div><div><br></div><div>For the record, I'm not just trying to break stuff. i'm looking at deep changes to the pass manager, and the profiling passes don't always use it cleanly at the moment. One fewer strange user will make the new design easier and better. The strangeness appears to have to do with legacy and lack of maintenance, nothing to do with the nature of profiling passes.</div>
<div><br></div><div><br></div><div>Thanks!</div><div>-Chandler</div><div class="gmail_extra"><br><br><div class="gmail_quote">On Tue, Jul 17, 2012 at 11:40 PM, Alastair Murray <span dir="ltr"><<a href="mailto:alastairmurray42@gmail.com" target="_blank">alastairmurray42@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Alok,<br>
<div class="im"><br>
On 17/07/12 13:00, #ALOK PRAKASH# wrote:<br>
> Profile.pl is understandably a very simple script, but it does make<br>
> it easier to see some preliminary profile results and identity the<br>
> hot portions of a program which are suitable for hardware<br>
> acceleration. llvm-prof also helps in the same way.<br>
<br>
</div>Ok, great -- using llvm-prof for detecting hot portions is the obvious<br>
use, but I just wanted to check.<br>
<br>
I'll keep llvm-prof in mind when writing the new loader and update<br>
llvm-prof to use it. I also don't have an issue with <a href="http://profile.pl" target="_blank">profile.pl</a> so I<br>
can update it as required, it is tiny after all. I'm not sure if others<br>
want it sticking around though.<br>
<div class="HOEnZb"><div class="h5"><br>
Regards,<br>
Alastair.<br>
_______________________________________________<br>
LLVM Developers mailing list<br>
<a href="mailto:LLVMdev@cs.uiuc.edu">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev</a><br>
</div></div></blockquote></div><br></div>