Use auto profiler to set branch weights

[Diego Novillo]

On Fri, Oct 18, 2013 at 7:30 PM, Eric Christopher <echristo at gmail.com> wrote:
> A few general comments:
>
> A comment would be nice for dumpInstrumentedInsns that we don't need
> DEBUG for the dbg() statements.

I think I actually need to use errs() here. It's now only called by a
DEBUG block, but it doesn't have to be. Or am I off track?

> It's handy if the test case has the C/C++/etc code used to reproduce
> inside it in case something changes or someone wants to take a fresh

You mean, have the C/C++ code in comments? Or you mean start from the
C++ code itself?

> look. Also a comment with what you're testing and where so that people
> don't have to guess.

Sorry.  What would they have to guess?  The command needed to run the
test is embedded in the .ll file itself.

>
> It looks like you're doing the sample information based on line?

Yes. That's why we need debug information.

I've coalesced all the llvm patches into one.  It's hard to keep track
of them, otherwise. And they are actually all related.

The name is still in flux.  I will change it once we reach consensus.
Right now, I'd like a full review on the actual implementation.


Thanks.  Diego.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-AutoProfile-pass.-Initial-setup.patch
Type: application/octet-stream
Size: 35852 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131022/2e783c24/attachment.obj>