[cfe-dev] latest linux/clang patches

[Chris Lattner]

On Apr 4, 2011, at 4:37 PM, pageexec at freemail.hu wrote:

> hello everyone,
> 
> attached are the updated patches for llvm/clang trunk and linux 2.6.38.2(-pax) to build both
> an i386 and amd64 kernel as well. since my last post several issues were reported in bugzilla,
> some of them even got fixed while others have patches and some are lingering, so if anyone
> feels like helping out, check out PR4068 that links to them.
> 
> the changes since last time:
> 
> 1. amd64/vdso builds without patching the linux side now (must have been fixed in llvm/clang
>   while i wasn't looking ;)
> 
> 2. clang needs a patch to fix mcount instrumentation, right now it looks like that inlined
>   functions get instrumented first and then inlined, resulting in several calls to mcount
>   in the outer function. furthermore the no_instrument_function attribute doesn't affect
>   mcount whereas it does in gcc. the attached clang patch tries to fix these but i'm pretty
>   sure the inline handling is a bad hack only, so feel free to help :).
> 
>   note that -finstrument-functions is meant to apply to inlined functions, it's only mcount
>   that needs special handling.
> 
> 3. i also fixed a small typo in how -fno-merge-all-constants was passed down from the driver
>   (used to be needed for building the vdso, that's why i even noticed).
> 
> 4. i changed the __builtin_object_size prototype to take a const pointer, otherwise there's
>   lots of noise when const pointers are passed to it (not sure why gcc ever defined it as
>   non-const since this builtin isn't meant to modify the target object).
> 
> so that's all for now, the work continues.

Wow, it looks like you have a lot of great patches here. Can you split these up into individual patches?  I see that #4 went in, and the fast isel changes look fine.  I applied #3, and the other patches look fine too  but should be reviewed individually.  Thanks for all the great work!

-Chris