Sure, I will change summary, thanks for review!

For this example:

ping

Well yes, but it would be nicer not having to modify the scripts which we currently use in our build systems.

I have something like this in many places:

Comments Addressed. Thanks for the review.

ping

ping

ping

ping

ping

Comments addressed

Regarding secure plt in general I was using this document: http://devpit.org/wiki/Debugging_PowerPC_ELF_Binaries. This patch just expands secure-plt support for TLS symbols, basically the same thing as for functions symbols or global symbols in the last patch.
Also as a reference I'm using gcc.
I haven't run into example that fails because of shrink-wrapping. If this is a potential issue would not it have appeared so far, since
getGlobalBaseReg() is existing function and it was used before secure-plt support? Do you have some test case that potentially can reproduce issue related with shrink-wrapping?

ping

-mbss-plt is currently default in LLVM, once secure plt support is finished we can set secure plt as default in LLVM, but not for now.

ping

Yes, secure PLT is PowerPC specific feature.

Is this patch OK to commit now ?

Updated patch so it covers cases where we have just function call, without global variables as arguments or global variables used within function.

ping

ping

I tried to compile some important libraries for X86 and MIPS64 within Chromium with clang/llvm. I have compared results between LLVM trunk, and LLVM trunk with my patch. There is code size improvement on many libraries, here are some results:

ping

I was looking if I can reslove this problem in backend. Example:

Well, basically I'm just expanding the existing algorithm, why should we split fields just in case when current field is integer,
I'm not resolving specific problem with unaligned loads/stores on MIPS.

Indentations fixed.

Thanks for the review! I will check indentations with clang format.

Comments addressed.

Comments addressed.

ping

Comments addressed.

Comments addressed.

Hi Bruno,