Addresses comments.

Gentle ping.

Rebase.
Clean the failure in test llvm/test/CodeGen/PowerPC/aix-xcoff-visibility.ll.

Addressed @DiggerLin's comments, THX.

1. Simplify tests.
2. Test dwarf64.

Addressed comments and improved tests.

Hi @DiggerLin, Sorry for the late update, I have verified that the SignAndSize value should be 15 for both fixup_ppc_half16ds and fixup_ppc_half16dq. Although there are zeros being implied, the totally size should be a 16-bit fixup. I confirmed this by testing LIT and test-suite, and the test-suite passes only if the value is 15, otherwise there will be an error like this

Relocation overflow in reference to: L..CPI168_7 (entry 3199)
        Address: 0x00082c12; RLD type: R_TOC; RLD length: 14

Improved the format of run lines in tests.

Addressed @DiggerLin 's comments.

Thanks for your comments! @jhenderson
It seems inappropriate to support the symbol table before the sections and relocations, due to their offset order, or some other reasons. So I enabled all these fields in this patch and also modified related test cases.

Addressed comments.

Addressed @DiggerLin 's comments, ie. auxiliaryHeaderSize() is kept.

Addressed comments.

Addressed comments.

Updated comments for consistency with parameters' names.

Addressed @DiggerLin and @jhenderson 's comments. Thanks!

Addressed @DiggerLin 's comments. Thanks

In D97656#3342197, @shchenz wrote:

LGTM with one nit. Thanks for adding this support.

Have you tried this patch with objects generated by other tools, like XLC?

Revised this patch on @avl 's work, i.e. D88827, D120345, D114664 and D114429.

Addressed comments.

Make the SymbolRef::Type in a convenient symbol-priority order.

In D117642#3286951, @jhenderson wrote:

The problem is, as I understand it, that the XCOFF symbol type enum isn't ordered in a convenient symbol-priority order, right?

1. Addressed comments.
2. Replaced binary inputs with YAMLs for testing.
3. Updated tests for error paths.

Gently ping.

In D113552#3197920, @DiggerLin wrote:

the x_auxtype is only for 64bits. in the implement, it means we always need to specific the AUX_type no matter it is 32bit or 64bits.

Thanks! @jhenderson @DiggerLin @Higuoxing
Address comments.

In D115503#3195171, @dblaikie wrote:

Ah, cool - could you include % growth on those rows

Address comments.

Thank you so much for your reviewing. @jhenderson @Higuoxing
Do you have more comments for this patch?
Btw. there are 2 accepted patches (D113825 and D114434) depends on this patch.

1. Address comments.
2. Skip recording branches for function calls in collectLocalBranchTargets();

In D115503#3192840, @dblaikie wrote:

Thanks for the data - looks good to me. Maybe include some of that data (summary of total binary size change/total debug info size change - and if you could include the flags (was this an -O0 -g build? Optimized (at what level)? Compressed debug info (-gz)? etc) that'd be helpful to better understand the comparison.

Address comment.

In D115503#3188302, @dblaikie wrote:

Got any data on how much this (combined with the LLVM patch) increases debug info size of, say, a clang self-host build? I assume not much, but wouldn't hurt to know.

Add a case to show that the default access flag is not recorded.

Thanks! @dblaikie
Separate into Clang part (this patch) and LLVM part (D115606).

Address comments.

Address comments.

Should unit tests be added for to_hexString(), to_string() and enumToString() (from D114840)? Although they are not part of the ScopedPrinter class but just in ScopedPrinter.h.

Address comments.

Address comments.

Address comments.