[llvm-dev] Emitting a local alias

[Leonard Chan via llvm-dev]

Hi,

I wanted to see if there was a way in IR to emit a local alias such that I
would get:

```
        .type   _ZTVSt13bad_exception, at object # @_ZTVSt13bad_exception
        .globl  _ZTVSt13bad_exception
_ZTVSt13bad_exception:
*.L_ZTVSt13bad_exception:*
        .long   0                       # 0x0
        .long   (_ZTISt13bad_exception.rtti_proxy-.L_ZTVSt13bad_exception)-8
        .long   (_ZNSt13bad_exceptionD2Ev.stub at PLT-
.L_ZTVSt13bad_exception)-8
        .long   (_ZNSt13bad_exceptionD0Ev.stub at PLT-
.L_ZTVSt13bad_exception)-8
        .long   (_ZNKSt13bad_exception4whatEv.stub at PLT-
.L_ZTVSt13bad_exception)-8
```

rather than:

```
        .type   _ZTVSt13bad_exception, at object # @_ZTVSt13bad_exception
        .globl  _ZTVSt13bad_exception
_ZTVSt13bad_exception:
        .long   0                       # 0x0
        .long   (_ZTISt13bad_exception.rtti_proxy-.L_ZTVSt13bad_exception)-8
        .long   (_ZNSt13bad_exceptionD2Ev.stub at PLT-
.L_ZTVSt13bad_exception)-8
        .long   (_ZNSt13bad_exceptionD0Ev.stub at PLT-
.L_ZTVSt13bad_exception)-8
        .long   (_ZNKSt13bad_exception4whatEv.stub at PLT-
.L_ZTVSt13bad_exception)-8

*.set .L_ZTVSt13bad_exception, _ZTVSt13bad_exception*
```

The latter is what's generated when using a GlobalAlias with private
linkage. I'm asking because in either case, I want to be able to place
_ZTVSt13bad_exception in .rodata instead of .data.rel.ro which requires
that the object does not need a relocation.

According to `Constant::needsRelocation()`, a reloc isn't needed if the
offset I'm taking (.long symbol_A - symbol_B) contains symbols that are
both dso_local. I can guarantee that `symbol_A` will be, but `symbol_B`
needs to be a global with default visibility. My solution around this is to
emit a local alias, but the issue is that with optimizations (-O3), the
latter snippet seems to "resolve" the aliases and instead replaces them
with my aliasee (_ZTVSt13bad_exception) which is not dso_local. An
alternative solution to my problem would also be finding a way to somehow
prevent the aliases from being "optimized out".

Any ideas on either (1) IR that can codegen into my first snippet or (2)
prevent my alias in the second snippet from being "optimized out"?

- Leonard
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200506/e113ab4a/attachment.html>