Subzero: Fix emission of global initializers.

Subzero: Fix emission of global initializers.

Also changes the szbuild.py script to add -fdata-sections, and entirely removes the -disable-globals option.

The global initializer emission basically copies what llc does, based on 3 properties of the global: constant vs non-constant, internal vs external, and full zero-initializer vs non-trivial initializer.

BUG= none
R=jvoung@chromium.org, kschimpf@google.com

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=2a063e2b8e38b49cc5ee4aa33841ce39378de567

[Jim Stichnoth, 2014-10-07 23:28:38]

stichnot@chromium.org changed reviewers:
+ jvoung@chromium.org, kschimpf@google.com

[Jim Stichnoth, 2014-10-07 23:43:11]

Karl - I tried to look at your TODO about not mangling uninitialized external
global references.  E.g. when running "llvm2ice --prefix Subzero_", this
bitcode:

@unmangled_global = external global [4 x i8], align 4
@blah = internal global i32 ptrtoint ([4 x i8]* @unmangled_global to i32), align
4

define internal i32 @retBlah() {
  %expanded = ptrtoint i32* @blah to i32
  ret i32 %expanded
}

should produce an initialization of blah containing:
  .long unmangled_global
and not:
  .long Subzero_unmangled_global

However, the name of unmangled_global seems not to be available.  Hopefully that
just a bug in the bitcode reader hookup.

[Karl, 2014-10-08 15:54:09]

lgtm.

On the notion with --prefix, this is a known problem (and the reason for the
TODO). As soon as I fix how function addresses are represented, this problem
will be fixed.

[Karl, 2014-10-08 15:54:09]

lgtm.

On the notion with --prefix, this is a known problem (and the reason for the
TODO). As soon as I fix how function addresses are represented, this problem
will be fixed.

[jvoung (off chromium), 2014-10-08 16:52:06]

LGTM too

https://codereview.chromium.org/631383003/diff/2/tests_lit/llvm2ice_tests/glo...
File tests_lit/llvm2ice_tests/globalinit.pnacl.ll (right):

https://codereview.chromium.org/631383003/diff/2/tests_lit/llvm2ice_tests/glo...
tests_lit/llvm2ice_tests/globalinit.pnacl.ll:18: ; CHECK-NEXT: .section
.rodata,"a",@progbits
There are some cases where initialized constant data of a specific alignment and
per-element size can be "aM" (mergeable) (e.g., strings that share a suffix) but
that doesn't seem worth worrying about.

[Jim Stichnoth, 2014-10-08 18:24:54]

Committed patchset #2 (id:2) manually as
2a063e2b8e38b49cc5ee4aa33841ce39378de567 (presubmit successful).