Subzero: Fix the standalone build to work with the LLVM cmake build.

Subzero: Fix the standalone build to work with the LLVM cmake build.

Autoconf is the default.  Use "make -f Makefile.standalong CMAKE=1" to use the cmake build.

BUG= none
R=jvoung@chromium.org, mtrofin@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=6c6adf11b12a3529d87587445c1ce82c0dfdbf8c

[Jim Stichnoth, 2015-03-11 04:56:52]

stichnot@chromium.org changed reviewers:
+ jvoung@chromium.org

[Jim Stichnoth, 2015-03-11 04:59:05]

Jan - This compiles OK with the LLVM cmake build, but gets link errors
presumably because of -std= and -stdlib= issues and the implementation of
std::string in particular.  Any idea how to make the link work?

[Jim Stichnoth, 2015-03-11 04:59:05]

Jan - This compiles OK with the LLVM cmake build, but gets link errors
presumably because of -std= and -stdlib= issues and the implementation of
std::string in particular.  Any idea how to make the link work?

[Jim Stichnoth, 2015-03-11 13:31:07]

Update.  pnacl-sz links if Makefile.standalone is changed to remove
-stdlib=libc++ and the -I for libc++, and change -std to c++11.  There's a
problem then with llvm::cl static initialization, which gets fixed by
restricting the -l library list in the link line.

After that, 2 of the lit tests fail with different output: elf_container.ll and
randomize-regalloc.ll.

[jvoung (off chromium), 2015-03-11 16:54:59]

On 2015/03/11 13:31:07, stichnot wrote:
> Update.  pnacl-sz links if Makefile.standalone is changed to remove
> -stdlib=libc++ and the -I for libc++, and change -std to c++11.  There's a
> problem then with llvm::cl static initialization, which gets fixed by
> restricting the -l library list in the link line.
> 
> After that, 2 of the lit tests fail with different output: elf_container.ll
and
> randomize-regalloc.ll.

Hmm, yes there is originally the libc++ vs libstdc++ error. I tried to fix that
with https://codereview.chromium.org/978963002/.

There is this path issue you fixed.

I also, ran into a problem finding the llvm shared libraries (no longer a static
link?). If I set LD_LIBRARY_PATH or something to get past that I get:

ln -s build/Release+Asserts/pnacl-sz
Build Attributes:
<premain>: CommandLine Error: Option 'o' registered more than once!
LLVM ERROR: inconsistency in registered CommandLine options
make: *** [make_symlink] Error 1

Is that the llvm::cl issue you mentioned?

Odd that there are the other two failures =/

Path change LGTM though.

[Jim Stichnoth, 2015-04-06 19:34:51]

stichnot@chromium.org changed reviewers:
+ mtrofin@chromium.org

[Jim Stichnoth, 2015-04-07 21:14:24]

Not LGTM.  PTAL.

The (main) problem with the previous patchset is that the llvm_x86_64_linux_work
directory persists when you switch between cmake and autoconf, even if you nuke
pnacl_newlib_raw.  This means that work dir can't be used to auto-detect cmake
vs autoconf.

We work around this by setting CMAKE, e.g.:
  make -f Makefile.standalong CMAKE=1 check

There are two failing lit tests with the cmake build.  Those will be handled by
a different CL.

randomize-regalloc.ll: Different implementations of std::random_shuffle produce
different results.

elf_container.ll: Constant pool entries are emitted in hash table order. 
Different unordered_map implementations produce different results.

[Mircea Trofin, 2015-04-07 21:16:49]

lgtm

[Jim Stichnoth, 2015-04-07 21:22:29]

Committed patchset #4 (id:60001) manually as
6c6adf11b12a3529d87587445c1ce82c0dfdbf8c (presubmit successful).