Remove i686-hosted Linux build and start switching to new directory layout

Remove i686-hosted Linux build and start switching to new directory layout

Switch the default host for the toolchain on Linux to x86_64, and stop
building the i686 version.

Also stop using the host_x86_{32,64} directories for host binaries and
instead put everything (including the driver scripts) into bin/
(This also reverts r13694, removing the pnacl-llc driver script, as that
was the only script that had a name clash with an LLVM binary. It was
only added as a proxy to the pnacl-llc binary anyway).

This moves us closer to a layout that is more direct-to-nacl-friendly
and more like existing gnu-style toolchains.
Also update the README file which describes the layout. It reflects the
intended eventual state (including the placement of the PNaCl target
libraries and translator libraries which will move in a separate CL), but
does not yet include direct-to-nacl-related items.

R=jvoung@chromium.org, noelallen@chromium.org
BUG= https://code.google.com/p/nativeclient/issues/detail?id=3946

Committed: https://src.chromium.org/viewvc/native_client?view=rev&revision=13753

[Derek Schuff, 2014-09-17 23:40:45]

PTAL.
The x86-32 toolchain buildbot will continue to build and use (but not upload)
the i686 packages, so it can keep working.
As for the rest of the x86-32 bots (e.g. main waterfall and chrome), they can
already run x86_64 binaries because Chrome's clang is built the same way.

[Derek Schuff, 2014-09-18 04:52:59]

Patchset #6 (id:100001) has been deleted

[Derek Schuff, 2014-09-18 04:53:05]

Patchset #5 (id:80001) has been deleted

[jvoung (off chromium), 2014-09-18 16:17:34]

https://codereview.chromium.org/580983002/diff/120001/pnacl/README
File pnacl/README (right):

https://codereview.chromium.org/580983002/diff/120001/pnacl/README#newcode18
pnacl/README:18: le32-nacl/
This part hasn't changed yet (It seems only the host_* parts)?

https://codereview.chromium.org/580983002/diff/120001/pnacl/README#newcode28
pnacl/README:28: *These are not generally needed directly by by developers of
PNaCl modules
** instead of * to match the ** above.

"by by" -> "by"

https://codereview.chromium.org/580983002/diff/120001/toolchain_build/toolcha...
File toolchain_build/toolchain_build_pnacl.py (right):

https://codereview.chromium.org/580983002/diff/120001/toolchain_build/toolcha...
toolchain_build/toolchain_build_pnacl.py:131: def DefaultHostForTargetLibs():
This can be removed now, that it's no longer used.

[Derek Schuff, 2014-09-18 17:03:14]

https://codereview.chromium.org/580983002/diff/120001/pnacl/README
File pnacl/README (right):

https://codereview.chromium.org/580983002/diff/120001/pnacl/README#newcode18
pnacl/README:18: le32-nacl/
On 2014/09/18 16:17:33, jvoung wrote:
> This part hasn't changed yet (It seems only the host_* parts)?

Yes, this file documents the intended final state (of the PNaCl parts). I'm
working a CL right now to make this change. It didn't seem worthwhile to do an
intermediate version of this document.

https://codereview.chromium.org/580983002/diff/120001/pnacl/README#newcode28
pnacl/README:28: *These are not generally needed directly by by developers of
PNaCl modules
On 2014/09/18 16:17:33, jvoung wrote:
> ** instead of * to match the ** above.
> 
> "by by" -> "by"

Done.

https://codereview.chromium.org/580983002/diff/120001/toolchain_build/toolcha...
File toolchain_build/toolchain_build_pnacl.py (right):

https://codereview.chromium.org/580983002/diff/120001/toolchain_build/toolcha...
toolchain_build/toolchain_build_pnacl.py:131: def DefaultHostForTargetLibs():
On 2014/09/18 16:17:33, jvoung wrote:
> This can be removed now, that it's no longer used.

Done.

[jvoung (off chromium), 2014-09-18 17:24:45]

lgtm

https://codereview.chromium.org/580983002/diff/140001/buildbot/buildbot_pnacl...
File buildbot/buildbot_pnacl_toolchain_tests.sh (right):

https://codereview.chromium.org/580983002/diff/140001/buildbot/buildbot_pnacl...
buildbot/buildbot_pnacl_toolchain_tests.sh:295: # tools such as fpcmp (used for
llvm test suite). For some reason it matters
Is this comment still necessary. (There's only the 64-bit build for linux now)

https://codereview.chromium.org/580983002/diff/140001/pnacl/README
File pnacl/README (right):

https://codereview.chromium.org/580983002/diff/140001/pnacl/README#newcode20
pnacl/README:20: tools-arm/
Could remove tools-arm and tools-x86 now as well.

[Derek Schuff, 2014-09-18 17:29:13]

https://codereview.chromium.org/580983002/diff/140001/buildbot/buildbot_pnacl...
File buildbot/buildbot_pnacl_toolchain_tests.sh (right):

https://codereview.chromium.org/580983002/diff/140001/buildbot/buildbot_pnacl...
buildbot/buildbot_pnacl_toolchain_tests.sh:295: # tools such as fpcmp (used for
llvm test suite). For some reason it matters
On 2014/09/18 17:24:45, jvoung wrote:
> Is this comment still necessary. (There's only the 64-bit build for linux now)

removed.

https://codereview.chromium.org/580983002/diff/140001/pnacl/README
File pnacl/README (right):

https://codereview.chromium.org/580983002/diff/140001/pnacl/README#newcode20
pnacl/README:20: tools-arm/
On 2014/09/18 17:24:45, jvoung wrote:
> Could remove tools-arm and tools-x86 now as well.

Done.

[noelallen1, 2014-09-18 19:28:05]

lgtm

[Derek Schuff, 2014-09-18 19:40:59]

Committed patchset #7 (id:160001) manually as 13753 (presubmit successful).

[jvoung - send to chromium..., 2014-09-19 15:15:43]

jvoung@google.com changed reviewers:
+ jvoung@google.com

[jvoung - send to chromium..., 2014-09-19 15:15:44]

FYI: developer tools may be a bit faster now:
- One bump from the switch to clang (newer compiler?)
- One bump from the switch to 64-bit.

https://chromeperf.appspot.com/report?masters=NativeClient&bots=lucid_64-newl...

[Derek Schuff, 2014-09-19 15:50:12]

On 2014/09/19 15:15:44, jvoung - send to chromium... wrote:
> FYI: developer tools may be a bit faster now:
> - One bump from the switch to clang (newer compiler?)
> - One bump from the switch to 64-bit.
> 
>
https://chromeperf.appspot.com/report?masters=NativeClient&bots=lucid_64-newl...

cool :)
odd that in most of the benchmarks, the sandboxed translator is actually faster
than the host translator? Could that be because of the LTO?

[jvoung (off chromium), 2014-09-19 16:17:25]

On 2014/09/19 15:50:12, Derek Schuff wrote:
> On 2014/09/19 15:15:44, jvoung - send to chromium... wrote:
> > FYI: developer tools may be a bit faster now:
> > - One bump from the switch to clang (newer compiler?)
> > - One bump from the switch to 64-bit.
> > 
> >
>
https://chromeperf.appspot.com/report?masters=NativeClient&bots=lucid_64-newl...
> 
> cool :)
> odd that in most of the benchmarks, the sandboxed translator is actually
faster
> than the host translator? Could that be because of the LTO?

That's probably one of the things. There's also asserts vs no-asserts. PIC vs
non-PIC.