clang: add support for asmjs arch and Emscripten OS

clang: add support for asmjs arch and Emscripten OS

This only adds support for the arch/OS and doesn't allow anything else in LLVM
for now. There's a corresponding LLVM patch which adds the Triple values.

LLVM patch:
  https://codereview.chromium.org/1024073002/

R=jvoung@chromium.org, azakai@mozilla.com, sunfish@mozilla.com
BUG= https://code.google.com/p/nativeclient/issues/detail?id=4102
TEST= make check-all

Committed: 96b3da27dcefc9d152e51cf54280989b2206d789

[jvoung (off chromium), 2015-03-20 17:07:07]

https://codereview.chromium.org/1022123003/diff/1/include/clang/Basic/TargetC...
File include/clang/Basic/TargetCXXABI.h (right):

https://codereview.chromium.org/1022123003/diff/1/include/clang/Basic/TargetC...
include/clang/Basic/TargetCXXABI.h:163: case GenericItanium:
Hmm, so LE32 uses GenericItanium directly and doesn't create a separate CXXABI. 
What it does is in
https://codereview.chromium.org/1022123003/diff/1/lib/CodeGen/ItaniumCXXABI.cpp,
it specially sets /* UseARMMethodPtrABI = */ true.

This means that LE32 almost gets everything about the ARM method pointer ABI,
except for this arePointersToMemberFunctionsAligned().

Is there a way to have the LE32 version be similar? Seems a bit unfortunate that
there are two places that control this. So, if possible it would be nice if
there was just one place that controls this.

Also, because "arePointersToMemberFunctionsAligned" didn't exist previously,
this is the first time that GenericARM/GenericAArch64 will stop aligning method
pointers, so might want be careful to run some ARM tests.

Grepping around, there's also a "getMemberPointerWidthAndAlign", which may need
to take this into account? That's at the lib/AST level, and I'm not sure of the
implications.

https://codereview.chromium.org/1022123003/diff/1/lib/Basic/Targets.cpp
File lib/Basic/Targets.cpp (right):

https://codereview.chromium.org/1022123003/diff/1/lib/Basic/Targets.cpp#newco...
lib/Basic/Targets.cpp:6060: DescriptionString =
"e-p:32:32-i64:64-v128:32:128-n32-S128";
Also test this in test/CodeGen/target-data.c

https://codereview.chromium.org/1022123003/diff/1/lib/CodeGen/CodeGenModule.cpp
File lib/CodeGen/CodeGenModule.cpp (left):

https://codereview.chromium.org/1022123003/diff/1/lib/CodeGen/CodeGenModule.c...
lib/CodeGen/CodeGenModule.cpp:3560: 
nit: extra change?

https://codereview.chromium.org/1022123003/diff/1/lib/Driver/ToolChains.h
File lib/Driver/ToolChains.h (right):

https://codereview.chromium.org/1022123003/diff/1/lib/Driver/ToolChains.h#new...
lib/Driver/ToolChains.h:726: bool IsMathErrnoDefault() const;
Indicate these methods are "override" ?

Why not just inline the bodies? Most of them do that except for TCEToolChain.

https://codereview.chromium.org/1022123003/diff/1/test/CodeGen/emscripten-arg...
File test/CodeGen/emscripten-arguments.c (right):

https://codereview.chromium.org/1022123003/diff/1/test/CodeGen/emscripten-arg...
test/CodeGen/emscripten-arguments.c:7: 
check "long double" and "long long" too?

https://codereview.chromium.org/1022123003/diff/1/test/Driver/asmjs-unknown-e...
File test/Driver/asmjs-unknown-emscripten.cpp (right):

https://codereview.chromium.org/1022123003/diff/1/test/Driver/asmjs-unknown-e...
test/Driver/asmjs-unknown-emscripten.cpp:55: void __EMSCRIPTEN__defined() {}
I guess there is also the legacy EMSCRIPTEN without the __ around it.

[JF, 2015-03-20 18:49:39]

https://codereview.chromium.org/1022123003/diff/1/include/clang/Basic/TargetC...
File include/clang/Basic/TargetCXXABI.h (right):

https://codereview.chromium.org/1022123003/diff/1/include/clang/Basic/TargetC...
include/clang/Basic/TargetCXXABI.h:163: case GenericItanium:
On 2015/03/20 17:07:07, jvoung wrote:
> Hmm, so LE32 uses GenericItanium directly and doesn't create a separate
CXXABI. 
> What it does is in
>
https://codereview.chromium.org/1022123003/diff/1/lib/CodeGen/ItaniumCXXABI.cpp,
> it specially sets /* UseARMMethodPtrABI = */ true.
> 
> This means that LE32 almost gets everything about the ARM method pointer ABI,
> except for this arePointersToMemberFunctionsAligned().
> 
> Is there a way to have the LE32 version be similar? Seems a bit unfortunate
that
> there are two places that control this. So, if possible it would be nice if
> there was just one place that controls this.
> 
> Also, because "arePointersToMemberFunctionsAligned" didn't exist previously,
> this is the first time that GenericARM/GenericAArch64 will stop aligning
method
> pointers, so might want be careful to run some ARM tests.
> 
> Grepping around, there's also a "getMemberPointerWidthAndAlign", which may
need
> to take this into account? That's at the lib/AST level, and I'm not sure of
the
> implications.

As discussed on IRC I left ARM as it was, with a TODO, and we'll have only
Emscripten be different for now.

https://codereview.chromium.org/1022123003/diff/1/lib/Basic/Targets.cpp
File lib/Basic/Targets.cpp (right):

https://codereview.chromium.org/1022123003/diff/1/lib/Basic/Targets.cpp#newco...
lib/Basic/Targets.cpp:6060: DescriptionString =
"e-p:32:32-i64:64-v128:32:128-n32-S128";
On 2015/03/20 17:07:07, jvoung wrote:
> Also test this in test/CodeGen/target-data.c

Done.

https://codereview.chromium.org/1022123003/diff/1/lib/CodeGen/CodeGenModule.cpp
File lib/CodeGen/CodeGenModule.cpp (left):

https://codereview.chromium.org/1022123003/diff/1/lib/CodeGen/CodeGenModule.c...
lib/CodeGen/CodeGenModule.cpp:3560: 
On 2015/03/20 17:07:07, jvoung wrote:
> nit: extra change?

Oops, emacs decided to delete the extra line.

https://codereview.chromium.org/1022123003/diff/1/lib/Driver/ToolChains.h
File lib/Driver/ToolChains.h (right):

https://codereview.chromium.org/1022123003/diff/1/lib/Driver/ToolChains.h#new...
lib/Driver/ToolChains.h:726: bool IsMathErrnoDefault() const;
On 2015/03/20 17:07:07, jvoung wrote:
> Indicate these methods are "override" ?
> 
> Why not just inline the bodies? Most of them do that except for TCEToolChain.

Done.

https://codereview.chromium.org/1022123003/diff/1/test/CodeGen/emscripten-arg...
File test/CodeGen/emscripten-arguments.c (right):

https://codereview.chromium.org/1022123003/diff/1/test/CodeGen/emscripten-arg...
test/CodeGen/emscripten-arguments.c:7: 
On 2015/03/20 17:07:07, jvoung wrote:
> check "long double" and "long long" too?

Done.

https://codereview.chromium.org/1022123003/diff/1/test/Driver/asmjs-unknown-e...
File test/Driver/asmjs-unknown-emscripten.cpp (right):

https://codereview.chromium.org/1022123003/diff/1/test/Driver/asmjs-unknown-e...
test/Driver/asmjs-unknown-emscripten.cpp:55: void __EMSCRIPTEN__defined() {}
On 2015/03/20 17:07:07, jvoung wrote:
> I guess there is also the legacy EMSCRIPTEN without the __ around it.

Done.

[jvoung (off chromium), 2015-03-20 19:22:28]

lgtm

[JF, 2015-03-20 20:19:15]

Committed patchset #2 (id:20001) manually as
96b3da27dcefc9d152e51cf54280989b2206d789 (presubmit successful).