ARM EH: Extend DWARF_FRAME_REGISTERS count to 48 and map reg# 256 -> 16.

ARM EH: Extend DWARF_FRAME_REGISTERS count to 48 and map reg# 256 -> 16.

This avoids indexing out of bounds without using too much memory.
This won't work with general dwarf info, but LLVM happens to
only generate reg #256->287 for now, so the mapping may be okay.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=3689
TEST=./scons bitcode=1 platform=arm \
run_eh_floating_point_opt_frame_test \
run_eh_floating_point_opt_noframe_test \
run_eh_floating_point_noopt_frame_test \
run_eh_floating_point_noopt_noframe_test

R=dschuff@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-gcc.git;a=commit;h=2933722

[Derek Schuff, 2013-09-18 15:45:28]

On 2013/09/18 00:08:02, jvoung (cr) wrote:

this lgtm (unless roland has some objection... well, other than the obvious ones
:)
Presumably we should switch to libunwind along with libcxxabi anyway.

[jvoung (off chromium), 2013-09-18 18:46:49]

On 2013/09/18 15:45:28, Derek Schuff wrote:
> On 2013/09/18 00:08:02, jvoung (cr) wrote:
> 
> this lgtm (unless roland has some objection... well, other than the obvious
ones
> :)
> Presumably we should switch to libunwind along with libcxxabi anyway.

Sure we could try libunwind -- I wanted to see if this could fix the problem for
one of our users =)

We'd also want to make sure libunwind handles the test correctly.

[Derek Schuff, 2013-09-18 19:16:34]

On 2013/09/18 18:46:49, jvoung (cr) wrote:
> On 2013/09/18 15:45:28, Derek Schuff wrote:
> > On 2013/09/18 00:08:02, jvoung (cr) wrote:
> > 
> > this lgtm (unless roland has some objection... well, other than the obvious
> ones
> > :)
> > Presumably we should switch to libunwind along with libcxxabi anyway.
> 
> Sure we could try libunwind -- I wanted to see if this could fix the problem
for
> one of our users =)
> 
> We'd also want to make sure libunwind handles the test correctly.

right, I don't mean we need to switch to fix this, just that changes here will
need to be cleaned up for stable EH in any case, and we will probably want to
switch away from libgcc_eh and all its many undocumented gcc builtins, etc

[jvoung (off chromium), 2013-09-18 19:55:01]

Committed patchset #2 manually as r2933722 (presubmit successful).