Fixed deopt reasons in TaggedToI.

Fixed deopt reasons in TaggedToI.

Every DeoptimizeIf should be preceded by a RecordComment explaining
the reason. In the long run, the reason should be an argument of
DeoptimizeIf, but we're not there yet.

On x87, things are a bit ugly due to some pushing/popping, so if
somebody feels inclined to improve this: Feel free. Probably the right
approach would be reloading instead of the push/pop horror.

Another thing is our inconsistent handling of the "done" continuation
of deferred code on the various platforms, I made tiny changes on the
way, but this should better be unified somehow, with all those
micro-optimizations removed.

R=jkummerow@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=23996

[Sven Panne, 2014-09-17 08:51:19]

Patchset #1 (id:1) has been deleted

[Sven Panne, 2014-09-17 08:57:03]

svenpanne@chromium.org changed reviewers:
+ jkummerow@chromium.org

[Jakob Kummerow, 2014-09-17 09:22:35]

LGTM with comments (which apply to all platforms).

https://codereview.chromium.org/578583002/diff/20001/src/ia32/lithium-codegen...
File src/ia32/lithium-codegen-ia32.cc (right):

https://codereview.chromium.org/578583002/diff/20001/src/ia32/lithium-codegen...
src/ia32/lithium-codegen-ia32.cc:4766: __ jmp(done);
As discussed, we don't need this jump (as it would jump to the very next
instruction, which is the jump to the deferred code's exit label).

https://codereview.chromium.org/578583002/diff/20001/src/ia32/lithium-codegen...
src/ia32/lithium-codegen-ia32.cc:4790: __ jmp(done);
This can go too (see above).

[Sven Panne, 2014-09-17 09:51:33]

Committed patchset #2 (id:40001) manually as 23996 (presubmit successful).