Fix missing x87 tracking for deferred code

Fix missing x87 tracking for deferred code.

When compiling deferred code we have to remember the corresponding stack state and ensure that the deferred code does not manipulate the order of elements.

BUG=
R=mstarzinger@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=16433

[Michael Starzinger, 2013-08-29 08:55:56]

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
File src/ia32/lithium-codegen-ia32.cc (right):

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.cc:1030: if (x87_stack_.depth() > 0 ||
(FLAG_debug_code && FLAG_enable_slow_asserts)) {
I don't understand why we need to enable flushing when --debug-code is passed.
Can we drop this?

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
File src/ia32/lithium-codegen-ia32.h (right):

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.h:470: void CommitWrite(X87Register reg);
I think both, "PrepareToWrite" and "CommitWrite" should also check the
mutability predicate explicitly.

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.h:492: bool copy_;
Let's call this "is_mutable_" instead and reverse the semantics. I think that's
easier to understand. WDYT?

[oliv, 2013-08-29 09:23:25]

ok thanks, ptal

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
File src/ia32/lithium-codegen-ia32.cc (right):

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.cc:1030: if (x87_stack_.depth() > 0 ||
(FLAG_debug_code && FLAG_enable_slow_asserts)) {
On 2013/08/29 08:55:56, Michael Starzinger wrote:
> I don't understand why we need to enable flushing when --debug-code is passed.
> Can we drop this?

The idea would have been to verify that the stack is empty at runtime.
Unfortunately this blows up some near-labels by now, since the verify is like 15
instructions, so i have to drop it for now...

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
File src/ia32/lithium-codegen-ia32.h (right):

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.h:470: void CommitWrite(X87Register reg);
On 2013/08/29 08:55:56, Michael Starzinger wrote:
> I think both, "PrepareToWrite" and "CommitWrite" should also check the
> mutability predicate explicitly.

Done.

https://codereview.chromium.org/23135018/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.h:492: bool copy_;
On 2013/08/29 08:55:56, Michael Starzinger wrote:
> Let's call this "is_mutable_" instead and reverse the semantics. I think
that's
> easier to understand. WDYT?

Done.

[Michael Starzinger, 2013-08-29 14:24:14]

LGTM.

[oliv, 2013-08-29 16:00:02]

Committed patchset #3 manually as r16433 (presubmit successful).