And Zaheed, you didn't upload new version of the patch.


http://codereview.chromium.org/6170001/diff/1/src/arm/macro-assembler-arm.cc
File src/arm/macro-assembler-arm.cc (right):

http://codereview.chromium.org/6170001/diff/1/src/arm/macro-assembler-arm.cc#newcode1414
src/arm/macro-assembler-arm.cc:1414: mov(scratch,
Operand(unwind_space));
Zaheer,

I am not sure it's something which cannot be solved.

I do agree with Sergey and think we should allocate v8::Arguments below
exit frame (exactly like we do for ia32)---one of the reasons I asked
you to refactor the code was to make these things immediately visible.

We deal with alignment on x64 as well and it looks fine (in the worst
case you would need the same and_ trick, see PrepareCallCFunction).

And so on.

On 2011/01/11 15:44:35, zaheer wrote:
On 2011/01/11 14:11:33, antonm wrote:
> it looks like a lot of code below is shared with
MacroAssembler::EnterExitFrame.
>  Is it possible to refactor the common code and make it more like
ia32
> implementation (which just invokes EnterApiExitFrame + some magic
for some
> platforms?
The behavior is different from EnterExitFrame
- v8::arguments has to be allocated above exit frame to miss gc
- The alignment has to happen after allocating space for the args and
it should
handle aligned/unaligned case based on arg_stack_space (unlike
EnterExitFrame
which knows the number of pushes - 5)
- exit frame pc patching has to be handled
- argv and builtin function is n/a in this case

http://codereview.chromium.org/6170001/diff/1/src/arm/stub-cache-arm.cc
File src/arm/stub-cache-arm.cc (right):

http://codereview.chromium.org/6170001/diff/1/src/arm/stub-cache-arm.cc#newcode2319
src/arm/stub-cache-arm.cc:2319: #ifdef USE_SIMULATOR
On 2011/01/11 15:44:35, zaheer wrote:
On 2011/01/11 14:11:33, antonm wrote:
> do we need a special case for simulator?
currently the simulator interface to the native api expects a argc,
argv and
direct call breaks that (it has a v8::Arguments&). I wasnt sure how to
fix this
in the simulator, but it would be great if we can do it.

I see, let's ask Erik, maybe he can suggest something.  I am concerned
with the testability of this change---I think most of the people use
simulator to run tests so we can miss errors.

http://codereview.chromium.org/6170001/

--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev

Reply via email to