Fix ABI for API calls on ia32.

Fix ABI for API calls on ia32.

Instead of relying on eax value after the call load returned handle value directly from the slot that was preallocated for it.

R=vitalyr@chromium.org

Committed: http://code.google.com/p/v8/source/detail?r=8533

[Vyacheslav Egorov (Chromium), 2011-07-04 18:20:18]

MSVC generates the code slightly differently and does not pop the implicit 0th
argument from the stack. 

I have change to code to make it more portable.

[Vitaly Repeshko, 2011-07-04 22:41:36]

Almost LG.

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
File src/ia32/macro-assembler-ia32.cc (right):

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.cc:1453: // Save out slot address in the
callee-save register.
Let's be consistent with "slot" vs. "cell" here in the function above. I prefer
"slot".

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.cc:1455: lea(esi, Operand(esp, (argc + 1) *
kPointerSize));
This is tightly coupled to PrepareCallApiFunction. Its comments already
(incorrectly it seems) say it uses esi to restore the original stack state. How
about moving this block to PrepareCallApiFunction? This way we can avoid passing
the confusing argc here.
Also for debugging purposes it may be useful to clobber esi even if
kReturnHandlesDirectly is true (in debug mode).

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.cc:1467: mov(eax, Operand(esi, 0));
Some comment here would be helpful.

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia32.h
File src/ia32/macro-assembler-ia32.h (right):

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.h:571: int argc,
See the .cc file comments. If you still think passing argc is better, please
update the comments here.

[Vyacheslav Egorov (Chromium), 2011-07-05 11:03:50]

PTAL

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
File src/ia32/macro-assembler-ia32.cc (right):

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.cc:1453: // Save out slot address in the
callee-save register.
On 2011/07/04 22:41:36, Vitaly Repeshko wrote:
> Let's be consistent with "slot" vs. "cell" here in the function above. I
prefer
> "slot".

Done.

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.cc:1455: lea(esi, Operand(esp, (argc + 1) *
kPointerSize));
On 2011/07/04 22:41:36, Vitaly Repeshko wrote:
> This is tightly coupled to PrepareCallApiFunction. Its comments already
> (incorrectly it seems) say it uses esi to restore the original stack state.
How
> about moving this block to PrepareCallApiFunction? This way we can avoid
passing
> the confusing argc here.
> Also for debugging purposes it may be useful to clobber esi even if
> kReturnHandlesDirectly is true (in debug mode).

I though about that but did not want to prolong esi liverange. 

Done.

http://codereview.chromium.org/7307004/diff/3001/src/ia32/macro-assembler-ia3...
src/ia32/macro-assembler-ia32.cc:1467: mov(eax, Operand(esi, 0));
On 2011/07/04 22:41:36, Vitaly Repeshko wrote:
> Some comment here would be helpful.

Done.

[Vitaly Repeshko, 2011-07-05 11:08:29]

LGTM!