Use kRegisterSize when passing arguments to the C++ runtime for X64

Use kRegisterSize when passing arguments to the C++ runtime for X64

R=verwaest@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=18674

[haitao.feng, 2014-01-14 08:40:39]

Toon, please have a look.

https://codereview.chromium.org/137443004/diff/1/src/x64/deoptimizer-x64.cc
File src/x64/deoptimizer-x64.cc (right):

https://codereview.chromium.org/137443004/diff/1/src/x64/deoptimizer-x64.cc#n...
src/x64/deoptimizer-x64.cc:159: const int kSavedRegistersAreaSize =
kNumberOfRegisters * kRegisterSize +
x32 uses the 64-bit push instruction to save all the registers when
deoptimization happens.

https://codereview.chromium.org/137443004/diff/1/src/x64/deoptimizer-x64.cc#n...
src/x64/deoptimizer-x64.cc:172: __ movq(arg_reg_4, Operand(rsp,
kSavedRegistersAreaSize + 1 * kRegisterSize));
x32 uses 64-bit push instruction for the bailout id to reduce the code size of
deoptimization prologue.

[haitao.feng, 2014-01-14 12:29:24]

Forgot to mention x32 ABI uses registers first to pass argument (%rdi, %rsi,
%rdx, %rcx, %r8, %r9), and then use 8-byte stack slots. The ABI applies when we
call a C++ x32 function from JavaScript, so the code is modified to use
kRegisterSize instead kPointerSize.

[Toon Verwaest, 2014-01-16 09:50:27]

https://codereview.chromium.org/137443004/diff/1/src/x64/code-stubs-x64.cc
File src/x64/code-stubs-x64.cc (right):

https://codereview.chromium.org/137443004/diff/1/src/x64/code-stubs-x64.cc#ne...
src/x64/code-stubs-x64.cc:5176: __ push(arg_reg_2);
Shouldn't we mov here?

https://codereview.chromium.org/137443004/diff/1/src/x64/code-stubs-x64.cc#ne...
src/x64/code-stubs-x64.cc:5180: Operand(rsp, kNumSavedRegisters * kPointerSize +
kPCOnStackSize));
In that case, this would be "* kRegisterSize" rather than "* kPointerSize".

[haitao.feng, 2014-01-16 11:54:41]

Toon, thanks for the review.

https://codereview.chromium.org/137443004/diff/1/src/x64/code-stubs-x64.cc
File src/x64/code-stubs-x64.cc (right):

https://codereview.chromium.org/137443004/diff/1/src/x64/code-stubs-x64.cc#ne...
src/x64/code-stubs-x64.cc:5176: __ push(arg_reg_2);
Sorry I do not quite understand this comment.
On 2014/01/16 09:50:27, Toon Verwaest wrote:
> Shouldn't we mov here?

https://codereview.chromium.org/137443004/diff/1/src/x64/code-stubs-x64.cc#ne...
src/x64/code-stubs-x64.cc:5180: Operand(rsp, kNumSavedRegisters * kPointerSize +
kPCOnStackSize));
On 2014/01/16 09:50:27, Toon Verwaest wrote:
> In that case, this would be "* kRegisterSize" rather than "* kPointerSize".

Done.

"make x32.release.check" happens to pass as "push" is translated to "Push" which
is 4-byte pseudo instruction and the high 32-bit of arg_reg_1 and arg_reg_2 is
unused. You are right we should use 8-byte push in this function for safety. A
side note is that currently we only have "__k (keep)" annotation, all the other
annotations are gone. I will introduce pushp, addp, ... to eliminate the __k
annotation.

[Toon Verwaest, 2014-01-17 14:28:29]

lgtm

[haitao.feng, 2014-01-20 01:52:35]

Committed patchset #2 manually as r18674 (presubmit successful).