X64: Implement unary subtraction, ObjectPair. 350 tests still fail.

X64: Implement unary subtraction, ObjectPair. 350 tests still fail.

Committed: http://code.google.com/p/v8/source/detail?r=2304

[William Hesse, 2009-06-30 09:08:24]

I checked the generated assembly code, and ObjectPair does the right thing on
X64.  It returns two values, in rax and rdx.

[Lasse Reichstein, 2009-06-30 10:55:26]

LGTM w/ comments.

http://codereview.chromium.org/150093/diff/1/2
File src/runtime.cc (right):

http://codereview.chromium.org/150093/diff/1/2#newcode4491
Line 4491: return result;  // Should be returned in rax and rdx, in AMD-x64-abi.
Make comment saying that it hasn't been tested in MSVC.
Or use ifdefs using __GCC__ or __MSVC__ to not enable it.

Have you tested whether a struct also works in 32-bit mode? (I assume it doesn't
or it would have been used already).

http://codereview.chromium.org/150093/diff/1/3
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/150093/diff/1/3#newcode5471
Line 5471: __ j(overflow, &undo);
Will this work? If rax is a smi, then 0 - rax will not overflow a 64-bit signed
subtraction.

If we have a Smi, we only need to special case zero and kMinSmiValue (aka
-0x80000000). A single testl(rax, 0x7fffffff) should be able to do that. After
that, we can just use neg(rax).

http://codereview.chromium.org/150093/diff/1/3#newcode5475
Line 5475: __ j(zero, &done);
How can it not be smi tagged if we subtract a smi tagged number from zero?

[William Hesse, 2009-06-30 12:11:50]

http://codereview.chromium.org/150093/diff/1/2
File src/runtime.cc (right):

http://codereview.chromium.org/150093/diff/1/2#newcode4491
Line 4491: return result;  // Should be returned in rax and rdx, in AMD-x64-abi.
On 2009/06/30 10:55:26, Lasse Reichstein wrote:
> Make comment saying that it hasn't been tested in MSVC.
> Or use ifdefs using __GCC__ or __MSVC__ to not enable it.
> 
> Have you tested whether a struct also works in 32-bit mode? (I assume it
doesn't
> or it would have been used already).

Done.

http://codereview.chromium.org/150093/diff/1/3
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/150093/diff/1/3#newcode5471
Line 5471: __ j(overflow, &undo);
On 2009/06/30 10:55:26, Lasse Reichstein wrote:
> Will this work? If rax is a smi, then 0 - rax will not overflow a 64-bit
signed
> subtraction.
> 
> If we have a Smi, we only need to special case zero and kMinSmiValue (aka
> -0x80000000). A single testl(rax, 0x7fffffff) should be able to do that. After
> that, we can just use neg(rax).
> 
> 

Replaced with subl, which will overflow.  We need to be able to restore rax from
rdx, which is why we use rdx.

http://codereview.chromium.org/150093/diff/1/3#newcode5475
Line 5475: __ j(zero, &done);
On 2009/06/30 10:55:26, Lasse Reichstein wrote:
> How can it not be smi tagged if we subtract a smi tagged number from zero?

Done.