X64: Added register holding Smi::FromInt(1).

X64: Added register holding Smi::FromInt(1).
Don't use r15 for anything any more.

[William Hesse, 2010-07-02 12:24:07]

LGTM.

http://codereview.chromium.org/2885018/diff/1/3
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/2885018/diff/1/3#newcode9965
src/x64/codegen-x64.cc:9965: // TODO(X64): On Win64, if we ever use XMM6-XMM15,
the low low 64 bits are
the low low?

http://codereview.chromium.org/2885018/diff/1/5
File src/x64/macro-assembler-x64.cc (right):

http://codereview.chromium.org/2885018/diff/1/5#newcode559
src/x64/macro-assembler-x64.cc:559: //imul(dst, kSmiConstantRegister,
Immediate(value));
Commented code.  It's a bad thing.

http://codereview.chromium.org/2885018/diff/1/5#newcode727
src/x64/macro-assembler-x64.cc:727: // Make mask 0x8000000000000001 and test
that both bits are zero.
How about 
movq(scratch, src)
rolq(scratch, 1)
testb(scratch, 3).

or  (the best, because it is least data-dependent?)
mov(scratch, 3)  
ror(scratch, 1)
test(scratch, src)

or, weirdly:
rol(src, 1)
test(src, 3)
ror(src, 1)

http://codereview.chromium.org/2885018/diff/1/5#newcode775
src/x64/macro-assembler-x64.cc:775: cmpq(kScratchRegister, src);
Why not 

cmpq(src, kSmiConstantRegister)
return overflow

Unless you are also testing for Sminess at the same time, minSmi is the only Smi
that overflows when subtracting Smi 1.

http://codereview.chromium.org/2885018/diff/1/5#newcode886
src/x64/macro-assembler-x64.cc:886: subq(src1, kScratchRegister);
How about
movq(scratch, src1)
subq(scratch, src2)
j(overflow)
movq(dst, scratch).

http://codereview.chromium.org/2885018/diff/1/9
File test/cctest/test-macro-assembler-x64.cc (right):

http://codereview.chromium.org/2885018/diff/1/9#newcode103
test/cctest/test-macro-assembler-x64.cc:103: __
pop(v8::internal::kSmiConstantRegister);
Why not check kSmiConstantRegister before popping the old value, since this is
testing code?

[Lasse Reichstein, 2010-07-02 13:06:52]

http://codereview.chromium.org/2885018/diff/1/5
File src/x64/macro-assembler-x64.cc (right):

http://codereview.chromium.org/2885018/diff/1/5#newcode559
src/x64/macro-assembler-x64.cc:559: //imul(dst, kSmiConstantRegister,
Immediate(value));
Ack, yes.

http://codereview.chromium.org/2885018/diff/1/5#newcode727
src/x64/macro-assembler-x64.cc:727: // Make mask 0x8000000000000001 and test
that both bits are zero.
I used lea instead of movl(.., Immediate(3)) since the latter is actually longer
(the immediate must be 4 bytes).
I don't want to use more rotates than necessary (although concern for the P4
might not be that necessary these days), and the second rotate clobbers the
flags from the src.

I think the first one is probably the best one.

http://codereview.chromium.org/2885018/diff/1/5#newcode775
src/x64/macro-assembler-x64.cc:775: cmpq(kScratchRegister, src);
Good one!

http://codereview.chromium.org/2885018/diff/1/5#newcode886
src/x64/macro-assembler-x64.cc:886: subq(src1, kScratchRegister);
The current version should be slightly better on an i7, due to cmp/j macro-op
fusion (sadly it doesn't exist on earlier chips in 64-bit mode).

http://codereview.chromium.org/2885018/diff/1/9
File test/cctest/test-macro-assembler-x64.cc (right):

http://codereview.chromium.org/2885018/diff/1/9#newcode103
test/cctest/test-macro-assembler-x64.cc:103: __
pop(v8::internal::kSmiConstantRegister);
Sure.