X64: Tweak code generation slightly.

X64: Tweak code generation slightly. 

Uses a shorter opcode for movl and movb with immediates. 
Make movl preferable to movq in Set(register, int64_t) and use Set in more places.

Committed: http://code.google.com/p/v8/source/detail?r=7589

[William Hesse, 2011-04-12 13:04:59]

Only two comments about the correctness.  All others are optimizations that you
don't need to add.

LGTM with the two comments addressed.

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc
File src/x64/code-stubs-x64.cc (right):

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:1284: __ Set(r8, 0);
Could an implementation as movl(r8, SmiOneRegister) be more efficient in this
case? (inside Set()?)

But probably the CPU knows xor(x,x) generates a constant 0, and can optimize it
internally earlier than a movl.

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:1461: __ Set(kScratchRegister, 0x01);
How about rol(rdx, 1), xor(rdx, 1), ror(rdx, 1)?

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:1466: __ movq(FieldOperand(rax,
HeapNumber::kValueOffset), rdx);
Or in this case, Set(rdx, 1), ror(rdx,1), xor(Operand(rax, HN:ValOff), rdx).
Or shl(SmiOne, 31), xor(Operand(rax, ...),SmiOne), shr(SmiOne, 31).

http://codereview.chromium.org/6824074/diff/1/src/x64/lithium-codegen-x64.cc
File src/x64/lithium-codegen-x64.cc (right):

http://codereview.chromium.org/6824074/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:2427: __ Set(result, scope()->num_parameters());
This will be wrong if result and InputAt(0) can be the same register.

http://codereview.chromium.org/6824074/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

http://codereview.chromium.org/6824074/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:2005: movl(counter_operand, Immediate(value));
Is this really safe?  Are we sure that counter_operand has its top bits clear
already?  And that value is non-negative?

http://codereview.chromium.org/6824074/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:2223: Set(kScratchRegister, -kFrameAlignment);
How about or_(rsp, kFrameAlignment - 1);
          sub(rsp, kFrameAlignment - 1);

This makes an easier constant.

[Lasse Reichstein, 2011-04-12 13:19:07]

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc
File src/x64/code-stubs-x64.cc (right):

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:1284: __ Set(r8, 0);
It can optimize xor better. The xorl is guaranteed to break dependencies, and it
also has a very short encoding. I don't think it can be optimized.
Also, I don't want to depend on the *value* of the smi constant register. I
really want to keep the door open for a different smi representation (even if it
might never happen).

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:1461: __ Set(kScratchRegister, 0x01);
Unlikely to be faster, and definitely not shorter. This Set is five bytes, a
rol/xor/rdx with Imm8 is three bytes.

If anything, maybe I can try
  xorl(reg,reg);
  stc();
  rcr(1);
but it doesn't seem that great, with partial flag dependencies and slow
dependent operations.

http://codereview.chromium.org/6824074/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:1466: __ movq(FieldOperand(rax,
HeapNumber::kValueOffset), rdx);
Again, I'd rather not depend on the value of the smi register for anything but
smi operations. Here it's a double operation.

http://codereview.chromium.org/6824074/diff/1/src/x64/lithium-codegen-x64.cc
File src/x64/lithium-codegen-x64.cc (right):

http://codereview.chromium.org/6824074/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:2427: __ Set(result, scope()->num_parameters());
Ack. Will fix.

http://codereview.chromium.org/6824074/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

http://codereview.chromium.org/6824074/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:2005: movl(counter_operand, Immediate(value));
The counter is of type int*. I think this is actually a bug-fix.

http://codereview.chromium.org/6824074/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:2223: Set(kScratchRegister, -kFrameAlignment);
Not really, -kFrameAlignment should be an imm8.
We should be able to do and_(rsp, Immediate(-kFrameAlignment)) directly.