X64 Crankshaft: Implement MathRound, MathFloor, MathSqrt, and MathPowHalf Una...

X64 Crankshaft: Implement MathRound, MathFloor, MathSqrt, and MathPowHalf UnaryMathOperations in optimizing code generator.  Remove unneeded bailout on Math.pow(-0, 0.5) from ia32 and x64 platforms - we produce the correct value without bailing out.

Committed: http://code.google.com/p/v8/source/detail?r=6856

[Lasse Reichstein, 2011-02-18 14:13:47]

LGTM.

http://codereview.chromium.org/6532054/diff/1/src/x64/lithium-codegen-x64.cc
File src/x64/lithium-codegen-x64.cc (right):

http://codereview.chromium.org/6532054/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:2097: __ cvttsd2si(output_reg, input_reg);
Use the "q" version, so you can properly handle min-int.

http://codereview.chromium.org/6532054/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:2113: __ movsd(xmm_scratch,
Operand(kScratchRegister, 0));
Why load from memory when we have 64-bit immediates. Just load the bit-pattern
into a scratch register and then to an XMM register.

[William Hesse, 2011-02-18 14:33:28]

http://codereview.chromium.org/6532054/diff/1/src/x64/lithium-codegen-x64.cc
File src/x64/lithium-codegen-x64.cc (right):

http://codereview.chromium.org/6532054/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:2097: __ cvttsd2si(output_reg, input_reg);
On 2011/02/18 14:13:47, Lasse Reichstein wrote:
> Use the "q" version, so you can properly handle min-int.
But then we have much more complicated tests, because integers not in the 32-bit
range are generated without overflow.  The output is an untagged int32.

http://codereview.chromium.org/6532054/diff/1/src/x64/lithium-codegen-x64.cc#...
src/x64/lithium-codegen-x64.cc:2113: __ movsd(xmm_scratch,
Operand(kScratchRegister, 0));
On 2011/02/18 14:13:47, Lasse Reichstein wrote:
> Why load from memory when we have 64-bit immediates. Just load the bit-pattern
> into a scratch register and then to an XMM register.

Done.